platform/upstream/gcc.git
2 years agotestsuite: Skip gcc.dg/Wno-frame-address.c for PRU
Dimitar Dimitrov [Sun, 3 Apr 2022 13:11:27 +0000 (16:11 +0300)]
testsuite: Skip gcc.dg/Wno-frame-address.c for PRU

Access to arbitrary stack frames is not supported on PRU.

gcc/testsuite/ChangeLog:

* gcc.dg/Wno-frame-address.c: Skip for PRU target.

Signed-off-by: Dimitar Dimitrov <dimitar@dinux.eu>
2 years agoPR tree-optimization/102950: Improved EVRP for signed BIT_XOR_EXPR.
Roger Sayle [Tue, 3 May 2022 18:38:50 +0000 (14:38 -0400)]
PR tree-optimization/102950: Improved EVRP for signed BIT_XOR_EXPR.

This patch fixes PR tree-optimization/102950, which is a P2 regression,
by providing better range bounds for BIT_XOR_EXPR, BIT_AND_EXPR and
BIT_IOR_EXPR on signed integer types.  In general terms, any binary
bitwise operation on sign-extended or zero-extended integer types will
produce results that are themselves sign-extended or zero-extended.
More precisely, we can derive signed bounds from the number of leading
redundant sign bit copies, from the equation:
clrsb(X op Y) >= min (clrsb (X), clrsb(Y))
and from the property that for any (signed or unsigned) range [lb, ub]
that clrsb([lb, ub]) >= min (clrsb(lb), clrsb(ub)).

These can be used to show that [-1, 0] op [-1, 0] is [-1, 0] or that
[-128, 127] op [-128, 127] is [-128, 127], even when tracking nonzero
bits would result in VARYING (as every bit can be 0 or 1).  This is
equivalent to determining the minimum type precision in which the
operation can be performed then sign extending the result.

One additional refinement is to observe that X ^ Y can never be
zero if the ranges of X and Y don't overlap, i.e. X can't be equal
to Y.

Previously, the expression "(int)(char)a ^ 233" in the PR was considered
VARYING, but with the above changes now has the range [-256, -1][1, 255],
which is sufficient to optimize away the call to foo.

2022-05-03  Roger Sayle  <roger@nextmovesoftware.com>

gcc/ChangeLog
PR tree-optimization/102950
* range-op.cc (wi_optimize_signed_bitwise_op): New function to
determine bounds of bitwise operations on signed types.
(operator_bitwise_and::wi_fold): Call the above function.
(operator_bitwise_or::wi_fold): Likewise.
(operator_bitwise_xor::wi_fold): Likewise.  Additionally, the
result can't be zero if the operands can't be equal.

gcc/testsuite/ChangeLog
PR tree-optimization/102950
* gcc.dg/pr102950.c: New test case.
* gcc.dg/tree-ssa/evrp10.c: New test case.

2 years agoObjective-C, NeXT: Adjust symbol marking to match host tools.
Iain Sandoe [Mon, 2 May 2022 18:42:49 +0000 (19:42 +0100)]
Objective-C, NeXT: Adjust symbol marking to match host tools.

Current host tools mark some additional symbols as 'no dead strip' and also
expose one additional group to the linker.  This does not affect older Darwin
versions or x86_64, but omitting these changes results in link errors for
aarch64.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
gcc/ChangeLog:

* config/darwin.cc (darwin_label_is_anonymous_local_objc_name): Make
protocol class methods linker-visible.

gcc/objc/ChangeLog:

* objc-next-runtime-abi-02.cc (next_runtime_abi_02_protocol_decl): Do
not dead-strip the runtime meta-data symbols.
(build_v2_classrefs_table): Likewise.
(build_v2_protocol_list_address_table): Likewise.

2 years ago[PR105324] libstdc++: testsuite: pr105324 requires FP from_char
Alexandre Oliva [Tue, 3 May 2022 17:57:15 +0000 (14:57 -0300)]
[PR105324] libstdc++: testsuite: pr105324 requires FP from_char

The floating-point overloads of from_char are only declared if
_GLIBCXX_HAVE_USELOCALE is #defined as nonzero.  That's exposed from
charconv as __cpp_lib_to_chars >= 201611L, so guard the test body with
that.

for  libstdc++-v3/ChangeLog

PR c++/105324
* testsuite/20_util/from_chars/pr105324.cc: Guard test body
with conditional for floating-point overloads of from_char.

2 years agoi386: Optimize _mm_storeu_si16 w/o SSE4 [PR105079]
Uros Bizjak [Tue, 3 May 2022 15:59:40 +0000 (17:59 +0200)]
i386: Optimize _mm_storeu_si16 w/o SSE4 [PR105079]

Optimize _mm_storeu_si16 to use MOVD from a SSE to an integer register
instead of PEXTRW from a low word of the SSE register to an integer reg.

Avoid the transformation when optimizing for size for targets without
TARGET_INTER_UNIT_MOVES_FROM_VEC capability, where the transformation
results in two moves via a memory location.

2022-05-03  Uroš Bizjak  <ubizjak@gmail.com>

gcc/ChangeLog:

PR target/105079
* config/i386/sse.md (*vec_extract<mode>_0_mem): New pre-reload
define_insn_and_split pattern.

gcc/testsuite/ChangeLog:

PR target/105079
* gcc.target/i386/pr105079.c: New test.
* gcc.target/i386/pr95483-1.c (dg-options): Use -msse4.1.

2 years agodfp.m4: Add missing license header
Christophe Lyon [Fri, 29 Apr 2022 13:11:11 +0000 (14:11 +0100)]
dfp.m4: Add missing license header

It seems the license header was omitted when dfp.m4 was originally
contributed in 2010 (commit 3c39bca6bbb5, r0-102573 or svn r163815.

This copies the license from libdecnumber/configure.ac since dfp.m4
was originally extracted from that file.

2022-04-29  Christophe Lyon  <christophe.lyon@arm.com>

config/
* dfp.m4: Add license header.

2 years agomiddle-end/105083 - check prerequesites in scev_initialize
Richard Biener [Tue, 3 May 2022 10:48:43 +0000 (12:48 +0200)]
middle-end/105083 - check prerequesites in scev_initialize

2022-05-03  Richard Biener  <rguenther@suse.de>

PR middle-end/105083
* tree-scalar-evolution.cc (scev_initialize): Verify we
have appropriate loop state.
* tree-ssa-dce.cc (perform_tree_ssa_dce): Re-order SCEV and
loop init and finalization.

2 years agomiddle-end/105461 - opts processing of -fvar-tracking
Richard Biener [Tue, 3 May 2022 08:36:30 +0000 (10:36 +0200)]
middle-end/105461 - opts processing of -fvar-tracking

The flag_var_tracking reset in finish_options doesn't match the
condition in process_options, in particular we fail to reset it
when the option was specified on the command line.  The following
fixes this and also alters the debug info level guard to match
the one in process_options.

2022-05-03  Richard Biener  <rguenther@suse.de>

PR middle-end/105461
* opts.cc (finish_options): Match the condition to
disable flag_var_tracking to that of process_options.

* gcc.dg/pr105461.c: New testcase.

2 years agoFixup OPTION_SET_P usage in finish_options
Richard Biener [Tue, 3 May 2022 08:23:51 +0000 (10:23 +0200)]
Fixup OPTION_SET_P usage in finish_options

When some code was moved from process_options to finish_options,
uses of OPTION_SET_P were not replaced with references to the
opts_set option set.  The following fixes this.

2022-05-03  Richard Biener  <rguenther@suse.de>

* opts.cc: #undef OPTIONS_SET_P.
(finish_options): Use opts_set instead of OPTIONS_SET_P.

2 years agotree-optimization/105394 - vector lowering of compares
Richard Biener [Fri, 29 Apr 2022 09:54:45 +0000 (11:54 +0200)]
tree-optimization/105394 - vector lowering of compares

The following fixes missing handling of non-integer mode but
masked (SVE or MVE) compares in vector lowering by using the
appropriate mask element width to extract the components and
adjust the index.

2022-04-29  Richard Biener  <rguenther@suse.de>

PR tree-optimization/105394
* tree-vect-generic.cc (expand_vector_condition): Adjust
comp_width for non-integer mode masks as well.

2 years agoOpenMP, libgomp: Add new runtime routine omp_get_mapped_ptr.
Marcel Vollweiler [Tue, 3 May 2022 06:56:44 +0000 (23:56 -0700)]
OpenMP, libgomp: Add new runtime routine omp_get_mapped_ptr.

This patch adds the OpenMP runtime routine "omp_get_mapped_ptr" which was
introduced in OpenMP 5.1.

gcc/ChangeLog:

* omp-low.cc (omp_runtime_api_call): Added get_mapped_ptr to
omp_runtime_apis array.

libgomp/ChangeLog:

* libgomp.map: Added omp_get_mapped_ptr.
* libgomp.texi: Tagged omp_get_mapped_ptr as supported.
* omp.h.in: Added omp_get_mapped_ptr.
* omp_lib.f90.in: Added interface for omp_get_mapped_ptr.
* omp_lib.h.in: Likewise.
* target.c (omp_get_mapped_ptr): Added implementation of
omp_get_mapped_ptr.
* testsuite/libgomp.c-c++-common/get-mapped-ptr-1.c: New test.
* testsuite/libgomp.c-c++-common/get-mapped-ptr-2.c: New test.
* testsuite/libgomp.c-c++-common/get-mapped-ptr-3.c: New test.
* testsuite/libgomp.c-c++-common/get-mapped-ptr-4.c: New test.
* testsuite/libgomp.fortran/get-mapped-ptr-1.f90: New test.
* testsuite/libgomp.fortran/get-mapped-ptr-2.f90: New test.
* testsuite/libgomp.fortran/get-mapped-ptr-3.f90: New test.
* testsuite/libgomp.fortran/get-mapped-ptr-4.f90: New test.

2 years agotestsuite: vect: update unaligned message
Alexandre Oliva [Wed, 27 Apr 2022 21:23:01 +0000 (18:23 -0300)]
testsuite: vect: update unaligned message

gcc.dg/vect/costmodel/ppc/costmodel-vect-31a.c covers ppc variants
that accept and reject misaligned accesses.  The message that it
expects for rejection was removed in the gcc-11 development cycle by
commit r11-1969.  The patch adjusted multiple tests to use the message
introduced in r11-1945, but missed this one.

for  gcc/testsuite/ChangeLog

* gcc.dg/vect/costmodel/ppc/costmodel-vect-31a.c: Update
the expected message for the case in which unaligned accesses
are not allowed.

2 years agoDaily bump.
GCC Administrator [Tue, 3 May 2022 00:16:25 +0000 (00:16 +0000)]
Daily bump.

2 years agoc++: improve template-id location
Jason Merrill [Sat, 30 Apr 2022 09:45:02 +0000 (05:45 -0400)]
c++: improve template-id location

On PR102629 I noticed that we were giving the entire lambda as the location
for this template-id.

gcc/cp/ChangeLog:

* pt.cc (tsubst_copy_and_build) [TEMPLATE_ID_EXPR]: Copy location.
(do_auto_deduction): Use expr location.

gcc/testsuite/ChangeLog:

* g++.dg/cpp2a/lambda-pack-init7.C: Check column number.

2 years agoc++: also note name used in enclosing class
Jason Merrill [Thu, 31 Mar 2022 20:18:58 +0000 (16:18 -0400)]
c++: also note name used in enclosing class

While looking at PR96645 I noticed that while we were diagnosing names
changing meaning in the full class context, we weren't doing this for
lookups in nested class bodies.

Note that this breaks current range-v3; I've submitted a pull request to fix
its violation of the rule.

gcc/cp/ChangeLog:

* class.cc (maybe_note_name_used_in_class): Note in all enclosing
classes.  Remember location of use.
(note_name_declared_in_class): Adjust.

gcc/testsuite/ChangeLog:

* g++.dg/lookup/name-clash13.C: New test.
* g++.dg/lookup/name-clash14.C: New test.
* g++.dg/lookup/name-clash15.C: New test.
* g++.dg/lookup/name-clash16.C: New test.

2 years agox86: Add missing .note.GNU-stack to assembly source
H.J. Lu [Fri, 29 Apr 2022 17:26:24 +0000 (10:26 -0700)]
x86: Add missing .note.GNU-stack to assembly source

Add .note.GNU-stack assembly source to avoid linker warning:

ld: warning: /tmp/ccPZSZ7Z.o: missing .note.GNU-stack section implies executable stack
ld: NOTE: This behaviour is deprecated and will be removed in a future version of the linker
FAIL: gcc.target/i386/iamcu/test_3_element_struct_and_unions.c compilation,  -O0

PR testsuite/105433
* gcc.target/i386/iamcu/asm-support.S: Add .note.GNU-stack.
* gcc.target/x86_64/abi/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512f/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512fp16/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512fp16/m256h/asm-support.S: Likewise.
* gcc.target/x86_64/abi/avx512fp16/m512h/asm-support.S: Likewise.
* gcc.target/x86_64/abi/ms-sysv/do-test.S: Likewise.

2 years agoFortran: Add location info to OpenMP tree nodes
Sandra Loosemore [Fri, 25 Mar 2022 04:02:34 +0000 (21:02 -0700)]
Fortran: Add location info to OpenMP tree nodes

gcc/fortran/
* trans-openmp.cc (gfc_trans_omp_critical): Set location on OMP
tree node.
(gfc_trans_omp_do): Likewise.
(gfc_trans_omp_masked): Likewise.
(gfc_trans_omp_do_simd): Likewise.
(gfc_trans_omp_scope): Likewise.
(gfc_trans_omp_taskgroup): Likewise.
(gfc_trans_omp_taskwait): Likewise.
(gfc_trans_omp_distribute): Likewise.
(gfc_trans_omp_taskloop): Likewise.
(gfc_trans_omp_master_masked_taskloop): Likewise.

2 years agoc++: uses_template_parms cleanups
Marek Polacek [Wed, 27 Apr 2022 22:36:29 +0000 (18:36 -0400)]
c++: uses_template_parms cleanups

gcc/cp/ChangeLog:

* cp-tree.h (uses_template_parms): Adjust declaration.
* pt.cc (uses_template_parms): Return bool.  Use a RAII sentinel.

2 years agoAssume a call is expensive when it mismatches
Richard Biener [Wed, 13 Apr 2022 13:43:01 +0000 (15:43 +0200)]
Assume a call is expensive when it mismatches

This makes sure to not consider calls to builtin decls with
mismatching arguments as inexpensive.

2022-04-13  Richard Biener  <rguenther@suse.de>

* tree-scalar-evolution.cc (expression_expensive_p):
Never consider mismatched calls as cheap.

2 years agotree-optimization/104240 - SLP discovery with swapped comparisons
Richard Biener [Mon, 2 May 2022 10:01:51 +0000 (12:01 +0200)]
tree-optimization/104240 - SLP discovery with swapped comparisons

The following extends SLP discovery to handle swapped operands
in comparisons.

2022-05-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/104240
* tree-vect-slp.cc (op1_op0_map): New.
(vect_get_operand_map): Handle compares.
(vect_build_slp_tree_1): Support swapped operands for
tcc_comparison.

* gcc.dg/vect/bb-slp-pr104240.c: New testcase.

2 years agolibstdc++: Don't use std::tolower in <charconv> [PR103911]
Patrick Palka [Mon, 2 May 2022 11:01:33 +0000 (07:01 -0400)]
libstdc++: Don't use std::tolower in <charconv> [PR103911]

As with std::isdigit in r12-6281-gc83ecfbe74a5cf, we shouldn't be using
std::tolower in <charconv> either.

PR libstdc++/103911

libstdc++-v3/ChangeLog:

* src/c++17/floating_from_chars.cc (find_end_of_float): Accept
two delimeters for the exponent part in the form of a possibly
NULL string of length two.  Don't use std::tolower.
(pattern): Adjust calls to find_end_of_float accordingly.

2 years agolibstdc++: case-sensitivity in hexfloat std::from_chars [PR105441]
Patrick Palka [Mon, 2 May 2022 11:00:48 +0000 (07:00 -0400)]
libstdc++: case-sensitivity in hexfloat std::from_chars [PR105441]

The hexfloat parser for binary32/64 added in r12-6645-gcc3bf3404e4b1c
overlooked that the exponent part can also begin with an uppercase 'P'.

PR libstdc++/105441

libstdc++-v3/ChangeLog:

* src/c++17/floating_from_chars.cc (__floating_from_chars_hex):
Also accept 'P' as the start of the exponent.
* testsuite/20_util/from_chars/7.cc: Add corresponding testcase.

2 years agoexpand: Throw away non-external decls without varpool nodes [PR105415]
Jakub Jelinek [Mon, 2 May 2022 09:30:58 +0000 (11:30 +0200)]
expand: Throw away non-external decls without varpool nodes [PR105415]

The following testcase fails -fcompare-debug on aarch64-linux.  The problem
is that for the n variable we create a varpool node, then remove it again
because the var isn't really used, but it keeps being referenced in debug
stmts/insns with -g.  Later during sched1 pass we ask whether the n var
can be modified through some store to an anchored variable and with -g
we create a new varpool node for it again just so that we can find its
ultimate alias target.  Even later on we create some cgraph node for the
loop parallelization, but as there has been an extra varpool node creation
in between, we get higher node->order with -g than without.

The patch fixes that by throwing variables without varpool nodes away
during expansion time, they are very unlikely to actually end up with
useful debug info anyway.

I've bootstrapped/regtested the following on x86_64-linux and i686-linux,
then bootstrapped with the patch reverted, reapplied the patch and did make
cc1plus in stage3.  The debug section sizes are identical, .debug_info and
.debug_loc is identical too, so I think we don't lose any debug info through
it.
So at least on cc1plus it makes no difference.

2022-05-02  Jakub Jelinek  <jakub@redhat.com>

PR debug/105415
* cfgexpand.cc (expand_debug_expr): Don't make_decl_rtl_for_debug
if there is no symtab node for the VAR_DECL.

* gcc.dg/pr105415.c: New test.

2 years agogcov: Fix first time gcov info dump
Sebastian Huber [Mon, 2 May 2022 06:44:16 +0000 (08:44 +0200)]
gcov: Fix first time gcov info dump

This patch fixes an issue introduced by commit
ef9a53feae5701953da9161afef2aea0329ec8b2:

gcc --coverage main.c && ./a.out
libgcov profiling error:a-main.gcda:Error writing

gcc/ChangeLog:

* gcov-io.cc (gcov_rewrite):  Clear the file error status.

2 years agotree-optimization/105437 - BB vect with extern defs of throwing stmts
Richard Biener [Mon, 2 May 2022 07:30:27 +0000 (09:30 +0200)]
tree-optimization/105437 - BB vect with extern defs of throwing stmts

We have to watch out for vectorized stmt insert locations if the
def from the last stmt alters control flow.  We constrain region
building so we know the def is outside of the current region
and thus we can insert at the region start point.

2022-05-02  Richard Biener  <rguenther@suse.de>

PR tree-optimization/105437
* tree-vect-slp.cc (vect_schedule_slp_node): Handle the
case where last_stmt alters control flow.

* g++.dg/vect/pr105437.cc: New testcase.

2 years agoUse CASE_CONVERT in a few more cases
Richard Biener [Wed, 12 Jan 2022 09:42:32 +0000 (10:42 +0100)]
Use CASE_CONVERT in a few more cases

This uses CASE_CONVERT more which eases eventual removal of NOP_EXPR.

2022-04-29  Richard Biener  <rguenther@suse.de>

gcc/cp/
* constexpr.cc (fold_simple_1): Use CASE_CONVERT.
* cp-gimplify.cc (cp_fold): Likewise.
* pt.cc (tsubst_copy): Likewise.

gcc/
* dojump.cc (do_jump): Use CASE_CONVERT.
* tree-ssa-dom.cc (edge_info::derive_equivalences): Likewise.

2 years agosystem.h: Include <initializer_list> in system.h unconditionally
Jakub Jelinek [Mon, 2 May 2022 05:10:14 +0000 (07:10 +0200)]
system.h: Include <initializer_list> in system.h unconditionally

On Sun, May 01, 2022 at 07:06:53PM +0100, Jonathan Wakely wrote:
> > >> the reason is that "gcc/analyzer/region-model.cc” uses initializer_lists, and it seems that <initializer_list>
> > >> is not transitively included by any used headers for _LIBCPP_VERSION < 4000.  I fixed that locally by
> > >> adding initializer_list into system.h (and adding INCLUDE_INITIALIZER_LIST to the top of gcc/analyzer/region-model.cc)
> > >> - with that change those versions do bootstrap and test OK***
> > >
> > > From what I can see, with libstdc++ it works because <utility> which is
> > > included by system.h includes <initializer_list>.
> > > If I rename initializer_list in analyzer/region-model.ii to initializer_listx, I
> > > also get:
> > > ../../gcc/analyzer/region-model.cc: In function ‘void ana::selftest::test_binop_svalue_folding()’:
> > > ../../gcc/analyzer/region-model.cc:4966:48: error: deducing from brace-enclosed initializer list requires ‘#include <initializer_list>’
> > > 4508 |
> > >  +++ |+#include <initializer_list>
> > > 4509 | static void
> > > ......
> > > 4966 |     for (auto op : {BIT_IOR_EXPR, TRUTH_OR_EXPR})
> > >      |                                                ^
> > > ../../gcc/analyzer/region-model.cc:4978:49: error: deducing from brace-enclosed initializer list requires ‘#include <initializer_list>’
> > > 4978 |     for (auto op : {BIT_AND_EXPR, TRUTH_AND_EXPR})
> > >      |                                                 ^
> > >
> > > I think we have 2 options, one is do what you wrote above,
> > > INCLUDE_INITIALIZER_LIST defined before system.h to get #include <initializer_list>.
> > > The other option is just to include that unconditionally, it is a very small
> > > header.  For libstdc++ it will make no difference as it is included anyway
> > > and the header is really small there, libc++ includes <cstddef> which isn't
> > > normally included and system.h includes <stddef.h> instead.
> >
> > I’d say unconditionally would be OK. I suppose the chance that any host
> > C++ is good enough to build GCC as-is but fails to provide
> > <initializer_list> is zero?
> >
>
> Yes, definitely.
>
> > I’d be OK to do this change without a new RC even.

2022-05-02  Jakub Jelinek  <jakub@redhat.com>

* system.h: Include initializer_list.

2 years agolibgcov: add ATTRIBUTE_UNUSED for dump_string
Martin Liska [Mon, 2 May 2022 04:46:05 +0000 (06:46 +0200)]
libgcov: add ATTRIBUTE_UNUSED for dump_string

Mitigates the following clang warning:
libgcc/libgcov-driver.c:416:1: warning: unused function 'dump_string' [-Wunused-function]

libgcc/ChangeLog:

* libgcov-driver.c: Add ATTRIBUTE_UNUSED.

2 years agoDaily bump.
GCC Administrator [Mon, 2 May 2022 00:16:24 +0000 (00:16 +0000)]
Daily bump.

2 years agors6000: "Y" is DS-form, not DQ-form
Segher Boessenkool [Tue, 26 Apr 2022 21:30:44 +0000 (21:30 +0000)]
rs6000: "Y" is DS-form, not DQ-form

2022-05-01  Segher Boessenkool  <segher@kernel.crashing.org>

* config/rs6000/constraints.md (Y constraint): Fix comment.

2 years agoDenormalize VR_VARYING to VR_RANGE before passing it to set_range_info_raw.
Aldy Hernandez [Fri, 29 Apr 2022 20:46:25 +0000 (22:46 +0200)]
Denormalize VR_VARYING to VR_RANGE before passing it to set_range_info_raw.

We are ICEing in set_range_info_raw because value_range_kind cannot be
VR_VARYING, since SSA_NAME_RANGE_TYPE can only hold VR_RANGE /
VR_ANTI_RANGE.  Most of the time setting a VR_VARYING as a global
range makes no sense.  However, we can have a range spanning the
entire domain (VR_RANGE of [MIN,MAX] which is essentially a
VR_VARYING), if the nonzero bits are set.

This was working before because set_range_info_raw allows setting
VR_RANGE of [MIN, MAX].  However, when going through an irange, we
normalize this to a VR_VARYING, thus causing the ICE.  It's
interesting that other calls to set_range_info with an irange haven't
triggered this.

One solution would be to just ignore VR_VARYING and bail, since
set_range_info* is really an update of the current range semantic
wise.  After all, we keep the nonzero bits which provide additional
info.  But this would be a change in behavior, so not suitable until
after GCC 12 is released.  So in order to keep with current behavior
we can just denormalize the varying to VR_RANGE.

Tested on x86-64 Linux.

    PR tree-optimization/105432

gcc/ChangeLog:

* tree-ssanames.cc (set_range_info): Denormalize VR_VARYING to
VR_RANGE before passing a piecewise range to set_range_info_raw.

2 years agoDaily bump.
GCC Administrator [Sun, 1 May 2022 00:16:18 +0000 (00:16 +0000)]
Daily bump.

2 years agogengtype: remove "tree_exp" special attribute
Patrick Palka [Sat, 30 Apr 2022 14:56:49 +0000 (10:56 -0400)]
gengtype: remove "tree_exp" special attribute

The function comment for adjust_field_tree_exp says this attribute is
for handling expression trees whose operands may contain pointers to RTL
instead of to trees.  But ever since r0-59671-gac45df5dba5804, which
fixed/removed the last two tree codes for which this was possible
(WITH_CLEANUP_EXPR and GOTO_SUBROUTINE_EXPR), this special attribute is
mostly a no-op.

This patch removes it and instead just annotates struct tree_exp
with the "length" attribute directly.  Not sure it makes a difference,
but I use %h instead of %0 in the attribute string to be consistent with
the other uses of the "length" attribute within tree-core.h.

This changes the code generated for TS_EXP handling in gt-cp-tree.h from:

  case TS_EXP:
    gt_ggc_m_9tree_node ((*x).generic.exp.typed.type);
    switch ((int) (TREE_CODE ((tree) &(*x))))
      {
      default:
{
  size_t i3;
  size_t l3 = (size_t)(TREE_OPERAND_LENGTH ((tree) &(*x)));
  for (i3 = 0; i3 != l3; i3++) {
    gt_ggc_m_9tree_node ((*x).generic.exp.operands[i3]);
  }
}
break;
      }
    break;

to:

  case TS_EXP:
    {
      size_t l3 = (size_t)(TREE_OPERAND_LENGTH ((tree)&((*x).generic.exp)));
      gt_ggc_m_9tree_node ((*x).generic.exp.typed.type);
      {
size_t i3;
for (i3 = 0; i3 != l3; i3++) {
  gt_ggc_m_9tree_node ((*x).generic.exp.operands[i3]);
}
      }
    }

which seems equivalent and simpler.

gcc/ChangeLog:

* gengtype.cc (adjust_field_tree_exp): Remove.
(adjust_field_type): Don't handle the "tree_exp" special attribute.
* tree-core.h (struct tree_exp): Remove "special" and "desc"
attributes.  Add "length" attribute.

2 years agoc-family: attribute ((aligned, mode)) [PR100545]
Jason Merrill [Fri, 15 Apr 2022 20:27:05 +0000 (16:27 -0400)]
c-family: attribute ((aligned, mode)) [PR100545]

The problem here was that handle_mode_attribute clobbered the changes of any
previous attribute, only copying type qualifiers to the new type.  And
common_handle_aligned_attribute had previously set up the typedef, so when
we later called set_underlying_type it saw DECL_ORIGINAL_TYPE set and just
returned, even though handle_mode_attribute had messed up the TREE_TYPE.
So, let's fix handle_mode_attribute to copy attributes, alignment, and
typedefness to the new type.

PR c/100545

gcc/c-family/ChangeLog:

* c-attribs.cc (handle_mode_attribute): Copy attributes, aligned,
and typedef.
* c-common.cc (set_underlying_type): Add assert.

gcc/testsuite/ChangeLog:

* c-c++-common/attr-mode-1.c: New test.
* c-c++-common/attr-mode-2.c: New test.

2 years agoDaily bump.
GCC Administrator [Sat, 30 Apr 2022 00:16:27 +0000 (00:16 +0000)]
Daily bump.

2 years agoc++: reorganize friend template matching [PR91618]
Jason Merrill [Fri, 8 Apr 2022 17:48:25 +0000 (13:48 -0400)]
c++: reorganize friend template matching [PR91618]

The the different calling of check_explicit_specialization for class and
namespace scope friends bothered me, so this patch combines them.

PR c++/91618
PR c++/96604

gcc/cp/ChangeLog:

* friend.cc (do_friend): Call check_explicit_specialization here.
* decl.cc (grokdeclarator): Not here.
* decl2.cc (check_classfn): Or here.

2 years agoc++: alias CTAD and member alias templates [PR104470]
Jason Merrill [Wed, 27 Apr 2022 15:13:24 +0000 (11:13 -0400)]
c++: alias CTAD and member alias templates [PR104470]

In this testcase, we were trying to substitute into
variant<Foo<T>>::__accepted_type, but failed to look it up because
variant<Foo<T>> doesn't exist.  In other cases we already rewrite such
things into a dependent reference; we need to do that for alias templates as
well.

This caused some testsuite regressions on alias uses outside of deduction
guides, so I've made all of this rewriting conditional on a new tf_dguide
tsubst flag.

PR c++/104470

gcc/cp/ChangeLog:

* cp-tree.h (enum tsubst_flags): Add tf_dguide.
* pt.cc (tsubst_aggr_type): Check it.
(tsubst_baselink, tsubst_copy): Check it.
(maybe_dependent_member_ref): Check it.
(instantiate_alias_template): Handle it.
(build_deduction_guide): Set it.

gcc/testsuite/ChangeLog:

* g++.dg/cpp2a/explicit11.C: Second example also ill-formed.
* g++.dg/cpp2a/class-deduction-alias12.C: New test.

2 years agoc++: lambda capture dependent type [PR82980]
Jason Merrill [Thu, 14 Apr 2022 18:09:13 +0000 (14:09 -0400)]
c++: lambda capture dependent type [PR82980]

The stage 4 patch limited direct propagation of dependent type to capture
field/proxy to the "current instantiation", but many more types should be
suitable as well.

PR c++/82980

gcc/cp/ChangeLog:

* lambda.cc (type_deducible_expression_p): Allow more types.

2 years agoc++: tidy auto deduction
Jason Merrill [Fri, 29 Apr 2022 18:21:47 +0000 (14:21 -0400)]
c++: tidy auto deduction

In r185768 I added the !FUNCTION_DECL check, but we might as well just check
for variable; nothing else should take this path, and asserting as much
doesn't regress anything.

gcc/cp/ChangeLog:

* decl.cc (cp_finish_decl): Only consider auto for vars.

2 years agoc++: pedwarn for empty unnamed enum in decl [PR67048]
Marek Polacek [Thu, 28 Apr 2022 20:50:06 +0000 (16:50 -0400)]
c++: pedwarn for empty unnamed enum in decl [PR67048]

[dcl.dcl]/5 says that

  enum { };

is ill-formed, and since r197742 we issue a pedwarn.  However, the
pedwarn also fires for

   enum { } x;

which is well-formed.  So only warn when {} is followed by a ;.  This
should be correct since you can't have "enum {}, <whatever>" -- that
produces "expected unqualified-id before ',' token".

PR c++/67048

gcc/cp/ChangeLog:

* parser.cc (cp_parser_enum_specifier): Warn about empty unnamed enum
only when it's followed by a semicolon.

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/enum42.C: New test.

2 years agoc++: check completeness after auto deduction [PR80351]
Jason Merrill [Thu, 24 Mar 2022 01:01:20 +0000 (18:01 -0700)]
c++: check completeness after auto deduction [PR80351]

Normally we check for incomplete type in start_decl, but that obviously
doesn't work for auto variables. Thanks to Pokechu22 for the analysis and
testcases:

"When cp_finish_decl calls cp_apply_type_quals_to_decl on a const auto or
constexpr auto variable, the type might not be complete the first time
(this happened when auto deduces to an initializer_list).
cp_apply_type_quals_to_decl removes the const qualifier if the type is
not complete, which is appropriate for grokdeclarator, on the assumption
that the type will be complete when called by cp_finish_decl."

PR c++/80351

gcc/cp/ChangeLog:

* decl.cc (cp_finish_decl): Check completeness of deduced type.

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/constexpr-77482.C: Adjust message.
* g++.dg/cpp1y/auto-fn27.C: Likewise.
* g++.dg/cpp1y/lambda-generic-variadic22.C: Likewise.
* g++.dg/cpp1z/decomp54.C: Likewise.
* g++.dg/cpp0x/initlist-const1.C: New test.
* g++.dg/warn/Wunused-var-37.C: New test.
* g++.dg/warn/Wunused-var-38.C: New test.
* g++.dg/warn/Wunused-var-39.C: New test.

2 years agoFix exchanged period and letter in gfortan.texi.
Thomas Koenig [Fri, 29 Apr 2022 17:57:05 +0000 (19:57 +0200)]
Fix exchanged period and letter in gfortan.texi.

gcc/fortran/ChangeLog:

* gfortran.texi: Fix exchanged period and letter.

2 years agoipa: Release body of clone_of when removing its last clone (PR 100413)
Martin Jambor [Fri, 29 Apr 2022 15:38:15 +0000 (17:38 +0200)]
ipa: Release body of clone_of when removing its last clone (PR 100413)

In the PR, the verifier complains that we did not manage to remove the
body of a node and it is right.  The node is kept for materialization
of two clones but after one is materialized, the other one is removed
as unneeded (as a part of delete_unreachable_blocks_update_callgraph).
The problem is that the node removal does not check for this situation
and can leave the clone_of node there with a body attached to it even
though there is no use for it any more.  This patch does checks for it
and handles the situation in a simlar way that
cgraph_node::materialize_clone does it, except that it also has to be
careful that the removed node itself does not have any clones, which
would still need the clone_of's body.  Failing to do that results in a
bootstrap failure.

gcc/ChangeLog:

2022-04-27  Martin Jambor  <mjambor@suse.cz>

PR ipa/100413
* cgraph.cc (cgraph_node::remove): Release body of the node this
is clone_of if appropriate.

gcc/testsuite/ChangeLog:

2022-04-27  Martin Jambor  <mjambor@suse.cz>

PR ipa/100413
* g++.dg/ipa/pr100413.C: New test.

2 years agoaarch64: Add option to pr105219 testcase
Andre Vieira [Fri, 29 Apr 2022 15:31:08 +0000 (16:31 +0100)]
aarch64: Add option to pr105219 testcase

Add option that originally caused testcase to fail for aarch64.

gcc/testsuite/ChangeLog:

PR tree-optimization/105219
* gcc.dg/vect/pr105219.c: Add aarch64 target option.

2 years agoc++: Add fixed test [PR83596]
Marek Polacek [Fri, 29 Apr 2022 13:52:55 +0000 (09:52 -0400)]
c++: Add fixed test [PR83596]

This was fixed by r269078.

PR c++/83596

gcc/testsuite/ChangeLog:

* g++.dg/cpp1z/nontype5.C: New test.

2 years agolibstdc++: Add missing exports for ppc64le --with-long-double-format=ibm [PR105417]
Jonathan Wakely [Fri, 29 Apr 2022 11:17:13 +0000 (12:17 +0100)]
libstdc++: Add missing exports for ppc64le --with-long-double-format=ibm [PR105417]

The --with-long-double-abi=ibm build is missing some exports that are
present in the --with-long-double-abi=ieee build. Those symbols never
should have been exported at all, but now that they have been, they
should be exported consistently by both ibm and ieee.

This simply defines them as aliases for equivalent symbols that are
already present. The abi-tag on num_get::_M_extract_int isn't really
needed, because it only uses a std::string as a local variable, not in
the return type or function parameters, so it's safe to define the
_M_extract_int[abi:cxx11] symbols as aliases for the corresponding
function without the abi-tag.

This causes some new symbols to be added to the GLIBCXX_3.4.29 version
for the ibm long double build mode, but there is no advantage to adding
them to 3.4.30 for that build. That would just create more
inconsistencies.

libstdc++-v3/ChangeLog:

PR libstdc++/105417
* config/abi/post/powerpc64-linux-gnu/baseline_symbols.txt:
Regenerate.
* src/c++11/compatibility-ldbl-alt128.cc [_GLIBCXX_USE_DUAL_ABI]:
Define __gnu_ieee128::num_get<C>::_M_extract_int[abi:cxx11]<I>
symbols as aliases for corresponding symbols without abi-tag.

2 years agoc++: Add fixed test [PR78244]
Marek Polacek [Fri, 29 Apr 2022 13:38:40 +0000 (09:38 -0400)]
c++: Add fixed test [PR78244]

This was finally fixed for GCC 11 by r11-434.

PR c++/78244

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/Wnarrowing20.C: New test.

2 years agoc++: Add fixed test [PR71424]
Marek Polacek [Fri, 29 Apr 2022 13:33:08 +0000 (09:33 -0400)]
c++: Add fixed test [PR71424]

This was fixed by r12-6329-g4f6bc28fc7dd86.

PR c++/71424

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/initlist-array15.C: New test.

2 years agoc++: using in diagnostics [PR102987]
Jason Merrill [Thu, 14 Apr 2022 21:35:35 +0000 (17:35 -0400)]
c++: using in diagnostics [PR102987]

The decl pretty-printing code wasn't looking at the flags parameter, so we
were printing 'using' in the middle of an expression.

PR c++/102987

gcc/cp/ChangeLog:

* error.cc (dump_decl) [USING_DECL]: Respect flags.

gcc/testsuite/ChangeLog:

* g++.dg/diagnostic/using1.C: Check pretty-printing.

2 years agoc++: dump alias-declaration scope
Jason Merrill [Thu, 31 Mar 2022 22:15:24 +0000 (18:15 -0400)]
c++: dump alias-declaration scope

An alias can't be declared with a qualified-id in actual code, but in
diagnostics we want to know which scope it belongs to, and I think a
nested-name-specifier is the best way to provide that.

gcc/cp/ChangeLog:

* error.cc (dump_decl): Check TFF_UNQUALIFIED_NAME.

gcc/testsuite/ChangeLog:

* g++.dg/cpp0x/alias-decl-1.C: Expect qualified name.

2 years agoc++: Improve diagnostics for template args terminated with >= or >>= [PR104319]
Jakub Jelinek [Fri, 29 Apr 2022 11:50:10 +0000 (13:50 +0200)]
c++: Improve diagnostics for template args terminated with >= or >>= [PR104319]

As mentioned in the PR, for C++98 we have diagnostics that expect
>> terminating template arguments to be a mistake for > > (C++11
said it has to be treated that way), while if user trying to spare the
spacebar doesn't separate > from following = or >> from following =,
the diagnostics is confusing, while clang suggests adding space in between.

The following patch does that for >= and >>= too.

For some strange reason the error recovery emits further errors,
not really sure what's going on because I overwrite the token->type
like the code does for the C++11 >> case or for the C++98 >> cases,
but at least the first error is nicer (well, for the C++98 nested
template case and >>= I need to overwrite it to > and so the = is lost,
so perhaps some follow-up errors are needed for that case).

2022-04-29  Jakub Jelinek  <jakub@redhat.com>

PR c++/104319
* parser.cc (cp_parser_template_argument): Treat >= like C++98 >>
after a type id by setting maybe_type_id and aborting tentative
parse.
(cp_parser_enclosed_template_argument_list): Handle
CPP_GREATER_EQ like misspelled CPP_GREATER CPP_RQ and
CPP_RSHIFT_EQ like misspelled CPP_GREATER CPP_GREATER_EQ
or CPP_RSHIFT CPP_EQ or CPP_GREATER CPP_GREATER CPP_EQ.
(cp_parser_next_token_ends_template_argument_p): Return true
also for CPP_GREATER_EQ and CPP_RSHIFT_EQ.

* g++.dg/parse/template28.C: Adjust expected diagnostics.
* g++.dg/parse/template30.C: New test.

2 years agolibstdc++: Update Solaris baselines for GCC 12.1
Rainer Orth [Fri, 29 Apr 2022 11:44:49 +0000 (13:44 +0200)]
libstdc++: Update Solaris baselines for GCC 12.1

The following patch updates the Solaris baselines for GCC 12.1.

Tested on i386-pc-solaris2.11 and sparc-sun-solaris2.11 (Solaris 11.3
and 11.4 in each case).

The only (expected) difference between the 11.3 and 11.4 versions is

--- baseline_symbols.txt.s113s 2022-04-28 10:37:11.464068450 +0000
+++ baseline_symbols.txt.s114s 2022-04-27 16:54:31.995636805 +0000
@@ -4070,3 +4070,3 @@
-FUNC:_ZSt10from_charsPKcS0_RdSt12chars_format@@GLIBCXX_3.4.30
-FUNC:_ZSt10from_charsPKcS0_ReSt12chars_format@@GLIBCXX_3.4.30
-FUNC:_ZSt10from_charsPKcS0_RfSt12chars_format@@GLIBCXX_3.4.30
+FUNC:_ZSt10from_charsPKcS0_RdSt12chars_format@@GLIBCXX_3.4.29
+FUNC:_ZSt10from_charsPKcS0_ReSt12chars_format@@GLIBCXX_3.4.29
+FUNC:_ZSt10from_charsPKcS0_RfSt12chars_format@@GLIBCXX_3.4.29

which is handled by the fix for PR libstdc++/103407.  I'm using the 11.4
version here.

2022-04-27  Rainer Orth  <ro@CeBiTec.Uni-Bielefeld.DE>

libstdc++-v3:
* config/abi/post/i386-solaris/baseline_symbols.txt: Regenerate.
* config/abi/post/i386-solaris/amd64/baseline_symbols.txt:
Likewise.
* config/abi/post/sparc-solaris/baseline_symbols.txt: Likewise.
* config/abi/post/sparc-solaris/sparcv9/baseline_symbols.txt:
Likewise.

2 years agoi386: Optimize double-word negation [PR51954]
Uros Bizjak [Fri, 29 Apr 2022 11:27:48 +0000 (13:27 +0200)]
i386: Optimize double-word negation [PR51954]

Introduce peephole2 pattern to convert from:

   mov %esi, %edx
   negl %eax
   adcl $0, %edx
   negl %edx
 to:
   xorl %edx, %edx
   negl %eax
   sbbl %esi, %edx

This conversion is profitable only when initial move is found.  Otherwise,
additional move to a temporary together with clearing xor is needed.

2022-04-29  Uroš Bizjak  <ubizjak@gmail.com>

gcc/ChangeLog:

PR target/51954
* config/i386/i386.md (adcl/neg -> sbb peephole): New peephole2.
gcc/testsuite/ChangeLog:

PR target/51954
* gcc.target/i386/pr51954.c: New test.

2 years agoFix is_gimple_condexpr vs is_gimple_condexpr_for_cond
Richard Biener [Mon, 11 Apr 2022 10:18:48 +0000 (12:18 +0200)]
Fix is_gimple_condexpr vs is_gimple_condexpr_for_cond

The following fixes wrongly used is_gimple_condexpr and makes
canonicalize_cond_expr_cond honor either, delaying final checking
to callers where all but two in ifcombine are doing the correct
thing already.

This fixes bugs but is now mainly in preparation for making
COND_EXPRs in GIMPLE assignments no longer have a GENERIC expression
as condition operand like we already transitioned VEC_COND_EXPR earlier.

2022-04-11  Richard Biener  <rguenther@suse.de>

* gimple-expr.cc (is_gimple_condexpr): Adjust comment.
(canonicalize_cond_expr_cond): Move here from gimple.cc,
allow both COND_EXPR and GIMPLE_COND forms.
* gimple-expr.h (canonicalize_cond_expr_cond): Declare.
* gimple.cc (canonicalize_cond_expr_cond): Remove here.
* gimple.h (canonicalize_cond_expr_cond): Likewise.
* gimple-loop-versioning.cc (loop_versioning::version_loop):
Use is_gimple_condexpr_for_cond.
* tree-parloops.cc (gen_parallel_loop): Likewise.
* tree-ssa-ifcombine.cc (ifcombine_ifandif): Check for
a proper cond expr after canonicalize_cond_expr_cond.
Use is_gimple_condexpr_for_cond where appropriate.
* tree-ssa-loop-manip.cc (determine_exit_conditions): Likewise.
* tree-vect-loop-manip.cc (slpeel_add_loop_guard): Likewise.

2 years agoAdd gsi_after_labels overload for gimple_seq
Richard Biener [Tue, 1 Feb 2022 09:48:07 +0000 (10:48 +0100)]
Add gsi_after_labels overload for gimple_seq

The following adds gsi_after_labels for gimple_seq so I do not have
to open-code it.  I took the liberty to remove the two #defines
wrapping gsi_start_1 and gsi_last_1 as we now have C++ references.

2022-02-01  Richard Biener  <rguenther@suse.de>

* gimple-iterator.h (gsi_after_labels): Add overload for
gimple_seq.
(gsi_start_1): Rename to gsi_start and take a reference.
(gsi_last_1): Likewise.
* gimple-iterator.cc (gsi_for_stmt): Use gsi_start.
* omp-low.cc (lower_rec_input_clauses): Likewise.
(lower_omp_scan): Likewise.

2 years agotree-optimization/105431 - another overflow in powi handling
Richard Biener [Fri, 29 Apr 2022 06:45:48 +0000 (08:45 +0200)]
tree-optimization/105431 - another overflow in powi handling

This avoids undefined signed overflow when calling powi_as_mults_1.

2022-04-29  Richard Biener  <rguenther@suse.de>

PR tree-optimization/105431
* tree-ssa-math-opts.cc (powi_as_mults_1): Make n unsigned.
(powi_as_mults): Use absu_hwi.
(gimple_expand_builtin_powi): Remove now pointless n != -n
check.

2 years agoMove common code from range-op.cc to header files.
Aldy Hernandez [Thu, 28 Apr 2022 18:19:14 +0000 (20:19 +0200)]
Move common code from range-op.cc to header files.

In preparation for the agnostication of ranger, this patch moves
common code that can be shared between non-integer ranges (initially
pointers) into the relevant header files.

This is a relatively non-invasive change, as any changes that would
need to be ported to GCC 12, would occur in the range-op entries
themselves, not in the supporting glue which I'm moving.

Tested and benchmarked on x86-64 Linux.

gcc/ChangeLog:

* range-op.cc (empty_range_varying): Move to range-op.h.
(range_true): Move to range.h.
(range_false): Same.
(range_true_and_false): Same.
(enum bool_range_state): Move to range-op.h.
(relop_early_resolve): Same.
(operator_equal::op1_op2_relation): Abstract code to...
(equal_op1_op2_relation): ...here.
(operator_not_equal::op1_op2_relation): Abstract code to...
(not_equal_op1_op2_relation): ...here.
(operator_lt::op1_op2_relation): Abstract code to...
(lt_op1_op2_relation): ...here.
(operator_le::op1_op2_relation): Abstract code to...
(le_op1_op2_relation): ...here.
(operator_gt::op1_op2_relation): Abstract code to...
(gt_op1_op2_relation): ...here.
(operator_ge::op1_op2_relation): Abstract code to...
(ge_op1_op2_relation): ...here.
(class range_op_table): Move to range-op.h.
* range-op.h (equal_op1_op2_relation): Moved from range-op.cc.
(not_equal_op1_op2_relation): Same.
(lt_op1_op2_relation): Same.
(le_op1_op2_relation): Same.
(gt_op1_op2_relation): Same.
(ge_op1_op2_relation): Same.
(enum bool_range_state): Same.
(get_bool_state): Same.
(empty_range_varying): Same.
(relop_early_resolve): Same.
(class range_op_table): Same.
* range.h (range_true): Same.
(range_false): Same.
(range_true_and_false): Same.

2 years agoRemove various deprecated methods in class irange.
Aldy Hernandez [Mon, 14 Mar 2022 08:57:48 +0000 (09:57 +0100)]
Remove various deprecated methods in class irange.

This patch cleans up some irange methods in preparation for other
cleanups later in the cycle.

First, we prefer the reference overloads for union and intersect as
the pointer versions have been deprecated for a couple releases.

Also, I've renamed the legacy union/intersect whose only function was
to provide additional verbosity for VRP into
legacy_verbose_{union,intersect}.  This is a temporary rename to serve
as a visual reminder of which of the methods are bound for the chopping
block when the legacy code gets removed later this cycle.

Tested on x86-64 Linux.

gcc/ChangeLog:

* gimple-fold.cc (size_must_be_zero_p): Use reference
instead of pointer
* gimple-ssa-evrp-analyze.cc
(evrp_range_analyzer::record_ranges_from_incoming_edge): Rename
intersect to legacy_verbose_intersect.
* ipa-cp.cc (ipcp_vr_lattice::meet_with_1): Use reference instead
of pointer.
* tree-ssa-dom.cc (dom_jt_simplifier::simplify): Use value_range
instead of value_range_equiv.
* tree-vrp.cc (extract_range_from_plus_minus_expr): Use reference
instead of pointer.
(find_case_label_range): Same.
* value-range-equiv.cc (value_range_equiv::intersect): Rename to...
(value_range_equiv::legacy_verbose_intersect): ...this.
(value_range_equiv::union_): Rename to...
(value_range_equiv::legacy_verbose_union_): ...this.
* value-range-equiv.h (class value_range_equiv): Rename union and
intersect to legacy_verbose_{intersect,union}.
* value-range.cc (irange::union_): Rename to...
(irange::legacy_verbose_union_): ...this.
(irange::intersect): Rename to...
(irange::legacy_verbose_intersect): ...this.
* value-range.h (irange::union_): Rename union_ to
legacy_verbose_union.
(irange::intersect): Rename intersect to legacy_verbose_intersect.
* vr-values.cc (vr_values::update_value_range): Same.
(vr_values::extract_range_for_var_from_comparison_expr): Same.
(vr_values::extract_range_from_cond_expr): Rename union_ to
legacy_verbose_union.
(vr_values::extract_range_from_phi_node): Same.

2 years agoPrefer global range info setters that take a range.
Aldy Hernandez [Mon, 7 Mar 2022 13:56:34 +0000 (14:56 +0100)]
Prefer global range info setters that take a range.

This patch consolidates the multiple ways we have of storing global
ranges into one accepting a range.

In an upcoming patch series later this cycle we will be providing a
way to store iranges globally, as opposed to the mechanism we have now
which squishes wider ranges into value_range's.  This is preparation
for such work.

Tested and benchmarked on x86-64 Linux.

gcc/ChangeLog:

* gimple-ssa-evrp-analyze.cc
(evrp_range_analyzer::set_ssa_range_info): Use *range_info methods
that take a range.
* gimple-ssa-sprintf.cc (try_substitute_return_value): Same.
* ipa-prop.cc (ipcp_update_vr): Same.
* tree-inline.cc (remap_ssa_name): Same.
* tree-ssa-copy.cc (fini_copy_prop): Same.
* tree-ssa-math-opts.cc (optimize_spaceship): Same.
* tree-ssa-phiopt.cc (replace_phi_edge_with_variable): Same.
* tree-ssa-pre.cc (insert_into_preds_of_block): Same.
* tree-ssa-sccvn.cc (eliminate_dom_walker::eliminate_stmt): Same.
* tree-ssa-strlen.cc (set_strlen_range): Same.
(strlen_pass::handle_builtin_string_cmp): Same.
* tree-ssanames.cc (set_range_info): Make static.
(duplicate_ssa_name_range_info): Make static and add a new variant
calling the static.
* tree-ssanames.h (set_range_info): Remove version taking wide ints.
(duplicate_ssa_name_range_info): Remove version taking a
range_info_def and replace with a version taking SSA names.
* tree-vect-loop-manip.cc (vect_gen_vector_loop_niters): Use *range_info methods
that take a range.
(vect_do_peeling): Same.
* tree-vrp.cc (vrp_asserts::remove_range_assertions): Same.
* vr-values.cc (simplify_truth_ops_using_ranges): Same.

2 years agoCall set_undefined from irange constructor.
Aldy Hernandez [Mon, 7 Mar 2022 13:49:57 +0000 (14:49 +0100)]
Call set_undefined from irange constructor.

Small clean up to use set_undefined instead of duplicating the
functionality therein.

Tested on x86-64 Linux.

gcc/ChangeLog:

* value-range.h (irange::irange): Use set_undefined.

2 years agoMake irange::intersect(wide_int, wide_int) private.
Aldy Hernandez [Mon, 7 Mar 2022 13:48:58 +0000 (14:48 +0100)]
Make irange::intersect(wide_int, wide_int) private.

This method should have been private, and somehow seeped into the API.

Tested and benchmarked on x86-64 Linux.

gcc/ChangeLog:

* gimple-range-cache.h (non_null_ref::adjust_range): Do not use
irange::intersect (wide_int, wide_int).
* gimple-range-fold.cc (adjust_pointer_diff_expr): Same.
(adjust_imagpart_expr): Same.
* value-range.h (irange::intersect (wide_int, wide_int)): Make
private.

2 years agotree-optimization/104322 - remove dead code in vectorizable_reduction
Richard Biener [Tue, 1 Feb 2022 10:32:11 +0000 (11:32 +0100)]
tree-optimization/104322 - remove dead code in vectorizable_reduction

The PR points out dead code after previous refactoring.

2022-02-01  Richard Biener  <rguenther@suse.de>

PR tree-optimization/104322
* tree-vect-loop.cc (vectorizable_reduction): Remove dead code.

2 years agoc++, coroutines: Partial reversion of r12-8308-g15a176a833f23e [PR105426].
Iain Sandoe [Thu, 28 Apr 2022 19:06:29 +0000 (20:06 +0100)]
c++, coroutines: Partial reversion of r12-8308-g15a176a833f23e [PR105426].

The changes to fix PR 105287 included a tightening of the constraints on which
variables are promoted to frame copies.  This has exposed that we are failing
to name some variables that should be promoted.

We avoid the use of DECL_UID to build anonymous symbols since that might not
be stable for -fcompare-debug.

The long-term fix is to address the cases where the naming has been missed,
but for the short-term (and for the GCC-12 branch) backing out the additional
constraint is proposed.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
PR c++/105426

gcc/cp/ChangeLog:

* coroutines.cc (register_local_var_uses): Allow promotion of unnamed
temporaries to coroutine frame copies.

2 years agomiddle-end/105376 - invalid REAL_CST for DFP constant
Richard Biener [Wed, 27 Apr 2022 06:28:31 +0000 (08:28 +0200)]
middle-end/105376 - invalid REAL_CST for DFP constant

We are eventually ICEing in decimal_to_decnumber on non-decimal
REAL_VALUE_TYPE that creep in from uses of build_real (..., dconst*)
for DFP types.  The following extends the decimal_to_decnumber
special-casing of dconst* to build_real, avoiding the bogus REAL_CSTs
from creeping into the IL and modified to ones not handled by
the decimal_to_decnumber special casing.  It also makes sure to
ICE for not handled dconst* values at the point we build the REAL_CST.

2022-04-27  Richard Biener  <rguenther@suse.de>

PR middle-end/105376
* tree.cc (build_real): Special case dconst* arguments
for decimal floating point types.

* gcc.dg/pr105376.c: New testcase.

2 years agoada: bump Library_Version to 13.
Martin Liska [Fri, 29 Apr 2022 04:34:02 +0000 (06:34 +0200)]
ada: bump Library_Version to 13.

gcc/ada/ChangeLog:

* gnatvsn.ads: Bump Library_Version to 13.

2 years agoc++: traits, array of unknown bound of incomplete
Jason Merrill [Wed, 23 Mar 2022 16:25:18 +0000 (12:25 -0400)]
c++: traits, array of unknown bound of incomplete

My r161129 changed check_trait_type to reject arrays of unknown bound of
incomplete type, but I can't find a rationale for that, and now think it's
wrong: the standard just requires that the type be "complete, cv void, or an
array of unknown bound."  I imagine that allowing arrays of unknown bound is
because an array of unknown bound can't change from incomplete to complete
later in the translation unit, so there's no caching problem.

gcc/cp/ChangeLog:

* semantics.cc (check_trait_type): Don't check completeness
of element type of array of unknown bound.

gcc/testsuite/ChangeLog:

* g++.dg/ext/unary_trait_incomplete.C: Adjust.

2 years agoc++: typeid and instantiation [PR102651]
Jason Merrill [Fri, 15 Apr 2022 04:11:00 +0000 (00:11 -0400)]
c++: typeid and instantiation [PR102651]

PR49387 was a problem with initially asking for a typeid for a class
template specialization before it was complete, and later actually filling
in the descriptor when the class was complete, and thus disagreeing on the
form of the descriptor.  I fixed that by forcing the class to be complete,
but this testcase shows why that approach is problematic.  So instead let's
adjust the type of the descriptor later if needed.

PR c++/102651
PR c++/49387

gcc/cp/ChangeLog:

* rtti.cc (get_tinfo_decl_direct): Don't complete_type.
(emit_tinfo_decl): Update tdesc type if needed.

gcc/testsuite/ChangeLog:

* g++.dg/rtti/typeid-complete1.C: New test.

2 years agoc++: Add diagnostic when operator= is used as truth cond [PR25689]
Zhao Wei Liew [Tue, 15 Feb 2022 09:44:29 +0000 (17:44 +0800)]
c++: Add diagnostic when operator= is used as truth cond [PR25689]

When compiling the following code with g++ -Wparentheses, GCC does not
warn on the if statement. For example, there is no warning for this code:

struct A {
A& operator=(int);
operator bool();
};

void f(A a) {
if (a = 0); // no warning
}

This is because a = 0 is a call to operator=, which GCC does not handle.

This patch fixes this issue by handling calls to operator= when deciding
to warn.

Bootstrapped and regression tested on x86_64-pc-linux-gnu.

PR c++/25689

gcc/cp/ChangeLog:

* call.cc (extract_call_expr): Return a NULL_TREE on failure
instead of asserting.
(build_new_method_call): Suppress -Wparentheses diagnostic for
MODIFY_EXPR.
* semantics.cc (is_assignment_op_expr_p): Add function to check
if an expression is a call to an op= operator expression.
(maybe_convert_cond): Handle the case of a op= operator expression
for the -Wparentheses diagnostic.

gcc/testsuite/ChangeLog:

* g++.dg/warn/Wparentheses-31.C: New test.

Signed-off-by: Zhao Wei Liew <zhaoweiliew@gmail.com>
2 years agoDaily bump.
GCC Administrator [Fri, 29 Apr 2022 00:16:26 +0000 (00:16 +0000)]
Daily bump.

2 years agogcov: Add section for freestanding environments
Sebastian Huber [Fri, 8 Apr 2022 15:16:09 +0000 (17:16 +0200)]
gcov: Add section for freestanding environments

gcc/

* doc/gcov.texi (Profiling and Test Coverage in Freestanding
Environments): New section.

2 years agogcov: Use xstrerror()
Sebastian Huber [Thu, 31 Mar 2022 08:50:50 +0000 (10:50 +0200)]
gcov: Use xstrerror()

libgcc/

* libgcov-util.c (ftw_read_file): Improve notice using xstrerror().
(gcov_profile_merge_stream): Likewise.

2 years agogcov-tool: Add merge-stream subcommand
Sebastian Huber [Wed, 30 Mar 2022 14:53:29 +0000 (16:53 +0200)]
gcov-tool: Add merge-stream subcommand

gcc/

* doc/gcov-tool.texi: Document merge-stream subcommand.
* doc/invoke.texi (fprofile-info-section): Mention merge-stream
subcommand of gcov-tool.
* gcov-tool.cc (gcov_profile_merge_stream): Declare.
(print_merge_stream_usage_message): New.
(merge_stream_usage): Likewise.
(do_merge_stream): Likewise.
(print_usage): Call print_merge_stream_usage_message().
(main): Call do_merge_stream() to execute merge-stream subcommand.

libgcc/

* libgcov-util.c (consume_stream): New.
(get_target_profiles_for_merge): Likewise.
(gcov_profile_merge_stream): Likewise.

2 years agogcov: Record EOF error during read
Sebastian Huber [Thu, 31 Mar 2022 08:10:02 +0000 (10:10 +0200)]
gcov: Record EOF error during read

Use an enum for file error codes.

gcc/

* gcov-io.cc (gcov_file_error): New enum.
(gcov_var): Use gcov_file_error enum for the error member.
(gcov_open): Use GCOV_FILE_NO_ERROR.
(gcov_close): Use GCOV_FILE_WRITE_ERROR.
(gcov_write): Likewise.
(gcov_write_unsigned): Likewise.
(gcov_write_string): Likewise.
(gcov_read_bytes): Set error code if EOF is reached.
(gcov_read_counter): Use GCOV_FILE_COUNTER_OVERFLOW.

2 years agogcov: Fix integer types in ftw_read_file()
Sebastian Huber [Wed, 30 Mar 2022 20:05:12 +0000 (22:05 +0200)]
gcov: Fix integer types in ftw_read_file()

libgcc/

* libgcov-util.c (ftw_read_file): Use size_t for strlen() variables.

2 years agogcov: Move gcov_open() to caller of read_gcda_file()
Sebastian Huber [Wed, 30 Mar 2022 19:59:43 +0000 (21:59 +0200)]
gcov: Move gcov_open() to caller of read_gcda_file()

This allows to reuse read_gcda_file() to read multiple objects from a single
file.

libgcc/

* libgcov-util.c (read_gcda_file): Do not open file.
(ftw_read_file): Open file here.

2 years agogcov: Move prepend to list to read_gcda_file()
Sebastian Huber [Wed, 30 Mar 2022 19:54:36 +0000 (21:54 +0200)]
gcov: Move prepend to list to read_gcda_file()

This helps to reuse read_gcda_file().

libgcc/

* libgcov-util.c (read_gcda_file): Prepend new info object to global
list.
(ftw_read_file): Remove list append here.

2 years agogcov: Use xstrdup()
Sebastian Huber [Wed, 30 Mar 2022 19:49:51 +0000 (21:49 +0200)]
gcov: Use xstrdup()

Move duplication of filename to caller and use xstrdup() instead of custom
code.  This helps to reuse read_gcda_file() for other purposes.

libgcc/

* libgcov-util.c (read_gcda_file): Do not duplicate filename.
(ftw_read_file): Duplicate filename for read_gcda_file().

2 years agogcov-tool: Support file input from stdin
Sebastian Huber [Wed, 30 Mar 2022 19:40:55 +0000 (21:40 +0200)]
gcov-tool: Support file input from stdin

gcc/

* gcov-io.cc (GCOV_MODE_STDIN): Define.
(gcov_position): For gcov-tool, return calculated position if file is
stdin.
(gcov_open):  For gcov-tool, use stdin if filename is NULL.
(gcov_close): For gcov-tool, do not close stdin.
(gcov_read_bytes): For gcov-tool, update position if file is stdin.
(gcov_sync): For gcov-tool, discard input if file is stdin.

2 years agogcov: Add __gcov_filename_to_gcfn()
Sebastian Huber [Wed, 30 Mar 2022 19:45:23 +0000 (21:45 +0200)]
gcov: Add __gcov_filename_to_gcfn()

gcc/

* doc/invoke.texi (fprofile-info-section): Mention
__gcov_filename_to_gcfn().  Use "freestanding" to match with C11
standard language.  Fix minor example code issues.
* gcov-io.h (GCOV_FILENAME_MAGIC): Define and document.

gcc/testsuite/

* gcc.dg/gcov-info-to-gcda.c: Test __gcov_filename_to_gcfn().

libgcc/

* gcov.h (__gcov_info_to_gcda): Mention __gcov_filename_to_gcfn().
(__gcov_filename_to_gcfn): Declare and document.
* libgcov-driver.c (dump_string): New.
(__gcov_filename_to_gcfn): Likewise.
(__gcov_info_to_gcda): Adjust comment to match C11 standard language.

2 years agogcov: Make gcov_seek() static
Sebastian Huber [Mon, 28 Mar 2022 10:50:44 +0000 (12:50 +0200)]
gcov: Make gcov_seek() static

This function is only used by gcov_write_length() in the gcov-io.cc file.

gcc/

* gcov-io.cc (gcov_seek): Make it static.
* gcov-io.h (struct gcov_summary): Do not mention gcov_seek().

libgcc/

* libgcov.h (gcov_seek): Remove define and declaration.

2 years agogcov: Add open mode parameter to gcov_do_dump()
Sebastian Huber [Thu, 31 Mar 2022 09:37:56 +0000 (11:37 +0200)]
gcov: Add open mode parameter to gcov_do_dump()

gcc/

* gcov-tool.cc (gcov_do_dump): Add mode parameter.
(gcov_output_files): Open files for reading and writing.

libgcc/

* libgcov-driver-system.c (gcov_exit_open_gcda_file): Add mode
parameter.  Pass mode to gcov_open() calls.
* libgcov-driver.c (dump_one_gcov):  Add mode parameter.  Pass mode to
gcov_exit_open_gcda_file() call.
(gcov_do_dump): Add mode parameter.  Pass mode to dump_one_gcov()
calls.
(__gcov_dump_one):  Open file for reading and writing.

2 years agogcov: Add mode to all gcov_open()
Sebastian Huber [Thu, 24 Mar 2022 20:59:21 +0000 (21:59 +0100)]
gcov: Add mode to all gcov_open()

gcc/

* gcov-io.cc (gcov_open): Always use the mode parameter.
* gcov-io.h (gcov_open): Declare it unconditionally.

libgcc/

* libgcov-driver-system.c (gcov_exit_open_gcda_file): Open file for
reading and writing.
* libgcov-util.c (read_gcda_file): Open file for reading.
* libgcov.h (gcov_open): Delete declaration.

2 years agogcov-tool: Allow merging of empty profile lists
Sebastian Huber [Wed, 23 Mar 2022 09:20:56 +0000 (10:20 +0100)]
gcov-tool: Allow merging of empty profile lists

The gcov_profile_merge() already had code to deal with profile information
which had no counterpart to merge with.  For profile information from files
with no associated counterpart, the profile information is simply used as is
with the weighting transformation applied.  Make sure that gcov_profile_merge()
works with an empty target profile list.  Return the merged profile list.

gcc/
* gcov-tool.cc (gcov_profile_merge): Adjust return type.
(profile_merge): Allow merging of directories which contain no profile
files.

libgcc/
* libgcov-util.c (gcov_profile_merge): Return the list of merged
profiles.  Accept empty target and source profile lists.

2 years agoanalyzer: handle repeated accesses after init of unknown size [PR105285]
David Malcolm [Thu, 28 Apr 2022 17:49:59 +0000 (13:49 -0400)]
analyzer: handle repeated accesses after init of unknown size [PR105285]

PR analyzer/105285 reports a false positive from
-Wanalyzer-null-dereference on git.git's reftable/reader.c.

A reduced version of the problem can be seen in test_1a of
gcc.dg/analyzer/symbolic-12.c in the following:

void test_1a (void *p, unsigned next_off)
{
  struct st_1 *r = p;

  external_fn();

  if (next_off >= r->size)
    return;

  if (next_off >= r->size)
    /* We should have already returned if this is the case.  */
    __analyzer_dump_path (); /* { dg-bogus "path" } */
}

where the analyzer erroneously considers this path, where
(next_off >= r->size) is both false and then true:

symbolic-12.c: In function ‘test_1a’:
symbolic-12.c:22:5: note: path
   22 |     __analyzer_dump_path (); /* { dg-bogus "path" } */
      |     ^~~~~~~~~~~~~~~~~~~~~~~
  ‘test_1a’: events 1-5
    |
    |   17 |   if (next_off >= r->size)
    |      |      ^
    |      |      |
    |      |      (1) following ‘false’ branch...
    |......
    |   20 |   if (next_off >= r->size)
    |      |      ~            ~~~~~~~
    |      |      |             |
    |      |      |             (2) ...to here
    |      |      (3) following ‘true’ branch...
    |   21 |     /* We should have already returned if this is the case.  */
    |   22 |     __analyzer_dump_path (); /* { dg-bogus "path" } */
    |      |     ~~~~~~~~~~~~~~~~~~~~~~~
    |      |     |
    |      |     (4) ...to here
    |      |     (5) here
    |

The root cause is that, at the call to the external function, the
analyzer considers the cluster for *p to have been touched, binding it
to a conjured_svalue, but because p is void * no particular size is
known for the write, and so the cluster is bound using a symbolic key
covering the base region.  Later, the accesses to r->size are handled by
binding_cluster::get_any_binding, but binding_cluster::get_binding fails
to find a match for the concrete field lookup, due to the key for the
binding being symbolic, and reaching this code:

1522  /* If this cluster has been touched by a symbolic write, then the content
1523     of any subregion not currently specifically bound is "UNKNOWN".  */
1524  if (m_touched)
1525    {
1526      region_model_manager *rmm_mgr = mgr->get_svalue_manager ();
1527      return rmm_mgr->get_or_create_unknown_svalue (reg->get_type ());
1528    }

Hence each access to r->size is an unknown svalue, and thus the
condition (next_off >= r->size) isn't tracked, leading to the path with
contradictory conditions being treated as satisfiable.

In the original reproducer in git's reftable/reader.c, the call to the
external fn is:
  reftable_record_type(rec)
which is considered to possibly write to *rec, which is *tab, where tab
is the void * argument to reftable_reader_seek_void, and thus after the
call to reftable_record_type some arbitrary amount of *rec could have
been written to.

This patch fixes things by detecting the "this cluster has been 'filled'
with a conjured value of unknown size" case, and handling
get_any_binding on it by returning a sub_svalue of the conjured_svalue,
so that repeated accesses to r->size give the same symbolic value, so
that the constraint manager rejects the bogus execution path, fixing the
false positive.

gcc/analyzer/ChangeLog:
PR analyzer/105285
* store.cc (binding_cluster::get_any_binding): Handle accessing
sub_svalues of clusters where the base region has a symbolic
binding.

gcc/testsuite/ChangeLog:
PR analyzer/105285
* gcc.dg/analyzer/symbolic-12.c: New test.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
2 years agoanalyzer: add .fpath.txt dumps to -fdump-analyzer-feasibility
David Malcolm [Thu, 28 Apr 2022 17:46:15 +0000 (13:46 -0400)]
analyzer: add .fpath.txt dumps to -fdump-analyzer-feasibility

I found this extension to -fdump-analyzer-feasibility very helpful when
debugging PR analyzer/105285.

gcc/analyzer/ChangeLog:
* diagnostic-manager.cc (epath_finder::process_worklist_item):
Call dump_feasible_path when a path that reaches the the target
enode is found.
(epath_finder::dump_feasible_path): New.
* engine.cc (feasibility_state::dump_to_pp): New.
* exploded-graph.h (feasibility_state::dump_to_pp): New decl.
* feasible-graph.cc (feasible_graph::dump_feasible_path): New.
* feasible-graph.h (feasible_graph::dump_feasible_path): New
decls.
* program-point.cc (function_point::print): Fix missing trailing
newlines.
* program-point.h (program_point::print_source_line): Remove
unimplemented decl.

gcc/ChangeLog:
* doc/invoke.texi (-fdump-analyzer-feasibility): Mention the
fpath.txt output.

Signed-off-by: David Malcolm <dmalcolm@redhat.com>
2 years agoc++: partial ordering and dependent operator expr [PR105425]
Patrick Palka [Thu, 28 Apr 2022 17:10:56 +0000 (13:10 -0400)]
c++: partial ordering and dependent operator expr [PR105425]

Here ever since r12-6022-gbb2a7f80a98de3 we stopped deeming the partial
specialization #2 to be more specialized than #1 ultimately because
dependent operator expressions now have a DEPENDENT_OPERATOR_TYPE type
instead of an empty type, and this made unify stop deducing T(2) == 1
for K during partial ordering for #1 and #2.

This minimal patch fixes this by making the relevant logic in unify
treat DEPENDENT_OPERATOR_TYPE like an empty type.

PR c++/105425

gcc/cp/ChangeLog:

* pt.cc (unify) <case TEMPLATE_PARM_INDEX>: Treat
DEPENDENT_OPERATOR_TYPE like an empty type.

gcc/testsuite/ChangeLog:

* g++.dg/template/partial-specialization13.C: New test.

2 years agoDocument changes to CONVERT for -mabi-ieeelongdouble for POWER.
Thomas Koenig [Thu, 28 Apr 2022 16:23:16 +0000 (18:23 +0200)]
Document changes to CONVERT for -mabi-ieeelongdouble for POWER.

gcc/fortran/ChangeLog:

* gfortran.texi: Mention r16_ieee and r16_ibm.
* invoke.texi: Likewise.

2 years ago[committed] Fix more problems with new linker warnings
Jeff Law [Thu, 28 Apr 2022 16:03:52 +0000 (12:03 -0400)]
[committed] Fix more problems with new linker warnings

gcc/testsuite
* gcc.dg/lto/pr94157_0.c: Revert last change.
* lib/prune.exp (prune_gcc_output): Prune new linker warning.

2 years agoi386: Improve ix86_expand_int_movcc
Jakub Jelinek [Thu, 28 Apr 2022 15:41:49 +0000 (17:41 +0200)]
i386: Improve ix86_expand_int_movcc

When working on PR105338, I've noticed that in some cases we emit
unnecessarily long sequence which has then higher seq_cost than necessary.

E.g. when ix86_expand_int_movcc is called with
operands[0] (reg/v:SI 83 [ i ])
operands[1] (eq (reg/v:SI 83 [ i ]) (const_int 0 [0]))
operands[2] (reg/v:SI 83 [ i ])
operands[3] (const_int -2 [0xfffffffffffffffe])
i.e. r83 = r83 == 0 ? r83 : -2 which with my PR105338 patch is equivalent to
r83 = r83 == 0 ? 0 : -2, we emit:
(insn 24 0 25 (set (reg:CC 17 flags)
        (compare:CC (reg/v:SI 83 [ i ])
            (const_int 1 [0x1]))) 11 {*cmpsi_1}
     (nil))
(insn 25 24 26 (parallel [
            (set (reg:SI 85)
                (if_then_else:SI (ltu:SI (reg:CC 17 flags)
                        (const_int 0 [0]))
                    (const_int -1 [0xffffffffffffffff])
                    (const_int 0 [0])))
            (clobber (reg:CC 17 flags))
        ]) 1192 {*x86_movsicc_0_m1}
     (nil))
(insn 26 25 27 (set (reg:SI 85)
        (not:SI (reg:SI 85))) 683 {*one_cmplsi2_1}
     (nil))
(insn 27 26 28 (parallel [
            (set (reg:SI 85)
                (and:SI (reg:SI 85)
                    (const_int -2 [0xfffffffffffffffe])))
            (clobber (reg:CC 17 flags))
        ]) 533 {*andsi_1}
     (nil))
(insn 28 27 0 (set (reg/v:SI 83 [ i ])
        (reg:SI 85)) 81 {*movsi_internal}
     (nil))
which has seq_cost (seq, true) 24.  But it could have just cost 20
if we didn't decide to use a fresh temporary r85 and used r83 instead
- we could avoid the copy at the end.
The reason for it is in the 2 reg_overlap_mentioned_p calls,
the destination (out) indeed overlaps op0 - it is the same register,
but I don't see why that is a problem, this is in a code path where
we've already called
ix86_expand_carry_flag_compare (code, op0, op1, &compare_op)
earlier, so the fact that we've out overlaps op0 or op1 shouldn't matter
because insn 24 above is already emitted, we should just care if
it overlaps whatever we got from that ix86_expand_carry_flag_compare
call, i.e. compare_op, otherwise we can overwrite out just fine;
we also know at that point that the last 2 operands of ?: are constants.

2022-04-28  Jakub Jelinek  <jakub@redhat.com>

* config/i386/i386-expand.cc (ix86_expand_int_movcc): Create a
temporary only if out overlaps compare_op, not when it overlaps
op0 or op1.

2 years agoUpdate crontab and git_update_version.py
Jakub Jelinek [Thu, 28 Apr 2022 14:22:42 +0000 (16:22 +0200)]
Update crontab and git_update_version.py

2022-04-28  Jakub Jelinek  <jakub@redhat.com>

maintainer-scripts/
* crontab: Snapshots from trunk are now GCC 13 related.
Add GCC 12 snapshots from the respective branch.
contrib/
* gcc-changelog/git_update_version.py (active_refs): Add
releases/gcc-12.

2 years agoBump BASE-VER.
Jakub Jelinek [Thu, 28 Apr 2022 13:58:21 +0000 (15:58 +0200)]
Bump BASE-VER.

2022-04-28  Jakub Jelinek  <jakub@redhat.com>

* BASE-VER: Set to 13.0.0.

2 years agocgraph: Don't verify semantic_interposition flag for aliases [PR105399]
Jakub Jelinek [Thu, 28 Apr 2022 13:45:33 +0000 (15:45 +0200)]
cgraph: Don't verify semantic_interposition flag for aliases [PR105399]

The following testcase ICEs, because the ctors during cc1plus all have
!opt_for_fn (decl, flag_semantic_interposition) - they have NULL
DECL_FUNCTION_SPECIFIC_OPTIMIZATION (decl) and optimization_default_node
is for -Ofast and so has flag_semantic_interposition cleared.
During free lang data, we set DECL_FUNCTION_SPECIFIC_OPTIMIZATION (decl)
for the ctor which has body (or for thunks), but don't touch it for
aliases.
During lto1 optimization_default_node reflects the lto1 flags which
are -O2 rather than -Ofast and so has flag_semantic_interposition
set, for the ctor which has body that makes no difference, but as the
alias doesn't still have DECL_FUNCTION_SPECIFIC_OPTIMIZATION (decl) set,
we now trigger this verification check.

The following patch just doesn't verify it for aliases during lto1.
Another possibility would be to set DECL_FUNCTION_SPECIFIC_OPTIMIZATION (decl)
during free lang data even for aliases.

2022-04-28  Jakub Jelinek  <jakub@redhat.com>

PR lto/105399
* cgraph.cc (cgraph_node::verify_node): Don't verify
semantic_interposition flag against
opt_for_fn (decl, flag_semantic_interposition) for aliases in lto1.

* g++.dg/lto/pr105399_0.C: New test.

2 years agoFix up 'libgomp.oacc-fortran/print-1.f90' GCN offloading compilation [PR104717]
Thomas Schwinge [Tue, 26 Apr 2022 16:41:23 +0000 (18:41 +0200)]
Fix up 'libgomp.oacc-fortran/print-1.f90' GCN offloading compilation [PR104717]

That got broken by recent commit b2202431910e30d8505c94d1cb9341cac7080d10
"fortran: Fix up gfc_trans_oacc_construct [PR104717]".

PR fortran/104717
libgomp/
* testsuite/libgomp.oacc-fortran/print-1.f90: Add OpenACC
privatization scanning.  For GCN offloading compilation, raise
'-mgang-private-size'.

2 years agoc++, coroutines: Improve check for throwing final await [PR104051].
Iain Sandoe [Mon, 18 Apr 2022 15:23:30 +0000 (16:23 +0100)]
c++, coroutines: Improve check for throwing final await [PR104051].

We check that the final_suspend () method returns a sane type (i.e. a class
or structure) but, unfortunately, that check has to be later than the one
for a throwing case.  If the use returns some nonsensical type from the
method, we need to handle that in the checking for noexcept.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
PR c++/104051

gcc/cp/ChangeLog:

* coroutines.cc (coro_diagnose_throwing_final_aw_expr): Handle
non-target expression inputs.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/pr104051.C: New test.

2 years agoc++, coroutines: Account for overloaded promise return_value() [PR105301].
Iain Sandoe [Mon, 18 Apr 2022 08:21:52 +0000 (09:21 +0100)]
c++, coroutines: Account for overloaded promise return_value() [PR105301].

Whether it was intended or not, it is possible to define a coroutine promise
with multiple return_value() methods [which need not even have the same type].

We were not accounting for this possibility in the check to see whether both
return_value and return_void are specifier (which is prohibited by the
standard).  Fixed thus and provided an adjusted diagnostic for the case that
multiple return_value() methods are present.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
PR c++/105301

gcc/cp/ChangeLog:

* coroutines.cc (coro_promise_type_found_p): Account for possible
mutliple overloads of the promise return_value() method.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/pr105301.C: New test.

2 years agoc++, coroutines: Make sure our temporaries are in a bind expr [PR105287]
Iain Sandoe [Sun, 17 Apr 2022 19:58:28 +0000 (20:58 +0100)]
c++, coroutines: Make sure our temporaries are in a bind expr [PR105287]

There are a few cases where we can generate a temporary that does not need
to be added to the coroutine frame (i.e. these are genuinely ephemeral).  The
intent was that unnamed temporaries should not be 'promoted' to coroutine
frame entries.  However there was a thinko and these were not actually ever
added to the bind expressions being generated for the expanded awaits.  This
meant that they were showing in the global namspace, leading to an empty
DECL_CONTEXT and the ICE reported.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
PR c++/105287

gcc/cp/ChangeLog:

* coroutines.cc (maybe_promote_temps): Ensure generated temporaries
are added to the bind expr.
(add_var_to_bind): Fix local var naming to use portable punctuation.
(register_local_var_uses): Do not add synthetic names to unnamed
temporaries.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/pr105287.C: New test.

2 years agoc++, coroutines: Avoid expanding within templates [PR103868]
Nathan Sidwell [Sun, 3 Apr 2022 10:35:03 +0000 (11:35 +0100)]
c++, coroutines: Avoid expanding within templates [PR103868]

This is a forward-port of a patch by Nathan (against 10.x) which fixes an open
PR.

We are ICEing because we ended up tsubst_copying something that had already
been tsubst, leading to an assert failure (mostly such repeated tsubsting is
harmless).

We had a non-dependent co_await in a non-dependent-type template fn, so we
processed it at definition time, and then reprocessed at instantiation time.
We fix this here by deferring substitution while processing templates.

Additional observations (for a better future fix, in the GCC13 timescale):

Exprs only have dependent type if at least one operand is dependent which was
what the current code was intending to do.  Coroutines have the additional
wrinkle, that the current fn's type is an implicit operand.

So, if the coroutine function's type is not dependent, and the operand is not
dependent, we should determine the type of the co_await expression using the
DEPENDENT_EXPR wrapper machinery.  That allows us to determine the
subexpression type, but leave its operand unchanged and then instantiate it
later.

PR c++/103868

gcc/cp/ChangeLog:

* coroutines.cc (finish_co_await_expr): Do not process non-dependent
coroutine expressions at template definition time.
(finish_co_yield_expr): Likewise.
(finish_co_return_stmt): Likewise.

gcc/testsuite/ChangeLog:

* g++.dg/coroutines/pr103868.C: New test.

Co-Authored-by: Iain Sandoe <iain@sandoe.co.uk>
2 years agotestsuite,X86: Fix missing USER_LABEL_PREFIX cases.
Iain Sandoe [Fri, 15 Apr 2022 14:51:22 +0000 (15:51 +0100)]
testsuite,X86: Fix missing USER_LABEL_PREFIX cases.

Yet another set of testcases that do not account for targets that
use __USER_LABEL_PREFIX__.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
gcc/testsuite/ChangeLog:

* gcc.target/i386/memcpy-strategy-10.c: Account for
__USER_LABEL_PREFIX__.
* gcc.target/i386/memcpy-strategy-5.c: Likewise.
* gcc.target/i386/memset-strategy-5.c: Likewise.
* gcc.target/i386/memset-strategy-7.c: Likewise.

2 years agotestsuite: Add target requires for ifuncs to mv31.C.
Iain Sandoe [Thu, 28 Apr 2022 07:51:48 +0000 (08:51 +0100)]
testsuite: Add target requires for ifuncs to mv31.C.

g++.target/i386/mv31.C fails on targets without ifuncs support so add
the necessary target supports guard.

Signed-off-by: Iain Sandoe <iain@sandoe.co.uk>
gcc/testsuite/ChangeLog:

* g++.target/i386/mv31.C: Add target supports guard for ifuncs.

2 years agoc++: global-namespace-qualified var after class def [PR90107]
Marek Polacek [Wed, 27 Apr 2022 22:17:54 +0000 (18:17 -0400)]
c++: global-namespace-qualified var after class def [PR90107]

Here we wrongly reject the definition of "::N::a"

  struct A;
  namespace N { extern A a; }
  struct A {} ::N::a;

because our code to diagnose a missing ; after a class definition doesn't
realize that :: can follow a class definition.

PR c++/90107

gcc/cp/ChangeLog:

* parser.cc (cp_parser_class_specifier_1): Accept :: after a class
definition.

gcc/testsuite/ChangeLog:

* g++.dg/parse/qualified6.C: New test.