* omp-low.c (lower_rec_input_clauses): For lastprivate conditional
references, lookup in in hash map MEM_REF operand instead of the
MEM_REF itself.
(lower_omp_1): When looking for lastprivate conditional assignments,
handle MEM_REFs with REFERENCE_TYPE operands.
* testsuite/libgomp.c++/lastprivate-conditional-1.C: New test.
* testsuite/libgomp.c++/lastprivate-conditional-2.C: New test.
From-SVN: r271948
* omp-low.c (lower_rec_input_clauses): Force max_vf if is_simd and
on privatization clauses OMP_CLAUSE_DECL is privatized by reference
and references a VLA. Handle references to non-VLAs if is_simd
all privatization clauses like reductions.
(lower_rec_input_clauses) <case do_private, case do_firstprivate>:
If omp_is_reference, use always omp simd arrays and set
DECL_VALUE_EXPR in that case, if lower_rec_simd_input_clauses
fails, emit reference initialization.
* g++.dg/vect/simd-1.cc: New test.
From-SVN: r271947
When converting a constant to interface, such as interface{}(42)
or interface{}("hello"), if the interface escapes, we currently
generate a heap allocation to hold the constant value.
This CL changes it to generate a static allocation instead, as
the gc compiler does. This reduces allocations in such cases.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/180277
From-SVN: r271945
This fixes some testcases that the last fifteen or so patches broke.
In all these cases we no longer need to set VSX_REG_ATTR: the default
value of "wa" is correct.
gcc/testsuite/
* gcc.target/powerpc/direct-move-double1.c (VSX_REG_ATTR): Delete.
* gcc.target/powerpc/direct-move-double2.c: Ditto.
* gcc.target/powerpc/direct-move-float1.c: Ditto.
* gcc.target/powerpc/direct-move-float2.c: Ditto.
* gcc.target/powerpc/direct-move-vint1.c: Ditto.
* gcc.target/powerpc/direct-move-vint2.c: Ditto.
From-SVN: r271940
wp becomes wa with isa p9tf, and wq is replaced by wa with isa p9kf.
To manage to do that, there is the new mode attribute VSisa.
* config/rs6000/constraints.md (define_register_constraint "wp"):
Delete.
(define_register_constraint "wq"): Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wp and RS6000_CONSTRAINT_wq.
* config/rs6000/vsx.md (define_mode_attr VSr3): Delete.
(define_mode_attr VSa): Delete.
(define_mode_attr VSisa): New.
(rest of file): Adjust.
* doc/md.texi (Machine Constraints): Adjust.
From-SVN: r271939
This adds "p9kf" and "p9tf" isa values, to be used for instruction
alternatives where KFmode resp. TFmode is used.
* config/rs6000/rs6000.md (define_attr "isa"): Add p9kf and p9tf.
(define_attr "enabled"): Handle those new isa values.
From-SVN: r271938
We don't need the <VSs> mode attribute, if we make <sd> work for V4SF
and V2DF just like for SF and DF.
* config/rs6000/rs6000.md (define_mode_attr sd): Add values for V4SF
and V2DF.
* config/rs6000/vsx.md (define_mode_attr VSs): Delete.
(rest of file): Adjust.
From-SVN: r271936
I should have factored this series better. Oh well. Near the end,
let's call it loose ends.
gcc/testsuite/
* gcc.target/powerpc/direct-move-float1.c: Use "wa" instead of "ww"
constraint.
From-SVN: r271935
This removes the <Ftrad>, <Fvsx>, and <Fs> mode attributes, and creates
new <sd> and <s> mode attributes instead. <sd> is either "s" or "d",
depending on whether the mode is single-precision or double-precision
floating point; and <s> is either "s" or nothing.
* config/rs6000/rs6000.md (SFDF, SFDF2): Adjust comments.
(define_mode_attr sd): New.
(define_mode_attr s): New.
(define_mode_attr Ftrad): Delete.
(define_mode_attr Fvsx): Delete.
(define_mode_attr Fs): Delete.
(rest of file): Use the new mode attributes.
* config.rs6000/vsx.md: Use the new mode attributes.
From-SVN: r271931
When used in VSX_B, VSX_D, or VSX_F, both <VSr> and <VSa> are always
just "wa" now. Similarly <VSr2> and <VSr3>. The former of those is
always "wa", so we can remove the mode attribute completely.
* config/rs6000/vsx.md (define_mode_attr VSr2): Delete.
(rest of file): Replace all <VSa>, <VSr>, <VSr2>, and <VSr3> that are
used with VSX_B, VSX_D, or VSX_F, with just "wa".
From-SVN: r271929
aarch64_asan_shadow_offset is using the wrong
offset for ILP32. Change it to be a decent one.
ChangeLog:
* config/aarch64/aarch64.c (aarch64_asan_shadow_offset):
Fix ILP32 value.
From-SVN: r271920
<VS_64reg> now always is "wa". Make that simplification.
* config/rs6000/vsx.md (define_mode_attr VS_64reg): Delete.
(*vsx_extract_<P:mode>_<VSX_D:mode>_load): Adjust.
(vsx_splat_<mode>_reg): Adjust.
From-SVN: r271917
"wv" is "v", but only if VSX is enabled (otherwise it's NO_REGS). So
this patch sets "isa" "p7v" to all alternatives that used "wv" before
(and that do not already need a later ISA), and changes the constraint.
* config/rs6000/constraints.md (define_register_constraint "wv"):
Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wv.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.
From-SVN: r271915
"wi" and "wt" mean just the same as "wa" these days. Change them to
the simpler name.
* config/rs6000/constraints.md (define_register_constraint "wi"):
Delete.
(define_register_constraint "wt"): Delete.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wi and RS6000_CONSTRAINT_wt.
* config/rs6000/rs6000.md: Adjust.
* config/rs6000/vsx.md: Adjust.
* doc/md.texi (Machine Constraints): Adjust.
From-SVN: r271914
Commit r271869 broke visibility declarations in asm for extern symbols, because
the new ASM_OUTPUT_EXTERNAL hook failed to call the default hook for elf.
gcc/ChangeLog:
* config/aarch64/aarch64-protos.h (aarch64_asm_output_external): Remove
const.
* config/aarch64/aarch64.c (aarch64_asm_output_external): Call
default_elf_asm_output_external.
From-SVN: r271913
For PR60531, GCC wrongly rejects function templates with explicitly
specified template arguments as overloaded. They are resolved by
resolve_nondeduced_context, which is normally called by
cp_default_conversion through decay_conversion, but the latter have
extra effects making them unusable here. Calling the former directly
does work.
* typeck.c (cp_build_binary_op): See if overload can be resolved.
(cp_build_unary_op): Ditto.
* g++.dg/template/operator15.C: New test.
From-SVN: r271910
We want to evaluate the arguments to a call before looking into the cache so
that we have constant values, but if we then find the call in the cache we
end up with a TREE_LIST that we don't end up using; in highly recursive
constexpr evaluation this ends up being a large proportion of the garbage
generated.
The cxx_eval_increment_expression hunk is less important, but it's an easy
tweak; we only use the MODIFY_EXPR to evaluate it, so after that it's
garbage.
* constexpr.c (cxx_eval_call_expression): ggc_free any bindings we
don't save.
(cxx_eval_increment_expression): ggc_free the MODIFY_EXPR after
evaluating it.
From-SVN: r271909
* gimplify.c (gimplify_scan_omp_clauses): Don't sorry_at on lastprivate
conditional on combined for simd.
* omp-low.c (struct omp_context): Add combined_into_simd_safelen0
member.
(lower_rec_input_clauses): For gimple_omp_for_combined_into_p max_vf 1
constructs, don't remove lastprivate_conditional_map, but instead set
ctx->combined_into_simd_safelen0 and adjust hash_map, so that it points
to parent construct temporaries.
(lower_lastprivate_clauses): Handle ctx->combined_into_simd_safelen0
like !ctx->lastprivate_conditional_map.
(lower_omp_1) <case GIMPLE_ASSIGN>: If up->combined_into_simd_safelen0,
use up->outer context instead of up.
* omp-expand.c (expand_omp_for_generic): Perform cond_var bump even if
gimple_omp_for_combined_p.
(expand_omp_for_static_nochunk): Likewise.
(expand_omp_for_static_chunk): Add forgotten cond_var bump that was
probably moved over into expand_omp_for_generic rather than being copied
there.
gcc/cp/
* cp-tree.h (CP_OMP_CLAUSE_INFO): Allow for any clauses up to _condvar_
instead of only up to linear.
gcc/testsuite/
* c-c++-common/gomp/lastprivate-conditional-2.c (foo): Don't expect
a sorry_at on any of the clauses.
libgomp/
* testsuite/libgomp.c-c++-common/lastprivate-conditional-7.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-8.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-9.c: New test.
* testsuite/libgomp.c-c++-common/lastprivate-conditional-10.c: New test.
From-SVN: r271907
2019-06-04 Richard Biener <rguenther@suse.de>
PR tree-optimization/90738
Revert
2019-06-03 Richard Biener <rguenther@suse.de>
* tree-ssa-sccvn.c (ao_ref_init_from_vn_reference): Get original
full reference tree and record in ref->ref.
(vn_reference_lookup_3): Pass in original ref to
ao_ref_init_from_vn_reference.
(vn_reference_lookup): Likewise.
* tree-ssa-sccvn.h (ao_ref_init_from_vn_reference): Adjust prototype.
* tree-ssa-alias.c (nonoverlapping_component_refs_of_decl_p):
Handle non-decl bases in the original reference.
* gcc.dg/tree-ssa/alias-access-path-1.c: Scan fre1.
* gcc.dg/torture/pr90738.c: New testcase.
From-SVN: r271902
2019-06-04 Martin Liska <mliska@suse.cz>
* ipa-icf.c (sem_item_optimizer::add_item_to_class): Count
number of references.
(sem_item_optimizer::do_congruence_step):
(sem_item_optimizer::worklist_push): Dump how references
a class has.
(sem_item_optimizer::worklist_pop): Use heap.
(sem_item_optimizer::process_cong_reduction): Likewise.
* ipa-icf.h: Use fibonacci_heap insteam of std::list.
From-SVN: r271901
2019-06-04 Martin Liska <mliska@suse.cz>
* ipa-icf.h (struct sem_usage_pair_hash): New.
(sem_usage_pair_hash::hash): Likewise.
(sem_usage_pair_hash::equal): Likewise.
(struct sem_usage_hash): Likewise.
* ipa-icf.c (sem_item::sem_item): Initialize
referenced_by_count.
(sem_item::add_reference): Register a reference
in ref_map and not in target->usages.
(sem_item::setup): Remove initialization of
dead vectors.
(sem_item::~sem_item): Remove usage of dead vectors.
(sem_item::dump): Remove dump of references.
(sem_item_optimizer::sem_item_optimizer): Initialize
m_references.
(sem_item_optimizer::read_section): Remove useless
dump.
(sem_item_optimizer::parse_funcs_and_vars): Likewise here.
(sem_item_optimizer::build_graph): Pass m_references
to ::add_reference.
(sem_item_optimizer::verify_classes): Remove usage of dead
vectors.
(sem_item_optimizer::traverse_congruence_split): Return true
when a class is split.
(sem_item_optimizer::do_congruence_step_for_index): Use
hash_map for look up of (sem_item *, index). That brings
significant speed up.
(sem_item_optimizer::do_congruence_step): Return true
when a split is done.
(congruence_class::is_class_used): Use referenced_by_count.
2019-06-04 Martin Liska <mliska@suse.cz>
* c-c++-common/goacc/acc-icf.c: Change scanned pattern.
* gfortran.dg/goacc/pr78027.f90: Likewise.
From-SVN: r271900
Currently, the compiler already generates common symbols for type
descriptors, so the type descriptors are unique. However, when a
type is created through reflection, it is not deduplicated with
compiler-generated types. As a consequence, we cannot assume type
descriptors are unique, and cannot use pointer equality to
compare them. Also, when constructing a reflect.Type, it has to
go through a canonicalization map, which introduces overhead to
reflect.TypeOf, and lock contentions in concurrent programs.
In order for the reflect package to deduplicate types with
compiler-created types, we register all the compiler-created type
descriptors at startup time. The reflect package, when it needs
to create a type, looks up the registry of compiler-created types
before creates a new one. There is no lock contention since the
registry is read-only after initialization.
This lets us get rid of the canonicalization map, and also makes
it possible to compare type descriptors with pointer equality.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179598
From-SVN: r271894
When the runtime collects a stack trace to associate it with some
profiling event (mem alloc, mutex, etc) there is a skip count passed
to runtime.Callers (or equivalent) to skip some known count of frames
in order to get to the "interesting" frame corresponding to the
profile event. Now that the profiling mechanism uses lazy fixup (when
removing compiler artifacts like thunks, morestack calls etc), we also
need to move the frame skipping logic after the fixup, so as to insure
that the skip count isn't thrown off by these artifacts.
Fixesgolang/go#32290.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/179740
From-SVN: r271892
This requires tracking all references to unexported variables, so that
we can make them global symbols in the object file, and can export
them so that other compilations can see the right definition for their
own inline bodies.
This introduces a syntax for referencing names defined in other
packages: a <pNN> prefix, where NN is the package index. This will
need to be added to gccgoimporter, but I didn't do it yet since it
isn't yet possible to create an object for which gccgoimporter will
see a <pNN> prefix.
This increases the number of inlinable functions in the standard
library from 181 to 215, adding functions like context.Background.
Reviewed-on: https://go-review.googlesource.com/c/gofrontend/+/177920
From-SVN: r271891
The "wg" constraint is used for the floating point side on mfpgpr
instructions. Those instructions do not exist on any relevant
hardware. This patch deletes the constraint and the insns using it.
* config/rs6000/constraints.md (define_register_constraint "wg"):
Delete.
* config/rs6000/rs6000.h (enum r6000_reg_class_enum): Delete
RS6000_CONSTRAINT_wg.
* config/rs6000/rs6000.c (rs6000_debug_reg_global): Adjust.
(rs6000_init_hard_regno_mode_ok): Adjust.
* config/rs6000/rs6000.md (*mov<mode>_softfloat32, *movdi_internal64):
Delete "wg" alternatives.
* doc/md.texi (Machine Constraints): Adjust.
From-SVN: r271888
