tree-ssa-loop-ch.c: New file.
* tree-ssa-loop-ch.c: New file. * Makefile.in (tree-ssa-loop-ch.o): Add. (tree-into-ssa.o): Add GGC_H dependency. * tree-cfg.c (tree_duplicate_bb): Copy virtual arguments. * tree-flow.h (rewrite_into_ssa): Declaration changed. (rewrite_ssa_into_ssa, compute_global_livein, duplicate_ssa_name): Declare. * tree-into-ssa.c: Include ggc.h. (struct def_blocks_d): Add phi_blocks field. (struct mark_def_sites_global_data): Add names_to_rename field. (struct ssa_name_info): New. (compute_global_livein): Export. (set_def_block, insert_phi_nodes, mark_def_sites, set_livein_block, insert_phi_nodes_1, rewrite_finalize_block, insert_phi_nodes_for, register_new_def, get_reaching_def, def_blocks_free, get_def_blocks_for, rewrite_into_ssa): Modified to work with rewrite_ssa_into_ssa. (get_ssa_name_ann, get_phi_state, set_phi_state, get_current_def, set_current_def, ssa_mark_def_sites_initialize_block, ssa_mark_phi_uses, ssa_mark_def_sites, duplicate_ssa_name, ssa_register_new_def, ssa_rewrite_initialize_block, ssa_rewrite_phi_arguments, ssa_rewrite_finalize_block, ssa_rewrite_stmt, rewrite_ssa_into_ssa, rewrite_all_into_ssa): New functions. (pass_build_ssa): Call rewrite_all_into_ssa. * tree-optimize.c (execute_todo, execute_one_pass, tree_rest_of_compilation): Allocate vars_to_rename only once. * tree-ssa-dom.c (tree_ssa_dominator_optimize): Provide parameter to rewrite_into_ssa. * tree-ssa-loop.c (should_duplicate_loop_header_p, mark_defs_for_rewrite, duplicate_blocks, do_while_loop_p, copy_loop_headers, gate_ch, pass_ch): Moved to tree-ssa-loop-ch.c. Use rewrite_ssa_into_ssa. * tree-ssa-operands.c (copy_virtual_operands): New function. * tree-ssa-operands.h (copy_virtual_operands): Declare. * tree.h (struct tree_ssa_name): Add aux field. (SSA_NAME_AUX): New macro to access it. From-SVN: r83932
This commit is contained in:
parent
43e7557b6e
commit
5f240ec46e
|
@ -1,3 +1,43 @@
|
|||
2004-06-30 Zdenek Dvorak <rakdver@atrey.karlin.mff.cuni.cz>
|
||||
|
||||
* tree-ssa-loop-ch.c: New file.
|
||||
* Makefile.in (tree-ssa-loop-ch.o): Add.
|
||||
(tree-into-ssa.o): Add GGC_H dependency.
|
||||
* tree-cfg.c (tree_duplicate_bb): Copy virtual arguments.
|
||||
* tree-flow.h (rewrite_into_ssa): Declaration changed.
|
||||
(rewrite_ssa_into_ssa, compute_global_livein, duplicate_ssa_name):
|
||||
Declare.
|
||||
* tree-into-ssa.c: Include ggc.h.
|
||||
(struct def_blocks_d): Add phi_blocks field.
|
||||
(struct mark_def_sites_global_data): Add names_to_rename field.
|
||||
(struct ssa_name_info): New.
|
||||
(compute_global_livein): Export.
|
||||
(set_def_block, insert_phi_nodes, mark_def_sites, set_livein_block,
|
||||
insert_phi_nodes_1, rewrite_finalize_block, insert_phi_nodes_for,
|
||||
register_new_def, get_reaching_def, def_blocks_free,
|
||||
get_def_blocks_for, rewrite_into_ssa): Modified to work with
|
||||
rewrite_ssa_into_ssa.
|
||||
(get_ssa_name_ann, get_phi_state, set_phi_state, get_current_def,
|
||||
set_current_def, ssa_mark_def_sites_initialize_block,
|
||||
ssa_mark_phi_uses, ssa_mark_def_sites, duplicate_ssa_name,
|
||||
ssa_register_new_def, ssa_rewrite_initialize_block,
|
||||
ssa_rewrite_phi_arguments, ssa_rewrite_finalize_block,
|
||||
ssa_rewrite_stmt, rewrite_ssa_into_ssa, rewrite_all_into_ssa): New
|
||||
functions.
|
||||
(pass_build_ssa): Call rewrite_all_into_ssa.
|
||||
* tree-optimize.c (execute_todo, execute_one_pass,
|
||||
tree_rest_of_compilation): Allocate vars_to_rename only once.
|
||||
* tree-ssa-dom.c (tree_ssa_dominator_optimize): Provide parameter
|
||||
to rewrite_into_ssa.
|
||||
* tree-ssa-loop.c (should_duplicate_loop_header_p,
|
||||
mark_defs_for_rewrite, duplicate_blocks, do_while_loop_p,
|
||||
copy_loop_headers, gate_ch, pass_ch): Moved to tree-ssa-loop-ch.c.
|
||||
Use rewrite_ssa_into_ssa.
|
||||
* tree-ssa-operands.c (copy_virtual_operands): New function.
|
||||
* tree-ssa-operands.h (copy_virtual_operands): Declare.
|
||||
* tree.h (struct tree_ssa_name): Add aux field.
|
||||
(SSA_NAME_AUX): New macro to access it.
|
||||
|
||||
2004-05-28 Aaron W. LaFramboise <aaronraolete36@aaronwl.com>
|
||||
|
||||
* prefix.c (lookup_key): Cast buffer to LPBYTE.
|
||||
|
|
|
@ -898,7 +898,7 @@ OBJS-common = \
|
|||
cfg.o cfganal.o cfgbuild.o cfgcleanup.o cfglayout.o cfgloop.o \
|
||||
cfgloopanal.o cfgloopmanip.o loop-init.o loop-unswitch.o loop-unroll.o \
|
||||
cfgrtl.o combine.o conflict.o convert.o coverage.o cse.o cselib.o \
|
||||
dbxout.o ddg.o loop-invariant.o \
|
||||
dbxout.o ddg.o tree-ssa-loop-ch.o loop-invariant.o \
|
||||
debug.o df.o diagnostic.o dojump.o dominance.o loop-doloop.o \
|
||||
dwarf2asm.o dwarf2out.o emit-rtl.o except.o explow.o loop-iv.o \
|
||||
expmed.o expr.o final.o flow.o fold-const.o function.o gcse.o \
|
||||
|
@ -1598,7 +1598,8 @@ tree-ssa.o : tree-ssa.c $(TREE_FLOW_H) $(CONFIG_H) $(SYSTEM_H) \
|
|||
tree-into-ssa.o : tree-into-ssa.c $(TREE_FLOW_H) $(CONFIG_H) $(SYSTEM_H) \
|
||||
$(RTL_H) $(TREE_H) $(TM_P_H) $(EXPR_H) output.h diagnostic.h \
|
||||
errors.h toplev.h function.h $(TIMEVAR_H) tree-alias-common.h \
|
||||
$(TM_H) coretypes.h $(TREE_DUMP_H) langhooks.h domwalk.h tree-pass.h
|
||||
$(TM_H) coretypes.h $(TREE_DUMP_H) langhooks.h domwalk.h tree-pass.h \
|
||||
$(GGC_H)
|
||||
tree-outof-ssa.o : tree-outof-ssa.c $(TREE_FLOW_H) $(CONFIG_H) $(SYSTEM_H) \
|
||||
$(RTL_H) $(TREE_H) $(TM_P_H) $(EXPR_H) output.h diagnostic.h \
|
||||
errors.h toplev.h function.h $(TIMEVAR_H) tree-alias-common.h \
|
||||
|
@ -1674,6 +1675,10 @@ tree-ssa-loop.o : tree-ssa-loop.c $(TREE_FLOW_H) $(CONFIG_H) \
|
|||
$(SYSTEM_H) $(RTL_H) $(TREE_H) $(TM_P_H) $(CFGLOOP_H) \
|
||||
output.h diagnostic.h $(TIMEVAR_H) $(TM_H) coretypes.h $(TREE_DUMP_H) \
|
||||
tree-pass.h $(FLAGS_H) tree-inline.h
|
||||
tree-ssa-loop-ch.o : tree-ssa-loop-ch.c $(TREE_FLOW_H) $(CONFIG_H) \
|
||||
$(SYSTEM_H) $(RTL_H) $(TREE_H) $(TM_P_H) $(CFGLOOP_H) tree-inline.h \
|
||||
output.h diagnostic.h $(TIMEVAR_H) $(TM_H) coretypes.h $(TREE_DUMP_H) \
|
||||
tree-pass.h flags.h
|
||||
tree-ssa-alias.o : tree-ssa-alias.c $(TREE_FLOW_H) $(CONFIG_H) $(SYSTEM_H) \
|
||||
$(RTL_H) $(TREE_H) $(TM_P_H) $(EXPR_H) $(GGC_H) tree-inline.h $(FLAGS_H) \
|
||||
function.h $(TIMEVAR_H) tree-alias-common.h convert.h $(TM_H) coretypes.h \
|
||||
|
|
|
@ -4283,11 +4283,18 @@ tree_duplicate_bb (basic_block bb)
|
|||
for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
|
||||
{
|
||||
tree stmt = bsi_stmt (bsi);
|
||||
tree copy;
|
||||
|
||||
if (TREE_CODE (stmt) == LABEL_EXPR)
|
||||
continue;
|
||||
|
||||
bsi_insert_after (&bsi_tgt, unshare_expr (stmt), BSI_NEW_STMT);
|
||||
copy = unshare_expr (stmt);
|
||||
|
||||
/* Copy also the virtual operands. */
|
||||
get_stmt_ann (copy);
|
||||
copy_virtual_operands (copy, stmt);
|
||||
|
||||
bsi_insert_after (&bsi_tgt, copy, BSI_NEW_STMT);
|
||||
}
|
||||
|
||||
return new_bb;
|
||||
|
|
|
@ -575,7 +575,11 @@ extern void register_new_def (tree, varray_type *);
|
|||
extern void walk_use_def_chains (tree, walk_use_def_chains_fn, void *);
|
||||
|
||||
/* In tree-into-ssa.c */
|
||||
extern void rewrite_into_ssa (void);
|
||||
extern void rewrite_into_ssa (bool);
|
||||
extern void rewrite_ssa_into_ssa (bitmap);
|
||||
|
||||
void compute_global_livein (bitmap, bitmap);
|
||||
tree duplicate_ssa_name (tree, tree);
|
||||
|
||||
/* In tree-ssa-ccp.c */
|
||||
bool fold_stmt (tree *);
|
||||
|
|
File diff suppressed because it is too large
Load Diff
|
@ -349,9 +349,8 @@ execute_todo (unsigned int flags)
|
|||
{
|
||||
if (flags & TODO_rename_vars)
|
||||
{
|
||||
if (bitmap_first_set_bit (vars_to_rename) >= 0)
|
||||
rewrite_into_ssa ();
|
||||
BITMAP_XFREE (vars_to_rename);
|
||||
rewrite_into_ssa (false);
|
||||
bitmap_clear (vars_to_rename);
|
||||
}
|
||||
|
||||
if ((flags & TODO_dump_func) && dump_file)
|
||||
|
@ -407,10 +406,6 @@ execute_one_pass (struct tree_opt_pass *pass)
|
|||
if (pass->tv_id)
|
||||
timevar_push (pass->tv_id);
|
||||
|
||||
/* If the pass is requesting ssa variable renaming, allocate the bitmap. */
|
||||
if (pass->todo_flags_finish & TODO_rename_vars)
|
||||
vars_to_rename = BITMAP_XMALLOC ();
|
||||
|
||||
/* Do it! */
|
||||
if (pass->execute)
|
||||
pass->execute ();
|
||||
|
@ -509,6 +504,9 @@ tree_rest_of_compilation (tree fndecl, bool nested_p)
|
|||
}
|
||||
}
|
||||
|
||||
if (!vars_to_rename)
|
||||
vars_to_rename = BITMAP_XMALLOC ();
|
||||
|
||||
/* If this is a nested function, protect the local variables in the stack
|
||||
above us from being collected while we're compiling this function. */
|
||||
if (nested_p)
|
||||
|
|
|
@ -631,7 +631,7 @@ tree_ssa_dominator_optimize (void)
|
|||
if (cfg_altered
|
||||
&& bitmap_first_set_bit (vars_to_rename) >= 0)
|
||||
{
|
||||
rewrite_into_ssa ();
|
||||
rewrite_into_ssa (false);
|
||||
bitmap_clear (vars_to_rename);
|
||||
|
||||
/* The into SSA translation may have created new SSA_NAMES whic
|
||||
|
|
|
@ -0,0 +1,349 @@
|
|||
/* Loop header copying on trees.
|
||||
Copyright (C) 2004 Free Software Foundation, Inc.
|
||||
|
||||
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 2, 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 COPYING. If not, write to the Free
|
||||
Software Foundation, 59 Temple Place - Suite 330, Boston, MA
|
||||
02111-1307, USA. */
|
||||
|
||||
#include "config.h"
|
||||
#include "system.h"
|
||||
#include "coretypes.h"
|
||||
#include "tm.h"
|
||||
#include "tree.h"
|
||||
#include "rtl.h"
|
||||
#include "tm_p.h"
|
||||
#include "hard-reg-set.h"
|
||||
#include "basic-block.h"
|
||||
#include "output.h"
|
||||
#include "diagnostic.h"
|
||||
#include "tree-flow.h"
|
||||
#include "tree-dump.h"
|
||||
#include "tree-pass.h"
|
||||
#include "timevar.h"
|
||||
#include "cfgloop.h"
|
||||
#include "tree-inline.h"
|
||||
#include "flags.h"
|
||||
#include "tree-inline.h"
|
||||
|
||||
/* Duplicates headers of loops if they are small enough, so that the statements
|
||||
in the loop body are always executed when the loop is entered. This
|
||||
increases effectivity of code motion optimizations, and reduces the need
|
||||
for loop preconditioning. */
|
||||
|
||||
/* Check whether we should duplicate HEADER of LOOP. At most *LIMIT
|
||||
instructions should be duplicated, limit is decreased by the actual
|
||||
amount. */
|
||||
|
||||
static bool
|
||||
should_duplicate_loop_header_p (basic_block header, struct loop *loop,
|
||||
int *limit)
|
||||
{
|
||||
block_stmt_iterator bsi;
|
||||
tree last;
|
||||
|
||||
/* Do not copy one block more than once (we do not really want to do
|
||||
loop peeling here). */
|
||||
if (header->aux)
|
||||
return false;
|
||||
|
||||
if (!header->succ)
|
||||
abort ();
|
||||
if (!header->succ->succ_next)
|
||||
return false;
|
||||
if (header->succ->succ_next->succ_next)
|
||||
return false;
|
||||
if (flow_bb_inside_loop_p (loop, header->succ->dest)
|
||||
&& flow_bb_inside_loop_p (loop, header->succ->succ_next->dest))
|
||||
return false;
|
||||
|
||||
/* If this is not the original loop header, we want it to have just
|
||||
one predecessor in order to match the && pattern. */
|
||||
if (header != loop->header
|
||||
&& header->pred->pred_next)
|
||||
return false;
|
||||
|
||||
last = last_stmt (header);
|
||||
if (TREE_CODE (last) != COND_EXPR)
|
||||
return false;
|
||||
|
||||
/* Approximately copy the conditions that used to be used in jump.c --
|
||||
at most 20 insns and no calls. */
|
||||
for (bsi = bsi_start (header); !bsi_end_p (bsi); bsi_next (&bsi))
|
||||
{
|
||||
last = bsi_stmt (bsi);
|
||||
|
||||
if (TREE_CODE (last) == LABEL_EXPR)
|
||||
continue;
|
||||
|
||||
if (get_call_expr_in (last))
|
||||
return false;
|
||||
|
||||
*limit -= estimate_num_insns (last);
|
||||
if (*limit < 0)
|
||||
return false;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/* Marks variables defined in basic block BB for rewriting. */
|
||||
|
||||
static void
|
||||
mark_defs_for_rewrite (basic_block bb)
|
||||
{
|
||||
tree stmt, var;
|
||||
block_stmt_iterator bsi;
|
||||
stmt_ann_t ann;
|
||||
def_optype defs;
|
||||
v_may_def_optype v_may_defs;
|
||||
v_must_def_optype v_must_defs;
|
||||
unsigned i;
|
||||
|
||||
for (stmt = phi_nodes (bb); stmt; stmt = TREE_CHAIN (stmt))
|
||||
{
|
||||
var = PHI_RESULT (stmt);
|
||||
bitmap_set_bit (vars_to_rename, SSA_NAME_VERSION (var));
|
||||
}
|
||||
|
||||
for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
|
||||
{
|
||||
stmt = bsi_stmt (bsi);
|
||||
get_stmt_operands (stmt);
|
||||
ann = stmt_ann (stmt);
|
||||
|
||||
defs = DEF_OPS (ann);
|
||||
for (i = 0; i < NUM_DEFS (defs); i++)
|
||||
{
|
||||
var = DEF_OP (defs, i);
|
||||
bitmap_set_bit (vars_to_rename, SSA_NAME_VERSION (var));
|
||||
}
|
||||
|
||||
v_may_defs = V_MAY_DEF_OPS (ann);
|
||||
for (i = 0; i < NUM_V_MAY_DEFS (v_may_defs); i++)
|
||||
{
|
||||
var = V_MAY_DEF_RESULT (v_may_defs, i);
|
||||
bitmap_set_bit (vars_to_rename, SSA_NAME_VERSION (var));
|
||||
}
|
||||
|
||||
v_must_defs = V_MUST_DEF_OPS (ann);
|
||||
for (i = 0; i < NUM_V_MUST_DEFS (v_must_defs); i++)
|
||||
{
|
||||
var = V_MUST_DEF_OP (v_must_defs, i);
|
||||
bitmap_set_bit (vars_to_rename, SSA_NAME_VERSION (var));
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* Duplicates destinations of edges in BBS_TO_DUPLICATE. */
|
||||
|
||||
static void
|
||||
duplicate_blocks (varray_type bbs_to_duplicate)
|
||||
{
|
||||
unsigned i;
|
||||
edge preheader_edge, e, e1;
|
||||
basic_block header, new_header;
|
||||
tree phi, new_phi, var;
|
||||
|
||||
/* TODO: It should be quite easy to keep the dominance information
|
||||
up-to-date. */
|
||||
free_dominance_info (CDI_DOMINATORS);
|
||||
|
||||
for (i = 0; i < VARRAY_ACTIVE_SIZE (bbs_to_duplicate); i++)
|
||||
{
|
||||
preheader_edge = VARRAY_GENERIC_PTR_NOGC (bbs_to_duplicate, i);
|
||||
header = preheader_edge->dest;
|
||||
|
||||
/* It is sufficient to rewrite the definitions, since the uses of
|
||||
the operands defined outside of the duplicated basic block are
|
||||
still valid (every basic block that dominates the original block
|
||||
also dominates the duplicate). */
|
||||
mark_defs_for_rewrite (header);
|
||||
}
|
||||
|
||||
for (i = 0; i < VARRAY_ACTIVE_SIZE (bbs_to_duplicate); i++)
|
||||
{
|
||||
preheader_edge = VARRAY_GENERIC_PTR_NOGC (bbs_to_duplicate, i);
|
||||
header = preheader_edge->dest;
|
||||
|
||||
if (!header->aux)
|
||||
abort ();
|
||||
header->aux = NULL;
|
||||
|
||||
new_header = duplicate_block (header, preheader_edge);
|
||||
|
||||
/* Create the phi nodes on on entry to new_header. */
|
||||
for (phi = phi_nodes (header), var = PENDING_STMT (preheader_edge);
|
||||
phi;
|
||||
phi = TREE_CHAIN (phi), var = TREE_CHAIN (var))
|
||||
{
|
||||
new_phi = create_phi_node (PHI_RESULT (phi), new_header);
|
||||
add_phi_arg (&new_phi, TREE_VALUE (var), preheader_edge);
|
||||
}
|
||||
PENDING_STMT (preheader_edge) = NULL;
|
||||
|
||||
/* Add the phi arguments to the outgoing edges. */
|
||||
for (e = header->succ; e; e = e->succ_next)
|
||||
{
|
||||
for (e1 = new_header->succ; e1->dest != e->dest; e1 = e1->succ_next)
|
||||
continue;
|
||||
|
||||
for (phi = phi_nodes (e->dest); phi; phi = TREE_CHAIN (phi))
|
||||
{
|
||||
tree def = PHI_ARG_DEF_FROM_EDGE (phi, e);
|
||||
add_phi_arg (&phi, def, e1);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
calculate_dominance_info (CDI_DOMINATORS);
|
||||
|
||||
rewrite_ssa_into_ssa (vars_to_rename);
|
||||
bitmap_clear (vars_to_rename);
|
||||
}
|
||||
|
||||
/* Checks whether LOOP is a do-while style loop. */
|
||||
|
||||
static bool
|
||||
do_while_loop_p (struct loop *loop)
|
||||
{
|
||||
tree stmt = last_stmt (loop->latch);
|
||||
|
||||
/* If the latch of the loop is not empty, it is not a do-while loop. */
|
||||
if (stmt
|
||||
&& TREE_CODE (stmt) != LABEL_EXPR)
|
||||
return false;
|
||||
|
||||
/* If the header contains just a condition, it is not a do-while loop. */
|
||||
stmt = last_and_only_stmt (loop->header);
|
||||
if (stmt
|
||||
&& TREE_CODE (stmt) == COND_EXPR)
|
||||
return false;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/* For all loops, copy the condition at the end of the loop body in front
|
||||
of the loop. This is beneficial since it increases efficiency of
|
||||
code motion optimizations. It also saves one jump on entry to the loop. */
|
||||
|
||||
static void
|
||||
copy_loop_headers (void)
|
||||
{
|
||||
struct loops *loops;
|
||||
unsigned i;
|
||||
struct loop *loop;
|
||||
basic_block header;
|
||||
edge preheader_edge;
|
||||
varray_type bbs_to_duplicate = NULL;
|
||||
|
||||
loops = loop_optimizer_init (dump_file);
|
||||
if (!loops)
|
||||
return;
|
||||
|
||||
/* We do not try to keep the information about irreducible regions
|
||||
up-to-date. */
|
||||
loops->state &= ~LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS;
|
||||
|
||||
#ifdef ENABLE_CHECKING
|
||||
verify_loop_structure (loops);
|
||||
#endif
|
||||
|
||||
for (i = 1; i < loops->num; i++)
|
||||
{
|
||||
/* Copy at most 20 insns. */
|
||||
int limit = 20;
|
||||
|
||||
loop = loops->parray[i];
|
||||
preheader_edge = loop_preheader_edge (loop);
|
||||
header = preheader_edge->dest;
|
||||
|
||||
/* If the loop is already a do-while style one (either because it was
|
||||
written as such, or because jump threading transformed it into one),
|
||||
we might be in fact peeling the first iteration of the loop. This
|
||||
in general is not a good idea. */
|
||||
if (do_while_loop_p (loop))
|
||||
continue;
|
||||
|
||||
/* Iterate the header copying up to limit; this takes care of the cases
|
||||
like while (a && b) {...}, where we want to have both of the conditions
|
||||
copied. TODO -- handle while (a || b) - like cases, by not requiring
|
||||
the header to have just a single successor and copying up to
|
||||
postdominator.
|
||||
|
||||
We do not really copy the blocks immediately, so that we do not have
|
||||
to worry about updating loop structures, and also so that we do not
|
||||
have to rewrite variables out of and into ssa form for each block.
|
||||
Instead we just record the block into worklist and duplicate all of
|
||||
them at once. */
|
||||
while (should_duplicate_loop_header_p (header, loop, &limit))
|
||||
{
|
||||
if (!bbs_to_duplicate)
|
||||
VARRAY_GENERIC_PTR_NOGC_INIT (bbs_to_duplicate, 10,
|
||||
"bbs_to_duplicate");
|
||||
VARRAY_PUSH_GENERIC_PTR_NOGC (bbs_to_duplicate, preheader_edge);
|
||||
header->aux = &header->aux;
|
||||
|
||||
if (dump_file && (dump_flags & TDF_DETAILS))
|
||||
fprintf (dump_file,
|
||||
"Scheduled basic block %d for duplication.\n",
|
||||
header->index);
|
||||
|
||||
/* Find a successor of header that is inside a loop; i.e. the new
|
||||
header after the condition is copied. */
|
||||
if (flow_bb_inside_loop_p (loop, header->succ->dest))
|
||||
preheader_edge = header->succ;
|
||||
else
|
||||
preheader_edge = header->succ->succ_next;
|
||||
header = preheader_edge->dest;
|
||||
}
|
||||
}
|
||||
|
||||
loop_optimizer_finalize (loops, NULL);
|
||||
|
||||
if (bbs_to_duplicate)
|
||||
{
|
||||
duplicate_blocks (bbs_to_duplicate);
|
||||
VARRAY_FREE (bbs_to_duplicate);
|
||||
}
|
||||
|
||||
/* Run cleanup_tree_cfg here regardless of whether we have done anything, so
|
||||
that we cleanup the blocks created in order to get the loops into a
|
||||
canonical shape. */
|
||||
cleanup_tree_cfg ();
|
||||
}
|
||||
|
||||
static bool
|
||||
gate_ch (void)
|
||||
{
|
||||
return flag_tree_ch != 0;
|
||||
}
|
||||
|
||||
struct tree_opt_pass pass_ch =
|
||||
{
|
||||
"ch", /* name */
|
||||
gate_ch, /* gate */
|
||||
copy_loop_headers, /* execute */
|
||||
NULL, /* sub */
|
||||
NULL, /* next */
|
||||
0, /* static_pass_number */
|
||||
TV_TREE_CH, /* tv_id */
|
||||
PROP_cfg | PROP_ssa, /* properties_required */
|
||||
0, /* properties_provided */
|
||||
0, /* properties_destroyed */
|
||||
0, /* todo_flags_start */
|
||||
(TODO_dump_func
|
||||
| TODO_verify_ssa) /* todo_flags_finish */
|
||||
};
|
|
@ -39,335 +39,3 @@ Software Foundation, 59 Temple Place - Suite 330, Boston, MA
|
|||
#include "tree-inline.h"
|
||||
|
||||
|
||||
/* Check whether we should duplicate HEADER of LOOP. At most *LIMIT
|
||||
instructions should be duplicated, limit is decreased by the actual
|
||||
amount. */
|
||||
|
||||
static bool
|
||||
should_duplicate_loop_header_p (basic_block header, struct loop *loop,
|
||||
int *limit)
|
||||
{
|
||||
block_stmt_iterator bsi;
|
||||
tree last;
|
||||
|
||||
/* Do not copy one block more than once (we do not really want to do
|
||||
loop peeling here). */
|
||||
if (header->aux)
|
||||
return false;
|
||||
|
||||
if (!header->succ)
|
||||
abort ();
|
||||
if (!header->succ->succ_next)
|
||||
return false;
|
||||
if (header->succ->succ_next->succ_next)
|
||||
return false;
|
||||
if (flow_bb_inside_loop_p (loop, header->succ->dest)
|
||||
&& flow_bb_inside_loop_p (loop, header->succ->succ_next->dest))
|
||||
return false;
|
||||
|
||||
/* If this is not the original loop header, we want it to have just
|
||||
one predecessor in order to match the && pattern. */
|
||||
if (header != loop->header
|
||||
&& header->pred->pred_next)
|
||||
return false;
|
||||
|
||||
last = last_stmt (header);
|
||||
if (TREE_CODE (last) != COND_EXPR)
|
||||
return false;
|
||||
|
||||
/* Approximately copy the conditions that used to be used in jump.c --
|
||||
at most 20 insns and no calls. */
|
||||
for (bsi = bsi_start (header); !bsi_end_p (bsi); bsi_next (&bsi))
|
||||
{
|
||||
last = bsi_stmt (bsi);
|
||||
|
||||
if (TREE_CODE (last) == LABEL_EXPR)
|
||||
continue;
|
||||
|
||||
if (get_call_expr_in (last))
|
||||
return false;
|
||||
|
||||
*limit -= estimate_num_insns (last);
|
||||
if (*limit < 0)
|
||||
return false;
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/* Marks variables defined in basic block BB for rewriting. */
|
||||
|
||||
static void
|
||||
mark_defs_for_rewrite (basic_block bb)
|
||||
{
|
||||
tree stmt, var;
|
||||
block_stmt_iterator bsi;
|
||||
stmt_ann_t ann;
|
||||
def_optype defs;
|
||||
v_may_def_optype v_may_defs;
|
||||
vuse_optype vuses;
|
||||
v_must_def_optype v_must_defs;
|
||||
unsigned i;
|
||||
|
||||
for (stmt = phi_nodes (bb); stmt; stmt = PHI_CHAIN (stmt))
|
||||
{
|
||||
var = SSA_NAME_VAR (PHI_RESULT (stmt));
|
||||
bitmap_set_bit (vars_to_rename, var_ann (var)->uid);
|
||||
|
||||
/* If we have a type_mem_tag, add it as well. Due to rewriting the
|
||||
variable out of ssa, we lose its name tag, so we use type_mem_tag
|
||||
instead. */
|
||||
var = var_ann (var)->type_mem_tag;
|
||||
if (var)
|
||||
bitmap_set_bit (vars_to_rename, var_ann (var)->uid);
|
||||
}
|
||||
|
||||
for (bsi = bsi_start (bb); !bsi_end_p (bsi); bsi_next (&bsi))
|
||||
{
|
||||
stmt = bsi_stmt (bsi);
|
||||
get_stmt_operands (stmt);
|
||||
ann = stmt_ann (stmt);
|
||||
|
||||
defs = DEF_OPS (ann);
|
||||
for (i = 0; i < NUM_DEFS (defs); i++)
|
||||
{
|
||||
var = SSA_NAME_VAR (DEF_OP (defs, i));
|
||||
bitmap_set_bit (vars_to_rename, var_ann (var)->uid);
|
||||
|
||||
/* If we have a type_mem_tag, add it as well. Due to rewriting the
|
||||
variable out of ssa, we lose its name tag, so we use type_mem_tag
|
||||
instead. */
|
||||
var = var_ann (var)->type_mem_tag;
|
||||
if (var)
|
||||
bitmap_set_bit (vars_to_rename, var_ann (var)->uid);
|
||||
}
|
||||
|
||||
v_may_defs = V_MAY_DEF_OPS (ann);
|
||||
for (i = 0; i < NUM_V_MAY_DEFS (v_may_defs); i++)
|
||||
{
|
||||
var = SSA_NAME_VAR (V_MAY_DEF_RESULT (v_may_defs, i));
|
||||
bitmap_set_bit (vars_to_rename, var_ann (var)->uid);
|
||||
}
|
||||
|
||||
v_must_defs = V_MUST_DEF_OPS (ann);
|
||||
for (i = 0; i < NUM_V_MUST_DEFS (v_must_defs); i++)
|
||||
{
|
||||
var = SSA_NAME_VAR (V_MUST_DEF_OP (v_must_defs, i));
|
||||
bitmap_set_bit (vars_to_rename, var_ann (var)->uid);
|
||||
}
|
||||
|
||||
/* We also need to rewrite vuses, since we will copy the statements
|
||||
and the ssa versions could not be recovered in the copy. We do
|
||||
not have to do this for operands of V_MAY_DEFS explicitly, since
|
||||
they have the same underlying variable as the results. */
|
||||
vuses = VUSE_OPS (ann);
|
||||
for (i = 0; i < NUM_VUSES (vuses); i++)
|
||||
{
|
||||
var = SSA_NAME_VAR (VUSE_OP (vuses, i));
|
||||
bitmap_set_bit (vars_to_rename, var_ann (var)->uid);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* Duplicates destinations of edges in BBS_TO_DUPLICATE. */
|
||||
|
||||
static void
|
||||
duplicate_blocks (varray_type bbs_to_duplicate)
|
||||
{
|
||||
unsigned i;
|
||||
edge preheader_edge, e, e1;
|
||||
basic_block header, new_header;
|
||||
tree phi;
|
||||
size_t old_num_referenced_vars = num_referenced_vars;
|
||||
|
||||
for (i = 0; i < VARRAY_ACTIVE_SIZE (bbs_to_duplicate); i++)
|
||||
{
|
||||
preheader_edge = VARRAY_GENERIC_PTR_NOGC (bbs_to_duplicate, i);
|
||||
header = preheader_edge->dest;
|
||||
|
||||
/* It is sufficient to rewrite the definitions, since the uses of
|
||||
the operands defined outside of the duplicated basic block are
|
||||
still valid (every basic block that dominates the original block
|
||||
also dominates the duplicate). */
|
||||
mark_defs_for_rewrite (header);
|
||||
}
|
||||
|
||||
rewrite_vars_out_of_ssa (vars_to_rename);
|
||||
|
||||
for (i = old_num_referenced_vars; i < num_referenced_vars; i++)
|
||||
{
|
||||
bitmap_set_bit (vars_to_rename, i);
|
||||
var_ann (referenced_var (i))->out_of_ssa_tag = 0;
|
||||
}
|
||||
|
||||
for (i = 0; i < VARRAY_ACTIVE_SIZE (bbs_to_duplicate); i++)
|
||||
{
|
||||
preheader_edge = VARRAY_GENERIC_PTR_NOGC (bbs_to_duplicate, i);
|
||||
header = preheader_edge->dest;
|
||||
|
||||
/* We might have split the edge into the loop header when we have
|
||||
eliminated the phi nodes, so find the edge to that we want to
|
||||
copy the header. */
|
||||
while (!header->aux)
|
||||
{
|
||||
preheader_edge = header->succ;
|
||||
header = preheader_edge->dest;
|
||||
}
|
||||
header->aux = NULL;
|
||||
|
||||
new_header = duplicate_block (header, preheader_edge);
|
||||
|
||||
/* Add the phi arguments to the outgoing edges. */
|
||||
for (e = header->succ; e; e = e->succ_next)
|
||||
{
|
||||
for (e1 = new_header->succ; e1->dest != e->dest; e1 = e1->succ_next)
|
||||
continue;
|
||||
|
||||
for (phi = phi_nodes (e->dest); phi; phi = PHI_CHAIN (phi))
|
||||
{
|
||||
tree def = PHI_ARG_DEF_FROM_EDGE (phi, e);
|
||||
add_phi_arg (&phi, def, e1);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
/* Checks whether LOOP is a do-while style loop. */
|
||||
|
||||
static bool
|
||||
do_while_loop_p (struct loop *loop)
|
||||
{
|
||||
tree stmt = last_stmt (loop->latch);
|
||||
|
||||
/* If the latch of the loop is not empty, it is not a do-while loop. */
|
||||
if (stmt
|
||||
&& TREE_CODE (stmt) != LABEL_EXPR)
|
||||
return false;
|
||||
|
||||
/* If the header contains just a condition, it is not a do-while loop. */
|
||||
stmt = last_and_only_stmt (loop->header);
|
||||
if (stmt
|
||||
&& TREE_CODE (stmt) == COND_EXPR)
|
||||
return false;
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
/* For all loops, copy the condition at the end of the loop body in front
|
||||
of the loop. This is beneficial since it increases effectivity of
|
||||
code motion optimizations. It also saves one jump on entry to the loop. */
|
||||
|
||||
static void
|
||||
copy_loop_headers (void)
|
||||
{
|
||||
struct loops *loops;
|
||||
unsigned i;
|
||||
struct loop *loop;
|
||||
basic_block header;
|
||||
edge preheader_edge;
|
||||
varray_type bbs_to_duplicate = NULL;
|
||||
|
||||
loops = loop_optimizer_init (dump_file);
|
||||
if (!loops)
|
||||
return;
|
||||
|
||||
/* We are not going to need or update dominators. */
|
||||
free_dominance_info (CDI_DOMINATORS);
|
||||
|
||||
create_preheaders (loops, CP_SIMPLE_PREHEADERS);
|
||||
|
||||
/* We do not try to keep the information about irreducible regions
|
||||
up-to-date. */
|
||||
loops->state &= ~LOOPS_HAVE_MARKED_IRREDUCIBLE_REGIONS;
|
||||
|
||||
#ifdef ENABLE_CHECKING
|
||||
verify_loop_structure (loops);
|
||||
#endif
|
||||
|
||||
for (i = 1; i < loops->num; i++)
|
||||
{
|
||||
/* Copy at most 20 insns. */
|
||||
int limit = 20;
|
||||
|
||||
loop = loops->parray[i];
|
||||
preheader_edge = loop_preheader_edge (loop);
|
||||
header = preheader_edge->dest;
|
||||
|
||||
/* If the loop is already a do-while style one (either because it was
|
||||
written as such, or because jump threading transformed it into one),
|
||||
we might be in fact peeling the first iteration of the loop. This
|
||||
in general is not a good idea. */
|
||||
if (do_while_loop_p (loop))
|
||||
continue;
|
||||
|
||||
/* Iterate the header copying up to limit; this takes care of the cases
|
||||
like while (a && b) {...}, where we want to have both of the conditions
|
||||
copied. TODO -- handle while (a || b) - like cases, by not requiring
|
||||
the header to have just a single successor and copying up to
|
||||
postdominator.
|
||||
|
||||
We do not really copy the blocks immediately, so that we do not have
|
||||
to worry about updating loop structures, and also so that we do not
|
||||
have to rewrite variables out of and into ssa form for each block.
|
||||
Instead we just record the block into worklist and duplicate all of
|
||||
them at once. */
|
||||
while (should_duplicate_loop_header_p (header, loop, &limit))
|
||||
{
|
||||
if (!bbs_to_duplicate)
|
||||
VARRAY_GENERIC_PTR_NOGC_INIT (bbs_to_duplicate, 10,
|
||||
"bbs_to_duplicate");
|
||||
VARRAY_PUSH_GENERIC_PTR_NOGC (bbs_to_duplicate, preheader_edge);
|
||||
header->aux = &header->aux;
|
||||
|
||||
if (dump_file && (dump_flags & TDF_DETAILS))
|
||||
fprintf (dump_file,
|
||||
"Scheduled basic block %d for duplication.\n",
|
||||
header->index);
|
||||
|
||||
/* Find a successor of header that is inside a loop; i.e. the new
|
||||
header after the condition is copied. */
|
||||
if (flow_bb_inside_loop_p (loop, header->succ->dest))
|
||||
preheader_edge = header->succ;
|
||||
else
|
||||
preheader_edge = header->succ->succ_next;
|
||||
header = preheader_edge->dest;
|
||||
}
|
||||
}
|
||||
|
||||
loop_optimizer_finalize (loops, NULL);
|
||||
|
||||
if (bbs_to_duplicate)
|
||||
{
|
||||
duplicate_blocks (bbs_to_duplicate);
|
||||
VARRAY_FREE (bbs_to_duplicate);
|
||||
}
|
||||
|
||||
/* Run cleanup_tree_cfg here regardless of whether we have done anything, so
|
||||
that we cleanup the blocks created in order to get the loops into a
|
||||
canonical shape. */
|
||||
cleanup_tree_cfg ();
|
||||
}
|
||||
|
||||
static bool
|
||||
gate_ch (void)
|
||||
{
|
||||
return flag_tree_ch != 0;
|
||||
}
|
||||
|
||||
struct tree_opt_pass pass_ch =
|
||||
{
|
||||
"ch", /* name */
|
||||
gate_ch, /* gate */
|
||||
copy_loop_headers, /* execute */
|
||||
NULL, /* sub */
|
||||
NULL, /* next */
|
||||
0, /* static_pass_number */
|
||||
TV_TREE_CH, /* tv_id */
|
||||
PROP_cfg | PROP_ssa, /* properties_required */
|
||||
0, /* properties_provided */
|
||||
0, /* properties_destroyed */
|
||||
0, /* todo_flags_start */
|
||||
(TODO_rename_vars
|
||||
| TODO_dump_func
|
||||
| TODO_verify_ssa) /* todo_flags_finish */
|
||||
};
|
||||
|
|
|
@ -1446,4 +1446,43 @@ add_call_read_ops (tree stmt, voperands_t prev_vops)
|
|||
}
|
||||
}
|
||||
|
||||
/* Copies virtual operands from SRC to DST. */
|
||||
|
||||
void
|
||||
copy_virtual_operands (tree dst, tree src)
|
||||
{
|
||||
vuse_optype vuses = STMT_VUSE_OPS (src);
|
||||
v_may_def_optype v_may_defs = STMT_V_MAY_DEF_OPS (src);
|
||||
v_must_def_optype v_must_defs = STMT_V_MUST_DEF_OPS (src);
|
||||
vuse_optype *vuses_new = &stmt_ann (dst)->vuse_ops;
|
||||
v_may_def_optype *v_may_defs_new = &stmt_ann (dst)->v_may_def_ops;
|
||||
v_must_def_optype *v_must_defs_new = &stmt_ann (dst)->v_must_def_ops;
|
||||
unsigned i;
|
||||
|
||||
if (vuses)
|
||||
{
|
||||
*vuses_new = allocate_vuse_optype (NUM_VUSES (vuses));
|
||||
for (i = 0; i < NUM_VUSES (vuses); i++)
|
||||
SET_VUSE_OP (*vuses_new, i, VUSE_OP (vuses, i));
|
||||
}
|
||||
|
||||
if (v_may_defs)
|
||||
{
|
||||
*v_may_defs_new = allocate_v_may_def_optype (NUM_V_MAY_DEFS (v_may_defs));
|
||||
for (i = 0; i < NUM_V_MAY_DEFS (v_may_defs); i++)
|
||||
{
|
||||
SET_V_MAY_DEF_OP (*v_may_defs_new, i, V_MAY_DEF_OP (v_may_defs, i));
|
||||
SET_V_MAY_DEF_RESULT (*v_may_defs_new, i,
|
||||
V_MAY_DEF_RESULT (v_may_defs, i));
|
||||
}
|
||||
}
|
||||
|
||||
if (v_must_defs)
|
||||
{
|
||||
*v_must_defs_new = allocate_v_must_def_optype (NUM_V_MUST_DEFS (v_must_defs));
|
||||
for (i = 0; i < NUM_V_MUST_DEFS (v_must_defs); i++)
|
||||
SET_V_MUST_DEF_OP (*v_must_defs_new, i, V_MUST_DEF_OP (v_must_defs, i));
|
||||
}
|
||||
}
|
||||
|
||||
#include "gt-tree-ssa-operands.h"
|
||||
|
|
|
@ -164,5 +164,6 @@ extern void get_stmt_operands (tree);
|
|||
extern void remove_vuses (tree);
|
||||
extern void remove_v_may_defs (tree);
|
||||
extern void remove_v_must_defs (tree);
|
||||
extern void copy_virtual_operands (tree, tree);
|
||||
|
||||
#endif /* GCC_TREE_SSA_OPERANDS_H */
|
||||
|
|
|
@ -1235,6 +1235,10 @@ struct tree_exp GTY(())
|
|||
#define SSA_NAME_VALUE(N) \
|
||||
SSA_NAME_CHECK (N)->ssa_name.value_handle
|
||||
|
||||
/* Auxiliary pass-specific data. */
|
||||
#define SSA_NAME_AUX(N) \
|
||||
SSA_NAME_CHECK (N)->ssa_name.aux
|
||||
|
||||
#ifndef _TREE_FLOW_H
|
||||
struct ptr_info_def;
|
||||
#endif
|
||||
|
@ -1254,6 +1258,9 @@ struct tree_ssa_name GTY(())
|
|||
|
||||
/* Value for SSA name used by GVN. */
|
||||
tree GTY((skip)) value_handle;
|
||||
|
||||
/* Auxiliary information stored with the ssa name. */
|
||||
PTR GTY((skip)) aux;
|
||||
};
|
||||
|
||||
/* In a PHI_NODE node. */
|
||||
|
|
Loading…
Reference in New Issue