With later patches, we're able to vectorise the epilogues of these tests
on AArch64 and so get two instances of "vectorizing stmts using SLP".
Although it would be possible with a bit of effort to predict when
this happens, it doesn't seem important whether we get 1 vs. 2
occurrences. All that matters is zero vs. nonzero.
2019-11-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/testsuite/
* gcc.dg/vect/slp-9.c: Use scan-tree-dump rather than
scan-tree-dump-times.
* gcc.dg/vect/slp-widen-mult-s16.c: Likewise.
* gcc.dg/vect/slp-widen-mult-u8.c: Likewise.
From-SVN: r277881
The number of iterations of an epilogue loop is always smaller than the
VF of the main loop. vect_analyze_loop_costing was taking this into
account when deciding whether the loop is cheap enough to vectorise,
but that has no effect with the unlimited cost model. We need to use
a separate check for correctness as well.
This can happen if the sizes returned by autovectorize_vector_sizes
happen to be out of order, e.g. because the target prefers smaller
vectors. It can also happen with later patches if two vectorisation
attempts happen to end up with the same VF.
2019-11-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-loop.c (vect_analyze_loop_2): When vectorizing an
epilogue loop, make sure that the VF is small enough or that
the epilogue loop can be fully-masked.
From-SVN: r277880
Once vect_analyze_loop has found a valid loop_vec_info X, we carry
on searching for alternatives if (1) X doesn't satisfy simdlen or
(2) we want to vectorize the epilogue of X. I have a patch that
optionally adds a third reason: we want to see if there are cheaper
alternatives to X.
This patch restructures vect_analyze_loop so that it's easier
to add more reasons for continuing. There's supposed to be no
behavioural change.
If we wanted to, we could allow vectorisation of epilogues once
loop->simdlen has been reached by changing "loop->simdlen" to
"simdlen" in the new vect_epilogues condition. That should be
a separate change though.
2019-11-06 Richard Sandiford <richard.sandiford@arm.com>
gcc/
* tree-vect-loop.c (vect_analyze_loop): Break out of the main
loop when we've finished, rather than returning directly from
the loop. Use a local variable to track whether we're still
searching for the preferred simdlen. Make vect_epilogues
record whether the next iteration should try to treat the
loop as an epilogue.
From-SVN: r277879
Currently for hard float we need to check for
__ARC_FPU_SP__ || __ARC_FPU_DP__ and for soft float inverse of that.
So define single convenience macros for either cases.
gcc/
xxxx-xx-xx Vineet Gupta <vgupta@synopsyscom>
* config/arc/arc-c.c (arc_cpu_cpp_builtins): Add
__arc_hard_float__, __ARC_HARD_FLOAT__,
__arc_soft_float__, __ARC_SOFT_FLOAT__
From-SVN: r277878
This was first submitted many years ago
https://gcc.gnu.org/ml/gcc-patches/2010-10/msg02468.html
The command line option -fcallgraph-info is added and makes the
compiler generate another output file (xxx.ci) for each compilation
unit (or LTO partitoin), which is a valid VCG file (you can launch
your favorite VCG viewer on it unmodified) and contains the "final"
callgraph of the unit. "final" is a bit of a misnomer as this is
actually the callgraph at RTL expansion time, but since most
high-level optimizations are done at the Tree level and RTL doesn't
usually fiddle with calls, it's final in almost all cases. Moreover,
the nodes can be decorated with additional info: -fcallgraph-info=su
adds stack usage info and -fcallgraph-info=da dynamic allocation info.
for gcc/ChangeLog
From Eric Botcazou <ebotcazou@adacore.com>, Alexandre Oliva <oliva@adacore.com>
* common.opt (-fcallgraph-info[=]): New option.
* doc/invoke.texi (Developer options): Document it.
* opts.c (common_handle_option): Handle it.
* builtins.c (expand_builtin_alloca): Record allocation if
-fcallgraph-info=da.
* calls.c (expand_call): If -fcallgraph-info, record the call.
(emit_library_call_value_1): Likewise.
* flag-types.h (enum callgraph_info_type): New type.
* explow.c: Include stringpool.h.
(set_stack_check_libfunc): Set SET_SYMBOL_REF_DECL on the symbol.
* function.c (allocate_stack_usage_info): New.
(allocate_struct_function): Call it for -fcallgraph-info.
(prepare_function_start): Call it otherwise.
(record_final_call, record_dynamic_alloc): New.
* function.h (struct callinfo_callee): New.
(CALLEE_FROM_CGRAPH_P): New.
(struct callinfo_dalloc): New.
(struct stack_usage): Add callees and dallocs.
(record_final_call, record_dynamic_alloc): Declare.
* gimplify.c (gimplify_decl_expr): Record dynamically-allocated
object if -fcallgraph-info=da.
* optabs-libfuncs.c (build_libfunc_function): Keep SYMBOL_REF_DECL.
* print-tree.h (print_decl_identifier): Declare.
(PRINT_DECL_ORIGIN, PRINT_DECL_NAME, PRINT_DECL_UNIQUE_NAME): New.
* print-tree.c: Include print-tree.h.
(print_decl_identifier): New function.
* toplev.c: Include print-tree.h.
(callgraph_info_file): New global variable.
(callgraph_info_external_printed): Likewise.
(output_stack_usage): Rename to...
(output_stack_usage_1): ... this. Make it static, add cf
parameter. If -fcallgraph-info=su, print stack usage to cf.
If -fstack-usage, use print_decl_identifier for
pretty-printing.
(INDIRECT_CALL_NAME): New.
(dump_final_node_vcg_start): New.
(dump_final_callee_vcg, dump_final_node_vcg): New.
(output_stack_usage): New.
(lang_dependent_init): Open and start file if
-fcallgraph-info. Allocated callgraph_info_external_printed.
(finalize): If callgraph_info_file is not null, finish it,
close it, and release callgraph_info_external_printed.
for gcc/ada/ChangeLog
* gcc-interface/misc.c (callgraph_info_file): Delete.
Co-Authored-By: Alexandre Oliva <oliva@adacore.com>
From-SVN: r277876
OpenACC (cf. OpenACC 2.7, section 2.9.11. "reduction clause";
this was first clarified by OpenACC 2.6) requires that, if a
variable is used in reduction clauses on two nested loops, then
there must be reduction clauses for that variable on all loops
that are nested in between the two loops and all these reduction
clauses must use the same operator.
This commit introduces a check for that property which reports
warnings if it is violated.
2019-11-06 Gergö Barany <gergo@codesourcery.com>
Frederik Harwath <frederik@codesourcery.com>
Thomas Schwinge <thomas@codesourcery.com>
gcc/
* omp-low.c (struct omp_context): New fields
local_reduction_clauses, outer_reduction_clauses.
(new_omp_context): Initialize these.
(scan_sharing_clauses): Record reduction clauses on OpenACC constructs.
(scan_omp_for): Check reduction clauses for incorrect nesting.
gcc/testsuite/
* c-c++-common/goacc/nested-reductions-warn.c: New test.
* c-c++-common/goacc/nested-reductions.c: New test.
* gfortran.dg/goacc/nested-reductions-warn.f90: New test.
* gfortran.dg/goacc/nested-reductions.f90: New test.
libgomp/
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-1.c:
Add expected warnings about missing reduction clauses.
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-2.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-3.c:
Likewise.
* testsuite/libgomp.oacc-c-c++-common/par-loop-comb-reduction-4.c:
Likewise.
Reviewed-by: Thomas Schwinge <thomas@codesourcery.com>
From-SVN: r277875
-finline-functions is enabled by default for O2 since r276469, update the
test cases with -fno-inline-functions.
c11-atomic-exec-5.c stills hit ICE of LRA on BE systems in PR92090.
This commit is NOT a fix for the bug and so it must NOT be closed.
gcc/testsuite/ChangeLog:
2019-11-06 Xiong Hu Luo <luoxhu@linux.ibm.com>
PR92090
* gcc.target/powerpc/pr72804.c: Add -fno-inline-functions --param
max-inline-insns-single-O2=200.
* gcc.target/powerpc/pr79439-1.c: Add -fno-inline-functions.
* gcc.target/powerpc/vsx-builtin-7.c: Likewise.
From-SVN: r277872
The combine pass is perfectly happy if a splitter splits to just one
instruction (instead of two).
* doc/md.texi (Insn Splitting): Fix combiner documentation.
From-SVN: r277866
There are three major pieces to this support: scalar operator<=>,
synthesis of comparison operators, and rewritten/reversed overload
resolution (e.g. a < b becomes 0 > b <=> a).
Unlike other defaulted functions, where we use synthesized_method_walk to
semi-simulate what the definition of the function will be like, this patch
determines the characteristics of a comparison operator by trying to define
it.
My handling of non-dependent rewritten operators in templates can still use
some work: build_min_non_dep_op_overload can't understand the rewrites and
crashes, so I'm avoiding it for now by clearing *overload. This means we'll
do name lookup again at instantiation time, which can incorrectly mean a
different result. I'll poke at this more in stage 3.
I'm leaving out a fourth section ("strong structural equality") even though
I've implemented it, because it seems likely to change radically tomorrow.
Thanks to Tim van Deurzen and Jakub for implementing lexing of the <=>
operator, and Jonathan for the initial <compare> header.
gcc/cp/
* cp-tree.h (struct lang_decl_fn): Add maybe_deleted bitfield.
(DECL_MAYBE_DELETED): New.
(enum special_function_kind): Add sfk_comparison.
(LOOKUP_REWRITTEN, LOOKUP_REVERSED): New.
* call.c (struct z_candidate): Add rewritten and reversed methods.
(add_builtin_candidate): Handle SPACESHIP_EXPR.
(add_builtin_candidates): Likewise.
(add_candidates): Don't add a reversed candidate if the parms are
the same.
(add_operator_candidates): Split out from build_new_op_1. Handle
rewritten and reversed candidates.
(add_candidate): Swap conversions of reversed candidate.
(build_new_op_1): Swap them back. Build a second operation for
rewritten candidates.
(extract_call_expr): Handle rewritten calls.
(same_fn_or_template): New.
(joust): Handle rewritten and reversed candidates.
* class.c (add_implicitly_declared_members): Add implicit op==.
(classtype_has_op, classtype_has_defaulted_op): New.
* constexpr.c (cxx_eval_binary_expression): Handle SPACESHIP_EXPR.
(cxx_eval_constant_expression, potential_constant_expression_1):
Likewise.
* cp-gimplify.c (genericize_spaceship): New.
(cp_genericize_r): Use it.
* cp-objcp-common.c (cp_common_init_ts): Handle SPACESHIP_EXPR.
* decl.c (finish_function): Handle deleted function.
* decl2.c (grokfield): SET_DECL_FRIEND_CONTEXT on defaulted friend.
(mark_used): Check DECL_MAYBE_DELETED. Remove assumption that
defaulted functions are non-static members.
* error.c (dump_expr): Handle SPACESHIP_EXPR.
* method.c (type_has_trivial_fn): False for sfk_comparison.
(enum comp_cat_tag, struct comp_cat_info_t): New types.
(comp_cat_cache): New array variable.
(lookup_comparison_result, lookup_comparison_category)
(is_cat, cat_tag_for, spaceship_comp_cat)
(spaceship_type, genericize_spaceship)
(common_comparison_type, early_check_defaulted_comparison)
(comp_info, build_comparison_op): New.
(synthesize_method): Handle sfk_comparison. Handle deleted.
(get_defaulted_eh_spec, maybe_explain_implicit_delete)
(explain_implicit_non_constexpr, implicitly_declare_fn)
(defaulted_late_check, defaultable_fn_check): Handle sfk_comparison.
* name-lookup.c (get_std_name_hint): Add comparison categories.
* tree.c (special_function_p): Add sfk_comparison.
* typeck.c (cp_build_binary_op): Handle SPACESHIP_EXPR.
2019-11-05 Tim van Deurzen <tim@kompiler.org>
Add new tree code for the spaceship operator.
gcc/cp/
* cp-tree.def: Add new tree code.
* operators.def: New binary operator.
* parser.c: Add new token and tree code.
libcpp/
* cpplib.h: Add spaceship operator for C++.
* lex.c: Implement conditional lexing of spaceship operator for C++20.
2019-11-05 Jonathan Wakely <jwakely@redhat.com>
libstdc++-v3/
* libsupc++/compare: New header.
* libsupc++/Makefile.am (std_HEADERS): Add compare.
* include/std/version: Define __cpp_lib_three_way_comparison.
* include/std/functional: #include <compare>.
From-SVN: r277865
While working on C++20 operator<=>, I noticed that build_new_op_1 was doing
too much conversion when a built-in candidate was selected; the standard
says it should only perform user-defined conversions, and then leave the
normal operator semantics to handle any standard conversions. This is
important for operator<=> because a comparison of two different unscoped
enums is ill-formed; if we promote the enums to int here, cp_build_binary_op
never gets to see the original operand types, so we can't give the error.
I'm also disabling -Wmaybe-uninitialized for expmed.c to avoid the bootstrap
failure from the last time I applied this patch.
* call.c (build_new_op_1): Don't apply any standard conversions to
the operands of a built-in operator. Don't suppress conversions in
cp_build_unary_op.
* typeck.c (cp_build_unary_op): Do integral promotions for enums.
PR tree-optimization/91825
* expmed.c: Reduce -Wmaybe-uninitialized to warning.
From-SVN: r277864
My operator<=> patch wants to split up build_new_op_1, which makes using a
tree array as well as the vec inconvenient. build_new_op_1 already has a
vec, and build_conditional_expr_1 can release its vec right away, so this
doesn't increase garbage at all.
* call.c (build_builtin_candidate): Take args in a vec.
(add_builtin_candidate, add_builtin_candidates): Likewise.
(build_conditional_expr_1, build_new_op_1): Adjust.
From-SVN: r277863
Wrappers for lookup_qualified_name and build_x_binary_op to make calling
them more convenient in places, and a function named contextual_conv_bool
for places that want contextual conversion to bool.
I noticed that we weren't showing the declaration location when we complain
about a call to a non-constexpr function where a constant expression is
required.
If maybe_instantiate_noexcept doesn't actually instantiate, there's no
reason for it to mess with clones.
* constexpr.c (explain_invalid_constexpr_fn): Show location of fn.
* pt.c (maybe_instantiate_noexcept): Only update clones if we
instantiated.
* typeck.c (contextual_conv_bool): New.
* name-lookup.c (lookup_qualified_name): Add wrapper overload taking
C string rather than identifier.
* parser.c (cp_parser_userdef_numeric_literal): Use it.
* rtti.c (emit_support_tinfos): Use it.
* cp-tree.h (ovl_op_identifier): Change to inline functions.
(build_x_binary_op): Add wrapper with fewer parms.
From-SVN: r277862
The RISC-V backend wants to use a libcall when optimizing for size if
more than 6 instructions are needed. Emit_move_complex asks for no
libcalls. This case requires 8 insns for rv64 and 16 insns for rv32,
so we get fallback code that emits a loop. Commit_one_edge_insertion
doesn't allow code inserted for a phi node on an edge to end with a
branch, and so this triggers an assertion. This problem goes away if
we allow libcalls when optimizing for size, which gives the code the
RISC-V backend wants, and avoids triggering the assert.
gcc/
PR middle-end/92263
* expr.c (emit_move_complex): Only use BLOCK_OP_NO_LIBCALL when
optimize_insn_for_speed_p is true.
gcc/testsuite/
PR middle-end/92263
* gcc.dg/pr92263.c: New.
From-SVN: r277861
While looking at CA378 I noticed that we weren't properly diagnosing two of
the three erroneous lines in the example.
* decl2.c (mark_used): Diagnose use of a function with unsatisfied
constraints here.
* typeck.c (cp_build_function_call_vec): Not here.
From-SVN: r277860
PR middle-end/92333 - missing variable name referencing VLA in warnings
PR middle-end/82608 - missing -Warray-bounds on an out-of-bounds VLA index
gcc/testsuite/ChangeLog:
PR middle-end/92333
PR middle-end/82608
* gcc.dg/Warray-bounds-51.c: New test.
gcc/ChangeLog:
PR middle-end/92333
PR middle-end/82608
* tree-vrp.c (vrp_prop::check_array_ref): Handle VLAs with constant
size.
* tree-ssa-ccp.c (fold_builtin_alloca_with_align): Use a meaninful
name and location for a temporary variable.
From-SVN: r277854
https://gcc.gnu.org/ml/gcc-patches/2019-11/msg00323.html
cp/
PR c++/92370
* parser.c (cp_parser_error_1): Check EOF and UNKNOWN_LOCATION
when skipping over version control marker.
testsuite/
PR c++/92370
* g++.dg/pr92370.C: New.
From-SVN: r277853
PR middle-end/92341 - missing -Warray-bounds indexing past the end of a compound literal
PR middle-end/82612 - missing -Warray-bounds on a non-zero offset from the address of a non-array object
gcc/testsuite/ChangeLog:
PR middle-end/92341
PR middle-end/82612
* g++.dg/warn/Warray-bounds-4.C: Adjust text of expected warning.
* gcc.dg/Warray-bounds-53.c: New test.
* gcc.dg/Warray-bounds-54.c: New test.
gcc/ChangeLog:
PR middle-end/92341
PR middle-end/82612
* tree-sra.c (get_access_for_expr): Fail for out-of-bounds offsets.
* tree-vrp.c (vrp_prop::check_array_ref): Correct index and text
of message printed in a warning for empty arrays.
(vrp_prop::check_mem_ref): Also handle function parameters and
empty arrays.
From-SVN: r277851
2019-11-05 Richard Biener <rguenther@suse.de>
PR tree-optimization/92371
* tree-vect-loop.c (vectorizable_reduction): Set STMT_VINFO_REDUC_DEF
on the original stmt of live stmts in the chain.
(vectorizable_live_operation): Look at the original stmt when
checking STMT_VINFO_REDUC_DEF.
* gcc.dg/torture/pr92371.c: New testcase.
From-SVN: r277850
The negative concept is required for subsumption to work, it's not a
bug.
* include/bits/iterator_concepts.h (__iter_without_nested_types):
Remove incorrect comment.
From-SVN: r277848
The base class for ranges is currently value_range_base, which is
rather long and cumbersome. It also occurs more often than the derived
class of value_range. To avoid confusion, and save typing, this
patch does a global rename from value_range to value_range_equiv,
and from value_range_base to value_range.
This way, the base class is simply value_range, and the derived
class is value_range_equiv which explicitly states what it does.
From-SVN: r277847
`build_personality_function` generates a declaration for a personality
function. The type it declares for these functions doesn't match the
type of the actual personality functions that are defined by the C++
unwinding ABI.
This doesn't cause any crashes since the compiler never generates a call
to these decl's, and hence the type of the function is never used.
Nonetheless, for the sake of consistency and readability we update the
type of this declaration.
gcc/ChangeLog:
2019-11-05 Matthew Malcomson <matthew.malcomson@arm.com>
* expr.c (build_personality_function): Fix generated type to
match actual personality functions.
From-SVN: r277846
-fsanitize=hwaddress found a one-byte overwrite when running the
testsuite here. aarch64_handle_attr_branch_protection allocates
`strlen(str)` bytes for an error string, which is populated by
`strcpy(..., str)` in the case where the branch protection string is
completely invalid.
Not tested -- I don't want to re-build and it seems obvious.
gcc/ChangeLog:
2019-11-05 Matthew Malcomson <matthew.malcomson@arm.com>
* config/aarch64/aarch64.c (aarch64_handle_attr_cpu): Allocate
enough bytes for the NULL character.
From-SVN: r277845
2019-11-05 Martin Liska <mliska@suse.cz>
* ubsan/ubsan_flags.cpp (InitializeFlags): Trunk decided to print
summary for all sanitizers, but we want to have UBSAN without it.
From-SVN: r277837
These tests check if loop peeling has been applied to avoid
having to vectorize unaligned loops. On Z we do not have any
alignment requirements for vectorization so we also don't need want
the loop peeling here.
2019-11-05 Andreas Krebbel <krebbel@linux.ibm.com>
* gcc.dg/tree-ssa/gen-vect-26.c: Disable loop peeling check for
IBM Z.
* gcc.dg/tree-ssa/gen-vect-28.c: Likewise.
From-SVN: r277833
2019-11-05 Richard Biener <rguenther@suse.de>
PR tree-optimization/92280
* match.pd (BIT_FIELD_REF of CTOR): Unless the original CTOR
had a single use do not create a new CTOR.
* tree-ssa-forwprop.c (simplify_bitfield_ref): Do not re-fold
BIT_FIELD_REF of a CTOR via GENERIC.
From-SVN: r277832
In the Z backend we still set min-vect-loop-bound to 2 to work around
corner cases where awkward epilogue code gets generated in the
vectorizer. This has a particular bad impact when vectorizing loops
with a low iteration count. Due to this we do not vectorize the loop
in gen-vect-11/32 - what actually is a pity.
The patch sets min-vect-loop-bound back to the default value of 0 in
order to enable vectorization.
2019-11-05 Andreas Krebbel <krebbel@linux.ibm.com>
* gcc.dg/tree-ssa/gen-vect-11.c: Add --param min-vect-loop-bound=0
for IBM Z.
* gcc.dg/tree-ssa/gen-vect-23.c: Likewise.
From-SVN: r277831
This fixes various issues with the useable_hw check in s390.exp. The
check is supposed to verify whether a testcase can be run on the
current hardware.
- the test never returned true for -m31 because vzero is not available
in ESA mode and -m31 defaults to -mesa
- the missing v0 clobber on the vzero instruction made the check fail
if the stack pointer got saved in f0
- the lcbb instruction used for checking whether we are on a z13
also requires vx. Replace it with an instruction from the generic
instruction set extensions.
- no support for z14 and z15 so far
gcc/testsuite/ChangeLog:
2019-11-05 Andreas Krebbel <krebbel@linux.ibm.com>
* gcc.target/s390/s390.exp
(check_effective_target_s390_useable_hw): Add inline asm for z14
and z15. Replace instruction for z13 with lochiz. Add register
clobbers. Check also for __zarch__ when doing the __VX__ test.
From-SVN: r277830
This fixes an ICE in gcc.dg/attr-vector_size.c testcase.
gcc/ChangeLog:
2019-11-05 Andreas Krebbel <krebbel@linux.ibm.com>
* config/s390/s390.c (s390_vector_alignment): Check if the value
fits into uhwi before using it.
From-SVN: r277829
Here is the one-liner that fixes the incorrect
vec_perm cost for thunderx2t99 chip.
With the patch applied 526.blender of CPU2017
gets ~5% improvement with no measurable changes
for other benchmarks.
2019-11-05 Anton Youdkevitch <anton.youdkevitch@bell-sw.com>
* config/aarch64/aarch64.c (thunderx2t99_vector_cost):
Change vec_perm field to 10.
From-SVN: r277826
2019-11-05 Richard Biener <rguenther@suse.de>
PR tree-optimization/92324
* tree-vect-loop.c (check_reduction_path): For MIN/MAX require
all signed or unsigned operations.
* gcc.dg/vect/pr92324-3.c: New testcase.
From-SVN: r277822
