tree-ssa-math-opts.c (nop_stats, [...]): Moved to ...
* tree-ssa-math-opts.c (nop_stats, bswap_stats, struct symbolic_number,
BITS_PER_MARKER, MARKER_MASK, MARKER_BYTE_UNKNOWN, HEAD_MARKER, CMPNOP,
CMPXCHG, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, pass_data_optimize_bswap,
class pass_optimize_bswap, bswap_replace,
pass_optimize_bswap::execute): Moved to ...
* gimple-ssa-store-merging.c: ... this file.
Include optabs-tree.h.
(nop_stats, bswap_stats, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, bswap_replace): Put into
anonymous namespace, remove static keywords.
(pass_optimize_bswap::gate): Test BITS_PER_UNIT == 8 here...
(pass_optimize_bswap::execute): ... rather than here. Formatting fix.
From-SVN: r254947
2017-11-20 11:08:48 +01:00
|
|
|
/* GIMPLE store merging and byte swapping passes.
|
|
|
|
Copyright (C) 2009-2017 Free Software Foundation, Inc.
|
2016-10-28 16:18:50 +02:00
|
|
|
Contributed by ARM Ltd.
|
|
|
|
|
|
|
|
This file is part of GCC.
|
|
|
|
|
|
|
|
GCC is free software; you can redistribute it and/or modify it
|
|
|
|
under the terms of the GNU General Public License as published by
|
|
|
|
the Free Software Foundation; either version 3, or (at your option)
|
|
|
|
any later version.
|
|
|
|
|
|
|
|
GCC is distributed in the hope that it will be useful, but
|
|
|
|
WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
|
|
|
|
General Public License for more details.
|
|
|
|
|
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
along with GCC; see the file COPYING3. If not see
|
|
|
|
<http://www.gnu.org/licenses/>. */
|
|
|
|
|
tree-ssa-math-opts.c (nop_stats, [...]): Moved to ...
* tree-ssa-math-opts.c (nop_stats, bswap_stats, struct symbolic_number,
BITS_PER_MARKER, MARKER_MASK, MARKER_BYTE_UNKNOWN, HEAD_MARKER, CMPNOP,
CMPXCHG, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, pass_data_optimize_bswap,
class pass_optimize_bswap, bswap_replace,
pass_optimize_bswap::execute): Moved to ...
* gimple-ssa-store-merging.c: ... this file.
Include optabs-tree.h.
(nop_stats, bswap_stats, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, bswap_replace): Put into
anonymous namespace, remove static keywords.
(pass_optimize_bswap::gate): Test BITS_PER_UNIT == 8 here...
(pass_optimize_bswap::execute): ... rather than here. Formatting fix.
From-SVN: r254947
2017-11-20 11:08:48 +01:00
|
|
|
/* The purpose of the store merging pass is to combine multiple memory
|
|
|
|
stores of constant values, values loaded from memory or bitwise operations
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
on those to consecutive memory locations into fewer wider stores.
|
2016-10-28 16:18:50 +02:00
|
|
|
For example, if we have a sequence peforming four byte stores to
|
|
|
|
consecutive memory locations:
|
|
|
|
[p ] := imm1;
|
|
|
|
[p + 1B] := imm2;
|
|
|
|
[p + 2B] := imm3;
|
|
|
|
[p + 3B] := imm4;
|
|
|
|
we can transform this into a single 4-byte store if the target supports it:
|
|
|
|
[p] := imm1:imm2:imm3:imm4 //concatenated immediates according to endianness.
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
Or:
|
|
|
|
[p ] := [q ];
|
|
|
|
[p + 1B] := [q + 1B];
|
|
|
|
[p + 2B] := [q + 2B];
|
|
|
|
[p + 3B] := [q + 3B];
|
|
|
|
if there is no overlap can be transformed into a single 4-byte
|
|
|
|
load followed by single 4-byte store.
|
|
|
|
|
|
|
|
Or:
|
|
|
|
[p ] := [q ] ^ imm1;
|
|
|
|
[p + 1B] := [q + 1B] ^ imm2;
|
|
|
|
[p + 2B] := [q + 2B] ^ imm3;
|
|
|
|
[p + 3B] := [q + 3B] ^ imm4;
|
|
|
|
if there is no overlap can be transformed into a single 4-byte
|
|
|
|
load, xored with imm1:imm2:imm3:imm4 and stored using a single 4-byte store.
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
The algorithm is applied to each basic block in three phases:
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
1) Scan through the basic block recording assignments to
|
2016-10-28 16:18:50 +02:00
|
|
|
destinations that can be expressed as a store to memory of a certain size
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
at a certain bit offset from expressions we can handle. For bit-fields
|
|
|
|
we also note the surrounding bit region, bits that could be stored in
|
|
|
|
a read-modify-write operation when storing the bit-field. Record store
|
|
|
|
chains to different bases in a hash_map (m_stores) and make sure to
|
|
|
|
terminate such chains when appropriate (for example when when the stored
|
|
|
|
values get used subsequently).
|
2016-10-28 16:18:50 +02:00
|
|
|
These stores can be a result of structure element initializers, array stores
|
|
|
|
etc. A store_immediate_info object is recorded for every such store.
|
|
|
|
Record as many such assignments to a single base as possible until a
|
|
|
|
statement that interferes with the store sequence is encountered.
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
Each store has up to 2 operands, which can be an immediate constant
|
|
|
|
or a memory load, from which the value to be stored can be computed.
|
|
|
|
At most one of the operands can be a constant. The operands are recorded
|
|
|
|
in store_operand_info struct.
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
2) Analyze the chain of stores recorded in phase 1) (i.e. the vector of
|
|
|
|
store_immediate_info objects) and coalesce contiguous stores into
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
merged_store_group objects. For bit-fields stores, we don't need to
|
|
|
|
require the stores to be contiguous, just their surrounding bit regions
|
|
|
|
have to be contiguous. If the expression being stored is different
|
|
|
|
between adjacent stores, such as one store storing a constant and
|
|
|
|
following storing a value loaded from memory, or if the loaded memory
|
|
|
|
objects are not adjacent, a new merged_store_group is created as well.
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
For example, given the stores:
|
|
|
|
[p ] := 0;
|
|
|
|
[p + 1B] := 1;
|
|
|
|
[p + 3B] := 0;
|
|
|
|
[p + 4B] := 1;
|
|
|
|
[p + 5B] := 0;
|
|
|
|
[p + 6B] := 0;
|
|
|
|
This phase would produce two merged_store_group objects, one recording the
|
|
|
|
two bytes stored in the memory region [p : p + 1] and another
|
|
|
|
recording the four bytes stored in the memory region [p + 3 : p + 6].
|
|
|
|
|
|
|
|
3) The merged_store_group objects produced in phase 2) are processed
|
|
|
|
to generate the sequence of wider stores that set the contiguous memory
|
|
|
|
regions to the sequence of bytes that correspond to it. This may emit
|
|
|
|
multiple stores per store group to handle contiguous stores that are not
|
|
|
|
of a size that is a power of 2. For example it can try to emit a 40-bit
|
|
|
|
store as a 32-bit store followed by an 8-bit store.
|
|
|
|
We try to emit as wide stores as we can while respecting STRICT_ALIGNMENT or
|
2017-09-12 15:27:55 +02:00
|
|
|
TARGET_SLOW_UNALIGNED_ACCESS rules.
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
Note on endianness and example:
|
|
|
|
Consider 2 contiguous 16-bit stores followed by 2 contiguous 8-bit stores:
|
|
|
|
[p ] := 0x1234;
|
|
|
|
[p + 2B] := 0x5678;
|
|
|
|
[p + 4B] := 0xab;
|
|
|
|
[p + 5B] := 0xcd;
|
|
|
|
|
|
|
|
The memory layout for little-endian (LE) and big-endian (BE) must be:
|
|
|
|
p |LE|BE|
|
|
|
|
---------
|
|
|
|
0 |34|12|
|
|
|
|
1 |12|34|
|
|
|
|
2 |78|56|
|
|
|
|
3 |56|78|
|
|
|
|
4 |ab|ab|
|
|
|
|
5 |cd|cd|
|
|
|
|
|
|
|
|
To merge these into a single 48-bit merged value 'val' in phase 2)
|
|
|
|
on little-endian we insert stores to higher (consecutive) bitpositions
|
|
|
|
into the most significant bits of the merged value.
|
|
|
|
The final merged value would be: 0xcdab56781234
|
|
|
|
|
|
|
|
For big-endian we insert stores to higher bitpositions into the least
|
|
|
|
significant bits of the merged value.
|
|
|
|
The final merged value would be: 0x12345678abcd
|
|
|
|
|
|
|
|
Then, in phase 3), we want to emit this 48-bit value as a 32-bit store
|
|
|
|
followed by a 16-bit store. Again, we must consider endianness when
|
|
|
|
breaking down the 48-bit value 'val' computed above.
|
|
|
|
For little endian we emit:
|
|
|
|
[p] (32-bit) := 0x56781234; // val & 0x0000ffffffff;
|
|
|
|
[p + 4B] (16-bit) := 0xcdab; // (val & 0xffff00000000) >> 32;
|
|
|
|
|
|
|
|
Whereas for big-endian we emit:
|
|
|
|
[p] (32-bit) := 0x12345678; // (val & 0xffffffff0000) >> 16;
|
|
|
|
[p + 4B] (16-bit) := 0xabcd; // val & 0x00000000ffff; */
|
|
|
|
|
|
|
|
#include "config.h"
|
|
|
|
#include "system.h"
|
|
|
|
#include "coretypes.h"
|
|
|
|
#include "backend.h"
|
|
|
|
#include "tree.h"
|
|
|
|
#include "gimple.h"
|
|
|
|
#include "builtins.h"
|
|
|
|
#include "fold-const.h"
|
|
|
|
#include "tree-pass.h"
|
|
|
|
#include "ssa.h"
|
|
|
|
#include "gimple-pretty-print.h"
|
|
|
|
#include "alias.h"
|
|
|
|
#include "fold-const.h"
|
|
|
|
#include "params.h"
|
|
|
|
#include "print-tree.h"
|
|
|
|
#include "tree-hash-traits.h"
|
|
|
|
#include "gimple-iterator.h"
|
|
|
|
#include "gimplify.h"
|
|
|
|
#include "stor-layout.h"
|
|
|
|
#include "timevar.h"
|
|
|
|
#include "tree-cfg.h"
|
|
|
|
#include "tree-eh.h"
|
|
|
|
#include "target.h"
|
2016-11-02 16:00:48 +01:00
|
|
|
#include "gimplify-me.h"
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
#include "rtl.h"
|
|
|
|
#include "expr.h" /* For get_bit_range. */
|
tree-ssa-math-opts.c (nop_stats, [...]): Moved to ...
* tree-ssa-math-opts.c (nop_stats, bswap_stats, struct symbolic_number,
BITS_PER_MARKER, MARKER_MASK, MARKER_BYTE_UNKNOWN, HEAD_MARKER, CMPNOP,
CMPXCHG, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, pass_data_optimize_bswap,
class pass_optimize_bswap, bswap_replace,
pass_optimize_bswap::execute): Moved to ...
* gimple-ssa-store-merging.c: ... this file.
Include optabs-tree.h.
(nop_stats, bswap_stats, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, bswap_replace): Put into
anonymous namespace, remove static keywords.
(pass_optimize_bswap::gate): Test BITS_PER_UNIT == 8 here...
(pass_optimize_bswap::execute): ... rather than here. Formatting fix.
From-SVN: r254947
2017-11-20 11:08:48 +01:00
|
|
|
#include "optabs-tree.h"
|
2016-11-08 17:04:20 +01:00
|
|
|
#include "selftest.h"
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
/* The maximum size (in bits) of the stores this pass should generate. */
|
|
|
|
#define MAX_STORE_BITSIZE (BITS_PER_WORD)
|
|
|
|
#define MAX_STORE_BYTES (MAX_STORE_BITSIZE / BITS_PER_UNIT)
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Limit to bound the number of aliasing checks for loads with the same
|
|
|
|
vuse as the corresponding store. */
|
|
|
|
#define MAX_STORE_ALIAS_CHECKS 64
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
namespace {
|
|
|
|
|
tree-ssa-math-opts.c (nop_stats, [...]): Moved to ...
* tree-ssa-math-opts.c (nop_stats, bswap_stats, struct symbolic_number,
BITS_PER_MARKER, MARKER_MASK, MARKER_BYTE_UNKNOWN, HEAD_MARKER, CMPNOP,
CMPXCHG, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, pass_data_optimize_bswap,
class pass_optimize_bswap, bswap_replace,
pass_optimize_bswap::execute): Moved to ...
* gimple-ssa-store-merging.c: ... this file.
Include optabs-tree.h.
(nop_stats, bswap_stats, do_shift_rotate, verify_symbolic_number_p,
init_symbolic_number, find_bswap_or_nop_load, perform_symbolic_merge,
find_bswap_or_nop_1, find_bswap_or_nop, bswap_replace): Put into
anonymous namespace, remove static keywords.
(pass_optimize_bswap::gate): Test BITS_PER_UNIT == 8 here...
(pass_optimize_bswap::execute): ... rather than here. Formatting fix.
From-SVN: r254947
2017-11-20 11:08:48 +01:00
|
|
|
struct
|
|
|
|
{
|
|
|
|
/* Number of hand-written 16-bit nop / bswaps found. */
|
|
|
|
int found_16bit;
|
|
|
|
|
|
|
|
/* Number of hand-written 32-bit nop / bswaps found. */
|
|
|
|
int found_32bit;
|
|
|
|
|
|
|
|
/* Number of hand-written 64-bit nop / bswaps found. */
|
|
|
|
int found_64bit;
|
|
|
|
} nop_stats, bswap_stats;
|
|
|
|
|
|
|
|
/* A symbolic number structure is used to detect byte permutation and selection
|
|
|
|
patterns of a source. To achieve that, its field N contains an artificial
|
|
|
|
number consisting of BITS_PER_MARKER sized markers tracking where does each
|
|
|
|
byte come from in the source:
|
|
|
|
|
|
|
|
0 - target byte has the value 0
|
|
|
|
FF - target byte has an unknown value (eg. due to sign extension)
|
|
|
|
1..size - marker value is the byte index in the source (0 for lsb).
|
|
|
|
|
|
|
|
To detect permutations on memory sources (arrays and structures), a symbolic
|
|
|
|
number is also associated:
|
|
|
|
- a base address BASE_ADDR and an OFFSET giving the address of the source;
|
|
|
|
- a range which gives the difference between the highest and lowest accessed
|
|
|
|
memory location to make such a symbolic number;
|
|
|
|
- the address SRC of the source element of lowest address as a convenience
|
|
|
|
to easily get BASE_ADDR + offset + lowest bytepos;
|
|
|
|
- number of expressions N_OPS bitwise ored together to represent
|
|
|
|
approximate cost of the computation.
|
|
|
|
|
|
|
|
Note 1: the range is different from size as size reflects the size of the
|
|
|
|
type of the current expression. For instance, for an array char a[],
|
|
|
|
(short) a[0] | (short) a[3] would have a size of 2 but a range of 4 while
|
|
|
|
(short) a[0] | ((short) a[0] << 1) would still have a size of 2 but this
|
|
|
|
time a range of 1.
|
|
|
|
|
|
|
|
Note 2: for non-memory sources, range holds the same value as size.
|
|
|
|
|
|
|
|
Note 3: SRC points to the SSA_NAME in case of non-memory source. */
|
|
|
|
|
|
|
|
struct symbolic_number {
|
|
|
|
uint64_t n;
|
|
|
|
tree type;
|
|
|
|
tree base_addr;
|
|
|
|
tree offset;
|
|
|
|
HOST_WIDE_INT bytepos;
|
|
|
|
tree src;
|
|
|
|
tree alias_set;
|
|
|
|
tree vuse;
|
|
|
|
unsigned HOST_WIDE_INT range;
|
|
|
|
int n_ops;
|
|
|
|
};
|
|
|
|
|
|
|
|
#define BITS_PER_MARKER 8
|
|
|
|
#define MARKER_MASK ((1 << BITS_PER_MARKER) - 1)
|
|
|
|
#define MARKER_BYTE_UNKNOWN MARKER_MASK
|
|
|
|
#define HEAD_MARKER(n, size) \
|
|
|
|
((n) & ((uint64_t) MARKER_MASK << (((size) - 1) * BITS_PER_MARKER)))
|
|
|
|
|
|
|
|
/* The number which the find_bswap_or_nop_1 result should match in
|
|
|
|
order to have a nop. The number is masked according to the size of
|
|
|
|
the symbolic number before using it. */
|
|
|
|
#define CMPNOP (sizeof (int64_t) < 8 ? 0 : \
|
|
|
|
(uint64_t)0x08070605 << 32 | 0x04030201)
|
|
|
|
|
|
|
|
/* The number which the find_bswap_or_nop_1 result should match in
|
|
|
|
order to have a byte swap. The number is masked according to the
|
|
|
|
size of the symbolic number before using it. */
|
|
|
|
#define CMPXCHG (sizeof (int64_t) < 8 ? 0 : \
|
|
|
|
(uint64_t)0x01020304 << 32 | 0x05060708)
|
|
|
|
|
|
|
|
/* Perform a SHIFT or ROTATE operation by COUNT bits on symbolic
|
|
|
|
number N. Return false if the requested operation is not permitted
|
|
|
|
on a symbolic number. */
|
|
|
|
|
|
|
|
inline bool
|
|
|
|
do_shift_rotate (enum tree_code code,
|
|
|
|
struct symbolic_number *n,
|
|
|
|
int count)
|
|
|
|
{
|
|
|
|
int i, size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
|
|
|
|
unsigned head_marker;
|
|
|
|
|
|
|
|
if (count % BITS_PER_UNIT != 0)
|
|
|
|
return false;
|
|
|
|
count = (count / BITS_PER_UNIT) * BITS_PER_MARKER;
|
|
|
|
|
|
|
|
/* Zero out the extra bits of N in order to avoid them being shifted
|
|
|
|
into the significant bits. */
|
|
|
|
if (size < 64 / BITS_PER_MARKER)
|
|
|
|
n->n &= ((uint64_t) 1 << (size * BITS_PER_MARKER)) - 1;
|
|
|
|
|
|
|
|
switch (code)
|
|
|
|
{
|
|
|
|
case LSHIFT_EXPR:
|
|
|
|
n->n <<= count;
|
|
|
|
break;
|
|
|
|
case RSHIFT_EXPR:
|
|
|
|
head_marker = HEAD_MARKER (n->n, size);
|
|
|
|
n->n >>= count;
|
|
|
|
/* Arithmetic shift of signed type: result is dependent on the value. */
|
|
|
|
if (!TYPE_UNSIGNED (n->type) && head_marker)
|
|
|
|
for (i = 0; i < count / BITS_PER_MARKER; i++)
|
|
|
|
n->n |= (uint64_t) MARKER_BYTE_UNKNOWN
|
|
|
|
<< ((size - 1 - i) * BITS_PER_MARKER);
|
|
|
|
break;
|
|
|
|
case LROTATE_EXPR:
|
|
|
|
n->n = (n->n << count) | (n->n >> ((size * BITS_PER_MARKER) - count));
|
|
|
|
break;
|
|
|
|
case RROTATE_EXPR:
|
|
|
|
n->n = (n->n >> count) | (n->n << ((size * BITS_PER_MARKER) - count));
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
/* Zero unused bits for size. */
|
|
|
|
if (size < 64 / BITS_PER_MARKER)
|
|
|
|
n->n &= ((uint64_t) 1 << (size * BITS_PER_MARKER)) - 1;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Perform sanity checking for the symbolic number N and the gimple
|
|
|
|
statement STMT. */
|
|
|
|
|
|
|
|
inline bool
|
|
|
|
verify_symbolic_number_p (struct symbolic_number *n, gimple *stmt)
|
|
|
|
{
|
|
|
|
tree lhs_type;
|
|
|
|
|
|
|
|
lhs_type = gimple_expr_type (stmt);
|
|
|
|
|
|
|
|
if (TREE_CODE (lhs_type) != INTEGER_TYPE)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (TYPE_PRECISION (lhs_type) != TYPE_PRECISION (n->type))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Initialize the symbolic number N for the bswap pass from the base element
|
|
|
|
SRC manipulated by the bitwise OR expression. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
init_symbolic_number (struct symbolic_number *n, tree src)
|
|
|
|
{
|
|
|
|
int size;
|
|
|
|
|
|
|
|
if (! INTEGRAL_TYPE_P (TREE_TYPE (src)))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
n->base_addr = n->offset = n->alias_set = n->vuse = NULL_TREE;
|
|
|
|
n->src = src;
|
|
|
|
|
|
|
|
/* Set up the symbolic number N by setting each byte to a value between 1 and
|
|
|
|
the byte size of rhs1. The highest order byte is set to n->size and the
|
|
|
|
lowest order byte to 1. */
|
|
|
|
n->type = TREE_TYPE (src);
|
|
|
|
size = TYPE_PRECISION (n->type);
|
|
|
|
if (size % BITS_PER_UNIT != 0)
|
|
|
|
return false;
|
|
|
|
size /= BITS_PER_UNIT;
|
|
|
|
if (size > 64 / BITS_PER_MARKER)
|
|
|
|
return false;
|
|
|
|
n->range = size;
|
|
|
|
n->n = CMPNOP;
|
|
|
|
n->n_ops = 1;
|
|
|
|
|
|
|
|
if (size < 64 / BITS_PER_MARKER)
|
|
|
|
n->n &= ((uint64_t) 1 << (size * BITS_PER_MARKER)) - 1;
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Check if STMT might be a byte swap or a nop from a memory source and returns
|
|
|
|
the answer. If so, REF is that memory source and the base of the memory area
|
|
|
|
accessed and the offset of the access from that base are recorded in N. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
find_bswap_or_nop_load (gimple *stmt, tree ref, struct symbolic_number *n)
|
|
|
|
{
|
|
|
|
/* Leaf node is an array or component ref. Memorize its base and
|
|
|
|
offset from base to compare to other such leaf node. */
|
|
|
|
HOST_WIDE_INT bitsize, bitpos;
|
|
|
|
machine_mode mode;
|
|
|
|
int unsignedp, reversep, volatilep;
|
|
|
|
tree offset, base_addr;
|
|
|
|
|
|
|
|
/* Not prepared to handle PDP endian. */
|
|
|
|
if (BYTES_BIG_ENDIAN != WORDS_BIG_ENDIAN)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (!gimple_assign_load_p (stmt) || gimple_has_volatile_ops (stmt))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
base_addr = get_inner_reference (ref, &bitsize, &bitpos, &offset, &mode,
|
|
|
|
&unsignedp, &reversep, &volatilep);
|
|
|
|
|
|
|
|
if (TREE_CODE (base_addr) == MEM_REF)
|
|
|
|
{
|
|
|
|
offset_int bit_offset = 0;
|
|
|
|
tree off = TREE_OPERAND (base_addr, 1);
|
|
|
|
|
|
|
|
if (!integer_zerop (off))
|
|
|
|
{
|
|
|
|
offset_int boff, coff = mem_ref_offset (base_addr);
|
|
|
|
boff = coff << LOG2_BITS_PER_UNIT;
|
|
|
|
bit_offset += boff;
|
|
|
|
}
|
|
|
|
|
|
|
|
base_addr = TREE_OPERAND (base_addr, 0);
|
|
|
|
|
|
|
|
/* Avoid returning a negative bitpos as this may wreak havoc later. */
|
|
|
|
if (wi::neg_p (bit_offset))
|
|
|
|
{
|
|
|
|
offset_int mask = wi::mask <offset_int> (LOG2_BITS_PER_UNIT, false);
|
|
|
|
offset_int tem = wi::bit_and_not (bit_offset, mask);
|
|
|
|
/* TEM is the bitpos rounded to BITS_PER_UNIT towards -Inf.
|
|
|
|
Subtract it to BIT_OFFSET and add it (scaled) to OFFSET. */
|
|
|
|
bit_offset -= tem;
|
|
|
|
tem >>= LOG2_BITS_PER_UNIT;
|
|
|
|
if (offset)
|
|
|
|
offset = size_binop (PLUS_EXPR, offset,
|
|
|
|
wide_int_to_tree (sizetype, tem));
|
|
|
|
else
|
|
|
|
offset = wide_int_to_tree (sizetype, tem);
|
|
|
|
}
|
|
|
|
|
|
|
|
bitpos += bit_offset.to_shwi ();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (bitpos % BITS_PER_UNIT)
|
|
|
|
return false;
|
|
|
|
if (bitsize % BITS_PER_UNIT)
|
|
|
|
return false;
|
|
|
|
if (reversep)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (!init_symbolic_number (n, ref))
|
|
|
|
return false;
|
|
|
|
n->base_addr = base_addr;
|
|
|
|
n->offset = offset;
|
|
|
|
n->bytepos = bitpos / BITS_PER_UNIT;
|
|
|
|
n->alias_set = reference_alias_ptr_type (ref);
|
|
|
|
n->vuse = gimple_vuse (stmt);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Compute the symbolic number N representing the result of a bitwise OR on 2
|
|
|
|
symbolic number N1 and N2 whose source statements are respectively
|
|
|
|
SOURCE_STMT1 and SOURCE_STMT2. */
|
|
|
|
|
|
|
|
gimple *
|
|
|
|
perform_symbolic_merge (gimple *source_stmt1, struct symbolic_number *n1,
|
|
|
|
gimple *source_stmt2, struct symbolic_number *n2,
|
|
|
|
struct symbolic_number *n)
|
|
|
|
{
|
|
|
|
int i, size;
|
|
|
|
uint64_t mask;
|
|
|
|
gimple *source_stmt;
|
|
|
|
struct symbolic_number *n_start;
|
|
|
|
|
|
|
|
tree rhs1 = gimple_assign_rhs1 (source_stmt1);
|
|
|
|
if (TREE_CODE (rhs1) == BIT_FIELD_REF
|
|
|
|
&& TREE_CODE (TREE_OPERAND (rhs1, 0)) == SSA_NAME)
|
|
|
|
rhs1 = TREE_OPERAND (rhs1, 0);
|
|
|
|
tree rhs2 = gimple_assign_rhs1 (source_stmt2);
|
|
|
|
if (TREE_CODE (rhs2) == BIT_FIELD_REF
|
|
|
|
&& TREE_CODE (TREE_OPERAND (rhs2, 0)) == SSA_NAME)
|
|
|
|
rhs2 = TREE_OPERAND (rhs2, 0);
|
|
|
|
|
|
|
|
/* Sources are different, cancel bswap if they are not memory location with
|
|
|
|
the same base (array, structure, ...). */
|
|
|
|
if (rhs1 != rhs2)
|
|
|
|
{
|
|
|
|
uint64_t inc;
|
|
|
|
HOST_WIDE_INT start_sub, end_sub, end1, end2, end;
|
|
|
|
struct symbolic_number *toinc_n_ptr, *n_end;
|
|
|
|
basic_block bb1, bb2;
|
|
|
|
|
|
|
|
if (!n1->base_addr || !n2->base_addr
|
|
|
|
|| !operand_equal_p (n1->base_addr, n2->base_addr, 0))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (!n1->offset != !n2->offset
|
|
|
|
|| (n1->offset && !operand_equal_p (n1->offset, n2->offset, 0)))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (n1->bytepos < n2->bytepos)
|
|
|
|
{
|
|
|
|
n_start = n1;
|
|
|
|
start_sub = n2->bytepos - n1->bytepos;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
n_start = n2;
|
|
|
|
start_sub = n1->bytepos - n2->bytepos;
|
|
|
|
}
|
|
|
|
|
|
|
|
bb1 = gimple_bb (source_stmt1);
|
|
|
|
bb2 = gimple_bb (source_stmt2);
|
|
|
|
if (dominated_by_p (CDI_DOMINATORS, bb1, bb2))
|
|
|
|
source_stmt = source_stmt1;
|
|
|
|
else
|
|
|
|
source_stmt = source_stmt2;
|
|
|
|
|
|
|
|
/* Find the highest address at which a load is performed and
|
|
|
|
compute related info. */
|
|
|
|
end1 = n1->bytepos + (n1->range - 1);
|
|
|
|
end2 = n2->bytepos + (n2->range - 1);
|
|
|
|
if (end1 < end2)
|
|
|
|
{
|
|
|
|
end = end2;
|
|
|
|
end_sub = end2 - end1;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
end = end1;
|
|
|
|
end_sub = end1 - end2;
|
|
|
|
}
|
|
|
|
n_end = (end2 > end1) ? n2 : n1;
|
|
|
|
|
|
|
|
/* Find symbolic number whose lsb is the most significant. */
|
|
|
|
if (BYTES_BIG_ENDIAN)
|
|
|
|
toinc_n_ptr = (n_end == n1) ? n2 : n1;
|
|
|
|
else
|
|
|
|
toinc_n_ptr = (n_start == n1) ? n2 : n1;
|
|
|
|
|
|
|
|
n->range = end - n_start->bytepos + 1;
|
|
|
|
|
|
|
|
/* Check that the range of memory covered can be represented by
|
|
|
|
a symbolic number. */
|
|
|
|
if (n->range > 64 / BITS_PER_MARKER)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Reinterpret byte marks in symbolic number holding the value of
|
|
|
|
bigger weight according to target endianness. */
|
|
|
|
inc = BYTES_BIG_ENDIAN ? end_sub : start_sub;
|
|
|
|
size = TYPE_PRECISION (n1->type) / BITS_PER_UNIT;
|
|
|
|
for (i = 0; i < size; i++, inc <<= BITS_PER_MARKER)
|
|
|
|
{
|
|
|
|
unsigned marker
|
|
|
|
= (toinc_n_ptr->n >> (i * BITS_PER_MARKER)) & MARKER_MASK;
|
|
|
|
if (marker && marker != MARKER_BYTE_UNKNOWN)
|
|
|
|
toinc_n_ptr->n += inc;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
n->range = n1->range;
|
|
|
|
n_start = n1;
|
|
|
|
source_stmt = source_stmt1;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!n1->alias_set
|
|
|
|
|| alias_ptr_types_compatible_p (n1->alias_set, n2->alias_set))
|
|
|
|
n->alias_set = n1->alias_set;
|
|
|
|
else
|
|
|
|
n->alias_set = ptr_type_node;
|
|
|
|
n->vuse = n_start->vuse;
|
|
|
|
n->base_addr = n_start->base_addr;
|
|
|
|
n->offset = n_start->offset;
|
|
|
|
n->src = n_start->src;
|
|
|
|
n->bytepos = n_start->bytepos;
|
|
|
|
n->type = n_start->type;
|
|
|
|
size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
|
|
|
|
|
|
|
|
for (i = 0, mask = MARKER_MASK; i < size; i++, mask <<= BITS_PER_MARKER)
|
|
|
|
{
|
|
|
|
uint64_t masked1, masked2;
|
|
|
|
|
|
|
|
masked1 = n1->n & mask;
|
|
|
|
masked2 = n2->n & mask;
|
|
|
|
if (masked1 && masked2 && masked1 != masked2)
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
n->n = n1->n | n2->n;
|
|
|
|
n->n_ops = n1->n_ops + n2->n_ops;
|
|
|
|
|
|
|
|
return source_stmt;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* find_bswap_or_nop_1 invokes itself recursively with N and tries to perform
|
|
|
|
the operation given by the rhs of STMT on the result. If the operation
|
|
|
|
could successfully be executed the function returns a gimple stmt whose
|
|
|
|
rhs's first tree is the expression of the source operand and NULL
|
|
|
|
otherwise. */
|
|
|
|
|
|
|
|
gimple *
|
|
|
|
find_bswap_or_nop_1 (gimple *stmt, struct symbolic_number *n, int limit)
|
|
|
|
{
|
|
|
|
enum tree_code code;
|
|
|
|
tree rhs1, rhs2 = NULL;
|
|
|
|
gimple *rhs1_stmt, *rhs2_stmt, *source_stmt1;
|
|
|
|
enum gimple_rhs_class rhs_class;
|
|
|
|
|
|
|
|
if (!limit || !is_gimple_assign (stmt))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
rhs1 = gimple_assign_rhs1 (stmt);
|
|
|
|
|
|
|
|
if (find_bswap_or_nop_load (stmt, rhs1, n))
|
|
|
|
return stmt;
|
|
|
|
|
|
|
|
/* Handle BIT_FIELD_REF. */
|
|
|
|
if (TREE_CODE (rhs1) == BIT_FIELD_REF
|
|
|
|
&& TREE_CODE (TREE_OPERAND (rhs1, 0)) == SSA_NAME)
|
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT bitsize = tree_to_uhwi (TREE_OPERAND (rhs1, 1));
|
|
|
|
unsigned HOST_WIDE_INT bitpos = tree_to_uhwi (TREE_OPERAND (rhs1, 2));
|
|
|
|
if (bitpos % BITS_PER_UNIT == 0
|
|
|
|
&& bitsize % BITS_PER_UNIT == 0
|
|
|
|
&& init_symbolic_number (n, TREE_OPERAND (rhs1, 0)))
|
|
|
|
{
|
|
|
|
/* Handle big-endian bit numbering in BIT_FIELD_REF. */
|
|
|
|
if (BYTES_BIG_ENDIAN)
|
|
|
|
bitpos = TYPE_PRECISION (n->type) - bitpos - bitsize;
|
|
|
|
|
|
|
|
/* Shift. */
|
|
|
|
if (!do_shift_rotate (RSHIFT_EXPR, n, bitpos))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Mask. */
|
|
|
|
uint64_t mask = 0;
|
|
|
|
uint64_t tmp = (1 << BITS_PER_UNIT) - 1;
|
|
|
|
for (unsigned i = 0; i < bitsize / BITS_PER_UNIT;
|
|
|
|
i++, tmp <<= BITS_PER_UNIT)
|
|
|
|
mask |= (uint64_t) MARKER_MASK << (i * BITS_PER_MARKER);
|
|
|
|
n->n &= mask;
|
|
|
|
|
|
|
|
/* Convert. */
|
|
|
|
n->type = TREE_TYPE (rhs1);
|
|
|
|
if (!n->base_addr)
|
|
|
|
n->range = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
|
|
|
|
|
|
|
|
return verify_symbolic_number_p (n, stmt) ? stmt : NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (TREE_CODE (rhs1) != SSA_NAME)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
code = gimple_assign_rhs_code (stmt);
|
|
|
|
rhs_class = gimple_assign_rhs_class (stmt);
|
|
|
|
rhs1_stmt = SSA_NAME_DEF_STMT (rhs1);
|
|
|
|
|
|
|
|
if (rhs_class == GIMPLE_BINARY_RHS)
|
|
|
|
rhs2 = gimple_assign_rhs2 (stmt);
|
|
|
|
|
|
|
|
/* Handle unary rhs and binary rhs with integer constants as second
|
|
|
|
operand. */
|
|
|
|
|
|
|
|
if (rhs_class == GIMPLE_UNARY_RHS
|
|
|
|
|| (rhs_class == GIMPLE_BINARY_RHS
|
|
|
|
&& TREE_CODE (rhs2) == INTEGER_CST))
|
|
|
|
{
|
|
|
|
if (code != BIT_AND_EXPR
|
|
|
|
&& code != LSHIFT_EXPR
|
|
|
|
&& code != RSHIFT_EXPR
|
|
|
|
&& code != LROTATE_EXPR
|
|
|
|
&& code != RROTATE_EXPR
|
|
|
|
&& !CONVERT_EXPR_CODE_P (code))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
source_stmt1 = find_bswap_or_nop_1 (rhs1_stmt, n, limit - 1);
|
|
|
|
|
|
|
|
/* If find_bswap_or_nop_1 returned NULL, STMT is a leaf node and
|
|
|
|
we have to initialize the symbolic number. */
|
|
|
|
if (!source_stmt1)
|
|
|
|
{
|
|
|
|
if (gimple_assign_load_p (stmt)
|
|
|
|
|| !init_symbolic_number (n, rhs1))
|
|
|
|
return NULL;
|
|
|
|
source_stmt1 = stmt;
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (code)
|
|
|
|
{
|
|
|
|
case BIT_AND_EXPR:
|
|
|
|
{
|
|
|
|
int i, size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
|
|
|
|
uint64_t val = int_cst_value (rhs2), mask = 0;
|
|
|
|
uint64_t tmp = (1 << BITS_PER_UNIT) - 1;
|
|
|
|
|
|
|
|
/* Only constants masking full bytes are allowed. */
|
|
|
|
for (i = 0; i < size; i++, tmp <<= BITS_PER_UNIT)
|
|
|
|
if ((val & tmp) != 0 && (val & tmp) != tmp)
|
|
|
|
return NULL;
|
|
|
|
else if (val & tmp)
|
|
|
|
mask |= (uint64_t) MARKER_MASK << (i * BITS_PER_MARKER);
|
|
|
|
|
|
|
|
n->n &= mask;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case LSHIFT_EXPR:
|
|
|
|
case RSHIFT_EXPR:
|
|
|
|
case LROTATE_EXPR:
|
|
|
|
case RROTATE_EXPR:
|
|
|
|
if (!do_shift_rotate (code, n, (int) TREE_INT_CST_LOW (rhs2)))
|
|
|
|
return NULL;
|
|
|
|
break;
|
|
|
|
CASE_CONVERT:
|
|
|
|
{
|
|
|
|
int i, type_size, old_type_size;
|
|
|
|
tree type;
|
|
|
|
|
|
|
|
type = gimple_expr_type (stmt);
|
|
|
|
type_size = TYPE_PRECISION (type);
|
|
|
|
if (type_size % BITS_PER_UNIT != 0)
|
|
|
|
return NULL;
|
|
|
|
type_size /= BITS_PER_UNIT;
|
|
|
|
if (type_size > 64 / BITS_PER_MARKER)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Sign extension: result is dependent on the value. */
|
|
|
|
old_type_size = TYPE_PRECISION (n->type) / BITS_PER_UNIT;
|
|
|
|
if (!TYPE_UNSIGNED (n->type) && type_size > old_type_size
|
|
|
|
&& HEAD_MARKER (n->n, old_type_size))
|
|
|
|
for (i = 0; i < type_size - old_type_size; i++)
|
|
|
|
n->n |= (uint64_t) MARKER_BYTE_UNKNOWN
|
|
|
|
<< ((type_size - 1 - i) * BITS_PER_MARKER);
|
|
|
|
|
|
|
|
if (type_size < 64 / BITS_PER_MARKER)
|
|
|
|
{
|
|
|
|
/* If STMT casts to a smaller type mask out the bits not
|
|
|
|
belonging to the target type. */
|
|
|
|
n->n &= ((uint64_t) 1 << (type_size * BITS_PER_MARKER)) - 1;
|
|
|
|
}
|
|
|
|
n->type = type;
|
|
|
|
if (!n->base_addr)
|
|
|
|
n->range = type_size;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return NULL;
|
|
|
|
};
|
|
|
|
return verify_symbolic_number_p (n, stmt) ? source_stmt1 : NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Handle binary rhs. */
|
|
|
|
|
|
|
|
if (rhs_class == GIMPLE_BINARY_RHS)
|
|
|
|
{
|
|
|
|
struct symbolic_number n1, n2;
|
|
|
|
gimple *source_stmt, *source_stmt2;
|
|
|
|
|
|
|
|
if (code != BIT_IOR_EXPR)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (TREE_CODE (rhs2) != SSA_NAME)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
rhs2_stmt = SSA_NAME_DEF_STMT (rhs2);
|
|
|
|
|
|
|
|
switch (code)
|
|
|
|
{
|
|
|
|
case BIT_IOR_EXPR:
|
|
|
|
source_stmt1 = find_bswap_or_nop_1 (rhs1_stmt, &n1, limit - 1);
|
|
|
|
|
|
|
|
if (!source_stmt1)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
source_stmt2 = find_bswap_or_nop_1 (rhs2_stmt, &n2, limit - 1);
|
|
|
|
|
|
|
|
if (!source_stmt2)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (TYPE_PRECISION (n1.type) != TYPE_PRECISION (n2.type))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (!n1.vuse != !n2.vuse
|
|
|
|
|| (n1.vuse && !operand_equal_p (n1.vuse, n2.vuse, 0)))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
source_stmt
|
|
|
|
= perform_symbolic_merge (source_stmt1, &n1, source_stmt2, &n2, n);
|
|
|
|
|
|
|
|
if (!source_stmt)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
if (!verify_symbolic_number_p (n, stmt))
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
return source_stmt;
|
|
|
|
}
|
|
|
|
return NULL;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Check if STMT completes a bswap implementation or a read in a given
|
|
|
|
endianness consisting of ORs, SHIFTs and ANDs and sets *BSWAP
|
|
|
|
accordingly. It also sets N to represent the kind of operations
|
|
|
|
performed: size of the resulting expression and whether it works on
|
|
|
|
a memory source, and if so alias-set and vuse. At last, the
|
|
|
|
function returns a stmt whose rhs's first tree is the source
|
|
|
|
expression. */
|
|
|
|
|
|
|
|
gimple *
|
|
|
|
find_bswap_or_nop (gimple *stmt, struct symbolic_number *n, bool *bswap)
|
|
|
|
{
|
|
|
|
unsigned rsize;
|
|
|
|
uint64_t tmpn, mask;
|
|
|
|
/* The number which the find_bswap_or_nop_1 result should match in order
|
|
|
|
to have a full byte swap. The number is shifted to the right
|
|
|
|
according to the size of the symbolic number before using it. */
|
|
|
|
uint64_t cmpxchg = CMPXCHG;
|
|
|
|
uint64_t cmpnop = CMPNOP;
|
|
|
|
|
|
|
|
gimple *ins_stmt;
|
|
|
|
int limit;
|
|
|
|
|
|
|
|
/* The last parameter determines the depth search limit. It usually
|
|
|
|
correlates directly to the number n of bytes to be touched. We
|
|
|
|
increase that number by log2(n) + 1 here in order to also
|
|
|
|
cover signed -> unsigned conversions of the src operand as can be seen
|
|
|
|
in libgcc, and for initial shift/and operation of the src operand. */
|
|
|
|
limit = TREE_INT_CST_LOW (TYPE_SIZE_UNIT (gimple_expr_type (stmt)));
|
|
|
|
limit += 1 + (int) ceil_log2 ((unsigned HOST_WIDE_INT) limit);
|
|
|
|
ins_stmt = find_bswap_or_nop_1 (stmt, n, limit);
|
|
|
|
|
|
|
|
if (!ins_stmt)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Find real size of result (highest non-zero byte). */
|
|
|
|
if (n->base_addr)
|
|
|
|
for (tmpn = n->n, rsize = 0; tmpn; tmpn >>= BITS_PER_MARKER, rsize++);
|
|
|
|
else
|
|
|
|
rsize = n->range;
|
|
|
|
|
|
|
|
/* Zero out the bits corresponding to untouched bytes in original gimple
|
|
|
|
expression. */
|
|
|
|
if (n->range < (int) sizeof (int64_t))
|
|
|
|
{
|
|
|
|
mask = ((uint64_t) 1 << (n->range * BITS_PER_MARKER)) - 1;
|
|
|
|
cmpxchg >>= (64 / BITS_PER_MARKER - n->range) * BITS_PER_MARKER;
|
|
|
|
cmpnop &= mask;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Zero out the bits corresponding to unused bytes in the result of the
|
|
|
|
gimple expression. */
|
|
|
|
if (rsize < n->range)
|
|
|
|
{
|
|
|
|
if (BYTES_BIG_ENDIAN)
|
|
|
|
{
|
|
|
|
mask = ((uint64_t) 1 << (rsize * BITS_PER_MARKER)) - 1;
|
|
|
|
cmpxchg &= mask;
|
|
|
|
cmpnop >>= (n->range - rsize) * BITS_PER_MARKER;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
mask = ((uint64_t) 1 << (rsize * BITS_PER_MARKER)) - 1;
|
|
|
|
cmpxchg >>= (n->range - rsize) * BITS_PER_MARKER;
|
|
|
|
cmpnop &= mask;
|
|
|
|
}
|
|
|
|
n->range = rsize;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* A complete byte swap should make the symbolic number to start with
|
|
|
|
the largest digit in the highest order byte. Unchanged symbolic
|
|
|
|
number indicates a read with same endianness as target architecture. */
|
|
|
|
if (n->n == cmpnop)
|
|
|
|
*bswap = false;
|
|
|
|
else if (n->n == cmpxchg)
|
|
|
|
*bswap = true;
|
|
|
|
else
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
/* Useless bit manipulation performed by code. */
|
|
|
|
if (!n->base_addr && n->n == cmpnop && n->n_ops == 1)
|
|
|
|
return NULL;
|
|
|
|
|
|
|
|
n->range *= BITS_PER_UNIT;
|
|
|
|
return ins_stmt;
|
|
|
|
}
|
|
|
|
|
|
|
|
const pass_data pass_data_optimize_bswap =
|
|
|
|
{
|
|
|
|
GIMPLE_PASS, /* type */
|
|
|
|
"bswap", /* name */
|
|
|
|
OPTGROUP_NONE, /* optinfo_flags */
|
|
|
|
TV_NONE, /* tv_id */
|
|
|
|
PROP_ssa, /* properties_required */
|
|
|
|
0, /* properties_provided */
|
|
|
|
0, /* properties_destroyed */
|
|
|
|
0, /* todo_flags_start */
|
|
|
|
0, /* todo_flags_finish */
|
|
|
|
};
|
|
|
|
|
|
|
|
class pass_optimize_bswap : public gimple_opt_pass
|
|
|
|
{
|
|
|
|
public:
|
|
|
|
pass_optimize_bswap (gcc::context *ctxt)
|
|
|
|
: gimple_opt_pass (pass_data_optimize_bswap, ctxt)
|
|
|
|
{}
|
|
|
|
|
|
|
|
/* opt_pass methods: */
|
|
|
|
virtual bool gate (function *)
|
|
|
|
{
|
|
|
|
return flag_expensive_optimizations && optimize && BITS_PER_UNIT == 8;
|
|
|
|
}
|
|
|
|
|
|
|
|
virtual unsigned int execute (function *);
|
|
|
|
|
|
|
|
}; // class pass_optimize_bswap
|
|
|
|
|
|
|
|
/* Perform the bswap optimization: replace the expression computed in the rhs
|
|
|
|
of CUR_STMT by an equivalent bswap, load or load + bswap expression.
|
|
|
|
Which of these alternatives replace the rhs is given by N->base_addr (non
|
|
|
|
null if a load is needed) and BSWAP. The type, VUSE and set-alias of the
|
|
|
|
load to perform are also given in N while the builtin bswap invoke is given
|
|
|
|
in FNDEL. Finally, if a load is involved, SRC_STMT refers to one of the
|
|
|
|
load statements involved to construct the rhs in CUR_STMT and N->range gives
|
|
|
|
the size of the rhs expression for maintaining some statistics.
|
|
|
|
|
|
|
|
Note that if the replacement involve a load, CUR_STMT is moved just after
|
|
|
|
SRC_STMT to do the load with the same VUSE which can lead to CUR_STMT
|
|
|
|
changing of basic block. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
bswap_replace (gimple *cur_stmt, gimple *ins_stmt, tree fndecl,
|
|
|
|
tree bswap_type, tree load_type, struct symbolic_number *n,
|
|
|
|
bool bswap)
|
|
|
|
{
|
|
|
|
gimple_stmt_iterator gsi;
|
|
|
|
tree src, tmp, tgt;
|
|
|
|
gimple *bswap_stmt;
|
|
|
|
|
|
|
|
gsi = gsi_for_stmt (cur_stmt);
|
|
|
|
src = n->src;
|
|
|
|
tgt = gimple_assign_lhs (cur_stmt);
|
|
|
|
|
|
|
|
/* Need to load the value from memory first. */
|
|
|
|
if (n->base_addr)
|
|
|
|
{
|
|
|
|
gimple_stmt_iterator gsi_ins = gsi_for_stmt (ins_stmt);
|
|
|
|
tree addr_expr, addr_tmp, val_expr, val_tmp;
|
|
|
|
tree load_offset_ptr, aligned_load_type;
|
|
|
|
gimple *addr_stmt, *load_stmt;
|
|
|
|
unsigned align;
|
|
|
|
HOST_WIDE_INT load_offset = 0;
|
|
|
|
basic_block ins_bb, cur_bb;
|
|
|
|
|
|
|
|
ins_bb = gimple_bb (ins_stmt);
|
|
|
|
cur_bb = gimple_bb (cur_stmt);
|
|
|
|
if (!dominated_by_p (CDI_DOMINATORS, cur_bb, ins_bb))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
align = get_object_alignment (src);
|
|
|
|
|
|
|
|
/* Move cur_stmt just before one of the load of the original
|
|
|
|
to ensure it has the same VUSE. See PR61517 for what could
|
|
|
|
go wrong. */
|
|
|
|
if (gimple_bb (cur_stmt) != gimple_bb (ins_stmt))
|
|
|
|
reset_flow_sensitive_info (gimple_assign_lhs (cur_stmt));
|
|
|
|
gsi_move_before (&gsi, &gsi_ins);
|
|
|
|
gsi = gsi_for_stmt (cur_stmt);
|
|
|
|
|
|
|
|
/* Compute address to load from and cast according to the size
|
|
|
|
of the load. */
|
|
|
|
addr_expr = build_fold_addr_expr (unshare_expr (src));
|
|
|
|
if (is_gimple_mem_ref_addr (addr_expr))
|
|
|
|
addr_tmp = addr_expr;
|
|
|
|
else
|
|
|
|
{
|
|
|
|
addr_tmp = make_temp_ssa_name (TREE_TYPE (addr_expr), NULL,
|
|
|
|
"load_src");
|
|
|
|
addr_stmt = gimple_build_assign (addr_tmp, addr_expr);
|
|
|
|
gsi_insert_before (&gsi, addr_stmt, GSI_SAME_STMT);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Perform the load. */
|
|
|
|
aligned_load_type = load_type;
|
|
|
|
if (align < TYPE_ALIGN (load_type))
|
|
|
|
aligned_load_type = build_aligned_type (load_type, align);
|
|
|
|
load_offset_ptr = build_int_cst (n->alias_set, load_offset);
|
|
|
|
val_expr = fold_build2 (MEM_REF, aligned_load_type, addr_tmp,
|
|
|
|
load_offset_ptr);
|
|
|
|
|
|
|
|
if (!bswap)
|
|
|
|
{
|
|
|
|
if (n->range == 16)
|
|
|
|
nop_stats.found_16bit++;
|
|
|
|
else if (n->range == 32)
|
|
|
|
nop_stats.found_32bit++;
|
|
|
|
else
|
|
|
|
{
|
|
|
|
gcc_assert (n->range == 64);
|
|
|
|
nop_stats.found_64bit++;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Convert the result of load if necessary. */
|
|
|
|
if (!useless_type_conversion_p (TREE_TYPE (tgt), load_type))
|
|
|
|
{
|
|
|
|
val_tmp = make_temp_ssa_name (aligned_load_type, NULL,
|
|
|
|
"load_dst");
|
|
|
|
load_stmt = gimple_build_assign (val_tmp, val_expr);
|
|
|
|
gimple_set_vuse (load_stmt, n->vuse);
|
|
|
|
gsi_insert_before (&gsi, load_stmt, GSI_SAME_STMT);
|
|
|
|
gimple_assign_set_rhs_with_ops (&gsi, NOP_EXPR, val_tmp);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
gimple_assign_set_rhs_with_ops (&gsi, MEM_REF, val_expr);
|
|
|
|
gimple_set_vuse (cur_stmt, n->vuse);
|
|
|
|
}
|
|
|
|
update_stmt (cur_stmt);
|
|
|
|
|
|
|
|
if (dump_file)
|
|
|
|
{
|
|
|
|
fprintf (dump_file,
|
|
|
|
"%d bit load in target endianness found at: ",
|
|
|
|
(int) n->range);
|
|
|
|
print_gimple_stmt (dump_file, cur_stmt, 0);
|
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
val_tmp = make_temp_ssa_name (aligned_load_type, NULL, "load_dst");
|
|
|
|
load_stmt = gimple_build_assign (val_tmp, val_expr);
|
|
|
|
gimple_set_vuse (load_stmt, n->vuse);
|
|
|
|
gsi_insert_before (&gsi, load_stmt, GSI_SAME_STMT);
|
|
|
|
}
|
|
|
|
src = val_tmp;
|
|
|
|
}
|
|
|
|
else if (!bswap)
|
|
|
|
{
|
|
|
|
gimple *g;
|
|
|
|
if (!useless_type_conversion_p (TREE_TYPE (tgt), TREE_TYPE (src)))
|
|
|
|
{
|
|
|
|
if (!is_gimple_val (src))
|
|
|
|
return false;
|
|
|
|
g = gimple_build_assign (tgt, NOP_EXPR, src);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
g = gimple_build_assign (tgt, src);
|
|
|
|
if (n->range == 16)
|
|
|
|
nop_stats.found_16bit++;
|
|
|
|
else if (n->range == 32)
|
|
|
|
nop_stats.found_32bit++;
|
|
|
|
else
|
|
|
|
{
|
|
|
|
gcc_assert (n->range == 64);
|
|
|
|
nop_stats.found_64bit++;
|
|
|
|
}
|
|
|
|
if (dump_file)
|
|
|
|
{
|
|
|
|
fprintf (dump_file,
|
|
|
|
"%d bit reshuffle in target endianness found at: ",
|
|
|
|
(int) n->range);
|
|
|
|
print_gimple_stmt (dump_file, cur_stmt, 0);
|
|
|
|
}
|
|
|
|
gsi_replace (&gsi, g, true);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
else if (TREE_CODE (src) == BIT_FIELD_REF)
|
|
|
|
src = TREE_OPERAND (src, 0);
|
|
|
|
|
|
|
|
if (n->range == 16)
|
|
|
|
bswap_stats.found_16bit++;
|
|
|
|
else if (n->range == 32)
|
|
|
|
bswap_stats.found_32bit++;
|
|
|
|
else
|
|
|
|
{
|
|
|
|
gcc_assert (n->range == 64);
|
|
|
|
bswap_stats.found_64bit++;
|
|
|
|
}
|
|
|
|
|
|
|
|
tmp = src;
|
|
|
|
|
|
|
|
/* Convert the src expression if necessary. */
|
|
|
|
if (!useless_type_conversion_p (TREE_TYPE (tmp), bswap_type))
|
|
|
|
{
|
|
|
|
gimple *convert_stmt;
|
|
|
|
|
|
|
|
tmp = make_temp_ssa_name (bswap_type, NULL, "bswapsrc");
|
|
|
|
convert_stmt = gimple_build_assign (tmp, NOP_EXPR, src);
|
|
|
|
gsi_insert_before (&gsi, convert_stmt, GSI_SAME_STMT);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Canonical form for 16 bit bswap is a rotate expression. Only 16bit values
|
|
|
|
are considered as rotation of 2N bit values by N bits is generally not
|
|
|
|
equivalent to a bswap. Consider for instance 0x01020304 r>> 16 which
|
|
|
|
gives 0x03040102 while a bswap for that value is 0x04030201. */
|
|
|
|
if (bswap && n->range == 16)
|
|
|
|
{
|
|
|
|
tree count = build_int_cst (NULL, BITS_PER_UNIT);
|
|
|
|
src = fold_build2 (LROTATE_EXPR, bswap_type, tmp, count);
|
|
|
|
bswap_stmt = gimple_build_assign (NULL, src);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
bswap_stmt = gimple_build_call (fndecl, 1, tmp);
|
|
|
|
|
|
|
|
tmp = tgt;
|
|
|
|
|
|
|
|
/* Convert the result if necessary. */
|
|
|
|
if (!useless_type_conversion_p (TREE_TYPE (tgt), bswap_type))
|
|
|
|
{
|
|
|
|
gimple *convert_stmt;
|
|
|
|
|
|
|
|
tmp = make_temp_ssa_name (bswap_type, NULL, "bswapdst");
|
|
|
|
convert_stmt = gimple_build_assign (tgt, NOP_EXPR, tmp);
|
|
|
|
gsi_insert_after (&gsi, convert_stmt, GSI_SAME_STMT);
|
|
|
|
}
|
|
|
|
|
|
|
|
gimple_set_lhs (bswap_stmt, tmp);
|
|
|
|
|
|
|
|
if (dump_file)
|
|
|
|
{
|
|
|
|
fprintf (dump_file, "%d bit bswap implementation found at: ",
|
|
|
|
(int) n->range);
|
|
|
|
print_gimple_stmt (dump_file, cur_stmt, 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
gsi_insert_after (&gsi, bswap_stmt, GSI_SAME_STMT);
|
|
|
|
gsi_remove (&gsi, true);
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Find manual byte swap implementations as well as load in a given
|
|
|
|
endianness. Byte swaps are turned into a bswap builtin invokation
|
|
|
|
while endian loads are converted to bswap builtin invokation or
|
|
|
|
simple load according to the target endianness. */
|
|
|
|
|
|
|
|
unsigned int
|
|
|
|
pass_optimize_bswap::execute (function *fun)
|
|
|
|
{
|
|
|
|
basic_block bb;
|
|
|
|
bool bswap32_p, bswap64_p;
|
|
|
|
bool changed = false;
|
|
|
|
tree bswap32_type = NULL_TREE, bswap64_type = NULL_TREE;
|
|
|
|
|
|
|
|
bswap32_p = (builtin_decl_explicit_p (BUILT_IN_BSWAP32)
|
|
|
|
&& optab_handler (bswap_optab, SImode) != CODE_FOR_nothing);
|
|
|
|
bswap64_p = (builtin_decl_explicit_p (BUILT_IN_BSWAP64)
|
|
|
|
&& (optab_handler (bswap_optab, DImode) != CODE_FOR_nothing
|
|
|
|
|| (bswap32_p && word_mode == SImode)));
|
|
|
|
|
|
|
|
/* Determine the argument type of the builtins. The code later on
|
|
|
|
assumes that the return and argument type are the same. */
|
|
|
|
if (bswap32_p)
|
|
|
|
{
|
|
|
|
tree fndecl = builtin_decl_explicit (BUILT_IN_BSWAP32);
|
|
|
|
bswap32_type = TREE_VALUE (TYPE_ARG_TYPES (TREE_TYPE (fndecl)));
|
|
|
|
}
|
|
|
|
|
|
|
|
if (bswap64_p)
|
|
|
|
{
|
|
|
|
tree fndecl = builtin_decl_explicit (BUILT_IN_BSWAP64);
|
|
|
|
bswap64_type = TREE_VALUE (TYPE_ARG_TYPES (TREE_TYPE (fndecl)));
|
|
|
|
}
|
|
|
|
|
|
|
|
memset (&nop_stats, 0, sizeof (nop_stats));
|
|
|
|
memset (&bswap_stats, 0, sizeof (bswap_stats));
|
|
|
|
calculate_dominance_info (CDI_DOMINATORS);
|
|
|
|
|
|
|
|
FOR_EACH_BB_FN (bb, fun)
|
|
|
|
{
|
|
|
|
gimple_stmt_iterator gsi;
|
|
|
|
|
|
|
|
/* We do a reverse scan for bswap patterns to make sure we get the
|
|
|
|
widest match. As bswap pattern matching doesn't handle previously
|
|
|
|
inserted smaller bswap replacements as sub-patterns, the wider
|
|
|
|
variant wouldn't be detected. */
|
|
|
|
for (gsi = gsi_last_bb (bb); !gsi_end_p (gsi);)
|
|
|
|
{
|
|
|
|
gimple *ins_stmt, *cur_stmt = gsi_stmt (gsi);
|
|
|
|
tree fndecl = NULL_TREE, bswap_type = NULL_TREE, load_type;
|
|
|
|
enum tree_code code;
|
|
|
|
struct symbolic_number n;
|
|
|
|
bool bswap;
|
|
|
|
|
|
|
|
/* This gsi_prev (&gsi) is not part of the for loop because cur_stmt
|
|
|
|
might be moved to a different basic block by bswap_replace and gsi
|
|
|
|
must not points to it if that's the case. Moving the gsi_prev
|
|
|
|
there make sure that gsi points to the statement previous to
|
|
|
|
cur_stmt while still making sure that all statements are
|
|
|
|
considered in this basic block. */
|
|
|
|
gsi_prev (&gsi);
|
|
|
|
|
|
|
|
if (!is_gimple_assign (cur_stmt))
|
|
|
|
continue;
|
|
|
|
|
|
|
|
code = gimple_assign_rhs_code (cur_stmt);
|
|
|
|
switch (code)
|
|
|
|
{
|
|
|
|
case LROTATE_EXPR:
|
|
|
|
case RROTATE_EXPR:
|
|
|
|
if (!tree_fits_uhwi_p (gimple_assign_rhs2 (cur_stmt))
|
|
|
|
|| tree_to_uhwi (gimple_assign_rhs2 (cur_stmt))
|
|
|
|
% BITS_PER_UNIT)
|
|
|
|
continue;
|
|
|
|
/* Fall through. */
|
|
|
|
case BIT_IOR_EXPR:
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
ins_stmt = find_bswap_or_nop (cur_stmt, &n, &bswap);
|
|
|
|
|
|
|
|
if (!ins_stmt)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
switch (n.range)
|
|
|
|
{
|
|
|
|
case 16:
|
|
|
|
/* Already in canonical form, nothing to do. */
|
|
|
|
if (code == LROTATE_EXPR || code == RROTATE_EXPR)
|
|
|
|
continue;
|
|
|
|
load_type = bswap_type = uint16_type_node;
|
|
|
|
break;
|
|
|
|
case 32:
|
|
|
|
load_type = uint32_type_node;
|
|
|
|
if (bswap32_p)
|
|
|
|
{
|
|
|
|
fndecl = builtin_decl_explicit (BUILT_IN_BSWAP32);
|
|
|
|
bswap_type = bswap32_type;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
case 64:
|
|
|
|
load_type = uint64_type_node;
|
|
|
|
if (bswap64_p)
|
|
|
|
{
|
|
|
|
fndecl = builtin_decl_explicit (BUILT_IN_BSWAP64);
|
|
|
|
bswap_type = bswap64_type;
|
|
|
|
}
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (bswap && !fndecl && n.range != 16)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (bswap_replace (cur_stmt, ins_stmt, fndecl, bswap_type, load_type,
|
|
|
|
&n, bswap))
|
|
|
|
changed = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
statistics_counter_event (fun, "16-bit nop implementations found",
|
|
|
|
nop_stats.found_16bit);
|
|
|
|
statistics_counter_event (fun, "32-bit nop implementations found",
|
|
|
|
nop_stats.found_32bit);
|
|
|
|
statistics_counter_event (fun, "64-bit nop implementations found",
|
|
|
|
nop_stats.found_64bit);
|
|
|
|
statistics_counter_event (fun, "16-bit bswap implementations found",
|
|
|
|
bswap_stats.found_16bit);
|
|
|
|
statistics_counter_event (fun, "32-bit bswap implementations found",
|
|
|
|
bswap_stats.found_32bit);
|
|
|
|
statistics_counter_event (fun, "64-bit bswap implementations found",
|
|
|
|
bswap_stats.found_64bit);
|
|
|
|
|
|
|
|
return (changed ? TODO_update_ssa : 0);
|
|
|
|
}
|
|
|
|
|
|
|
|
} // anon namespace
|
|
|
|
|
|
|
|
gimple_opt_pass *
|
|
|
|
make_pass_optimize_bswap (gcc::context *ctxt)
|
|
|
|
{
|
|
|
|
return new pass_optimize_bswap (ctxt);
|
|
|
|
}
|
|
|
|
|
|
|
|
namespace {
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Struct recording one operand for the store, which is either a constant,
|
|
|
|
then VAL represents the constant and all the other fields are zero,
|
|
|
|
or a memory load, then VAL represents the reference, BASE_ADDR is non-NULL
|
|
|
|
and the other fields also reflect the memory load. */
|
|
|
|
|
|
|
|
struct store_operand_info
|
|
|
|
{
|
|
|
|
tree val;
|
|
|
|
tree base_addr;
|
|
|
|
unsigned HOST_WIDE_INT bitsize;
|
|
|
|
unsigned HOST_WIDE_INT bitpos;
|
|
|
|
unsigned HOST_WIDE_INT bitregion_start;
|
|
|
|
unsigned HOST_WIDE_INT bitregion_end;
|
|
|
|
gimple *stmt;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
bool bit_not_p;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
store_operand_info ();
|
|
|
|
};
|
|
|
|
|
|
|
|
store_operand_info::store_operand_info ()
|
|
|
|
: val (NULL_TREE), base_addr (NULL_TREE), bitsize (0), bitpos (0),
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
bitregion_start (0), bitregion_end (0), stmt (NULL), bit_not_p (false)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
{
|
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Struct recording the information about a single store of an immediate
|
|
|
|
to memory. These are created in the first phase and coalesced into
|
|
|
|
merged_store_group objects in the second phase. */
|
|
|
|
|
|
|
|
struct store_immediate_info
|
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT bitsize;
|
|
|
|
unsigned HOST_WIDE_INT bitpos;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT bitregion_start;
|
|
|
|
/* This is one past the last bit of the bit region. */
|
|
|
|
unsigned HOST_WIDE_INT bitregion_end;
|
2016-10-28 16:18:50 +02:00
|
|
|
gimple *stmt;
|
|
|
|
unsigned int order;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* INTEGER_CST for constant stores, MEM_REF for memory copy or
|
|
|
|
BIT_*_EXPR for logical bitwise operation. */
|
|
|
|
enum tree_code rhs_code;
|
2017-11-10 15:43:13 +01:00
|
|
|
/* True if BIT_{AND,IOR,XOR}_EXPR result is inverted before storing. */
|
2017-11-09 20:24:59 +01:00
|
|
|
bool bit_not_p;
|
2017-11-10 15:43:13 +01:00
|
|
|
/* True if ops have been swapped and thus ops[1] represents
|
|
|
|
rhs1 of BIT_{AND,IOR,XOR}_EXPR and ops[0] represents rhs2. */
|
|
|
|
bool ops_swapped_p;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Operands. For BIT_*_EXPR rhs_code both operands are used, otherwise
|
|
|
|
just the first one. */
|
|
|
|
store_operand_info ops[2];
|
2016-11-02 14:05:07 +01:00
|
|
|
store_immediate_info (unsigned HOST_WIDE_INT, unsigned HOST_WIDE_INT,
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT, unsigned HOST_WIDE_INT,
|
2017-11-09 20:24:59 +01:00
|
|
|
gimple *, unsigned int, enum tree_code, bool,
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
const store_operand_info &,
|
|
|
|
const store_operand_info &);
|
2016-10-28 16:18:50 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
store_immediate_info::store_immediate_info (unsigned HOST_WIDE_INT bs,
|
2016-11-02 14:05:07 +01:00
|
|
|
unsigned HOST_WIDE_INT bp,
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT brs,
|
|
|
|
unsigned HOST_WIDE_INT bre,
|
2016-11-02 14:05:07 +01:00
|
|
|
gimple *st,
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
unsigned int ord,
|
|
|
|
enum tree_code rhscode,
|
2017-11-09 20:24:59 +01:00
|
|
|
bool bitnotp,
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
const store_operand_info &op0r,
|
|
|
|
const store_operand_info &op1r)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
: bitsize (bs), bitpos (bp), bitregion_start (brs), bitregion_end (bre),
|
2017-11-10 15:43:13 +01:00
|
|
|
stmt (st), order (ord), rhs_code (rhscode), bit_not_p (bitnotp),
|
|
|
|
ops_swapped_p (false)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
#if __cplusplus >= 201103L
|
|
|
|
, ops { op0r, op1r }
|
|
|
|
{
|
|
|
|
}
|
|
|
|
#else
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
ops[0] = op0r;
|
|
|
|
ops[1] = op1r;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
#endif
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
/* Struct representing a group of stores to contiguous memory locations.
|
|
|
|
These are produced by the second phase (coalescing) and consumed in the
|
|
|
|
third phase that outputs the widened stores. */
|
|
|
|
|
|
|
|
struct merged_store_group
|
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT start;
|
|
|
|
unsigned HOST_WIDE_INT width;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT bitregion_start;
|
|
|
|
unsigned HOST_WIDE_INT bitregion_end;
|
|
|
|
/* The size of the allocated memory for val and mask. */
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned HOST_WIDE_INT buf_size;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT align_base;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
unsigned HOST_WIDE_INT load_align_base[2];
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
unsigned int align;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
unsigned int load_align[2];
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned int first_order;
|
|
|
|
unsigned int last_order;
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
auto_vec<store_immediate_info *> stores;
|
2016-10-28 16:18:50 +02:00
|
|
|
/* We record the first and last original statements in the sequence because
|
|
|
|
we'll need their vuse/vdef and replacement position. It's easier to keep
|
|
|
|
track of them separately as 'stores' is reordered by apply_stores. */
|
|
|
|
gimple *last_stmt;
|
|
|
|
gimple *first_stmt;
|
|
|
|
unsigned char *val;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned char *mask;
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
merged_store_group (store_immediate_info *);
|
|
|
|
~merged_store_group ();
|
|
|
|
void merge_into (store_immediate_info *);
|
|
|
|
void merge_overlapping (store_immediate_info *);
|
|
|
|
bool apply_stores ();
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
private:
|
|
|
|
void do_merge (store_immediate_info *);
|
2016-10-28 16:18:50 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
/* Debug helper. Dump LEN elements of byte array PTR to FD in hex. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
dump_char_array (FILE *fd, unsigned char *ptr, unsigned int len)
|
|
|
|
{
|
|
|
|
if (!fd)
|
|
|
|
return;
|
|
|
|
|
|
|
|
for (unsigned int i = 0; i < len; i++)
|
|
|
|
fprintf (fd, "%x ", ptr[i]);
|
|
|
|
fprintf (fd, "\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Shift left the bytes in PTR of SZ elements by AMNT bits, carrying over the
|
|
|
|
bits between adjacent elements. AMNT should be within
|
|
|
|
[0, BITS_PER_UNIT).
|
|
|
|
Example, AMNT = 2:
|
|
|
|
00011111|11100000 << 2 = 01111111|10000000
|
|
|
|
PTR[1] | PTR[0] PTR[1] | PTR[0]. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
shift_bytes_in_array (unsigned char *ptr, unsigned int sz, unsigned int amnt)
|
|
|
|
{
|
|
|
|
if (amnt == 0)
|
|
|
|
return;
|
|
|
|
|
|
|
|
unsigned char carry_over = 0U;
|
2016-11-22 11:15:43 +01:00
|
|
|
unsigned char carry_mask = (~0U) << (unsigned char) (BITS_PER_UNIT - amnt);
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned char clear_mask = (~0U) << amnt;
|
|
|
|
|
|
|
|
for (unsigned int i = 0; i < sz; i++)
|
|
|
|
{
|
|
|
|
unsigned prev_carry_over = carry_over;
|
2016-11-22 11:15:43 +01:00
|
|
|
carry_over = (ptr[i] & carry_mask) >> (BITS_PER_UNIT - amnt);
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
ptr[i] <<= amnt;
|
|
|
|
if (i != 0)
|
|
|
|
{
|
|
|
|
ptr[i] &= clear_mask;
|
|
|
|
ptr[i] |= prev_carry_over;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Like shift_bytes_in_array but for big-endian.
|
|
|
|
Shift right the bytes in PTR of SZ elements by AMNT bits, carrying over the
|
|
|
|
bits between adjacent elements. AMNT should be within
|
|
|
|
[0, BITS_PER_UNIT).
|
|
|
|
Example, AMNT = 2:
|
|
|
|
00011111|11100000 >> 2 = 00000111|11111000
|
|
|
|
PTR[0] | PTR[1] PTR[0] | PTR[1]. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
shift_bytes_in_array_right (unsigned char *ptr, unsigned int sz,
|
|
|
|
unsigned int amnt)
|
|
|
|
{
|
|
|
|
if (amnt == 0)
|
|
|
|
return;
|
|
|
|
|
|
|
|
unsigned char carry_over = 0U;
|
|
|
|
unsigned char carry_mask = ~(~0U << amnt);
|
|
|
|
|
|
|
|
for (unsigned int i = 0; i < sz; i++)
|
|
|
|
{
|
|
|
|
unsigned prev_carry_over = carry_over;
|
2016-11-22 11:15:43 +01:00
|
|
|
carry_over = ptr[i] & carry_mask;
|
2016-10-28 16:18:50 +02:00
|
|
|
|
2017-02-28 23:39:11 +01:00
|
|
|
carry_over <<= (unsigned char) BITS_PER_UNIT - amnt;
|
|
|
|
ptr[i] >>= amnt;
|
|
|
|
ptr[i] |= prev_carry_over;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Clear out LEN bits starting from bit START in the byte array
|
|
|
|
PTR. This clears the bits to the *right* from START.
|
|
|
|
START must be within [0, BITS_PER_UNIT) and counts starting from
|
|
|
|
the least significant bit. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
clear_bit_region_be (unsigned char *ptr, unsigned int start,
|
|
|
|
unsigned int len)
|
|
|
|
{
|
|
|
|
if (len == 0)
|
|
|
|
return;
|
|
|
|
/* Clear len bits to the right of start. */
|
|
|
|
else if (len <= start + 1)
|
|
|
|
{
|
|
|
|
unsigned char mask = (~(~0U << len));
|
|
|
|
mask = mask << (start + 1U - len);
|
|
|
|
ptr[0] &= ~mask;
|
|
|
|
}
|
|
|
|
else if (start != BITS_PER_UNIT - 1)
|
|
|
|
{
|
|
|
|
clear_bit_region_be (ptr, start, (start % BITS_PER_UNIT) + 1);
|
|
|
|
clear_bit_region_be (ptr + 1, BITS_PER_UNIT - 1,
|
|
|
|
len - (start % BITS_PER_UNIT) - 1);
|
|
|
|
}
|
|
|
|
else if (start == BITS_PER_UNIT - 1
|
|
|
|
&& len > BITS_PER_UNIT)
|
|
|
|
{
|
|
|
|
unsigned int nbytes = len / BITS_PER_UNIT;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
memset (ptr, 0, nbytes);
|
2016-10-28 16:18:50 +02:00
|
|
|
if (len % BITS_PER_UNIT != 0)
|
|
|
|
clear_bit_region_be (ptr + nbytes, BITS_PER_UNIT - 1,
|
|
|
|
len % BITS_PER_UNIT);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
gcc_unreachable ();
|
|
|
|
}
|
|
|
|
|
|
|
|
/* In the byte array PTR clear the bit region starting at bit
|
|
|
|
START and is LEN bits wide.
|
|
|
|
For regions spanning multiple bytes do this recursively until we reach
|
|
|
|
zero LEN or a region contained within a single byte. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
clear_bit_region (unsigned char *ptr, unsigned int start,
|
|
|
|
unsigned int len)
|
|
|
|
{
|
|
|
|
/* Degenerate base case. */
|
|
|
|
if (len == 0)
|
|
|
|
return;
|
|
|
|
else if (start >= BITS_PER_UNIT)
|
|
|
|
clear_bit_region (ptr + 1, start - BITS_PER_UNIT, len);
|
|
|
|
/* Second base case. */
|
|
|
|
else if ((start + len) <= BITS_PER_UNIT)
|
|
|
|
{
|
2016-11-22 11:15:43 +01:00
|
|
|
unsigned char mask = (~0U) << (unsigned char) (BITS_PER_UNIT - len);
|
2016-10-28 16:18:50 +02:00
|
|
|
mask >>= BITS_PER_UNIT - (start + len);
|
|
|
|
|
|
|
|
ptr[0] &= ~mask;
|
|
|
|
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
/* Clear most significant bits in a byte and proceed with the next byte. */
|
|
|
|
else if (start != 0)
|
|
|
|
{
|
|
|
|
clear_bit_region (ptr, start, BITS_PER_UNIT - start);
|
2016-11-08 13:31:31 +01:00
|
|
|
clear_bit_region (ptr + 1, 0, len - (BITS_PER_UNIT - start));
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
/* Whole bytes need to be cleared. */
|
|
|
|
else if (start == 0 && len > BITS_PER_UNIT)
|
|
|
|
{
|
|
|
|
unsigned int nbytes = len / BITS_PER_UNIT;
|
2017-07-14 11:11:20 +02:00
|
|
|
/* We could recurse on each byte but we clear whole bytes, so a simple
|
|
|
|
memset will do. */
|
2016-11-22 11:15:43 +01:00
|
|
|
memset (ptr, '\0', nbytes);
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Clear the remaining sub-byte region if there is one. */
|
|
|
|
if (len % BITS_PER_UNIT != 0)
|
|
|
|
clear_bit_region (ptr + nbytes, 0, len % BITS_PER_UNIT);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
gcc_unreachable ();
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Write BITLEN bits of EXPR to the byte array PTR at
|
|
|
|
bit position BITPOS. PTR should contain TOTAL_BYTES elements.
|
|
|
|
Return true if the operation succeeded. */
|
|
|
|
|
|
|
|
static bool
|
|
|
|
encode_tree_to_bitpos (tree expr, unsigned char *ptr, int bitlen, int bitpos,
|
2016-11-22 11:15:43 +01:00
|
|
|
unsigned int total_bytes)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
unsigned int first_byte = bitpos / BITS_PER_UNIT;
|
|
|
|
tree tmp_int = expr;
|
2017-02-28 23:39:11 +01:00
|
|
|
bool sub_byte_op_p = ((bitlen % BITS_PER_UNIT)
|
|
|
|
|| (bitpos % BITS_PER_UNIT)
|
Make more use of int_mode_for_size
This patch converts more places that could use int_mode_for_size instead
of mode_for_size. This is in preparation for an upcoming patch that
makes mode_for_size itself return an opt_mode.
require () seems like the right choice in expand_builtin_powi
because we have got past the point of backing out. We go on to do:
op1 = expand_expr (arg1, NULL_RTX, mode2, EXPAND_NORMAL);
if (GET_MODE (op1) != mode2)
op1 = convert_to_mode (mode2, op1, 0);
which would be invalid for (and have failed for) BLKmode.
In get_builtin_sync_mode and expand_ifn_atomic_compare_exchange,
the possible bitsizes are {8, 16, 32, 64, 128}, all of which give
target-independent integer modes (up to TImode). The comment above
the call in get_builtin_sync_mode makes clear that an integer mode
must be found.
We can use require () in expand_builtin_atomic_clear and
expand_builtin_atomic_test_and_set because there's always an integer
mode for the boolean type. The same goes for the POINTER_SIZE request
in layout_type. Similarly we can use require () in combine_instructions
and gen_lowpart_common because there's always an integer mode for
HOST_BITS_PER_WIDE_INT (DImode when BITS_PER_UNIT == 8), and
HOST_BITS_PER_DOUBLE_INT (TImode).
The calls in aarch64_function_value, arm_function_value,
aapcs_allocate_return_reg and mips_function_value_1 are handling
cases in which a big-endian target passes or returns values at
the most significant end of a register. In each case the ABI
constrains the size to a small amount and does not handle
non-power-of-2 sizes wider than a word.
The calls in c6x_expand_movmem, i386.c:emit_memset,
lm32_block_move_inline, microblaze_block_move_straight and
mips_block_move_straight are dealing with expansions of
block memory operations using register-wise operations,
and those registers must have non-BLK mode.
The reason for using require () in ix86_expand_sse_cmp,
mips_expand_ins_as_unaligned_store, spu.c:adjust_operand and
spu_emit_branch_and_set is that we go on to emit non-call
instructions that use registers of that mode, which wouldn't
be valid for BLKmode.
2017-09-05 Richard Sandiford <richard.sandiford@linaro.org>
gcc/
* builtins.c (expand_builtin_powi): Use int_mode_for_size.
(get_builtin_sync_mode): Likewise.
(expand_ifn_atomic_compare_exchange): Likewise.
(expand_builtin_atomic_clear): Likewise.
(expand_builtin_atomic_test_and_set): Likewise.
(fold_builtin_atomic_always_lock_free): Likewise.
* calls.c (compute_argument_addresses): Likewise.
(emit_library_call_value_1): Likewise.
(store_one_arg): Likewise.
* combine.c (combine_instructions): Likewise.
* config/aarch64/aarch64.c (aarch64_function_value): Likewise.
* config/arm/arm.c (arm_function_value): Likewise.
(aapcs_allocate_return_reg): Likewise.
* config/c6x/c6x.c (c6x_expand_movmem): Likewise.
* config/i386/i386.c (construct_container): Likewise.
(ix86_gimplify_va_arg): Likewise.
(ix86_expand_sse_cmp): Likewise.
(emit_memmov): Likewise.
(emit_memset): Likewise.
(expand_small_movmem_or_setmem): Likewise.
(ix86_expand_pextr): Likewise.
(ix86_expand_pinsr): Likewise.
* config/lm32/lm32.c (lm32_block_move_inline): Likewise.
* config/microblaze/microblaze.c (microblaze_block_move_straight):
Likewise.
* config/mips/mips.c (mips_function_value_1) Likewise.
(mips_block_move_straight): Likewise.
(mips_expand_ins_as_unaligned_store): Likewise.
* config/powerpcspe/powerpcspe.c
(rs6000_darwin64_record_arg_advance_flush): Likewise.
(rs6000_darwin64_record_arg_flush): Likewise.
* config/rs6000/rs6000.c
(rs6000_darwin64_record_arg_advance_flush): Likewise.
(rs6000_darwin64_record_arg_flush): Likewise.
* config/sparc/sparc.c (sparc_function_arg_1): Likewise.
(sparc_function_value_1): Likewise.
* config/spu/spu.c (adjust_operand): Likewise.
(spu_emit_branch_or_set): Likewise.
(arith_immediate_p): Likewise.
* emit-rtl.c (gen_lowpart_common): Likewise.
* expr.c (expand_expr_real_1): Likewise.
* function.c (assign_parm_setup_block): Likewise.
* gimple-ssa-store-merging.c (encode_tree_to_bitpos): Likewise.
* reload1.c (alter_reg): Likewise.
* stor-layout.c (mode_for_vector): Likewise.
(layout_type): Likewise.
gcc/ada/
* gcc-interface/utils2.c (build_load_modify_store):
Use int_mode_for_size.
From-SVN: r251726
2017-09-05 21:57:01 +02:00
|
|
|
|| !int_mode_for_size (bitlen, 0).exists ());
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
if (!sub_byte_op_p)
|
2017-10-06 11:39:07 +02:00
|
|
|
return native_encode_expr (tmp_int, ptr + first_byte, total_bytes) != 0;
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
/* LITTLE-ENDIAN
|
|
|
|
We are writing a non byte-sized quantity or at a position that is not
|
|
|
|
at a byte boundary.
|
|
|
|
|--------|--------|--------| ptr + first_byte
|
|
|
|
^ ^
|
|
|
|
xxx xxxxxxxx xxx< bp>
|
|
|
|
|______EXPR____|
|
|
|
|
|
2016-11-22 11:15:43 +01:00
|
|
|
First native_encode_expr EXPR into a temporary buffer and shift each
|
2016-10-28 16:18:50 +02:00
|
|
|
byte in the buffer by 'bp' (carrying the bits over as necessary).
|
|
|
|
|00000000|00xxxxxx|xxxxxxxx| << bp = |000xxxxx|xxxxxxxx|xxx00000|
|
|
|
|
<------bitlen---->< bp>
|
|
|
|
Then we clear the destination bits:
|
|
|
|
|---00000|00000000|000-----| ptr + first_byte
|
|
|
|
<-------bitlen--->< bp>
|
|
|
|
|
|
|
|
Finally we ORR the bytes of the shifted EXPR into the cleared region:
|
|
|
|
|---xxxxx||xxxxxxxx||xxx-----| ptr + first_byte.
|
|
|
|
|
|
|
|
BIG-ENDIAN
|
|
|
|
We are writing a non byte-sized quantity or at a position that is not
|
|
|
|
at a byte boundary.
|
|
|
|
ptr + first_byte |--------|--------|--------|
|
|
|
|
^ ^
|
|
|
|
<bp >xxx xxxxxxxx xxx
|
|
|
|
|_____EXPR_____|
|
|
|
|
|
2016-11-22 11:15:43 +01:00
|
|
|
First native_encode_expr EXPR into a temporary buffer and shift each
|
2016-10-28 16:18:50 +02:00
|
|
|
byte in the buffer to the right by (carrying the bits over as necessary).
|
|
|
|
We shift by as much as needed to align the most significant bit of EXPR
|
|
|
|
with bitpos:
|
|
|
|
|00xxxxxx|xxxxxxxx| >> 3 = |00000xxx|xxxxxxxx|xxxxx000|
|
|
|
|
<---bitlen----> <bp ><-----bitlen----->
|
|
|
|
Then we clear the destination bits:
|
|
|
|
ptr + first_byte |-----000||00000000||00000---|
|
|
|
|
<bp ><-------bitlen----->
|
|
|
|
|
|
|
|
Finally we ORR the bytes of the shifted EXPR into the cleared region:
|
|
|
|
ptr + first_byte |---xxxxx||xxxxxxxx||xxx-----|.
|
|
|
|
The awkwardness comes from the fact that bitpos is counted from the
|
|
|
|
most significant bit of a byte. */
|
|
|
|
|
2017-11-01 12:49:34 +01:00
|
|
|
/* We must be dealing with fixed-size data at this point, since the
|
|
|
|
total size is also fixed. */
|
|
|
|
fixed_size_mode mode = as_a <fixed_size_mode> (TYPE_MODE (TREE_TYPE (expr)));
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Allocate an extra byte so that we have space to shift into. */
|
2017-11-01 12:49:34 +01:00
|
|
|
unsigned int byte_size = GET_MODE_SIZE (mode) + 1;
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned char *tmpbuf = XALLOCAVEC (unsigned char, byte_size);
|
2016-11-22 11:15:43 +01:00
|
|
|
memset (tmpbuf, '\0', byte_size);
|
2016-10-28 16:18:50 +02:00
|
|
|
/* The store detection code should only have allowed constants that are
|
|
|
|
accepted by native_encode_expr. */
|
2017-10-06 11:39:07 +02:00
|
|
|
if (native_encode_expr (expr, tmpbuf, byte_size - 1) == 0)
|
2016-10-28 16:18:50 +02:00
|
|
|
gcc_unreachable ();
|
|
|
|
|
|
|
|
/* The native_encode_expr machinery uses TYPE_MODE to determine how many
|
|
|
|
bytes to write. This means it can write more than
|
|
|
|
ROUND_UP (bitlen, BITS_PER_UNIT) / BITS_PER_UNIT bytes (for example
|
|
|
|
write 8 bytes for a bitlen of 40). Skip the bytes that are not within
|
|
|
|
bitlen and zero out the bits that are not relevant as well (that may
|
|
|
|
contain a sign bit due to sign-extension). */
|
|
|
|
unsigned int padding
|
|
|
|
= byte_size - ROUND_UP (bitlen, BITS_PER_UNIT) / BITS_PER_UNIT - 1;
|
2017-02-28 23:39:11 +01:00
|
|
|
/* On big-endian the padding is at the 'front' so just skip the initial
|
|
|
|
bytes. */
|
|
|
|
if (BYTES_BIG_ENDIAN)
|
|
|
|
tmpbuf += padding;
|
|
|
|
|
|
|
|
byte_size -= padding;
|
|
|
|
|
|
|
|
if (bitlen % BITS_PER_UNIT != 0)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
2016-11-02 10:28:35 +01:00
|
|
|
if (BYTES_BIG_ENDIAN)
|
2017-02-28 23:39:11 +01:00
|
|
|
clear_bit_region_be (tmpbuf, BITS_PER_UNIT - 1,
|
|
|
|
BITS_PER_UNIT - (bitlen % BITS_PER_UNIT));
|
|
|
|
else
|
|
|
|
clear_bit_region (tmpbuf, bitlen,
|
|
|
|
byte_size * BITS_PER_UNIT - bitlen);
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
2017-02-28 23:39:11 +01:00
|
|
|
/* Left shifting relies on the last byte being clear if bitlen is
|
|
|
|
a multiple of BITS_PER_UNIT, which might not be clear if
|
|
|
|
there are padding bytes. */
|
|
|
|
else if (!BYTES_BIG_ENDIAN)
|
|
|
|
tmpbuf[byte_size - 1] = '\0';
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
/* Clear the bit region in PTR where the bits from TMPBUF will be
|
2016-11-22 11:15:43 +01:00
|
|
|
inserted into. */
|
2016-10-28 16:18:50 +02:00
|
|
|
if (BYTES_BIG_ENDIAN)
|
|
|
|
clear_bit_region_be (ptr + first_byte,
|
|
|
|
BITS_PER_UNIT - 1 - (bitpos % BITS_PER_UNIT), bitlen);
|
|
|
|
else
|
|
|
|
clear_bit_region (ptr + first_byte, bitpos % BITS_PER_UNIT, bitlen);
|
|
|
|
|
|
|
|
int shift_amnt;
|
|
|
|
int bitlen_mod = bitlen % BITS_PER_UNIT;
|
|
|
|
int bitpos_mod = bitpos % BITS_PER_UNIT;
|
|
|
|
|
|
|
|
bool skip_byte = false;
|
|
|
|
if (BYTES_BIG_ENDIAN)
|
|
|
|
{
|
|
|
|
/* BITPOS and BITLEN are exactly aligned and no shifting
|
|
|
|
is necessary. */
|
|
|
|
if (bitpos_mod + bitlen_mod == BITS_PER_UNIT
|
|
|
|
|| (bitpos_mod == 0 && bitlen_mod == 0))
|
|
|
|
shift_amnt = 0;
|
|
|
|
/* |. . . . . . . .|
|
|
|
|
<bp > <blen >.
|
|
|
|
We always shift right for BYTES_BIG_ENDIAN so shift the beginning
|
|
|
|
of the value until it aligns with 'bp' in the next byte over. */
|
|
|
|
else if (bitpos_mod + bitlen_mod < BITS_PER_UNIT)
|
|
|
|
{
|
|
|
|
shift_amnt = bitlen_mod + bitpos_mod;
|
|
|
|
skip_byte = bitlen_mod != 0;
|
|
|
|
}
|
|
|
|
/* |. . . . . . . .|
|
|
|
|
<----bp--->
|
|
|
|
<---blen---->.
|
|
|
|
Shift the value right within the same byte so it aligns with 'bp'. */
|
|
|
|
else
|
|
|
|
shift_amnt = bitlen_mod + bitpos_mod - BITS_PER_UNIT;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
shift_amnt = bitpos % BITS_PER_UNIT;
|
|
|
|
|
|
|
|
/* Create the shifted version of EXPR. */
|
|
|
|
if (!BYTES_BIG_ENDIAN)
|
2016-11-22 11:15:43 +01:00
|
|
|
{
|
|
|
|
shift_bytes_in_array (tmpbuf, byte_size, shift_amnt);
|
|
|
|
if (shift_amnt == 0)
|
|
|
|
byte_size--;
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
else
|
|
|
|
{
|
|
|
|
gcc_assert (BYTES_BIG_ENDIAN);
|
|
|
|
shift_bytes_in_array_right (tmpbuf, byte_size, shift_amnt);
|
|
|
|
/* If shifting right forced us to move into the next byte skip the now
|
|
|
|
empty byte. */
|
|
|
|
if (skip_byte)
|
|
|
|
{
|
|
|
|
tmpbuf++;
|
|
|
|
byte_size--;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Insert the bits from TMPBUF. */
|
|
|
|
for (unsigned int i = 0; i < byte_size; i++)
|
|
|
|
ptr[first_byte + i] |= tmpbuf[i];
|
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Sorting function for store_immediate_info objects.
|
|
|
|
Sorts them by bitposition. */
|
|
|
|
|
|
|
|
static int
|
|
|
|
sort_by_bitpos (const void *x, const void *y)
|
|
|
|
{
|
|
|
|
store_immediate_info *const *tmp = (store_immediate_info * const *) x;
|
|
|
|
store_immediate_info *const *tmp2 = (store_immediate_info * const *) y;
|
|
|
|
|
2017-07-20 17:13:56 +02:00
|
|
|
if ((*tmp)->bitpos < (*tmp2)->bitpos)
|
2016-10-28 16:18:50 +02:00
|
|
|
return -1;
|
|
|
|
else if ((*tmp)->bitpos > (*tmp2)->bitpos)
|
|
|
|
return 1;
|
2017-07-20 17:13:56 +02:00
|
|
|
else
|
2017-09-13 15:49:50 +02:00
|
|
|
/* If they are the same let's use the order which is guaranteed to
|
|
|
|
be different. */
|
|
|
|
return (*tmp)->order - (*tmp2)->order;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Sorting function for store_immediate_info objects.
|
|
|
|
Sorts them by the order field. */
|
|
|
|
|
|
|
|
static int
|
|
|
|
sort_by_order (const void *x, const void *y)
|
|
|
|
{
|
|
|
|
store_immediate_info *const *tmp = (store_immediate_info * const *) x;
|
|
|
|
store_immediate_info *const *tmp2 = (store_immediate_info * const *) y;
|
|
|
|
|
|
|
|
if ((*tmp)->order < (*tmp2)->order)
|
|
|
|
return -1;
|
|
|
|
else if ((*tmp)->order > (*tmp2)->order)
|
|
|
|
return 1;
|
|
|
|
|
|
|
|
gcc_unreachable ();
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Initialize a merged_store_group object from a store_immediate_info
|
|
|
|
object. */
|
|
|
|
|
|
|
|
merged_store_group::merged_store_group (store_immediate_info *info)
|
|
|
|
{
|
|
|
|
start = info->bitpos;
|
|
|
|
width = info->bitsize;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
bitregion_start = info->bitregion_start;
|
|
|
|
bitregion_end = info->bitregion_end;
|
2016-10-28 16:18:50 +02:00
|
|
|
/* VAL has memory allocated for it in apply_stores once the group
|
|
|
|
width has been finalized. */
|
|
|
|
val = NULL;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
mask = NULL;
|
|
|
|
unsigned HOST_WIDE_INT align_bitpos = 0;
|
|
|
|
get_object_alignment_1 (gimple_assign_lhs (info->stmt),
|
|
|
|
&align, &align_bitpos);
|
|
|
|
align_base = start - align_bitpos;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
for (int i = 0; i < 2; ++i)
|
|
|
|
{
|
|
|
|
store_operand_info &op = info->ops[i];
|
|
|
|
if (op.base_addr == NULL_TREE)
|
|
|
|
{
|
|
|
|
load_align[i] = 0;
|
|
|
|
load_align_base[i] = 0;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
get_object_alignment_1 (op.val, &load_align[i], &align_bitpos);
|
|
|
|
load_align_base[i] = op.bitpos - align_bitpos;
|
|
|
|
}
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
stores.create (1);
|
|
|
|
stores.safe_push (info);
|
|
|
|
last_stmt = info->stmt;
|
|
|
|
last_order = info->order;
|
|
|
|
first_stmt = last_stmt;
|
|
|
|
first_order = last_order;
|
|
|
|
buf_size = 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
merged_store_group::~merged_store_group ()
|
|
|
|
{
|
|
|
|
if (val)
|
|
|
|
XDELETEVEC (val);
|
|
|
|
}
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
/* Helper method for merge_into and merge_overlapping to do
|
|
|
|
the common part. */
|
2016-10-28 16:18:50 +02:00
|
|
|
void
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
merged_store_group::do_merge (store_immediate_info *info)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
bitregion_start = MIN (bitregion_start, info->bitregion_start);
|
|
|
|
bitregion_end = MAX (bitregion_end, info->bitregion_end);
|
|
|
|
|
|
|
|
unsigned int this_align;
|
|
|
|
unsigned HOST_WIDE_INT align_bitpos = 0;
|
|
|
|
get_object_alignment_1 (gimple_assign_lhs (info->stmt),
|
|
|
|
&this_align, &align_bitpos);
|
|
|
|
if (this_align > align)
|
|
|
|
{
|
|
|
|
align = this_align;
|
|
|
|
align_base = info->bitpos - align_bitpos;
|
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
for (int i = 0; i < 2; ++i)
|
|
|
|
{
|
|
|
|
store_operand_info &op = info->ops[i];
|
|
|
|
if (!op.base_addr)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
get_object_alignment_1 (op.val, &this_align, &align_bitpos);
|
|
|
|
if (this_align > load_align[i])
|
|
|
|
{
|
|
|
|
load_align[i] = this_align;
|
|
|
|
load_align_base[i] = op.bitpos - align_bitpos;
|
|
|
|
}
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
gimple *stmt = info->stmt;
|
|
|
|
stores.safe_push (info);
|
|
|
|
if (info->order > last_order)
|
|
|
|
{
|
|
|
|
last_order = info->order;
|
|
|
|
last_stmt = stmt;
|
|
|
|
}
|
|
|
|
else if (info->order < first_order)
|
|
|
|
{
|
|
|
|
first_order = info->order;
|
|
|
|
first_stmt = stmt;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
/* Merge a store recorded by INFO into this merged store.
|
|
|
|
The store is not overlapping with the existing recorded
|
|
|
|
stores. */
|
|
|
|
|
|
|
|
void
|
|
|
|
merged_store_group::merge_into (store_immediate_info *info)
|
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT wid = info->bitsize;
|
|
|
|
/* Make sure we're inserting in the position we think we're inserting. */
|
|
|
|
gcc_assert (info->bitpos >= start + width
|
|
|
|
&& info->bitregion_start <= bitregion_end);
|
|
|
|
|
|
|
|
width += wid;
|
|
|
|
do_merge (info);
|
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Merge a store described by INFO into this merged store.
|
|
|
|
INFO overlaps in some way with the current store (i.e. it's not contiguous
|
|
|
|
which is handled by merged_store_group::merge_into). */
|
|
|
|
|
|
|
|
void
|
|
|
|
merged_store_group::merge_overlapping (store_immediate_info *info)
|
|
|
|
{
|
|
|
|
/* If the store extends the size of the group, extend the width. */
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (info->bitpos + info->bitsize > start + width)
|
2016-10-28 16:18:50 +02:00
|
|
|
width += info->bitpos + info->bitsize - (start + width);
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
do_merge (info);
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Go through all the recorded stores in this group in program order and
|
|
|
|
apply their values to the VAL byte array to create the final merged
|
|
|
|
value. Return true if the operation succeeded. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
merged_store_group::apply_stores ()
|
|
|
|
{
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
/* Make sure we have more than one store in the group, otherwise we cannot
|
|
|
|
merge anything. */
|
|
|
|
if (bitregion_start % BITS_PER_UNIT != 0
|
|
|
|
|| bitregion_end % BITS_PER_UNIT != 0
|
2016-10-28 16:18:50 +02:00
|
|
|
|| stores.length () == 1)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
stores.qsort (sort_by_order);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
store_immediate_info *info;
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned int i;
|
|
|
|
/* Create a buffer of a size that is 2 times the number of bytes we're
|
|
|
|
storing. That way native_encode_expr can write power-of-2-sized
|
|
|
|
chunks without overrunning. */
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
buf_size = 2 * ((bitregion_end - bitregion_start) / BITS_PER_UNIT);
|
|
|
|
val = XNEWVEC (unsigned char, 2 * buf_size);
|
|
|
|
mask = val + buf_size;
|
|
|
|
memset (val, 0, buf_size);
|
|
|
|
memset (mask, ~0U, buf_size);
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
FOR_EACH_VEC_ELT (stores, i, info)
|
|
|
|
{
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned int pos_in_buffer = info->bitpos - bitregion_start;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
tree cst = NULL_TREE;
|
|
|
|
if (info->ops[0].val && info->ops[0].base_addr == NULL_TREE)
|
|
|
|
cst = info->ops[0].val;
|
|
|
|
else if (info->ops[1].val && info->ops[1].base_addr == NULL_TREE)
|
|
|
|
cst = info->ops[1].val;
|
|
|
|
bool ret = true;
|
|
|
|
if (cst)
|
|
|
|
ret = encode_tree_to_bitpos (cst, val, info->bitsize,
|
|
|
|
pos_in_buffer, buf_size);
|
|
|
|
if (cst && dump_file && (dump_flags & TDF_DETAILS))
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
if (ret)
|
|
|
|
{
|
|
|
|
fprintf (dump_file, "After writing ");
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
print_generic_expr (dump_file, cst, 0);
|
2016-10-28 16:18:50 +02:00
|
|
|
fprintf (dump_file, " of size " HOST_WIDE_INT_PRINT_DEC
|
|
|
|
" at position %d the merged region contains:\n",
|
|
|
|
info->bitsize, pos_in_buffer);
|
|
|
|
dump_char_array (dump_file, val, buf_size);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
fprintf (dump_file, "Failed to merge stores\n");
|
|
|
|
}
|
|
|
|
if (!ret)
|
|
|
|
return false;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned char *m = mask + (pos_in_buffer / BITS_PER_UNIT);
|
|
|
|
if (BYTES_BIG_ENDIAN)
|
2017-10-30 17:20:24 +01:00
|
|
|
clear_bit_region_be (m, (BITS_PER_UNIT - 1
|
|
|
|
- (pos_in_buffer % BITS_PER_UNIT)),
|
|
|
|
info->bitsize);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
else
|
|
|
|
clear_bit_region (m, pos_in_buffer % BITS_PER_UNIT, info->bitsize);
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Structure describing the store chain. */
|
|
|
|
|
|
|
|
struct imm_store_chain_info
|
|
|
|
{
|
2017-03-17 00:31:01 +01:00
|
|
|
/* Doubly-linked list that imposes an order on chain processing.
|
|
|
|
PNXP (prev's next pointer) points to the head of a list, or to
|
|
|
|
the next field in the previous chain in the list.
|
|
|
|
See pass_store_merging::m_stores_head for more rationale. */
|
|
|
|
imm_store_chain_info *next, **pnxp;
|
2016-11-02 14:05:07 +01:00
|
|
|
tree base_addr;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
auto_vec<store_immediate_info *> m_store_info;
|
2016-10-28 16:18:50 +02:00
|
|
|
auto_vec<merged_store_group *> m_merged_store_groups;
|
|
|
|
|
2017-03-17 00:31:01 +01:00
|
|
|
imm_store_chain_info (imm_store_chain_info *&inspt, tree b_a)
|
|
|
|
: next (inspt), pnxp (&inspt), base_addr (b_a)
|
|
|
|
{
|
|
|
|
inspt = this;
|
|
|
|
if (next)
|
|
|
|
{
|
|
|
|
gcc_checking_assert (pnxp == next->pnxp);
|
|
|
|
next->pnxp = &next;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
~imm_store_chain_info ()
|
|
|
|
{
|
|
|
|
*pnxp = next;
|
|
|
|
if (next)
|
|
|
|
{
|
|
|
|
gcc_checking_assert (&next == next->pnxp);
|
|
|
|
next->pnxp = pnxp;
|
|
|
|
}
|
|
|
|
}
|
2016-11-02 14:05:07 +01:00
|
|
|
bool terminate_and_process_chain ();
|
2016-10-28 16:18:50 +02:00
|
|
|
bool coalesce_immediate_stores ();
|
2016-11-02 14:05:07 +01:00
|
|
|
bool output_merged_store (merged_store_group *);
|
|
|
|
bool output_merged_stores ();
|
2016-10-28 16:18:50 +02:00
|
|
|
};
|
|
|
|
|
|
|
|
const pass_data pass_data_tree_store_merging = {
|
|
|
|
GIMPLE_PASS, /* type */
|
|
|
|
"store-merging", /* name */
|
|
|
|
OPTGROUP_NONE, /* optinfo_flags */
|
|
|
|
TV_GIMPLE_STORE_MERGING, /* tv_id */
|
|
|
|
PROP_ssa, /* properties_required */
|
|
|
|
0, /* properties_provided */
|
|
|
|
0, /* properties_destroyed */
|
|
|
|
0, /* todo_flags_start */
|
|
|
|
TODO_update_ssa, /* todo_flags_finish */
|
|
|
|
};
|
|
|
|
|
|
|
|
class pass_store_merging : public gimple_opt_pass
|
|
|
|
{
|
|
|
|
public:
|
|
|
|
pass_store_merging (gcc::context *ctxt)
|
2017-03-17 14:14:42 +01:00
|
|
|
: gimple_opt_pass (pass_data_tree_store_merging, ctxt), m_stores_head ()
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
}
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
/* Pass not supported for PDP-endianness, nor for insane hosts
|
|
|
|
or target character sizes where native_{encode,interpret}_expr
|
|
|
|
doesn't work properly. */
|
2016-10-28 16:18:50 +02:00
|
|
|
virtual bool
|
|
|
|
gate (function *)
|
|
|
|
{
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
return flag_store_merging
|
|
|
|
&& WORDS_BIG_ENDIAN == BYTES_BIG_ENDIAN
|
|
|
|
&& CHAR_BIT == 8
|
|
|
|
&& BITS_PER_UNIT == 8;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
virtual unsigned int execute (function *);
|
|
|
|
|
|
|
|
private:
|
|
|
|
hash_map<tree_operand_hash, struct imm_store_chain_info *> m_stores;
|
|
|
|
|
2017-03-17 00:31:01 +01:00
|
|
|
/* Form a doubly-linked stack of the elements of m_stores, so that
|
|
|
|
we can iterate over them in a predictable way. Using this order
|
|
|
|
avoids extraneous differences in the compiler output just because
|
|
|
|
of tree pointer variations (e.g. different chains end up in
|
|
|
|
different positions of m_stores, so they are handled in different
|
|
|
|
orders, so they allocate or release SSA names in different
|
|
|
|
orders, and when they get reused, subsequent passes end up
|
|
|
|
getting different SSA names, which may ultimately change
|
|
|
|
decisions when going out of SSA). */
|
|
|
|
imm_store_chain_info *m_stores_head;
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
void process_store (gimple *);
|
2016-10-28 16:18:50 +02:00
|
|
|
bool terminate_and_process_all_chains ();
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
bool terminate_all_aliasing_chains (imm_store_chain_info **, gimple *);
|
2016-11-02 14:05:07 +01:00
|
|
|
bool terminate_and_release_chain (imm_store_chain_info *);
|
2016-10-28 16:18:50 +02:00
|
|
|
}; // class pass_store_merging
|
|
|
|
|
|
|
|
/* Terminate and process all recorded chains. Return true if any changes
|
|
|
|
were made. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
pass_store_merging::terminate_and_process_all_chains ()
|
|
|
|
{
|
|
|
|
bool ret = false;
|
2017-03-17 00:31:01 +01:00
|
|
|
while (m_stores_head)
|
|
|
|
ret |= terminate_and_release_chain (m_stores_head);
|
|
|
|
gcc_assert (m_stores.elements () == 0);
|
|
|
|
gcc_assert (m_stores_head == NULL);
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
/* Terminate all chains that are affected by the statement STMT.
|
|
|
|
CHAIN_INFO is the chain we should ignore from the checks if
|
|
|
|
non-NULL. */
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
bool
|
2016-11-07 13:26:00 +01:00
|
|
|
pass_store_merging::terminate_all_aliasing_chains (imm_store_chain_info
|
2016-11-02 14:05:07 +01:00
|
|
|
**chain_info,
|
2016-10-28 16:18:50 +02:00
|
|
|
gimple *stmt)
|
|
|
|
{
|
|
|
|
bool ret = false;
|
|
|
|
|
|
|
|
/* If the statement doesn't touch memory it can't alias. */
|
|
|
|
if (!gimple_vuse (stmt))
|
|
|
|
return false;
|
|
|
|
|
2017-11-10 11:31:34 +01:00
|
|
|
tree store_lhs = gimple_store_p (stmt) ? gimple_get_lhs (stmt) : NULL_TREE;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
for (imm_store_chain_info *next = m_stores_head, *cur = next; cur; cur = next)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
next = cur->next;
|
|
|
|
|
|
|
|
/* We already checked all the stores in chain_info and terminated the
|
|
|
|
chain if necessary. Skip it here. */
|
|
|
|
if (chain_info && *chain_info == cur)
|
|
|
|
continue;
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
store_immediate_info *info;
|
|
|
|
unsigned int i;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
FOR_EACH_VEC_ELT (cur->m_store_info, i, info)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
2017-11-10 11:31:34 +01:00
|
|
|
tree lhs = gimple_assign_lhs (info->stmt);
|
|
|
|
if (ref_maybe_used_by_stmt_p (stmt, lhs)
|
|
|
|
|| stmt_may_clobber_ref_p (stmt, lhs)
|
|
|
|
|| (store_lhs && refs_output_dependent_p (store_lhs, lhs)))
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
fprintf (dump_file, "stmt causes chain termination:\n");
|
|
|
|
print_gimple_stmt (dump_file, stmt, 0);
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
terminate_and_release_chain (cur);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
ret = true;
|
|
|
|
break;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Helper function. Terminate the recorded chain storing to base object
|
|
|
|
BASE. Return true if the merging and output was successful. The m_stores
|
|
|
|
entry is removed after the processing in any case. */
|
|
|
|
|
|
|
|
bool
|
2016-11-02 14:05:07 +01:00
|
|
|
pass_store_merging::terminate_and_release_chain (imm_store_chain_info *chain_info)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
2016-11-02 14:05:07 +01:00
|
|
|
bool ret = chain_info->terminate_and_process_chain ();
|
|
|
|
m_stores.remove (chain_info->base_addr);
|
|
|
|
delete chain_info;
|
2016-10-28 16:18:50 +02:00
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Return true if stmts in between FIRST (inclusive) and LAST (exclusive)
|
|
|
|
may clobber REF. FIRST and LAST must be in the same basic block and
|
|
|
|
have non-NULL vdef. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
stmts_may_clobber_ref_p (gimple *first, gimple *last, tree ref)
|
|
|
|
{
|
|
|
|
ao_ref r;
|
|
|
|
ao_ref_init (&r, ref);
|
|
|
|
unsigned int count = 0;
|
|
|
|
tree vop = gimple_vdef (last);
|
|
|
|
gimple *stmt;
|
|
|
|
|
|
|
|
gcc_checking_assert (gimple_bb (first) == gimple_bb (last));
|
|
|
|
do
|
|
|
|
{
|
|
|
|
stmt = SSA_NAME_DEF_STMT (vop);
|
|
|
|
if (stmt_may_clobber_ref_p_1 (stmt, &r))
|
|
|
|
return true;
|
|
|
|
/* Avoid quadratic compile time by bounding the number of checks
|
|
|
|
we perform. */
|
|
|
|
if (++count > MAX_STORE_ALIAS_CHECKS)
|
|
|
|
return true;
|
|
|
|
vop = gimple_vuse (stmt);
|
|
|
|
}
|
|
|
|
while (stmt != first);
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return true if INFO->ops[IDX] is mergeable with the
|
|
|
|
corresponding loads already in MERGED_STORE group.
|
|
|
|
BASE_ADDR is the base address of the whole store group. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
compatible_load_p (merged_store_group *merged_store,
|
|
|
|
store_immediate_info *info,
|
|
|
|
tree base_addr, int idx)
|
|
|
|
{
|
|
|
|
store_immediate_info *infof = merged_store->stores[0];
|
|
|
|
if (!info->ops[idx].base_addr
|
|
|
|
|| (info->ops[idx].bitpos - infof->ops[idx].bitpos
|
|
|
|
!= info->bitpos - infof->bitpos)
|
|
|
|
|| !operand_equal_p (info->ops[idx].base_addr,
|
|
|
|
infof->ops[idx].base_addr, 0))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
store_immediate_info *infol = merged_store->stores.last ();
|
|
|
|
tree load_vuse = gimple_vuse (info->ops[idx].stmt);
|
|
|
|
/* In this case all vuses should be the same, e.g.
|
|
|
|
_1 = s.a; _2 = s.b; _3 = _1 | 1; t.a = _3; _4 = _2 | 2; t.b = _4;
|
|
|
|
or
|
|
|
|
_1 = s.a; _2 = s.b; t.a = _1; t.b = _2;
|
|
|
|
and we can emit the coalesced load next to any of those loads. */
|
|
|
|
if (gimple_vuse (infof->ops[idx].stmt) == load_vuse
|
|
|
|
&& gimple_vuse (infol->ops[idx].stmt) == load_vuse)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
/* Otherwise, at least for now require that the load has the same
|
|
|
|
vuse as the store. See following examples. */
|
|
|
|
if (gimple_vuse (info->stmt) != load_vuse)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (gimple_vuse (infof->stmt) != gimple_vuse (infof->ops[idx].stmt)
|
|
|
|
|| (infof != infol
|
|
|
|
&& gimple_vuse (infol->stmt) != gimple_vuse (infol->ops[idx].stmt)))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* If the load is from the same location as the store, already
|
|
|
|
the construction of the immediate chain info guarantees no intervening
|
|
|
|
stores, so no further checks are needed. Example:
|
|
|
|
_1 = s.a; _2 = _1 & -7; s.a = _2; _3 = s.b; _4 = _3 & -7; s.b = _4; */
|
|
|
|
if (info->ops[idx].bitpos == info->bitpos
|
|
|
|
&& operand_equal_p (info->ops[idx].base_addr, base_addr, 0))
|
|
|
|
return true;
|
|
|
|
|
|
|
|
/* Otherwise, we need to punt if any of the loads can be clobbered by any
|
|
|
|
of the stores in the group, or any other stores in between those.
|
|
|
|
Previous calls to compatible_load_p ensured that for all the
|
|
|
|
merged_store->stores IDX loads, no stmts starting with
|
|
|
|
merged_store->first_stmt and ending right before merged_store->last_stmt
|
|
|
|
clobbers those loads. */
|
|
|
|
gimple *first = merged_store->first_stmt;
|
|
|
|
gimple *last = merged_store->last_stmt;
|
|
|
|
unsigned int i;
|
|
|
|
store_immediate_info *infoc;
|
|
|
|
/* The stores are sorted by increasing store bitpos, so if info->stmt store
|
|
|
|
comes before the so far first load, we'll be changing
|
|
|
|
merged_store->first_stmt. In that case we need to give up if
|
|
|
|
any of the earlier processed loads clobber with the stmts in the new
|
|
|
|
range. */
|
|
|
|
if (info->order < merged_store->first_order)
|
|
|
|
{
|
|
|
|
FOR_EACH_VEC_ELT (merged_store->stores, i, infoc)
|
|
|
|
if (stmts_may_clobber_ref_p (info->stmt, first, infoc->ops[idx].val))
|
|
|
|
return false;
|
|
|
|
first = info->stmt;
|
|
|
|
}
|
|
|
|
/* Similarly, we could change merged_store->last_stmt, so ensure
|
|
|
|
in that case no stmts in the new range clobber any of the earlier
|
|
|
|
processed loads. */
|
|
|
|
else if (info->order > merged_store->last_order)
|
|
|
|
{
|
|
|
|
FOR_EACH_VEC_ELT (merged_store->stores, i, infoc)
|
|
|
|
if (stmts_may_clobber_ref_p (last, info->stmt, infoc->ops[idx].val))
|
|
|
|
return false;
|
|
|
|
last = info->stmt;
|
|
|
|
}
|
|
|
|
/* And finally, we'd be adding a new load to the set, ensure it isn't
|
|
|
|
clobbered in the new range. */
|
|
|
|
if (stmts_may_clobber_ref_p (first, last, info->ops[idx].val))
|
|
|
|
return false;
|
|
|
|
|
|
|
|
/* Otherwise, we are looking for:
|
|
|
|
_1 = s.a; _2 = _1 ^ 15; t.a = _2; _3 = s.b; _4 = _3 ^ 15; t.b = _4;
|
|
|
|
or
|
|
|
|
_1 = s.a; t.a = _1; _2 = s.b; t.b = _2; */
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Go through the candidate stores recorded in m_store_info and merge them
|
|
|
|
into merged_store_group objects recorded into m_merged_store_groups
|
|
|
|
representing the widened stores. Return true if coalescing was successful
|
|
|
|
and the number of widened stores is fewer than the original number
|
|
|
|
of stores. */
|
|
|
|
|
|
|
|
bool
|
|
|
|
imm_store_chain_info::coalesce_immediate_stores ()
|
|
|
|
{
|
|
|
|
/* Anything less can't be processed. */
|
|
|
|
if (m_store_info.length () < 2)
|
|
|
|
return false;
|
|
|
|
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
fprintf (dump_file, "Attempting to coalesce %u stores in chain.\n",
|
|
|
|
m_store_info.length ());
|
|
|
|
|
|
|
|
store_immediate_info *info;
|
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
/* Order the stores by the bitposition they write to. */
|
|
|
|
m_store_info.qsort (sort_by_bitpos);
|
|
|
|
|
|
|
|
info = m_store_info[0];
|
|
|
|
merged_store_group *merged_store = new merged_store_group (info);
|
|
|
|
|
|
|
|
FOR_EACH_VEC_ELT (m_store_info, i, info)
|
|
|
|
{
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
{
|
|
|
|
fprintf (dump_file, "Store %u:\nbitsize:" HOST_WIDE_INT_PRINT_DEC
|
|
|
|
" bitpos:" HOST_WIDE_INT_PRINT_DEC " val:\n",
|
|
|
|
i, info->bitsize, info->bitpos);
|
2017-05-16 16:51:02 +02:00
|
|
|
print_generic_expr (dump_file, gimple_assign_rhs1 (info->stmt));
|
2016-10-28 16:18:50 +02:00
|
|
|
fprintf (dump_file, "\n------------\n");
|
|
|
|
}
|
|
|
|
|
|
|
|
if (i == 0)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
/* |---store 1---|
|
|
|
|
|---store 2---|
|
|
|
|
Overlapping stores. */
|
|
|
|
unsigned HOST_WIDE_INT start = info->bitpos;
|
|
|
|
if (IN_RANGE (start, merged_store->start,
|
|
|
|
merged_store->start + merged_store->width - 1))
|
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Only allow overlapping stores of constants. */
|
|
|
|
if (info->rhs_code == INTEGER_CST
|
|
|
|
&& merged_store->stores[0]->rhs_code == INTEGER_CST)
|
|
|
|
{
|
|
|
|
merged_store->merge_overlapping (info);
|
|
|
|
continue;
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* |---store 1---||---store 2---|
|
|
|
|
This store is consecutive to the previous one.
|
|
|
|
Merge it into the current store group. There can be gaps in between
|
|
|
|
the stores, but there can't be gaps in between bitregions. */
|
|
|
|
else if (info->bitregion_start <= merged_store->bitregion_end
|
2017-11-13 11:26:13 +01:00
|
|
|
&& info->rhs_code == merged_store->stores[0]->rhs_code)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
store_immediate_info *infof = merged_store->stores[0];
|
|
|
|
|
|
|
|
/* All the rhs_code ops that take 2 operands are commutative,
|
|
|
|
swap the operands if it could make the operands compatible. */
|
|
|
|
if (infof->ops[0].base_addr
|
|
|
|
&& infof->ops[1].base_addr
|
|
|
|
&& info->ops[0].base_addr
|
|
|
|
&& info->ops[1].base_addr
|
|
|
|
&& (info->ops[1].bitpos - infof->ops[0].bitpos
|
|
|
|
== info->bitpos - infof->bitpos)
|
|
|
|
&& operand_equal_p (info->ops[1].base_addr,
|
|
|
|
infof->ops[0].base_addr, 0))
|
2017-11-10 15:43:13 +01:00
|
|
|
{
|
|
|
|
std::swap (info->ops[0], info->ops[1]);
|
|
|
|
info->ops_swapped_p = true;
|
|
|
|
}
|
2017-11-13 10:01:24 +01:00
|
|
|
if ((infof->ops[0].base_addr
|
|
|
|
? compatible_load_p (merged_store, info, base_addr, 0)
|
|
|
|
: !info->ops[0].base_addr)
|
|
|
|
&& (infof->ops[1].base_addr
|
|
|
|
? compatible_load_p (merged_store, info, base_addr, 1)
|
|
|
|
: !info->ops[1].base_addr))
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
{
|
|
|
|
merged_store->merge_into (info);
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* |---store 1---| <gap> |---store 2---|.
|
|
|
|
Gap between stores or the rhs not compatible. Start a new group. */
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Try to apply all the stores recorded for the group to determine
|
|
|
|
the bitpattern they write and discard it if that fails.
|
|
|
|
This will also reject single-store groups. */
|
|
|
|
if (!merged_store->apply_stores ())
|
|
|
|
delete merged_store;
|
|
|
|
else
|
|
|
|
m_merged_store_groups.safe_push (merged_store);
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
merged_store = new merged_store_group (info);
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
/* Record or discard the last store group. */
|
|
|
|
if (!merged_store->apply_stores ())
|
|
|
|
delete merged_store;
|
|
|
|
else
|
|
|
|
m_merged_store_groups.safe_push (merged_store);
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
gcc_assert (m_merged_store_groups.length () <= m_store_info.length ());
|
|
|
|
bool success
|
|
|
|
= !m_merged_store_groups.is_empty ()
|
|
|
|
&& m_merged_store_groups.length () < m_store_info.length ();
|
|
|
|
|
|
|
|
if (success && dump_file)
|
|
|
|
fprintf (dump_file, "Coalescing successful!\n"
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
"Merged into %u stores\n",
|
|
|
|
m_merged_store_groups.length ());
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
return success;
|
|
|
|
}
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Return the type to use for the merged stores or loads described by STMTS.
|
|
|
|
This is needed to get the alias sets right. If IS_LOAD, look for rhs,
|
|
|
|
otherwise lhs. Additionally set *CLIQUEP and *BASEP to MR_DEPENDENCE_*
|
|
|
|
of the MEM_REFs if any. */
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
static tree
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
get_alias_type_for_stmts (vec<gimple *> &stmts, bool is_load,
|
|
|
|
unsigned short *cliquep, unsigned short *basep)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
gimple *stmt;
|
|
|
|
unsigned int i;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
tree type = NULL_TREE;
|
|
|
|
tree ret = NULL_TREE;
|
|
|
|
*cliquep = 0;
|
|
|
|
*basep = 0;
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
FOR_EACH_VEC_ELT (stmts, i, stmt)
|
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
tree ref = is_load ? gimple_assign_rhs1 (stmt)
|
|
|
|
: gimple_assign_lhs (stmt);
|
|
|
|
tree type1 = reference_alias_ptr_type (ref);
|
|
|
|
tree base = get_base_address (ref);
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (i == 0)
|
|
|
|
{
|
|
|
|
if (TREE_CODE (base) == MEM_REF)
|
|
|
|
{
|
|
|
|
*cliquep = MR_DEPENDENCE_CLIQUE (base);
|
|
|
|
*basep = MR_DEPENDENCE_BASE (base);
|
|
|
|
}
|
|
|
|
ret = type = type1;
|
|
|
|
continue;
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
if (!alias_ptr_types_compatible_p (type, type1))
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
ret = ptr_type_node;
|
|
|
|
if (TREE_CODE (base) != MEM_REF
|
|
|
|
|| *cliquep != MR_DEPENDENCE_CLIQUE (base)
|
|
|
|
|| *basep != MR_DEPENDENCE_BASE (base))
|
|
|
|
{
|
|
|
|
*cliquep = 0;
|
|
|
|
*basep = 0;
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
return ret;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Return the location_t information we can find among the statements
|
|
|
|
in STMTS. */
|
|
|
|
|
|
|
|
static location_t
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
get_location_for_stmts (vec<gimple *> &stmts)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
gimple *stmt;
|
|
|
|
unsigned int i;
|
|
|
|
|
|
|
|
FOR_EACH_VEC_ELT (stmts, i, stmt)
|
|
|
|
if (gimple_has_location (stmt))
|
|
|
|
return gimple_location (stmt);
|
|
|
|
|
|
|
|
return UNKNOWN_LOCATION;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Used to decribe a store resulting from splitting a wide store in smaller
|
|
|
|
regularly-sized stores in split_group. */
|
|
|
|
|
|
|
|
struct split_store
|
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT bytepos;
|
|
|
|
unsigned HOST_WIDE_INT size;
|
|
|
|
unsigned HOST_WIDE_INT align;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
auto_vec<store_immediate_info *> orig_stores;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
/* True if there is a single orig stmt covering the whole split store. */
|
|
|
|
bool orig;
|
2016-10-28 16:18:50 +02:00
|
|
|
split_store (unsigned HOST_WIDE_INT, unsigned HOST_WIDE_INT,
|
|
|
|
unsigned HOST_WIDE_INT);
|
|
|
|
};
|
|
|
|
|
|
|
|
/* Simple constructor. */
|
|
|
|
|
|
|
|
split_store::split_store (unsigned HOST_WIDE_INT bp,
|
|
|
|
unsigned HOST_WIDE_INT sz,
|
|
|
|
unsigned HOST_WIDE_INT al)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
: bytepos (bp), size (sz), align (al), orig (false)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
orig_stores.create (0);
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Record all stores in GROUP that write to the region starting at BITPOS and
|
|
|
|
is of size BITSIZE. Record infos for such statements in STORES if
|
|
|
|
non-NULL. The stores in GROUP must be sorted by bitposition. Return INFO
|
|
|
|
if there is exactly one original store in the range. */
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
static store_immediate_info *
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
find_constituent_stores (struct merged_store_group *group,
|
|
|
|
vec<store_immediate_info *> *stores,
|
|
|
|
unsigned int *first,
|
|
|
|
unsigned HOST_WIDE_INT bitpos,
|
|
|
|
unsigned HOST_WIDE_INT bitsize)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
store_immediate_info *info, *ret = NULL;
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned int i;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
bool second = false;
|
|
|
|
bool update_first = true;
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned HOST_WIDE_INT end = bitpos + bitsize;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
for (i = *first; group->stores.iterate (i, &info); ++i)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT stmt_start = info->bitpos;
|
|
|
|
unsigned HOST_WIDE_INT stmt_end = stmt_start + info->bitsize;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (stmt_end <= bitpos)
|
|
|
|
{
|
|
|
|
/* BITPOS passed to this function never decreases from within the
|
|
|
|
same split_group call, so optimize and don't scan info records
|
|
|
|
which are known to end before or at BITPOS next time.
|
|
|
|
Only do it if all stores before this one also pass this. */
|
|
|
|
if (update_first)
|
|
|
|
*first = i + 1;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
update_first = false;
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
/* The stores in GROUP are ordered by bitposition so if we're past
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
the region for this group return early. */
|
|
|
|
if (stmt_start >= end)
|
|
|
|
return ret;
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (stores)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
stores->safe_push (info);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (ret)
|
|
|
|
{
|
|
|
|
ret = NULL;
|
|
|
|
second = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else if (ret)
|
|
|
|
return NULL;
|
|
|
|
if (!second)
|
|
|
|
ret = info;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
return ret;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
2017-11-09 14:08:41 +01:00
|
|
|
/* Return how many SSA_NAMEs used to compute value to store in the INFO
|
|
|
|
store have multiple uses. If any SSA_NAME has multiple uses, also
|
|
|
|
count statements needed to compute it. */
|
|
|
|
|
|
|
|
static unsigned
|
|
|
|
count_multiple_uses (store_immediate_info *info)
|
|
|
|
{
|
|
|
|
gimple *stmt = info->stmt;
|
|
|
|
unsigned ret = 0;
|
|
|
|
switch (info->rhs_code)
|
|
|
|
{
|
|
|
|
case INTEGER_CST:
|
|
|
|
return 0;
|
|
|
|
case BIT_AND_EXPR:
|
|
|
|
case BIT_IOR_EXPR:
|
|
|
|
case BIT_XOR_EXPR:
|
2017-11-09 20:24:59 +01:00
|
|
|
if (info->bit_not_p)
|
|
|
|
{
|
|
|
|
if (!has_single_use (gimple_assign_rhs1 (stmt)))
|
|
|
|
ret = 1; /* Fall through below to return
|
|
|
|
the BIT_NOT_EXPR stmt and then
|
|
|
|
BIT_{AND,IOR,XOR}_EXPR and anything it
|
|
|
|
uses. */
|
|
|
|
else
|
|
|
|
/* stmt is after this the BIT_NOT_EXPR. */
|
|
|
|
stmt = SSA_NAME_DEF_STMT (gimple_assign_rhs1 (stmt));
|
|
|
|
}
|
2017-11-09 14:08:41 +01:00
|
|
|
if (!has_single_use (gimple_assign_rhs1 (stmt)))
|
|
|
|
{
|
|
|
|
ret += 1 + info->ops[0].bit_not_p;
|
|
|
|
if (info->ops[1].base_addr)
|
|
|
|
ret += 1 + info->ops[1].bit_not_p;
|
|
|
|
return ret + 1;
|
|
|
|
}
|
|
|
|
stmt = SSA_NAME_DEF_STMT (gimple_assign_rhs1 (stmt));
|
|
|
|
/* stmt is now the BIT_*_EXPR. */
|
|
|
|
if (!has_single_use (gimple_assign_rhs1 (stmt)))
|
2017-11-10 15:43:13 +01:00
|
|
|
ret += 1 + info->ops[info->ops_swapped_p].bit_not_p;
|
|
|
|
else if (info->ops[info->ops_swapped_p].bit_not_p)
|
2017-11-09 14:08:41 +01:00
|
|
|
{
|
|
|
|
gimple *stmt2 = SSA_NAME_DEF_STMT (gimple_assign_rhs1 (stmt));
|
|
|
|
if (!has_single_use (gimple_assign_rhs1 (stmt2)))
|
|
|
|
++ret;
|
|
|
|
}
|
|
|
|
if (info->ops[1].base_addr == NULL_TREE)
|
2017-11-10 15:43:13 +01:00
|
|
|
{
|
|
|
|
gcc_checking_assert (!info->ops_swapped_p);
|
|
|
|
return ret;
|
|
|
|
}
|
2017-11-09 14:08:41 +01:00
|
|
|
if (!has_single_use (gimple_assign_rhs2 (stmt)))
|
2017-11-10 15:43:13 +01:00
|
|
|
ret += 1 + info->ops[1 - info->ops_swapped_p].bit_not_p;
|
|
|
|
else if (info->ops[1 - info->ops_swapped_p].bit_not_p)
|
2017-11-09 14:08:41 +01:00
|
|
|
{
|
|
|
|
gimple *stmt2 = SSA_NAME_DEF_STMT (gimple_assign_rhs2 (stmt));
|
|
|
|
if (!has_single_use (gimple_assign_rhs1 (stmt2)))
|
|
|
|
++ret;
|
|
|
|
}
|
|
|
|
return ret;
|
|
|
|
case MEM_REF:
|
|
|
|
if (!has_single_use (gimple_assign_rhs1 (stmt)))
|
|
|
|
return 1 + info->ops[0].bit_not_p;
|
|
|
|
else if (info->ops[0].bit_not_p)
|
|
|
|
{
|
|
|
|
stmt = SSA_NAME_DEF_STMT (gimple_assign_rhs1 (stmt));
|
|
|
|
if (!has_single_use (gimple_assign_rhs1 (stmt)))
|
|
|
|
return 1;
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
default:
|
|
|
|
gcc_unreachable ();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Split a merged store described by GROUP by populating the SPLIT_STORES
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
vector (if non-NULL) with split_store structs describing the byte offset
|
|
|
|
(from the base), the bit size and alignment of each store as well as the
|
|
|
|
original statements involved in each such split group.
|
2016-10-28 16:18:50 +02:00
|
|
|
This is to separate the splitting strategy from the statement
|
|
|
|
building/emission/linking done in output_merged_store.
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
Return number of new stores.
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
If ALLOW_UNALIGNED_STORE is false, then all stores must be aligned.
|
|
|
|
If ALLOW_UNALIGNED_LOAD is false, then all loads must be aligned.
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
If SPLIT_STORES is NULL, it is just a dry run to count number of
|
|
|
|
new stores. */
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
static unsigned int
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
split_group (merged_store_group *group, bool allow_unaligned_store,
|
|
|
|
bool allow_unaligned_load,
|
2017-11-09 14:08:41 +01:00
|
|
|
vec<struct split_store *> *split_stores,
|
|
|
|
unsigned *total_orig,
|
|
|
|
unsigned *total_new)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT pos = group->bitregion_start;
|
|
|
|
unsigned HOST_WIDE_INT size = group->bitregion_end - pos;
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned HOST_WIDE_INT bytepos = pos / BITS_PER_UNIT;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT group_align = group->align;
|
|
|
|
unsigned HOST_WIDE_INT align_base = group->align_base;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
unsigned HOST_WIDE_INT group_load_align = group_align;
|
2017-11-09 14:08:41 +01:00
|
|
|
bool any_orig = false;
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
gcc_assert ((size % BITS_PER_UNIT == 0) && (pos % BITS_PER_UNIT == 0));
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned int ret = 0, first = 0;
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned HOST_WIDE_INT try_pos = bytepos;
|
|
|
|
group->stores.qsort (sort_by_bitpos);
|
|
|
|
|
2017-11-09 14:08:41 +01:00
|
|
|
if (total_orig)
|
|
|
|
{
|
|
|
|
unsigned int i;
|
|
|
|
store_immediate_info *info = group->stores[0];
|
|
|
|
|
|
|
|
total_new[0] = 0;
|
|
|
|
total_orig[0] = 1; /* The orig store. */
|
|
|
|
info = group->stores[0];
|
|
|
|
if (info->ops[0].base_addr)
|
2017-11-13 11:26:13 +01:00
|
|
|
total_orig[0]++;
|
2017-11-09 14:08:41 +01:00
|
|
|
if (info->ops[1].base_addr)
|
2017-11-13 11:26:13 +01:00
|
|
|
total_orig[0]++;
|
2017-11-09 14:08:41 +01:00
|
|
|
switch (info->rhs_code)
|
|
|
|
{
|
|
|
|
case BIT_AND_EXPR:
|
|
|
|
case BIT_IOR_EXPR:
|
|
|
|
case BIT_XOR_EXPR:
|
|
|
|
total_orig[0]++; /* The orig BIT_*_EXPR stmt. */
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
total_orig[0] *= group->stores.length ();
|
|
|
|
|
|
|
|
FOR_EACH_VEC_ELT (group->stores, i, info)
|
2017-11-13 11:26:13 +01:00
|
|
|
{
|
|
|
|
total_new[0] += count_multiple_uses (info);
|
|
|
|
total_orig[0] += (info->bit_not_p
|
|
|
|
+ info->ops[0].bit_not_p
|
|
|
|
+ info->ops[1].bit_not_p);
|
|
|
|
}
|
2017-11-09 14:08:41 +01:00
|
|
|
}
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (!allow_unaligned_load)
|
|
|
|
for (int i = 0; i < 2; ++i)
|
|
|
|
if (group->load_align[i])
|
|
|
|
group_load_align = MIN (group_load_align, group->load_align[i]);
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
while (size > 0)
|
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if ((allow_unaligned_store || group_align <= BITS_PER_UNIT)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
&& group->mask[try_pos - bytepos] == (unsigned char) ~0U)
|
|
|
|
{
|
|
|
|
/* Skip padding bytes. */
|
|
|
|
++try_pos;
|
|
|
|
size -= BITS_PER_UNIT;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
unsigned HOST_WIDE_INT try_bitpos = try_pos * BITS_PER_UNIT;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned int try_size = MAX_STORE_BITSIZE, nonmasked;
|
|
|
|
unsigned HOST_WIDE_INT align_bitpos
|
|
|
|
= (try_bitpos - align_base) & (group_align - 1);
|
|
|
|
unsigned HOST_WIDE_INT align = group_align;
|
|
|
|
if (align_bitpos)
|
|
|
|
align = least_bit_hwi (align_bitpos);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (!allow_unaligned_store)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
try_size = MIN (try_size, align);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (!allow_unaligned_load)
|
|
|
|
{
|
|
|
|
/* If we can't do or don't want to do unaligned stores
|
|
|
|
as well as loads, we need to take the loads into account
|
|
|
|
as well. */
|
|
|
|
unsigned HOST_WIDE_INT load_align = group_load_align;
|
|
|
|
align_bitpos = (try_bitpos - align_base) & (load_align - 1);
|
|
|
|
if (align_bitpos)
|
|
|
|
load_align = least_bit_hwi (align_bitpos);
|
|
|
|
for (int i = 0; i < 2; ++i)
|
|
|
|
if (group->load_align[i])
|
|
|
|
{
|
|
|
|
align_bitpos = try_bitpos - group->stores[0]->bitpos;
|
|
|
|
align_bitpos += group->stores[0]->ops[i].bitpos;
|
|
|
|
align_bitpos -= group->load_align_base[i];
|
|
|
|
align_bitpos &= (group_load_align - 1);
|
|
|
|
if (align_bitpos)
|
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT a = least_bit_hwi (align_bitpos);
|
|
|
|
load_align = MIN (load_align, a);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
try_size = MIN (try_size, load_align);
|
|
|
|
}
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
store_immediate_info *info
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
= find_constituent_stores (group, NULL, &first, try_bitpos, try_size);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (info)
|
|
|
|
{
|
|
|
|
/* If there is just one original statement for the range, see if
|
|
|
|
we can just reuse the original store which could be even larger
|
|
|
|
than try_size. */
|
|
|
|
unsigned HOST_WIDE_INT stmt_end
|
|
|
|
= ROUND_UP (info->bitpos + info->bitsize, BITS_PER_UNIT);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
info = find_constituent_stores (group, NULL, &first, try_bitpos,
|
|
|
|
stmt_end - try_bitpos);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (info && info->bitpos >= try_bitpos)
|
|
|
|
{
|
|
|
|
try_size = stmt_end - try_bitpos;
|
|
|
|
goto found;
|
|
|
|
}
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
/* Approximate store bitsize for the case when there are no padding
|
|
|
|
bits. */
|
|
|
|
while (try_size > size)
|
|
|
|
try_size /= 2;
|
|
|
|
/* Now look for whole padding bytes at the end of that bitsize. */
|
|
|
|
for (nonmasked = try_size / BITS_PER_UNIT; nonmasked > 0; --nonmasked)
|
|
|
|
if (group->mask[try_pos - bytepos + nonmasked - 1]
|
|
|
|
!= (unsigned char) ~0U)
|
|
|
|
break;
|
|
|
|
if (nonmasked == 0)
|
|
|
|
{
|
|
|
|
/* If entire try_size range is padding, skip it. */
|
|
|
|
try_pos += try_size / BITS_PER_UNIT;
|
|
|
|
size -= try_size;
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
/* Otherwise try to decrease try_size if second half, last 3 quarters
|
|
|
|
etc. are padding. */
|
|
|
|
nonmasked *= BITS_PER_UNIT;
|
|
|
|
while (nonmasked <= try_size / 2)
|
|
|
|
try_size /= 2;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (!allow_unaligned_store && group_align > BITS_PER_UNIT)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
{
|
|
|
|
/* Now look for whole padding bytes at the start of that bitsize. */
|
|
|
|
unsigned int try_bytesize = try_size / BITS_PER_UNIT, masked;
|
|
|
|
for (masked = 0; masked < try_bytesize; ++masked)
|
|
|
|
if (group->mask[try_pos - bytepos + masked] != (unsigned char) ~0U)
|
|
|
|
break;
|
|
|
|
masked *= BITS_PER_UNIT;
|
|
|
|
gcc_assert (masked < try_size);
|
|
|
|
if (masked >= try_size / 2)
|
|
|
|
{
|
|
|
|
while (masked >= try_size / 2)
|
|
|
|
{
|
|
|
|
try_size /= 2;
|
|
|
|
try_pos += try_size / BITS_PER_UNIT;
|
|
|
|
size -= try_size;
|
|
|
|
masked -= try_size;
|
|
|
|
}
|
|
|
|
/* Need to recompute the alignment, so just retry at the new
|
|
|
|
position. */
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
found:
|
|
|
|
++ret;
|
2016-10-28 16:18:50 +02:00
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (split_stores)
|
|
|
|
{
|
|
|
|
struct split_store *store
|
|
|
|
= new split_store (try_pos, try_size, align);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
info = find_constituent_stores (group, &store->orig_stores,
|
|
|
|
&first, try_bitpos, try_size);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (info
|
|
|
|
&& info->bitpos >= try_bitpos
|
|
|
|
&& info->bitpos + info->bitsize <= try_bitpos + try_size)
|
2017-11-09 14:08:41 +01:00
|
|
|
{
|
|
|
|
store->orig = true;
|
|
|
|
any_orig = true;
|
|
|
|
}
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
split_stores->safe_push (store);
|
|
|
|
}
|
|
|
|
|
|
|
|
try_pos += try_size / BITS_PER_UNIT;
|
2016-10-28 16:18:50 +02:00
|
|
|
size -= try_size;
|
|
|
|
}
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
|
2017-11-09 14:08:41 +01:00
|
|
|
if (total_orig)
|
|
|
|
{
|
2017-11-13 11:26:13 +01:00
|
|
|
unsigned int i;
|
|
|
|
struct split_store *store;
|
2017-11-09 14:08:41 +01:00
|
|
|
/* If we are reusing some original stores and any of the
|
|
|
|
original SSA_NAMEs had multiple uses, we need to subtract
|
|
|
|
those now before we add the new ones. */
|
|
|
|
if (total_new[0] && any_orig)
|
|
|
|
{
|
|
|
|
FOR_EACH_VEC_ELT (*split_stores, i, store)
|
|
|
|
if (store->orig)
|
|
|
|
total_new[0] -= count_multiple_uses (store->orig_stores[0]);
|
|
|
|
}
|
|
|
|
total_new[0] += ret; /* The new store. */
|
|
|
|
store_immediate_info *info = group->stores[0];
|
|
|
|
if (info->ops[0].base_addr)
|
2017-11-13 11:26:13 +01:00
|
|
|
total_new[0] += ret;
|
2017-11-09 14:08:41 +01:00
|
|
|
if (info->ops[1].base_addr)
|
2017-11-13 11:26:13 +01:00
|
|
|
total_new[0] += ret;
|
2017-11-09 14:08:41 +01:00
|
|
|
switch (info->rhs_code)
|
|
|
|
{
|
|
|
|
case BIT_AND_EXPR:
|
|
|
|
case BIT_IOR_EXPR:
|
|
|
|
case BIT_XOR_EXPR:
|
|
|
|
total_new[0] += ret; /* The new BIT_*_EXPR stmt. */
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
break;
|
|
|
|
}
|
2017-11-13 11:26:13 +01:00
|
|
|
FOR_EACH_VEC_ELT (*split_stores, i, store)
|
|
|
|
{
|
|
|
|
unsigned int j;
|
|
|
|
bool bit_not_p[3] = { false, false, false };
|
|
|
|
/* If all orig_stores have certain bit_not_p set, then
|
|
|
|
we'd use a BIT_NOT_EXPR stmt and need to account for it.
|
|
|
|
If some orig_stores have certain bit_not_p set, then
|
|
|
|
we'd use a BIT_XOR_EXPR with a mask and need to account for
|
|
|
|
it. */
|
|
|
|
FOR_EACH_VEC_ELT (store->orig_stores, j, info)
|
|
|
|
{
|
|
|
|
if (info->ops[0].bit_not_p)
|
|
|
|
bit_not_p[0] = true;
|
|
|
|
if (info->ops[1].bit_not_p)
|
|
|
|
bit_not_p[1] = true;
|
|
|
|
if (info->bit_not_p)
|
|
|
|
bit_not_p[2] = true;
|
|
|
|
}
|
|
|
|
total_new[0] += bit_not_p[0] + bit_not_p[1] + bit_not_p[2];
|
|
|
|
}
|
|
|
|
|
2017-11-09 14:08:41 +01:00
|
|
|
}
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
return ret;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
2017-11-13 11:26:13 +01:00
|
|
|
/* Return the operation through which the operand IDX (if < 2) or
|
|
|
|
result (IDX == 2) should be inverted. If NOP_EXPR, no inversion
|
|
|
|
is done, if BIT_NOT_EXPR, all bits are inverted, if BIT_XOR_EXPR,
|
|
|
|
the bits should be xored with mask. */
|
|
|
|
|
|
|
|
static enum tree_code
|
|
|
|
invert_op (split_store *split_store, int idx, tree int_type, tree &mask)
|
|
|
|
{
|
|
|
|
unsigned int i;
|
|
|
|
store_immediate_info *info;
|
|
|
|
unsigned int cnt = 0;
|
|
|
|
FOR_EACH_VEC_ELT (split_store->orig_stores, i, info)
|
|
|
|
{
|
|
|
|
bool bit_not_p = idx < 2 ? info->ops[idx].bit_not_p : info->bit_not_p;
|
|
|
|
if (bit_not_p)
|
|
|
|
++cnt;
|
|
|
|
}
|
|
|
|
mask = NULL_TREE;
|
|
|
|
if (cnt == 0)
|
|
|
|
return NOP_EXPR;
|
|
|
|
if (cnt == split_store->orig_stores.length ())
|
|
|
|
return BIT_NOT_EXPR;
|
|
|
|
|
|
|
|
unsigned HOST_WIDE_INT try_bitpos = split_store->bytepos * BITS_PER_UNIT;
|
|
|
|
unsigned buf_size = split_store->size / BITS_PER_UNIT;
|
|
|
|
unsigned char *buf
|
|
|
|
= XALLOCAVEC (unsigned char, buf_size);
|
|
|
|
memset (buf, ~0U, buf_size);
|
|
|
|
FOR_EACH_VEC_ELT (split_store->orig_stores, i, info)
|
|
|
|
{
|
|
|
|
bool bit_not_p = idx < 2 ? info->ops[idx].bit_not_p : info->bit_not_p;
|
|
|
|
if (!bit_not_p)
|
|
|
|
continue;
|
|
|
|
/* Clear regions with bit_not_p and invert afterwards, rather than
|
|
|
|
clear regions with !bit_not_p, so that gaps in between stores aren't
|
|
|
|
set in the mask. */
|
|
|
|
unsigned HOST_WIDE_INT bitsize = info->bitsize;
|
|
|
|
unsigned int pos_in_buffer = 0;
|
|
|
|
if (info->bitpos < try_bitpos)
|
|
|
|
{
|
|
|
|
gcc_assert (info->bitpos + bitsize > try_bitpos);
|
|
|
|
bitsize -= (try_bitpos - info->bitpos);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
pos_in_buffer = info->bitpos - try_bitpos;
|
|
|
|
if (pos_in_buffer + bitsize > split_store->size)
|
|
|
|
bitsize = split_store->size - pos_in_buffer;
|
|
|
|
unsigned char *p = buf + (pos_in_buffer / BITS_PER_UNIT);
|
|
|
|
if (BYTES_BIG_ENDIAN)
|
|
|
|
clear_bit_region_be (p, (BITS_PER_UNIT - 1
|
|
|
|
- (pos_in_buffer % BITS_PER_UNIT)), bitsize);
|
|
|
|
else
|
|
|
|
clear_bit_region (p, pos_in_buffer % BITS_PER_UNIT, bitsize);
|
|
|
|
}
|
|
|
|
for (unsigned int i = 0; i < buf_size; ++i)
|
|
|
|
buf[i] = ~buf[i];
|
|
|
|
mask = native_interpret_expr (int_type, buf, buf_size);
|
|
|
|
return BIT_XOR_EXPR;
|
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Given a merged store group GROUP output the widened version of it.
|
|
|
|
The store chain is against the base object BASE.
|
|
|
|
Try store sizes of at most MAX_STORE_BITSIZE bits wide and don't output
|
|
|
|
unaligned stores for STRICT_ALIGNMENT targets or if it's too expensive.
|
|
|
|
Make sure that the number of statements output is less than the number of
|
|
|
|
original statements. If a better sequence is possible emit it and
|
|
|
|
return true. */
|
|
|
|
|
|
|
|
bool
|
2016-11-02 14:05:07 +01:00
|
|
|
imm_store_chain_info::output_merged_store (merged_store_group *group)
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
unsigned HOST_WIDE_INT start_byte_pos
|
|
|
|
= group->bitregion_start / BITS_PER_UNIT;
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
unsigned int orig_num_stmts = group->stores.length ();
|
|
|
|
if (orig_num_stmts < 2)
|
|
|
|
return false;
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
auto_vec<struct split_store *, 32> split_stores;
|
2016-10-28 16:18:50 +02:00
|
|
|
split_stores.create (0);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
bool allow_unaligned_store
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
= !STRICT_ALIGNMENT && PARAM_VALUE (PARAM_STORE_MERGING_ALLOW_UNALIGNED);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
bool allow_unaligned_load = allow_unaligned_store;
|
|
|
|
if (allow_unaligned_store)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
{
|
|
|
|
/* If unaligned stores are allowed, see how many stores we'd emit
|
|
|
|
for unaligned and how many stores we'd emit for aligned stores.
|
|
|
|
Only use unaligned stores if it allows fewer stores than aligned. */
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
unsigned aligned_cnt
|
2017-11-09 14:08:41 +01:00
|
|
|
= split_group (group, false, allow_unaligned_load, NULL, NULL, NULL);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
unsigned unaligned_cnt
|
2017-11-09 14:08:41 +01:00
|
|
|
= split_group (group, true, allow_unaligned_load, NULL, NULL, NULL);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (aligned_cnt <= unaligned_cnt)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
allow_unaligned_store = false;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
}
|
2017-11-09 14:08:41 +01:00
|
|
|
unsigned total_orig, total_new;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
split_group (group, allow_unaligned_store, allow_unaligned_load,
|
2017-11-09 14:08:41 +01:00
|
|
|
&split_stores, &total_orig, &total_new);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
|
|
|
|
if (split_stores.length () >= orig_num_stmts)
|
|
|
|
{
|
|
|
|
/* We didn't manage to reduce the number of statements. Bail out. */
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
2017-11-09 14:08:41 +01:00
|
|
|
fprintf (dump_file, "Exceeded original number of stmts (%u)."
|
|
|
|
" Not profitable to emit new sequence.\n",
|
|
|
|
orig_num_stmts);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
return false;
|
|
|
|
}
|
2017-11-09 14:08:41 +01:00
|
|
|
if (total_orig <= total_new)
|
|
|
|
{
|
|
|
|
/* If number of estimated new statements is above estimated original
|
|
|
|
statements, bail out too. */
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
fprintf (dump_file, "Estimated number of original stmts (%u)"
|
|
|
|
" not larger than estimated number of new"
|
|
|
|
" stmts (%u).\n",
|
|
|
|
total_orig, total_new);
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
gimple_stmt_iterator last_gsi = gsi_for_stmt (group->last_stmt);
|
|
|
|
gimple_seq seq = NULL;
|
|
|
|
tree last_vdef, new_vuse;
|
|
|
|
last_vdef = gimple_vdef (group->last_stmt);
|
|
|
|
new_vuse = gimple_vuse (group->last_stmt);
|
|
|
|
|
|
|
|
gimple *stmt = NULL;
|
|
|
|
split_store *split_store;
|
|
|
|
unsigned int i;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
auto_vec<gimple *, 32> orig_stmts;
|
2016-11-02 16:00:48 +01:00
|
|
|
tree addr = force_gimple_operand_1 (unshare_expr (base_addr), &seq,
|
|
|
|
is_gimple_mem_ref_addr, NULL_TREE);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
|
|
|
|
tree load_addr[2] = { NULL_TREE, NULL_TREE };
|
|
|
|
gimple_seq load_seq[2] = { NULL, NULL };
|
|
|
|
gimple_stmt_iterator load_gsi[2] = { gsi_none (), gsi_none () };
|
|
|
|
for (int j = 0; j < 2; ++j)
|
|
|
|
{
|
|
|
|
store_operand_info &op = group->stores[0]->ops[j];
|
|
|
|
if (op.base_addr == NULL_TREE)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
store_immediate_info *infol = group->stores.last ();
|
|
|
|
if (gimple_vuse (op.stmt) == gimple_vuse (infol->ops[j].stmt))
|
|
|
|
{
|
|
|
|
load_gsi[j] = gsi_for_stmt (op.stmt);
|
|
|
|
load_addr[j]
|
|
|
|
= force_gimple_operand_1 (unshare_expr (op.base_addr),
|
|
|
|
&load_seq[j], is_gimple_mem_ref_addr,
|
|
|
|
NULL_TREE);
|
|
|
|
}
|
|
|
|
else if (operand_equal_p (base_addr, op.base_addr, 0))
|
|
|
|
load_addr[j] = addr;
|
|
|
|
else
|
2017-11-06 17:29:11 +01:00
|
|
|
{
|
|
|
|
gimple_seq this_seq;
|
|
|
|
load_addr[j]
|
|
|
|
= force_gimple_operand_1 (unshare_expr (op.base_addr),
|
|
|
|
&this_seq, is_gimple_mem_ref_addr,
|
|
|
|
NULL_TREE);
|
|
|
|
gimple_seq_add_seq_without_update (&seq, this_seq);
|
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
FOR_EACH_VEC_ELT (split_stores, i, split_store)
|
|
|
|
{
|
|
|
|
unsigned HOST_WIDE_INT try_size = split_store->size;
|
|
|
|
unsigned HOST_WIDE_INT try_pos = split_store->bytepos;
|
|
|
|
unsigned HOST_WIDE_INT align = split_store->align;
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
tree dest, src;
|
|
|
|
location_t loc;
|
|
|
|
if (split_store->orig)
|
|
|
|
{
|
|
|
|
/* If there is just a single constituent store which covers
|
|
|
|
the whole area, just reuse the lhs and rhs. */
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
gimple *orig_stmt = split_store->orig_stores[0]->stmt;
|
|
|
|
dest = gimple_assign_lhs (orig_stmt);
|
|
|
|
src = gimple_assign_rhs1 (orig_stmt);
|
|
|
|
loc = gimple_location (orig_stmt);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
store_immediate_info *info;
|
|
|
|
unsigned short clique, base;
|
|
|
|
unsigned int k;
|
|
|
|
FOR_EACH_VEC_ELT (split_store->orig_stores, k, info)
|
|
|
|
orig_stmts.safe_push (info->stmt);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
tree offset_type
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
= get_alias_type_for_stmts (orig_stmts, false, &clique, &base);
|
|
|
|
loc = get_location_for_stmts (orig_stmts);
|
|
|
|
orig_stmts.truncate (0);
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
|
|
|
|
tree int_type = build_nonstandard_integer_type (try_size, UNSIGNED);
|
|
|
|
int_type = build_aligned_type (int_type, align);
|
|
|
|
dest = fold_build2 (MEM_REF, int_type, addr,
|
|
|
|
build_int_cst (offset_type, try_pos));
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (TREE_CODE (dest) == MEM_REF)
|
|
|
|
{
|
|
|
|
MR_DEPENDENCE_CLIQUE (dest) = clique;
|
|
|
|
MR_DEPENDENCE_BASE (dest) = base;
|
|
|
|
}
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
tree mask
|
|
|
|
= native_interpret_expr (int_type,
|
|
|
|
group->mask + try_pos - start_byte_pos,
|
|
|
|
group->buf_size);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
|
|
|
|
tree ops[2];
|
|
|
|
for (int j = 0;
|
|
|
|
j < 1 + (split_store->orig_stores[0]->ops[1].val != NULL_TREE);
|
|
|
|
++j)
|
|
|
|
{
|
|
|
|
store_operand_info &op = split_store->orig_stores[0]->ops[j];
|
|
|
|
if (op.base_addr)
|
|
|
|
{
|
|
|
|
FOR_EACH_VEC_ELT (split_store->orig_stores, k, info)
|
|
|
|
orig_stmts.safe_push (info->ops[j].stmt);
|
|
|
|
|
|
|
|
offset_type = get_alias_type_for_stmts (orig_stmts, true,
|
|
|
|
&clique, &base);
|
|
|
|
location_t load_loc = get_location_for_stmts (orig_stmts);
|
|
|
|
orig_stmts.truncate (0);
|
|
|
|
|
|
|
|
unsigned HOST_WIDE_INT load_align = group->load_align[j];
|
|
|
|
unsigned HOST_WIDE_INT align_bitpos
|
|
|
|
= (try_pos * BITS_PER_UNIT
|
|
|
|
- split_store->orig_stores[0]->bitpos
|
|
|
|
+ op.bitpos) & (load_align - 1);
|
|
|
|
if (align_bitpos)
|
|
|
|
load_align = least_bit_hwi (align_bitpos);
|
|
|
|
|
|
|
|
tree load_int_type
|
|
|
|
= build_nonstandard_integer_type (try_size, UNSIGNED);
|
|
|
|
load_int_type
|
|
|
|
= build_aligned_type (load_int_type, load_align);
|
|
|
|
|
|
|
|
unsigned HOST_WIDE_INT load_pos
|
|
|
|
= (try_pos * BITS_PER_UNIT
|
|
|
|
- split_store->orig_stores[0]->bitpos
|
|
|
|
+ op.bitpos) / BITS_PER_UNIT;
|
|
|
|
ops[j] = fold_build2 (MEM_REF, load_int_type, load_addr[j],
|
|
|
|
build_int_cst (offset_type, load_pos));
|
|
|
|
if (TREE_CODE (ops[j]) == MEM_REF)
|
|
|
|
{
|
|
|
|
MR_DEPENDENCE_CLIQUE (ops[j]) = clique;
|
|
|
|
MR_DEPENDENCE_BASE (ops[j]) = base;
|
|
|
|
}
|
|
|
|
if (!integer_zerop (mask))
|
|
|
|
/* The load might load some bits (that will be masked off
|
|
|
|
later on) uninitialized, avoid -W*uninitialized
|
|
|
|
warnings in that case. */
|
|
|
|
TREE_NO_WARNING (ops[j]) = 1;
|
|
|
|
|
|
|
|
stmt = gimple_build_assign (make_ssa_name (int_type),
|
|
|
|
ops[j]);
|
|
|
|
gimple_set_location (stmt, load_loc);
|
|
|
|
if (gsi_bb (load_gsi[j]))
|
|
|
|
{
|
|
|
|
gimple_set_vuse (stmt, gimple_vuse (op.stmt));
|
|
|
|
gimple_seq_add_stmt_without_update (&load_seq[j], stmt);
|
|
|
|
}
|
|
|
|
else
|
|
|
|
{
|
|
|
|
gimple_set_vuse (stmt, new_vuse);
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
}
|
|
|
|
ops[j] = gimple_assign_lhs (stmt);
|
2017-11-13 11:26:13 +01:00
|
|
|
tree xor_mask;
|
|
|
|
enum tree_code inv_op
|
|
|
|
= invert_op (split_store, j, int_type, xor_mask);
|
|
|
|
if (inv_op != NOP_EXPR)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
{
|
|
|
|
stmt = gimple_build_assign (make_ssa_name (int_type),
|
2017-11-13 11:26:13 +01:00
|
|
|
inv_op, ops[j], xor_mask);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
gimple_set_location (stmt, load_loc);
|
|
|
|
ops[j] = gimple_assign_lhs (stmt);
|
|
|
|
|
|
|
|
if (gsi_bb (load_gsi[j]))
|
|
|
|
gimple_seq_add_stmt_without_update (&load_seq[j],
|
|
|
|
stmt);
|
|
|
|
else
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
}
|
|
|
|
else
|
|
|
|
ops[j] = native_interpret_expr (int_type,
|
|
|
|
group->val + try_pos
|
|
|
|
- start_byte_pos,
|
|
|
|
group->buf_size);
|
|
|
|
}
|
|
|
|
|
|
|
|
switch (split_store->orig_stores[0]->rhs_code)
|
|
|
|
{
|
|
|
|
case BIT_AND_EXPR:
|
|
|
|
case BIT_IOR_EXPR:
|
|
|
|
case BIT_XOR_EXPR:
|
|
|
|
FOR_EACH_VEC_ELT (split_store->orig_stores, k, info)
|
|
|
|
{
|
|
|
|
tree rhs1 = gimple_assign_rhs1 (info->stmt);
|
|
|
|
orig_stmts.safe_push (SSA_NAME_DEF_STMT (rhs1));
|
|
|
|
}
|
|
|
|
location_t bit_loc;
|
|
|
|
bit_loc = get_location_for_stmts (orig_stmts);
|
|
|
|
orig_stmts.truncate (0);
|
|
|
|
|
|
|
|
stmt
|
|
|
|
= gimple_build_assign (make_ssa_name (int_type),
|
|
|
|
split_store->orig_stores[0]->rhs_code,
|
|
|
|
ops[0], ops[1]);
|
|
|
|
gimple_set_location (stmt, bit_loc);
|
|
|
|
/* If there is just one load and there is a separate
|
|
|
|
load_seq[0], emit the bitwise op right after it. */
|
|
|
|
if (load_addr[1] == NULL_TREE && gsi_bb (load_gsi[0]))
|
|
|
|
gimple_seq_add_stmt_without_update (&load_seq[0], stmt);
|
|
|
|
/* Otherwise, if at least one load is in seq, we need to
|
|
|
|
emit the bitwise op right before the store. If there
|
|
|
|
are two loads and are emitted somewhere else, it would
|
|
|
|
be better to emit the bitwise op as early as possible;
|
|
|
|
we don't track where that would be possible right now
|
|
|
|
though. */
|
|
|
|
else
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
src = gimple_assign_lhs (stmt);
|
2017-11-13 11:26:13 +01:00
|
|
|
tree xor_mask;
|
|
|
|
enum tree_code inv_op;
|
|
|
|
inv_op = invert_op (split_store, 2, int_type, xor_mask);
|
|
|
|
if (inv_op != NOP_EXPR)
|
2017-11-09 20:24:59 +01:00
|
|
|
{
|
|
|
|
stmt = gimple_build_assign (make_ssa_name (int_type),
|
2017-11-13 11:26:13 +01:00
|
|
|
inv_op, src, xor_mask);
|
2017-11-09 20:24:59 +01:00
|
|
|
gimple_set_location (stmt, bit_loc);
|
|
|
|
if (load_addr[1] == NULL_TREE && gsi_bb (load_gsi[0]))
|
|
|
|
gimple_seq_add_stmt_without_update (&load_seq[0], stmt);
|
|
|
|
else
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
src = gimple_assign_lhs (stmt);
|
|
|
|
}
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
break;
|
|
|
|
default:
|
|
|
|
src = ops[0];
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
if (!integer_zerop (mask))
|
|
|
|
{
|
|
|
|
tree tem = make_ssa_name (int_type);
|
|
|
|
tree load_src = unshare_expr (dest);
|
|
|
|
/* The load might load some or all bits uninitialized,
|
|
|
|
avoid -W*uninitialized warnings in that case.
|
|
|
|
As optimization, it would be nice if all the bits are
|
|
|
|
provably uninitialized (no stores at all yet or previous
|
|
|
|
store a CLOBBER) we'd optimize away the load and replace
|
|
|
|
it e.g. with 0. */
|
|
|
|
TREE_NO_WARNING (load_src) = 1;
|
|
|
|
stmt = gimple_build_assign (tem, load_src);
|
|
|
|
gimple_set_location (stmt, loc);
|
|
|
|
gimple_set_vuse (stmt, new_vuse);
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
|
|
|
|
/* FIXME: If there is a single chunk of zero bits in mask,
|
|
|
|
perhaps use BIT_INSERT_EXPR instead? */
|
|
|
|
stmt = gimple_build_assign (make_ssa_name (int_type),
|
|
|
|
BIT_AND_EXPR, tem, mask);
|
|
|
|
gimple_set_location (stmt, loc);
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
tem = gimple_assign_lhs (stmt);
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (TREE_CODE (src) == INTEGER_CST)
|
|
|
|
src = wide_int_to_tree (int_type,
|
|
|
|
wi::bit_and_not (wi::to_wide (src),
|
|
|
|
wi::to_wide (mask)));
|
|
|
|
else
|
|
|
|
{
|
|
|
|
tree nmask
|
|
|
|
= wide_int_to_tree (int_type,
|
|
|
|
wi::bit_not (wi::to_wide (mask)));
|
|
|
|
stmt = gimple_build_assign (make_ssa_name (int_type),
|
|
|
|
BIT_AND_EXPR, src, nmask);
|
|
|
|
gimple_set_location (stmt, loc);
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
src = gimple_assign_lhs (stmt);
|
|
|
|
}
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
stmt = gimple_build_assign (make_ssa_name (int_type),
|
|
|
|
BIT_IOR_EXPR, tem, src);
|
|
|
|
gimple_set_location (stmt, loc);
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
src = gimple_assign_lhs (stmt);
|
|
|
|
}
|
|
|
|
}
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
stmt = gimple_build_assign (dest, src);
|
|
|
|
gimple_set_location (stmt, loc);
|
|
|
|
gimple_set_vuse (stmt, new_vuse);
|
|
|
|
gimple_seq_add_stmt_without_update (&seq, stmt);
|
|
|
|
|
|
|
|
tree new_vdef;
|
|
|
|
if (i < split_stores.length () - 1)
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
new_vdef = make_ssa_name (gimple_vop (cfun), stmt);
|
2016-10-28 16:18:50 +02:00
|
|
|
else
|
|
|
|
new_vdef = last_vdef;
|
|
|
|
|
|
|
|
gimple_set_vdef (stmt, new_vdef);
|
|
|
|
SSA_NAME_DEF_STMT (new_vdef) = stmt;
|
|
|
|
new_vuse = new_vdef;
|
|
|
|
}
|
|
|
|
|
|
|
|
FOR_EACH_VEC_ELT (split_stores, i, split_store)
|
|
|
|
delete split_store;
|
|
|
|
|
|
|
|
gcc_assert (seq);
|
|
|
|
if (dump_file)
|
|
|
|
{
|
|
|
|
fprintf (dump_file,
|
|
|
|
"New sequence of %u stmts to replace old one of %u stmts\n",
|
re PR middle-end/22141 (Missing optimization when storing structures)
PR middle-end/22141
* gimple-ssa-store-merging.c: Include rtl.h and expr.h.
(struct store_immediate_info): Add bitregion_start and bitregion_end
fields.
(store_immediate_info::store_immediate_info): Add brs and bre
arguments and initialize bitregion_{start,end} from those.
(struct merged_store_group): Add bitregion_start, bitregion_end,
align_base and mask fields. Drop unnecessary struct keyword from
struct store_immediate_info. Add do_merge method.
(clear_bit_region_be): Use memset instead of loop storing zeros.
(merged_store_group::do_merge): New method.
(merged_store_group::merge_into): Use do_merge. Allow gaps in between
stores as long as the surrounding bitregions have no gaps.
(merged_store_group::merge_overlapping): Use do_merge.
(merged_store_group::apply_stores): Test that bitregion_{start,end}
is byte aligned, rather than requiring that start and width are
byte aligned. Drop unnecessary struct keyword from
struct store_immediate_info. Allocate and populate also mask array.
Make start of the arrays relative to bitregion_start rather than
start and size them according to bitregion_{end,start} difference.
(struct imm_store_chain_info): Drop unnecessary struct keyword from
struct store_immediate_info.
(pass_store_merging::gate): Punt if BITS_PER_UNIT or CHAR_BIT is not 8.
(pass_store_merging::terminate_all_aliasing_chains): Drop unnecessary
struct keyword from struct store_immediate_info.
(imm_store_chain_info::coalesce_immediate_stores): Allow gaps in
between stores as long as the surrounding bitregions have no gaps.
Formatting fixes.
(struct split_store): Add orig non-static data member.
(split_store::split_store): Initialize orig to false.
(find_constituent_stmts): Return store_immediate_info *, non-NULL
if there is exactly a single original stmt. Change stmts argument
to pointer from reference, if NULL, don't push anything to it. Add
first argument, use it to optimize skipping over orig stmts that
are known to be before bitpos already. Simplify.
(split_group): Return unsigned int count how many stores are or
would be needed rather than a bool. Add allow_unaligned argument.
Change split_stores argument from reference to pointer, if NULL,
only do a dry run computing how many stores would be produced.
Rewritten algorithm to use both alignment and misalign if
!allow_unaligned and handle bitfield stores with gaps.
(imm_store_chain_info::output_merged_store): Set start_byte_pos
from bitregion_start instead of start. Compute allow_unaligned
here, if true, do 2 split_group dry runs to compute which one
produces fewer stores and prefer aligned if equal. Punt if
new count is bigger or equal than original before emitting any
statements, rather than during that. Remove no longer needed
new_ssa_names tracking. Replace num_stmts with
split_stores.length (). Use 32-bit stack allocated entries
in split_stores auto_vec. Try to reuse original store lhs/rhs1
if possible. Handle bitfields with gaps.
(pass_store_merging::execute): Ignore bitsize == 0 stores.
Compute bitregion_{start,end} for the stores and construct
store_immediate_info with that. Formatting fixes.
* gcc.dg/store_merging_10.c: New test.
* gcc.dg/store_merging_11.c: New test.
* gcc.dg/store_merging_12.c: New test.
* g++.dg/pr71694.C: Add -fno-store-merging to dg-options.
From-SVN: r254213
2017-10-30 12:04:49 +01:00
|
|
|
split_stores.length (), orig_num_stmts);
|
2016-10-28 16:18:50 +02:00
|
|
|
if (dump_flags & TDF_DETAILS)
|
|
|
|
print_gimple_seq (dump_file, seq, 0, TDF_VOPS | TDF_MEMSYMS);
|
|
|
|
}
|
|
|
|
gsi_insert_seq_after (&last_gsi, seq, GSI_SAME_STMT);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
for (int j = 0; j < 2; ++j)
|
|
|
|
if (load_seq[j])
|
|
|
|
gsi_insert_seq_after (&load_gsi[j], load_seq[j], GSI_SAME_STMT);
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Process the merged_store_group objects created in the coalescing phase.
|
|
|
|
The stores are all against the base object BASE.
|
|
|
|
Try to output the widened stores and delete the original statements if
|
|
|
|
successful. Return true iff any changes were made. */
|
|
|
|
|
|
|
|
bool
|
2016-11-02 14:05:07 +01:00
|
|
|
imm_store_chain_info::output_merged_stores ()
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
unsigned int i;
|
|
|
|
merged_store_group *merged_store;
|
|
|
|
bool ret = false;
|
|
|
|
FOR_EACH_VEC_ELT (m_merged_store_groups, i, merged_store)
|
|
|
|
{
|
2016-11-02 14:05:07 +01:00
|
|
|
if (output_merged_store (merged_store))
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
unsigned int j;
|
|
|
|
store_immediate_info *store;
|
|
|
|
FOR_EACH_VEC_ELT (merged_store->stores, j, store)
|
|
|
|
{
|
|
|
|
gimple *stmt = store->stmt;
|
|
|
|
gimple_stmt_iterator gsi = gsi_for_stmt (stmt);
|
|
|
|
gsi_remove (&gsi, true);
|
|
|
|
if (stmt != merged_store->last_stmt)
|
|
|
|
{
|
|
|
|
unlink_stmt_vdef (stmt);
|
|
|
|
release_defs (stmt);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
ret = true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (ret && dump_file)
|
|
|
|
fprintf (dump_file, "Merging successful!\n");
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Coalesce the store_immediate_info objects recorded against the base object
|
|
|
|
BASE in the first phase and output them.
|
|
|
|
Delete the allocated structures.
|
|
|
|
Return true if any changes were made. */
|
|
|
|
|
|
|
|
bool
|
2016-11-02 14:05:07 +01:00
|
|
|
imm_store_chain_info::terminate_and_process_chain ()
|
2016-10-28 16:18:50 +02:00
|
|
|
{
|
|
|
|
/* Process store chain. */
|
|
|
|
bool ret = false;
|
|
|
|
if (m_store_info.length () > 1)
|
|
|
|
{
|
|
|
|
ret = coalesce_immediate_stores ();
|
|
|
|
if (ret)
|
2016-11-02 14:05:07 +01:00
|
|
|
ret = output_merged_stores ();
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
|
|
|
/* Delete all the entries we allocated ourselves. */
|
|
|
|
store_immediate_info *info;
|
|
|
|
unsigned int i;
|
|
|
|
FOR_EACH_VEC_ELT (m_store_info, i, info)
|
|
|
|
delete info;
|
|
|
|
|
|
|
|
merged_store_group *merged_info;
|
|
|
|
FOR_EACH_VEC_ELT (m_merged_store_groups, i, merged_info)
|
|
|
|
delete merged_info;
|
|
|
|
|
|
|
|
return ret;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return true iff LHS is a destination potentially interesting for
|
|
|
|
store merging. In practice these are the codes that get_inner_reference
|
|
|
|
can process. */
|
|
|
|
|
|
|
|
static bool
|
|
|
|
lhs_valid_for_store_merging_p (tree lhs)
|
|
|
|
{
|
|
|
|
tree_code code = TREE_CODE (lhs);
|
|
|
|
|
|
|
|
if (code == ARRAY_REF || code == ARRAY_RANGE_REF || code == MEM_REF
|
|
|
|
|| code == COMPONENT_REF || code == BIT_FIELD_REF)
|
|
|
|
return true;
|
|
|
|
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return true if the tree RHS is a constant we want to consider
|
|
|
|
during store merging. In practice accept all codes that
|
|
|
|
native_encode_expr accepts. */
|
|
|
|
|
|
|
|
static bool
|
|
|
|
rhs_valid_for_store_merging_p (tree rhs)
|
|
|
|
{
|
2017-10-06 11:39:07 +02:00
|
|
|
return native_encode_expr (rhs, NULL,
|
|
|
|
GET_MODE_SIZE (TYPE_MODE (TREE_TYPE (rhs)))) != 0;
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* If MEM is a memory reference usable for store merging (either as
|
|
|
|
store destination or for loads), return the non-NULL base_addr
|
|
|
|
and set *PBITSIZE, *PBITPOS, *PBITREGION_START and *PBITREGION_END.
|
|
|
|
Otherwise return NULL, *PBITPOS should be still valid even for that
|
|
|
|
case. */
|
|
|
|
|
|
|
|
static tree
|
|
|
|
mem_valid_for_store_merging (tree mem, unsigned HOST_WIDE_INT *pbitsize,
|
|
|
|
unsigned HOST_WIDE_INT *pbitpos,
|
|
|
|
unsigned HOST_WIDE_INT *pbitregion_start,
|
|
|
|
unsigned HOST_WIDE_INT *pbitregion_end)
|
|
|
|
{
|
|
|
|
HOST_WIDE_INT bitsize;
|
|
|
|
HOST_WIDE_INT bitpos;
|
|
|
|
unsigned HOST_WIDE_INT bitregion_start = 0;
|
|
|
|
unsigned HOST_WIDE_INT bitregion_end = 0;
|
|
|
|
machine_mode mode;
|
|
|
|
int unsignedp = 0, reversep = 0, volatilep = 0;
|
|
|
|
tree offset;
|
|
|
|
tree base_addr = get_inner_reference (mem, &bitsize, &bitpos, &offset, &mode,
|
|
|
|
&unsignedp, &reversep, &volatilep);
|
|
|
|
*pbitsize = bitsize;
|
|
|
|
if (bitsize == 0)
|
|
|
|
return NULL_TREE;
|
|
|
|
|
|
|
|
if (TREE_CODE (mem) == COMPONENT_REF
|
|
|
|
&& DECL_BIT_FIELD_TYPE (TREE_OPERAND (mem, 1)))
|
|
|
|
{
|
|
|
|
get_bit_range (&bitregion_start, &bitregion_end, mem, &bitpos, &offset);
|
|
|
|
if (bitregion_end)
|
|
|
|
++bitregion_end;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (reversep)
|
|
|
|
return NULL_TREE;
|
|
|
|
|
|
|
|
/* We do not want to rewrite TARGET_MEM_REFs. */
|
|
|
|
if (TREE_CODE (base_addr) == TARGET_MEM_REF)
|
|
|
|
return NULL_TREE;
|
|
|
|
/* In some cases get_inner_reference may return a
|
|
|
|
MEM_REF [ptr + byteoffset]. For the purposes of this pass
|
|
|
|
canonicalize the base_addr to MEM_REF [ptr] and take
|
|
|
|
byteoffset into account in the bitpos. This occurs in
|
|
|
|
PR 23684 and this way we can catch more chains. */
|
|
|
|
else if (TREE_CODE (base_addr) == MEM_REF)
|
|
|
|
{
|
|
|
|
offset_int bit_off, byte_off = mem_ref_offset (base_addr);
|
|
|
|
bit_off = byte_off << LOG2_BITS_PER_UNIT;
|
|
|
|
bit_off += bitpos;
|
|
|
|
if (!wi::neg_p (bit_off) && wi::fits_shwi_p (bit_off))
|
|
|
|
{
|
|
|
|
bitpos = bit_off.to_shwi ();
|
|
|
|
if (bitregion_end)
|
|
|
|
{
|
|
|
|
bit_off = byte_off << LOG2_BITS_PER_UNIT;
|
|
|
|
bit_off += bitregion_start;
|
|
|
|
if (wi::fits_uhwi_p (bit_off))
|
|
|
|
{
|
|
|
|
bitregion_start = bit_off.to_uhwi ();
|
|
|
|
bit_off = byte_off << LOG2_BITS_PER_UNIT;
|
|
|
|
bit_off += bitregion_end;
|
|
|
|
if (wi::fits_uhwi_p (bit_off))
|
|
|
|
bitregion_end = bit_off.to_uhwi ();
|
|
|
|
else
|
|
|
|
bitregion_end = 0;
|
|
|
|
}
|
|
|
|
else
|
|
|
|
bitregion_end = 0;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
else
|
|
|
|
return NULL_TREE;
|
|
|
|
base_addr = TREE_OPERAND (base_addr, 0);
|
|
|
|
}
|
|
|
|
/* get_inner_reference returns the base object, get at its
|
|
|
|
address now. */
|
|
|
|
else
|
|
|
|
{
|
|
|
|
if (bitpos < 0)
|
|
|
|
return NULL_TREE;
|
|
|
|
base_addr = build_fold_addr_expr (base_addr);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!bitregion_end)
|
|
|
|
{
|
|
|
|
bitregion_start = ROUND_DOWN (bitpos, BITS_PER_UNIT);
|
|
|
|
bitregion_end = ROUND_UP (bitpos + bitsize, BITS_PER_UNIT);
|
|
|
|
}
|
|
|
|
|
|
|
|
if (offset != NULL_TREE)
|
|
|
|
{
|
|
|
|
/* If the access is variable offset then a base decl has to be
|
|
|
|
address-taken to be able to emit pointer-based stores to it.
|
|
|
|
??? We might be able to get away with re-using the original
|
|
|
|
base up to the first variable part and then wrapping that inside
|
|
|
|
a BIT_FIELD_REF. */
|
|
|
|
tree base = get_base_address (base_addr);
|
|
|
|
if (! base
|
|
|
|
|| (DECL_P (base) && ! TREE_ADDRESSABLE (base)))
|
|
|
|
return NULL_TREE;
|
|
|
|
|
|
|
|
base_addr = build2 (POINTER_PLUS_EXPR, TREE_TYPE (base_addr),
|
|
|
|
base_addr, offset);
|
|
|
|
}
|
|
|
|
|
|
|
|
*pbitsize = bitsize;
|
|
|
|
*pbitpos = bitpos;
|
|
|
|
*pbitregion_start = bitregion_start;
|
|
|
|
*pbitregion_end = bitregion_end;
|
|
|
|
return base_addr;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Return true if STMT is a load that can be used for store merging.
|
|
|
|
In that case fill in *OP. BITSIZE, BITPOS, BITREGION_START and
|
|
|
|
BITREGION_END are properties of the corresponding store. */
|
|
|
|
|
|
|
|
static bool
|
|
|
|
handled_load (gimple *stmt, store_operand_info *op,
|
|
|
|
unsigned HOST_WIDE_INT bitsize, unsigned HOST_WIDE_INT bitpos,
|
|
|
|
unsigned HOST_WIDE_INT bitregion_start,
|
|
|
|
unsigned HOST_WIDE_INT bitregion_end)
|
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
if (!is_gimple_assign (stmt))
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
return false;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
if (gimple_assign_rhs_code (stmt) == BIT_NOT_EXPR)
|
|
|
|
{
|
|
|
|
tree rhs1 = gimple_assign_rhs1 (stmt);
|
|
|
|
if (TREE_CODE (rhs1) == SSA_NAME
|
|
|
|
&& handled_load (SSA_NAME_DEF_STMT (rhs1), op, bitsize, bitpos,
|
|
|
|
bitregion_start, bitregion_end))
|
|
|
|
{
|
2017-11-09 20:24:59 +01:00
|
|
|
/* Don't allow _1 = load; _2 = ~1; _3 = ~_2; which should have
|
|
|
|
been optimized earlier, but if allowed here, would confuse the
|
|
|
|
multiple uses counting. */
|
|
|
|
if (op->bit_not_p)
|
|
|
|
return false;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
op->bit_not_p = !op->bit_not_p;
|
|
|
|
return true;
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
if (gimple_vuse (stmt)
|
|
|
|
&& gimple_assign_load_p (stmt)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
&& !stmt_can_throw_internal (stmt)
|
|
|
|
&& !gimple_has_volatile_ops (stmt))
|
|
|
|
{
|
|
|
|
tree mem = gimple_assign_rhs1 (stmt);
|
|
|
|
op->base_addr
|
|
|
|
= mem_valid_for_store_merging (mem, &op->bitsize, &op->bitpos,
|
|
|
|
&op->bitregion_start,
|
|
|
|
&op->bitregion_end);
|
|
|
|
if (op->base_addr != NULL_TREE
|
|
|
|
&& op->bitsize == bitsize
|
|
|
|
&& ((op->bitpos - bitpos) % BITS_PER_UNIT) == 0
|
|
|
|
&& op->bitpos - op->bitregion_start >= bitpos - bitregion_start
|
|
|
|
&& op->bitregion_end - op->bitpos >= bitregion_end - bitpos)
|
|
|
|
{
|
|
|
|
op->stmt = stmt;
|
|
|
|
op->val = mem;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
op->bit_not_p = false;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
return true;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return false;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Record the store STMT for store merging optimization if it can be
|
|
|
|
optimized. */
|
|
|
|
|
|
|
|
void
|
|
|
|
pass_store_merging::process_store (gimple *stmt)
|
|
|
|
{
|
|
|
|
tree lhs = gimple_assign_lhs (stmt);
|
|
|
|
tree rhs = gimple_assign_rhs1 (stmt);
|
|
|
|
unsigned HOST_WIDE_INT bitsize, bitpos;
|
|
|
|
unsigned HOST_WIDE_INT bitregion_start;
|
|
|
|
unsigned HOST_WIDE_INT bitregion_end;
|
|
|
|
tree base_addr
|
|
|
|
= mem_valid_for_store_merging (lhs, &bitsize, &bitpos,
|
|
|
|
&bitregion_start, &bitregion_end);
|
|
|
|
if (bitsize == 0)
|
|
|
|
return;
|
|
|
|
|
|
|
|
bool invalid = (base_addr == NULL_TREE
|
|
|
|
|| ((bitsize > MAX_BITSIZE_MODE_ANY_INT)
|
|
|
|
&& (TREE_CODE (rhs) != INTEGER_CST)));
|
|
|
|
enum tree_code rhs_code = ERROR_MARK;
|
2017-11-09 20:24:59 +01:00
|
|
|
bool bit_not_p = false;
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
store_operand_info ops[2];
|
|
|
|
if (invalid)
|
|
|
|
;
|
|
|
|
else if (rhs_valid_for_store_merging_p (rhs))
|
|
|
|
{
|
|
|
|
rhs_code = INTEGER_CST;
|
|
|
|
ops[0].val = rhs;
|
|
|
|
}
|
2017-11-09 14:08:41 +01:00
|
|
|
else if (TREE_CODE (rhs) != SSA_NAME)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
invalid = true;
|
|
|
|
else
|
|
|
|
{
|
|
|
|
gimple *def_stmt = SSA_NAME_DEF_STMT (rhs), *def_stmt1, *def_stmt2;
|
|
|
|
if (!is_gimple_assign (def_stmt))
|
|
|
|
invalid = true;
|
|
|
|
else if (handled_load (def_stmt, &ops[0], bitsize, bitpos,
|
|
|
|
bitregion_start, bitregion_end))
|
|
|
|
rhs_code = MEM_REF;
|
2017-11-09 20:24:59 +01:00
|
|
|
else if (gimple_assign_rhs_code (def_stmt) == BIT_NOT_EXPR)
|
|
|
|
{
|
|
|
|
tree rhs1 = gimple_assign_rhs1 (def_stmt);
|
|
|
|
if (TREE_CODE (rhs1) == SSA_NAME
|
|
|
|
&& is_gimple_assign (SSA_NAME_DEF_STMT (rhs1)))
|
|
|
|
{
|
|
|
|
bit_not_p = true;
|
|
|
|
def_stmt = SSA_NAME_DEF_STMT (rhs1);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
if (rhs_code == ERROR_MARK && !invalid)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
switch ((rhs_code = gimple_assign_rhs_code (def_stmt)))
|
|
|
|
{
|
|
|
|
case BIT_AND_EXPR:
|
|
|
|
case BIT_IOR_EXPR:
|
|
|
|
case BIT_XOR_EXPR:
|
|
|
|
tree rhs1, rhs2;
|
|
|
|
rhs1 = gimple_assign_rhs1 (def_stmt);
|
|
|
|
rhs2 = gimple_assign_rhs2 (def_stmt);
|
|
|
|
invalid = true;
|
2017-11-09 14:08:41 +01:00
|
|
|
if (TREE_CODE (rhs1) != SSA_NAME)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
break;
|
|
|
|
def_stmt1 = SSA_NAME_DEF_STMT (rhs1);
|
|
|
|
if (!is_gimple_assign (def_stmt1)
|
|
|
|
|| !handled_load (def_stmt1, &ops[0], bitsize, bitpos,
|
|
|
|
bitregion_start, bitregion_end))
|
|
|
|
break;
|
|
|
|
if (rhs_valid_for_store_merging_p (rhs2))
|
|
|
|
ops[1].val = rhs2;
|
2017-11-09 14:08:41 +01:00
|
|
|
else if (TREE_CODE (rhs2) != SSA_NAME)
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
break;
|
|
|
|
else
|
|
|
|
{
|
|
|
|
def_stmt2 = SSA_NAME_DEF_STMT (rhs2);
|
|
|
|
if (!is_gimple_assign (def_stmt2))
|
|
|
|
break;
|
|
|
|
else if (!handled_load (def_stmt2, &ops[1], bitsize, bitpos,
|
|
|
|
bitregion_start, bitregion_end))
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
invalid = false;
|
|
|
|
break;
|
|
|
|
default:
|
|
|
|
invalid = true;
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (invalid)
|
|
|
|
{
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
terminate_all_aliasing_chains (NULL, stmt);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
struct imm_store_chain_info **chain_info = NULL;
|
|
|
|
if (base_addr)
|
|
|
|
chain_info = m_stores.get (base_addr);
|
|
|
|
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
store_immediate_info *info;
|
|
|
|
if (chain_info)
|
|
|
|
{
|
|
|
|
unsigned int ord = (*chain_info)->m_store_info.length ();
|
|
|
|
info = new store_immediate_info (bitsize, bitpos, bitregion_start,
|
|
|
|
bitregion_end, stmt, ord, rhs_code,
|
2017-11-09 20:24:59 +01:00
|
|
|
bit_not_p, ops[0], ops[1]);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
{
|
|
|
|
fprintf (dump_file, "Recording immediate store from stmt:\n");
|
|
|
|
print_gimple_stmt (dump_file, stmt, 0);
|
|
|
|
}
|
|
|
|
(*chain_info)->m_store_info.safe_push (info);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
terminate_all_aliasing_chains (chain_info, stmt);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* If we reach the limit of stores to merge in a chain terminate and
|
|
|
|
process the chain now. */
|
|
|
|
if ((*chain_info)->m_store_info.length ()
|
|
|
|
== (unsigned int) PARAM_VALUE (PARAM_MAX_STORES_TO_MERGE))
|
|
|
|
{
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
fprintf (dump_file,
|
|
|
|
"Reached maximum number of statements to merge:\n");
|
|
|
|
terminate_and_release_chain (*chain_info);
|
|
|
|
}
|
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Store aliases any existing chain? */
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c (struct store_operand_info): Add bit_not_p
data member.
(store_operand_info::store_operand_info): Initialize it to false.
(pass_store_merging::terminate_all_aliasing_chains): Rewritten to use
ref_maybe_used_by_stmt_p and stmt_may_clobber_ref_p on lhs of each
store in the group, and if chain_info is non-NULL, to ignore altogether
that chain.
(compatible_load_p): Fail if bit_not_p does not match.
(imm_store_chain_info::output_merged_store): Handle bit_not_p loads.
(handled_load): Fill in bit_not_p. Handle BIT_NOT_EXPR.
(pass_store_merging::process_store): Adjust
terminate_all_aliasing_chains calls to pass NULL in all current spots,
call terminate_all_aliasing_chains newly when adding a store into
a chain with non-NULL chain_info.
* gcc.dg/store_merging_2.c: Expect 3 store mergings instead of 2.
* gcc.dg/store_merging_13.c (f7, f8, f9, f10, f11, f12, f13): New
functions.
(main): Test also those. Expect 13 store mergings instead of 6.
* gcc.dg/store_merging_14.c (f7, f8, f9): New functions.
(main): Test also those. Expect 9 store mergings instead of 6.
From-SVN: r254536
2017-11-08 16:46:58 +01:00
|
|
|
terminate_all_aliasing_chains (NULL, stmt);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
/* Start a new chain. */
|
|
|
|
struct imm_store_chain_info *new_chain
|
|
|
|
= new imm_store_chain_info (m_stores_head, base_addr);
|
|
|
|
info = new store_immediate_info (bitsize, bitpos, bitregion_start,
|
|
|
|
bitregion_end, stmt, 0, rhs_code,
|
2017-11-09 20:24:59 +01:00
|
|
|
bit_not_p, ops[0], ops[1]);
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
new_chain->m_store_info.safe_push (info);
|
|
|
|
m_stores.put (base_addr, new_chain);
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
{
|
|
|
|
fprintf (dump_file, "Starting new chain with statement:\n");
|
|
|
|
print_gimple_stmt (dump_file, stmt, 0);
|
|
|
|
fprintf (dump_file, "The base object is:\n");
|
|
|
|
print_generic_expr (dump_file, base_addr);
|
|
|
|
fprintf (dump_file, "\n");
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
/* Entry point for the pass. Go over each basic block recording chains of
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
immediate stores. Upon encountering a terminating statement (as defined
|
|
|
|
by stmt_terminates_chain_p) process the recorded stores and emit the widened
|
|
|
|
variants. */
|
2016-10-28 16:18:50 +02:00
|
|
|
|
|
|
|
unsigned int
|
|
|
|
pass_store_merging::execute (function *fun)
|
|
|
|
{
|
|
|
|
basic_block bb;
|
|
|
|
hash_set<gimple *> orig_stmts;
|
|
|
|
|
|
|
|
FOR_EACH_BB_FN (bb, fun)
|
|
|
|
{
|
|
|
|
gimple_stmt_iterator gsi;
|
|
|
|
unsigned HOST_WIDE_INT num_statements = 0;
|
|
|
|
/* Record the original statements so that we can keep track of
|
|
|
|
statements emitted in this pass and not re-process new
|
|
|
|
statements. */
|
|
|
|
for (gsi = gsi_after_labels (bb); !gsi_end_p (gsi); gsi_next (&gsi))
|
|
|
|
{
|
|
|
|
if (is_gimple_debug (gsi_stmt (gsi)))
|
|
|
|
continue;
|
|
|
|
|
2017-10-06 11:39:07 +02:00
|
|
|
if (++num_statements >= 2)
|
2016-10-28 16:18:50 +02:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (num_statements < 2)
|
|
|
|
continue;
|
|
|
|
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
fprintf (dump_file, "Processing basic block <%d>:\n", bb->index);
|
|
|
|
|
|
|
|
for (gsi = gsi_after_labels (bb); !gsi_end_p (gsi); gsi_next (&gsi))
|
|
|
|
{
|
|
|
|
gimple *stmt = gsi_stmt (gsi);
|
|
|
|
|
2017-03-17 00:31:01 +01:00
|
|
|
if (is_gimple_debug (stmt))
|
|
|
|
continue;
|
|
|
|
|
2016-10-28 16:18:50 +02:00
|
|
|
if (gimple_has_volatile_ops (stmt))
|
|
|
|
{
|
|
|
|
/* Terminate all chains. */
|
|
|
|
if (dump_file && (dump_flags & TDF_DETAILS))
|
|
|
|
fprintf (dump_file, "Volatile access terminates "
|
|
|
|
"all chains\n");
|
|
|
|
terminate_and_process_all_chains ();
|
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (gimple_assign_single_p (stmt) && gimple_vdef (stmt)
|
|
|
|
&& !stmt_can_throw_internal (stmt)
|
|
|
|
&& lhs_valid_for_store_merging_p (gimple_assign_lhs (stmt)))
|
re PR tree-optimization/78821 (GCC7: Copying whole 32 bits structure field by field not optimised into copying whole 32 bits at once)
PR tree-optimization/78821
* gimple-ssa-store-merging.c: Update the file comment.
(MAX_STORE_ALIAS_CHECKS): Define.
(struct store_operand_info): New type.
(store_operand_info::store_operand_info): New constructor.
(struct store_immediate_info): Add rhs_code and ops data members.
(store_immediate_info::store_immediate_info): Add rhscode, op0r
and op1r arguments to the ctor, initialize corresponding data members.
(struct merged_store_group): Add load_align_base and load_align
data members.
(merged_store_group::merged_store_group): Initialize them.
(merged_store_group::do_merge): Update them.
(merged_store_group::apply_stores): Pick the constant for
encode_tree_to_bitpos from one of the two operands, or skip
encode_tree_to_bitpos if neither operand is a constant.
(class pass_store_merging): Add process_store method decl. Remove
bool argument from terminate_all_aliasing_chains method decl.
(pass_store_merging::terminate_all_aliasing_chains): Remove
var_offset_p argument and corresponding handling.
(stmts_may_clobber_ref_p): New function.
(compatible_load_p): New function.
(imm_store_chain_info::coalesce_immediate_stores): Terminate group
if there is overlap and rhs_code is not INTEGER_CST. For
non-overlapping stores terminate group if rhs is not mergeable.
(get_alias_type_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &. Add IS_LOAD, CLIQUEP and
BASEP arguments. If IS_LOAD is true, look at rhs1 of the stmts
instead of lhs. Compute *CLIQUEP and *BASEP in addition to the
alias type.
(get_location_for_stmts): Change first argument from
auto_vec<gimple *> & to vec<gimple *> &.
(struct split_store): Remove orig_stmts data member, add orig_stores.
(split_store::split_store): Create orig_stores rather than orig_stmts.
(find_constituent_stmts): Renamed to ...
(find_constituent_stores): ... this. Change second argument from
vec<gimple *> * to vec<store_immediate_info *> *, push pointers
to info structures rather than the statements.
(split_group): Rename ALLOW_UNALIGNED argument to
ALLOW_UNALIGNED_STORE, add ALLOW_UNALIGNED_LOAD argument and handle
it. Adjust find_constituent_stores caller.
(imm_store_chain_info::output_merged_store): Handle rhs_code other
than INTEGER_CST, adjust split_group, get_alias_type_for_stmts and
get_location_for_stmts callers. Set MR_DEPENDENCE_CLIQUE and
MR_DEPENDENCE_BASE on the MEM_REFs if they are the same in all stores.
(mem_valid_for_store_merging): New function.
(handled_load): New function.
(pass_store_merging::process_store): New method.
(pass_store_merging::execute): Use process_store method. Adjust
terminate_all_aliasing_chains caller.
* gcc.dg/store_merging_13.c: New test.
* gcc.dg/store_merging_14.c: New test.
From-SVN: r254391
2017-11-03 20:08:25 +01:00
|
|
|
process_store (stmt);
|
|
|
|
else
|
|
|
|
terminate_all_aliasing_chains (NULL, stmt);
|
2016-10-28 16:18:50 +02:00
|
|
|
}
|
|
|
|
terminate_and_process_all_chains ();
|
|
|
|
}
|
|
|
|
return 0;
|
|
|
|
}
|
|
|
|
|
|
|
|
} // anon namespace
|
|
|
|
|
|
|
|
/* Construct and return a store merging pass object. */
|
|
|
|
|
|
|
|
gimple_opt_pass *
|
|
|
|
make_pass_store_merging (gcc::context *ctxt)
|
|
|
|
{
|
|
|
|
return new pass_store_merging (ctxt);
|
|
|
|
}
|
2016-11-08 17:04:20 +01:00
|
|
|
|
|
|
|
#if CHECKING_P
|
|
|
|
|
|
|
|
namespace selftest {
|
|
|
|
|
|
|
|
/* Selftests for store merging helpers. */
|
|
|
|
|
|
|
|
/* Assert that all elements of the byte arrays X and Y, both of length N
|
|
|
|
are equal. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
verify_array_eq (unsigned char *x, unsigned char *y, unsigned int n)
|
|
|
|
{
|
|
|
|
for (unsigned int i = 0; i < n; i++)
|
|
|
|
{
|
|
|
|
if (x[i] != y[i])
|
|
|
|
{
|
|
|
|
fprintf (stderr, "Arrays do not match. X:\n");
|
|
|
|
dump_char_array (stderr, x, n);
|
|
|
|
fprintf (stderr, "Y:\n");
|
|
|
|
dump_char_array (stderr, y, n);
|
|
|
|
}
|
|
|
|
ASSERT_EQ (x[i], y[i]);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Test shift_bytes_in_array and that it carries bits across between
|
|
|
|
bytes correctly. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
verify_shift_bytes_in_array (void)
|
|
|
|
{
|
|
|
|
/* byte 1 | byte 0
|
|
|
|
00011111 | 11100000. */
|
|
|
|
unsigned char orig[2] = { 0xe0, 0x1f };
|
|
|
|
unsigned char in[2];
|
|
|
|
memcpy (in, orig, sizeof orig);
|
|
|
|
|
|
|
|
unsigned char expected[2] = { 0x80, 0x7f };
|
|
|
|
shift_bytes_in_array (in, sizeof (in), 2);
|
|
|
|
verify_array_eq (in, expected, sizeof (in));
|
|
|
|
|
|
|
|
memcpy (in, orig, sizeof orig);
|
|
|
|
memcpy (expected, orig, sizeof orig);
|
|
|
|
/* Check that shifting by zero doesn't change anything. */
|
|
|
|
shift_bytes_in_array (in, sizeof (in), 0);
|
|
|
|
verify_array_eq (in, expected, sizeof (in));
|
|
|
|
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Test shift_bytes_in_array_right and that it carries bits across between
|
|
|
|
bytes correctly. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
verify_shift_bytes_in_array_right (void)
|
|
|
|
{
|
|
|
|
/* byte 1 | byte 0
|
|
|
|
00011111 | 11100000. */
|
|
|
|
unsigned char orig[2] = { 0x1f, 0xe0};
|
|
|
|
unsigned char in[2];
|
|
|
|
memcpy (in, orig, sizeof orig);
|
|
|
|
unsigned char expected[2] = { 0x07, 0xf8};
|
|
|
|
shift_bytes_in_array_right (in, sizeof (in), 2);
|
|
|
|
verify_array_eq (in, expected, sizeof (in));
|
|
|
|
|
|
|
|
memcpy (in, orig, sizeof orig);
|
|
|
|
memcpy (expected, orig, sizeof orig);
|
|
|
|
/* Check that shifting by zero doesn't change anything. */
|
|
|
|
shift_bytes_in_array_right (in, sizeof (in), 0);
|
|
|
|
verify_array_eq (in, expected, sizeof (in));
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Test clear_bit_region that it clears exactly the bits asked and
|
|
|
|
nothing more. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
verify_clear_bit_region (void)
|
|
|
|
{
|
|
|
|
/* Start with all bits set and test clearing various patterns in them. */
|
|
|
|
unsigned char orig[3] = { 0xff, 0xff, 0xff};
|
|
|
|
unsigned char in[3];
|
|
|
|
unsigned char expected[3];
|
|
|
|
memcpy (in, orig, sizeof in);
|
|
|
|
|
|
|
|
/* Check zeroing out all the bits. */
|
|
|
|
clear_bit_region (in, 0, 3 * BITS_PER_UNIT);
|
|
|
|
expected[0] = expected[1] = expected[2] = 0;
|
|
|
|
verify_array_eq (in, expected, sizeof in);
|
|
|
|
|
|
|
|
memcpy (in, orig, sizeof in);
|
|
|
|
/* Leave the first and last bits intact. */
|
|
|
|
clear_bit_region (in, 1, 3 * BITS_PER_UNIT - 2);
|
|
|
|
expected[0] = 0x1;
|
|
|
|
expected[1] = 0;
|
|
|
|
expected[2] = 0x80;
|
|
|
|
verify_array_eq (in, expected, sizeof in);
|
|
|
|
}
|
|
|
|
|
|
|
|
/* Test verify_clear_bit_region_be that it clears exactly the bits asked and
|
|
|
|
nothing more. */
|
|
|
|
|
|
|
|
static void
|
|
|
|
verify_clear_bit_region_be (void)
|
|
|
|
{
|
|
|
|
/* Start with all bits set and test clearing various patterns in them. */
|
|
|
|
unsigned char orig[3] = { 0xff, 0xff, 0xff};
|
|
|
|
unsigned char in[3];
|
|
|
|
unsigned char expected[3];
|
|
|
|
memcpy (in, orig, sizeof in);
|
|
|
|
|
|
|
|
/* Check zeroing out all the bits. */
|
|
|
|
clear_bit_region_be (in, BITS_PER_UNIT - 1, 3 * BITS_PER_UNIT);
|
|
|
|
expected[0] = expected[1] = expected[2] = 0;
|
|
|
|
verify_array_eq (in, expected, sizeof in);
|
|
|
|
|
|
|
|
memcpy (in, orig, sizeof in);
|
|
|
|
/* Leave the first and last bits intact. */
|
|
|
|
clear_bit_region_be (in, BITS_PER_UNIT - 2, 3 * BITS_PER_UNIT - 2);
|
|
|
|
expected[0] = 0x80;
|
|
|
|
expected[1] = 0;
|
|
|
|
expected[2] = 0x1;
|
|
|
|
verify_array_eq (in, expected, sizeof in);
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
/* Run all of the selftests within this file. */
|
|
|
|
|
|
|
|
void
|
|
|
|
store_merging_c_tests (void)
|
|
|
|
{
|
|
|
|
verify_shift_bytes_in_array ();
|
|
|
|
verify_shift_bytes_in_array_right ();
|
|
|
|
verify_clear_bit_region ();
|
|
|
|
verify_clear_bit_region_be ();
|
|
|
|
}
|
|
|
|
|
|
|
|
} // namespace selftest
|
|
|
|
#endif /* CHECKING_P. */
|