In this PR case LRA needs to provide too many hard regs for insn
reloads, where some reload pseudos require 8 aligned regs for
themselves. As the last attempt, LRA tries to split live ranges of
hard regs for insn reload pseudos. It is a very rare case. An
inheritance pseudo involving a reload pseudo of the insn can be
spilled in the assignment sub-pass run right after splitting and we need
to run split sub-pass for the inheritance pseudo now.
gcc/ChangeLog:
PR target/115458
* lra-int.h (LRA_MAX_FAILED_SPLITS): Define and check its value.
(lra_split_hard_reg_for): Change prototype.
* lra.cc (lra): Try to split hard reg range several times after a
failure.
* lra-assigns.cc (lra_split_hard_reg_for): Add an arg, a flag of
giving up. Report asm error and nullify the asm insn depending on
the arg value.
HOST_WIDE_INT_PRINT* macros aren't supposed to be used in
gcc-internal-format format strings, we have the w modifier for HOST_WIDE_INT
in that case, the HOST_WIDE_INT_PRINT* macros might not work properly on
some hosts (e.g. mingw32 has HOST_LONG_LONG_FORMAT "I64" and that is
something pretty-print doesn't handle, while it handles "ll" for long long)
and also the use of macros in the middle of format strings breaks
translations (both that exgettext can't retrieve the string from there
and we get
#: config/pru/pru-pragma.cc:61
msgid "%<CTABLE_ENTRY%> index %"
msgstr ""
#: config/pru/pru-pragma.cc:64
msgid "redefinition of %<CTABLE_ENTRY %"
msgstr ""
in po/gcc.pot and also the macros are different on different hosts,
so even if exgettext extracted say "%<CTABLE_ENTRY%> index %lld is not valid"
it could be translated on some hosts but not e.g. mingw32).
So, the following patch just uses %wd instead.
Tested it before/after the
patch on
#pragma ctable_entry 12 0x48040000
#pragma ctable_entry 1024 0x48040000
#pragma ctable_entry 12 0x48040001
and the result is the same.
2025-02-25 Jakub Jelinek <jakub@redhat.com>
PR translation/118991
* config/pru/pru-pragma.cc (pru_pragma_ctable_entry): Use %wd
instead of %" HOST_WIDE_INT_PRINT "d to print a hwi in error.
Patrick Palka [Tue, 25 Feb 2025 18:35:04 +0000 (13:35 -0500)]
libstdc++: Implement LWG 4027 change to possibly-const-range [PR118083]
LWG 4027 effectively makes the const range access CPOs ranges::cfoo behave
more consistently across C++23 and C++20 (pre-P2278R4) and also more
consistently with the std::cfoo range accessors, as the below testcase
adjustments demonstrate (which mostly consist of reverting workarounds
added by r14-3771-gf12e26f3496275 and r13-7186-g0d94c6df183375).
In passing fix PR118083 which reports that the input_range constraint on
possibly-const-range is missing in our implementation. A consequence of
this is that the const range access CPOs now consistently reject a non-range
argument, and so in some our of tests we need to introduce otherwise
unused begin/end members.
PR libstdc++/118083
libstdc++-v3/ChangeLog:
* include/bits/ranges_base.h
(ranges::__access::__possibly_const_range): Adjust logic as per
LWG 4027. Add missing input_range constraint.
* testsuite/std/ranges/access/cbegin.cc (test05): Verify LWG
4027 testcases.
* testsuite/std/ranges/access/cdata.cc: Adjust, simplify and
consolidate some tests after the above.
* testsuite/std/ranges/access/cend.cc: Likewise.
* testsuite/std/ranges/access/crbegin.cc: Likewise.
* testsuite/std/ranges/access/crend.cc: Likewise.
* testsuite/std/ranges/adaptors/join.cc: Likewise.
* testsuite/std/ranges/adaptors/take_while.cc: Likewise.
* testsuite/std/ranges/adaptors/transform.cc: Likewise.
Iain Buclaw [Tue, 25 Feb 2025 17:58:27 +0000 (18:58 +0100)]
libphobos: Add scripts to update Makefile.am after an upstream merge
These two scripts have been used for updating Makefile.am whenever
there's been a file added/removed from either Druntime or Phobos since
the start, but never included in the source tree.
libphobos/ChangeLog:
* libdruntime/Makefile.am: Update comment.
* libdruntime/Makefile.in: Regenerate.
* src/Makefile.am: Update comment.
* src/Makefile.in: Regenerate.
* scripts/.gitignore: New file.
* scripts/README: New file.
* scripts/gen_druntime_sources.d: New file.
* scripts/gen_phobos_sources.d: New file.
Iain Buclaw [Tue, 25 Feb 2025 17:01:09 +0000 (18:01 +0100)]
d/i386: Add CET TargetInfo key and predefined version [PR118654]
Adds a new i386 d_target_info_spec entry to handle requests for
`__traits(getTargetInfo, "CET")', and add predefined target version
`GNU_CET' when the option `-fcf-protecton' is used.
Both TargetInfo key and predefined version have been added to the D
front-end documentation.
In the library, `GNU_CET' replaces the existing use of the user-defined
version flag `CET' when building libphobos.
Iain Buclaw [Mon, 24 Feb 2025 18:57:15 +0000 (19:57 +0100)]
d: Increase max parallelism of the D testsuite
It was noticed that when running the testsuite for gdc and libphobos in
parallel, this was capped at 10 simultaneous jobs each. Increase this
limit to 128, which enables running for example `make check-d -j48` to
complete in half the time.
gcc/d/ChangeLog:
* Make-lang.in (check_gdc_parallelize): Increase to 128.
Fortran: Fix detection of descriptor arrays in coarray [PR107635]
Look at the formal arguments generated type in the function declaration
to figure if an argument is a descriptor arrays. Fix handling of class
types while splitting coarray expressions.
PR fortran/107635
gcc/fortran/ChangeLog:
* coarray.cc (fixup_comp_refs): For class types set correct
component (class) type.
(split_expr_at_caf_ref): Provide location.
* trans-intrinsic.cc (conv_caf_send_to_remote): Look at
generated formal argument and not declared one to detect
descriptor arrays.
(conv_caf_sendget): Same.
Jakub Jelinek [Tue, 25 Feb 2025 08:33:21 +0000 (09:33 +0100)]
openmp: Mark OpenMP atomic write expression as read [PR119000]
The following testcase was emitting false positive warning that
the rhs of #pragma omp atomic write was stored but not read,
when the atomic actually does read it. The following patch
fixes that by calling default_function_array_read_conversion
on it, so that it is marked as read as well as converted from
lvalue to rvalue.
Furthermore, the code had
if (code == NOP_EXPR) ... else ... if (code == NOP_EXPR) ...
with none of ... parts changing code, so I've merged the two ifs.
2025-02-25 Jakub Jelinek <jakub@redhat.com>
PR c/119000
* c-parser.cc (c_parser_omp_atomic): For omp write call
default_function_array_read_conversion on the rhs expression.
Merge the two adjacent if (code == NOP_EXPR) blocks.
Jakub Jelinek [Tue, 25 Feb 2025 08:29:39 +0000 (09:29 +0100)]
openmp: Fix handling of declare target statics with array type which need destruction [PR118876]
The following testcase ICEs because it attempts to emit the __tcfa function twice,
once when handling the host destruction and once when handling nohost destruction.
This patch fixes it by using __omp_tcfa function for the nohost case and marks it
with the needed "omp declare target" and "omp declare target nohost" attributes.
2025-02-25 Jakub Jelinek <jakub@redhat.com>
PR c++/118876
* cp-tree.h (register_dtor_fn): Add a bool argument defaulted to false.
* decl.cc (start_cleanup_fn): Add OMP_TARGET argument, use
"__omp_tcf" prefix rather than "__tcf" in that case. Add
"omp declare target" and "omp declare target nohost" attributes
to the fndecl.
(register_dtor_fn): Add OMP_TARGET argument, pass it down to
start_cleanup_fn.
* decl2.cc (one_static_initialization_or_destruction): Add OMP_TARGET
argument, pass it down to register_dtor_fn.
(emit_partial_init_fini_fn): Pass omp_target to
one_static_initialization_or_destruction.
(handle_tls_init): Pass false to
one_static_initialization_or_destruction.
Jakub Jelinek [Tue, 25 Feb 2025 08:26:46 +0000 (09:26 +0100)]
c++: Fix range for with PMFs [PR118923]
The following testcases segfault because the new range for -frange-for-ext-temps
temporary extension extends even the internal TARGET_EXPRs created by
get_member_function_from_ptrfunc.
The following patch fixes that by using get_internal_target_expr for those
instead of force_target_expr (similarly in cp_finish_decl and
build_comparison_op) and using force_target_expr inside of
get_internal_target_expr.
2025-02-25 Jakub Jelinek <jakub@redhat.com>
PR c++/118923
* tree.cc (get_internal_target_expr): Use force_target_expr
instead of build_target_expr_with_type.
* typeck.cc (get_member_function_from_ptrfunc): Use
get_internal_target_expr instead of force_target_expr.
* decl.cc (cp_finish_decl): Likewise.
* method.cc (build_comparison_op): Likewise.
* g++.dg/cpp0x/pr118923.C: New test.
* g++.dg/cpp1y/pr118923.C: New test.
Robin Dapp [Fri, 21 Feb 2025 16:08:16 +0000 (17:08 +0100)]
RISC-V: Include pattern stmts for dynamic LMUL computation [PR114516].
When scanning for program points, i.e. vector statements, we're missing
pattern statements. In PR114516 this becomes obvious as we choose
LMUL=8 assuming there are only three statements but the divmod pattern
adds another three. Those push us beyond four registers so we need to
switch to LMUL=4.
This patch adds pattern statements to the program points which helps
calculate a better register pressure estimate.
PR target/114516
gcc/ChangeLog:
* config/riscv/riscv-vector-costs.cc (compute_estimated_lmul):
Add pattern statements to program points.
gcc/testsuite/ChangeLog:
* gcc.dg/vect/costmodel/riscv/rvv/pr114516.c: New test.
Robin Dapp [Fri, 21 Feb 2025 06:19:40 +0000 (07:19 +0100)]
vect: Use original LHS type for gather pattern [PR118950].
In PR118950 we do not zero masked elements in a gather load.
While recognizing a gather/scatter pattern we do not use the original
type of the LHS. This matters because the type can differ with bool
patterns (e.g. _Bool vs unsigned char) and we don't notice the need
for zeroing out the padding bytes.
This patch just uses the original LHS's type.
PR middle-end/118950
gcc/ChangeLog:
* tree-vect-patterns.cc (vect_recog_gather_scatter_pattern): Use
original LHS's type.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/pr118950.c: New test.
Jakub Jelinek [Mon, 24 Feb 2025 11:19:16 +0000 (12:19 +0100)]
reassoc: Fix up optimize_range_tests_to_bit_test [PR118915]
The following testcase is miscompiled due to a bug in
optimize_range_tests_to_bit_test. It is trying to optimize
check for a in [-34,-34] or [-26,-26] or [-6,-6] or [-4,inf] ranges.
Another reassoc optimization folds the the test for the first
two ranges into (a + 34U) & ~8U in [0U,0U] range, and extract_bit_test_mask
actually has code to virtually undo it and treat that again as test
for a being -34 or -26. The problem is that optimize_range_tests_to_bit_test
remembers in the type variable TREE_TYPE (ranges[i].exp); from the first
range. If extract_bit_test_mask doesn't do that virtual undoing of the
BIT_AND_EXPR handling, that is just fine, the returned exp is ranges[i].exp.
But if the first range is BIT_AND_EXPR, the type could be different, the
BIT_AND_EXPR form has the optional cast to corresponding unsigned type
in order to avoid introducing UB. Now, type was used to fill in the
max value if ranges[j].high was missing in subsequently tested range,
and so in this particular testcase the [-4,inf] range which was
signed int and so [-4,INT_MAX] was treated as [-4,UINT_MAX] instead.
And we were subtracting values of 2 different types and trying to make
sense out of that.
The following patch fixes this by using the type of the low bound
(which is always non-NULL) for the max value of the high bound instead.
2025-02-24 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/118915
* tree-ssa-reassoc.cc (optimize_range_tests_to_bit_test): For
highj == NULL_TREE use TYPE_MAX_VALUE (TREE_TYPE (lowj)) rather
than TYPE_MAX_VALUE (type).
Richard Biener [Mon, 24 Feb 2025 08:45:28 +0000 (09:45 +0100)]
tree-optimization/118973 - stray abnormal edge after DCE
DCE preserves stmts performing abnormal control flow transfer but
currently has an exception for replaceable allocations and cxa_atexit
calls. That results in a broken CFG since DCE isn't set up to prune
abnormal edges possibly hanging off those.
While we could try to add this handling, the following is the safe
fix at this point and more suitable for backporting.
PR tree-optimization/118973
* tree-ssa-dce.cc (mark_stmt_if_obviously_necessary): Calls
that alter control flow in unpredictable ways need to be
preserved.
Jakub Jelinek [Mon, 24 Feb 2025 08:25:34 +0000 (09:25 +0100)]
openmp: Fix diagnostics typo [PR118993]
There is a typo in one of the OpenMP gimplification diagnostics messages.
The following patch fixes that and adjusts tests which just copied that
message including typo to dg-warning regexps in 2 tests.
Jakub Jelinek [Mon, 24 Feb 2025 08:20:47 +0000 (09:20 +0100)]
Use nonnull_if_nonzero attribute rather than nonnull on various builtins [PR117023]
On top of the
https://gcc.gnu.org/pipermail/gcc-patches/2024-November/668554.html
https://gcc.gnu.org/pipermail/gcc-patches/2024-November/668699.html
https://gcc.gnu.org/pipermail/gcc-patches/2024-November/668700.html
patches the following patch adds nonnull_if_nonzero attribute(s) to
various builtins instead of or in addition to nonnull attribute.
The patch adjusts builtins (when we have them) corresponding to the APIs
mentioned in the C2Y N3322 paper:
1) strndup and memset get one nonnull_if_nonzero attribute instead of
nonnull
2) memcpy, memmove, strncpy, memcmp, strncmp get two nonnull_if_nonzero
attributes instead of nonnull
3) strncat has nonnull without argument changed to nonnull (1) and
gets one nonnull_if_nonzero for the src argument (maybe it needs
to be clarified in C2Y, but I really think first argument to strncat
and wcsncat shouldn't be NULL even for n == 0, because NULL doesn't
point to NULL terminated string and one can't append anything to it;
and various implementations in the wild including glibc will crash
with NULL first argument (x86_64 avx+ doesn't though)
Such changes are done also to the _chk suffixed counterparts of the
builtins.
Furthermore I've changed a couple of builtins for POSIX functions which
aren't covered by ISO C, but I'd expect if/when POSIX incorporates C2Y
it would do the same changes. In particular
4) strnlen gets one nonnull_if_nonzero instead of nonnull
5) mempcpy and stpncpy get two nonnull_if_nonzero instead of nonnull
and lose returns_nonnull attribute; this is kind of unfortunate
but I think in the spirit of N3322 mempcpy (NULL, src, 0) should
return NULL (i.e. dest + n aka NULL + 0, now valid) and it is hard to
express returns non-NULL if first argument is non-NULL or third argument
is non-zero
I'm not really sure about fread/fwrite, N3322 doesn't mention those,
can the first argument be NULL if third argument is 0? What about
if second argument is 0? Can the fourth argument be NULL in such cases?
And of course, when not using builtins the glibc headers will affect stuff
too, so we'll need to wait for N3322 implementation there too (possibly
by dropping the nonnull attributes and perhaps conditionally replacing them
with this new one if the compiler supports them).
2025-02-24 Jakub Jelinek <jakub@redhat.com>
PR c/117023
gcc/
* builtin-attrs.def (ATTR_NONNULL_IF_NONZERO): New DEF_ATTR_IDENT.
(ATTR_NOTHROW_NONNULL_IF12_LEAF, ATTR_NOTHROW_NONNULL_IF13_LEAF,
ATTR_NOTHROW_NONNULL_IF123_LEAF, ATTR_NOTHROW_NONNULL_IF23_LEAF,
ATTR_NOTHROW_NONNULL_1_IF23_LEAF, ATTR_PURE_NOTHROW_NONNULL_IF12_LEAF,
ATTR_PURE_NOTHROW_NONNULL_IF13_LEAF,
ATTR_PURE_NOTHROW_NONNULL_IF123_LEAF,
ATTR_WARN_UNUSED_RESULT_NOTHROW_NONNULL_IF12_LEAF,
ATTR_MALLOC_WARN_UNUSED_RESULT_NOTHROW_NONNULL_IF12_LEAF): New
DEF_ATTR_TREE_LIST.
* builtins.def (BUILT_IN_STRNDUP): Use
ATTR_MALLOC_WARN_UNUSED_RESULT_NOTHROW_NONNULL_IF12_LEAF instead of
ATTR_MALLOC_WARN_UNUSED_RESULT_NOTHROW_NONNULL_LEAF.
(BUILT_IN_STRNCAT, BUILT_IN_STRNCAT_CHK): Use
ATTR_NOTHROW_NONNULL_1_IF23_LEAF instead of ATTR_NOTHROW_NONNULL_LEAF.
(BUILT_IN_BCOPY, BUILT_IN_MEMCPY, BUILT_IN_MEMCPY_CHK,
BUILT_IN_MEMMOVE, BUILT_IN_MEMMOVE_CHK, BUILT_IN_STRNCPY,
BUILT_IN_STRNCPY_CHK): Use ATTR_NOTHROW_NONNULL_IF123_LEAF instead of
ATTR_NOTHROW_NONNULL_LEAF.
(BUILT_IN_MEMPCPY, BUILT_IN_MEMPCPY_CHK, BUILT_IN_STPNCPY,
BUILT_IN_STPNCPY_CHK): Use ATTR_NOTHROW_NONNULL_IF123_LEAF instead of
ATTR_RETNONNULL_NOTHROW_LEAF.
(BUILT_IN_BZERO, BUILT_IN_MEMSET, BUILT_IN_MEMSET_CHK): Use
ATTR_NOTHROW_NONNULL_IF13_LEAF instead of ATTR_NOTHROW_NONNULL_LEAF.
(BUILT_IN_BCMP, BUILT_IN_MEMCMP, BUILT_IN_STRNCASECMP,
BUILT_IN_STRNCMP): Use ATTR_PURE_NOTHROW_NONNULL_IF123_LEAF instead of
ATTR_PURE_NOTHROW_NONNULL_LEAF.
(BUILT_IN_STRNLEN): Use ATTR_PURE_NOTHROW_NONNULL_IF12_LEAF instead of
ATTR_PURE_NOTHROW_NONNULL_LEAF.
(BUILT_IN_MEMCHR): Use ATTR_PURE_NOTHROW_NONNULL_IF13_LEAF instead of
ATTR_PURE_NOTHROW_NONNULL_LEAF.
gcc/testsuite/
* gcc.dg/builtins-nonnull.c (test_memfuncs, test_memfuncs_chk,
test_strfuncs, test_strfuncs_chk): Add if (n == 0) return; at the
start of the functions.
* gcc.dg/Wnonnull-2.c: Copy __builtin_* call statements where
appropriate 3 times, once with 0 length, once with n and once with
non-zero constant and expect warning only in the third case.
Formatting fixes.
* gcc.dg/Wnonnull-3.c: Copy __builtin_* call statements where
appropriate 3 times, once with 0 length, once with n and once with
n guarded with n != 0 and expect warning only in the third case.
Formatting fixes.
* gcc.dg/nonnull-3.c (foo): Use 16 instead of 0 in the calls added
for PR80936.
* gcc.dg/nonnull-11.c: New test.
* c-c++-common/ubsan/nonnull-1.c: Don't expect runtime diagnostics
for the __builtin_memcpy call.
* gcc.dg/tree-ssa/pr78154.c (f): Add dn argument and return early
if it is NULL. Duplicate cases of builtins which have the first
argument changed from nonnull to nonnull_if_nonzero except stpncpy,
once with dn as first argument instead of d and once with constant
non-zero count rather than n. Disable the stpncpy non-null check.
* gcc.dg/Wbuiltin-declaration-mismatch-14.c (test_builtin_calls):
Triplicate the strncmp calls, once with 1 last argument and expect
warning, once with n last argument and don't expect warning and
once with 0 last argument and don't expect warning.
* gcc.dg/Wbuiltin-declaration-mismatch-15.c (test_builtin_calls_fe):
Likewise.
On top of the
https://gcc.gnu.org/pipermail/gcc-patches/2024-November/668554.html
patch which introduces the nonnull_if_nonzero attribute (because
C2Y is allowing NULL arguments on various calls like memcpy, memset,
strncpy etc. as long as the count is 0) the following patch adds just
limited handling of the attribute in the analyzer.
For nonnull attribute(s) we have the get_nonnull_args helper which
returns a bitmap, for nonnull_if_nonzero a function would need to
return a hash_map or something similar, I think it is better to
handle the attributes one by one. This patch just handles the
non-zero INTEGER_CST (integer_nonzerop) count arguments, in other places
the above patch uses ranger to some extent, but I'm not familiar enough
with the analyzer to know if one can use the ranger, or should somehow
explain in data structures the conditional nature of the nonnull property,
the argument is nonnull only if some other argument is nonzero.
Also, analyzer uses get_nonnull_args in another spot when entering a frame,
not sure if anything can be done there (note the conditional nonnull
somehow, pass from callers if the argument is nonzero, ...).
Note, the testsuite changes aren't strictly necessary with just
the above and this patch, but will be with a patch I'm going to post
soon.
2025-02-24 Jakub Jelinek <jakub@redhat.com>
PR c/117023
gcc/analyzer/
* sm-malloc.cc (malloc_state_machine::handle_nonnull): New private
method.
(malloc_state_machine::on_stmt): Use it for nonnull attribute arguments.
Handle also nonnull_if_nonzero attributes.
gcc/testsuite/
* c-c++-common/analyzer/call-summaries-malloc.c
(test_use_without_check): Pass 4 rather than sz to memset.
* c-c++-common/analyzer/strncpy-1.c (test_null_dst,
test_null_src): Pass 42 rather than count to strncpy.
Jonathan Wakely [Sat, 22 Feb 2025 21:02:29 +0000 (21:02 +0000)]
libstdc++: Fix bootstrap failure with --enable-vtable-verify [PR118981]
The compiler emits code with init_priority(99) for -fvtable-verify=std
and that creates a section conflict with the uses of init_priority(99)
that I recently added to src/c++20/tzdb.cc.
Change tzdb.cc to use a different priority to avoid the conflict.
libstdc++-v3/ChangeLog:
PR c++/118981
* src/c++20/tzdb.cc: Use init_priority(98) instead of
init_priority(99).
Thomas Schwinge [Fri, 21 Feb 2025 10:21:08 +0000 (11:21 +0100)]
Turn test cases into UNSUPPORTED if running into 'sorry, unimplemented: dynamic stack allocation not supported'
In Subversion r217296 (Git commit e2acc079ff125a869159be45371dc0a29b230e92)
"Testsuite alloca fixes for ptx", effective-target 'alloca' was added to mark
up test cases that run into the nvptx back end's non-support of dynamic stack
allocation. (Later, nvptx gained conditional support for that in
commit 3861d362ec7e3c50742fc43833fe9d8674f4070e
"nvptx: PTX 'alloca' for '-mptx=7.3'+, '-march=sm_52'+ [PR65181]", but on the
other hand, in commit f93a612fc4567652b75ffc916d31a446378e6613
"bpf: liberate R9 for general register allocation", the BPF back end joined
"the list of targets that do not support alloca in target-support.exp".
Manually maintaining the list of test cases requiring effective-target 'alloca'
is notoriously hard, gets out of date quickly: new test cases added to the test
suite may need to be analyzed and annotated, and over time annotations also may
need to be removed, in cases where the compiler learns to optimize out
'alloca'/VLA usage, for example. This commit replaces (99 % of) the manual
annotations with an automatic scheme: turn test cases into UNSUPPORTED if
running into 'sorry, unimplemented: dynamic stack allocation not supported'.
Thomas Schwinge [Fri, 21 Feb 2025 20:54:59 +0000 (21:54 +0100)]
Gracefully handle the case that 'gcc/testsuite/lib/gcc-dg.exp:find-dg-do-what' has not been called (indirectly) from 'dg-test'
No change in behavior intended.
gcc/testsuite/
* lib/gcc-dg.exp (find-dg-do-what): Gracefully handle the case
that we've not be called (indirectly) from 'dg-test'.
* lib/target-supports.exp (check_effective_target_stack_size)
(check_effective_target_alloca): Catch this.
Gaius Mulley [Sat, 22 Feb 2025 16:47:21 +0000 (16:47 +0000)]
PR modula2/118978 ICE when attempting to pass an incompatible parameter
This bugfix is for a an ICE which occurs if an incompatible parameter
is passed to a procedure. In particular if a REAL constant actual
parameter is passed to INTEGER formal parameter then M2Range is invoked
to check the type and then M2Range is called to check the value range.
The value range check causes an ICE. The bug fix introduces range
dependencies on type checks. If the type check fails an
error message is generated and any future range check cancelled.
These range and type checks are tightly coupled when generating
parameter quad intermediate code.
gcc/m2/ChangeLog:
PR modula2/118978
* gm2-compiler/M2Check.mod (checkConstMeta): Add check for
typed constants.
* gm2-compiler/M2Quads.mod (BoolFrame): New field RangeDep.
(CheckProcedureParameters): Call PutRangeDep to associate the
range dependency with the parameter on the quad stack.
Pass ParamCheckId to CheckParameter.
(CheckProcTypeAndProcedure): Add ParamCheckId parameter.
Pass ParamCheckId to BuildRange.
(CheckParameter): New parameter ParamCheckId.
Pass ParamCheckId to CheckProcTypeAndProcedure.
(CheckParameterOrdinals): Add extra range dep parameter to the
call of InitParameterRangeCheck.
(ConvertBooleanToVariable): Initialize RangeDep field.
(PushBacktok): Ditto.
(OperandRangeDep): New procedure.
(PutRangeDep): Ditto.
* gm2-compiler/M2Range.def (InitTypesParameterCheck): Add new
parameter depRangeId.
(InitParameterRangeCheck): Add new parameter parentRangeId.
(FoldRangeCheck): Add new parameter range.
* gm2-compiler/M2Range.mod (InitTypesParameterCheck): Add new
parameter depRangeId.
(InitParameterRangeCheck): Add new parameter parentRangeId.
(FoldRangeCheck): Add new parameter range and rewrite.
(FoldRangeCheckLower): New procedure.
(Range): New field cancelled.
New field dependantid.
(PutRangeParam): Initialize dependantid.
(PutRangeParamAssign): Ditto.
(CheckCancelled): New procedure.
(Cancel): Ditto.
(IsCancelled): New procedure function.
(FoldTypeParam): Add depRangeId parameter.
(WriteRangeCheck): Add dependent debugging.
gcc/testsuite/ChangeLog:
PR modula2/118978
* gm2/pim/fail/badparamtype.mod: New test.
Richard Biener [Fri, 21 Feb 2025 08:58:04 +0000 (09:58 +0100)]
tree-optimization/118954 - avoid UB on ref created by predcom
When predicitive commoning moves an invariant ref it makes sure to
not build a MEM_REF with a base that is negatively offsetted from
an object. But in trying to preserve some transforms it does not
consider association of a constant offset with the address computation
in DR_BASE_ADDRESS leading to exactly this problem again. This is
arguably a problem in data-ref analysis producing such an out-of-bound
DR_BASE_ADDRESS, but this looks quite involved to fix, so the
following avoids the association in one more case. This fixes the
testcase while preserving the desired transform in
gcc.dg/tree-ssa/predcom-1.c.
PR tree-optimization/118954
* tree-predcom.cc (ref_at_iteration): Make sure to not
associate the constant offset with DR_BASE_ADDRESS when
that is an offsetted pointer.
Tobias Burnus [Fri, 21 Feb 2025 07:54:12 +0000 (08:54 +0100)]
Fortran: Improve gfc_array_kind for assumed rank; gfc_tree_array_size on 'tree'
Improve the internal and debug representation of assumed-rank arrays by
honoring the pointer and allocatable property.
Permit obtaining the array size from only a tree (via the array descriptor)
besides obtaining it from the gfc_expr's array spec. This will be used
by a follow up OpenMP patch for mapping derived types with allocatable
components.
gcc/fortran/ChangeLog:
* trans-array.cc (gfc_full_array_size): Obtain the rank from
the array descriptor for assumed rank.
(gfc_tree_array_size): Likewise; permit expr = NULL to operate
only the tree.
(gfc_conv_descriptor_stride_get): Update for added assumed-rank
array types.
* trans-openmp.cc (gfc_omp_finish_clause): Likewise.
* trans-types.cc (gfc_build_array_type, gfc_get_derived_type,
gfc_get_array_descr_info): Likewise.
* trans.h (enum gfc_array_kind): Add
GFC_ARRAY_ASSUMED_RANK_{ALLOCATABLE,POINTER{,_CONT}}.
David Malcolm [Thu, 20 Feb 2025 22:10:53 +0000 (17:10 -0500)]
sarif-replay: improve error for unescaped braces in messages (§3.11.5)
Spotted via https://github.com/llvm/llvm-project/issues/128024
gcc/ChangeLog:
* libsarifreplay.cc
(sarif_replayer::make_plain_text_within_result_message): Capture
which json::string was used. When reporting on unescaped "{" or
"}" in SARIF message strings, use that string rather than the
message object, and refer the user to §3.11.5 ("Messages with
placeholders") rather than §3.11.11 ("arguments"). Ideally we'd
place the error at the precise character, but that can't be done
without reworking json-parsing.cc's lexer::lex_string, which is
too invasive for stage 4.
(sarif_replayer::get_plain_text_from_mfms): Capture which
json::string was used.
(sarif_replayer::lookup_plain_text_within_result_message):
Likewise.
gcc/testsuite/ChangeLog:
* sarif-replay.dg/2.1.0-invalid/3.11.11-malformed-placeholder.sarif:
Rename to...
* sarif-replay.dg/2.1.0-invalid/3.11.5-unescaped-braces.sarif:
...this. Update expected subsection in error message, and
expected underline in quoted JSON.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
aarch64: Remove old aarch64_expand_sve_vec_cmp_float code
While looking at PR118956, I noticed that we had some dead code
left over after the removal of the vcond patterns. The can_invert_p
path is no longer used.
gcc/
* config/aarch64/aarch64-protos.h (aarch64_expand_sve_vec_cmp_float):
Remove can_invert_p argument and change return type to void.
* config/aarch64/aarch64.cc (aarch64_expand_sve_vec_cmp_float):
Likewise.
* config/aarch64/aarch64-sve.md (vec_cmp<mode><vpred>): Update call
accordingly.
This test has been failing since r15-1619-g3b9b8d6cfdf593, which made
IRA prefer a call-clobbered register over a call-preserved register
for mem1 (the second load). In this particular case, that just
forces the variable p3 to be allocated to a call-preserved register
instead, leading to an extra predicate move from p3 to that register.
However, it was really pot luck that this worked before. Each argument
is used exactly once, so there isn't an obvious colouring order.
And mem0 and mem1 are passed by indirect reference, so they are not
REG_EQUIV to a stack slot in the way that some memory arguments are.
IIRC, the test was the result of some experimentation, and so I think
the best fix is to rework it to try to make it less sensitive to RA
decisions. This patch does that by enabling scheduling for the
function and using both memory arguments in the same instruction.
This gets rid of the distracting prologue and epilogue code and
restricts the test to the PCS parts.
gcc/testsuite/
PR testsuite/116604
* gcc.target/aarch64/sve/pcs/args_1.c (callee_pred): Enable scheduling
and use both memory arguments in the same instruction. Expect no
prologue and epilogue code.
This patch therefore changes the test to expect the new output
for var_stride_2.c.
The changes for var_stride_4.c were a wash, with both versions
having 18(!) arithmetic instructions before the alias check branch.
Both versions sign-extend the n and m arguments as part of this
sequence; the question is whether they do it first or later.
This patch therefore changes the test to accept either the old
or the new code for var_stride_4.c.
this does only get rectified by the next full folding which happens
in forwprop4 which is after the strlen pass emitting the unwanted
diagnostic. The following mitigates this case in a similar way as
r15-7472 did for PR118817 - by ensuring we have the IL folded.
This is done by simply folding all immediate uses of the former
PHI def that SCCP replaces. All other more general approaches have
too much fallout at this point.
PR tree-optimization/118521
* tree-scalar-evolution.cc (final_value_replacement_loop):
Fold uses of the replaced PHI def.
Jonathan Wakely [Thu, 13 Feb 2025 09:40:44 +0000 (09:40 +0000)]
libstdc++: Fix invalid signed arguments to <bit> functions
These should have been unsigned, but the static assertions are only in
the public std::bit_ceil and std::bit_width functions, not the internal
__bit_ceil and __bit_width ones.
libstdc++-v3/ChangeLog:
* include/experimental/bits/simd.h (__find_next_valid_abi): Cast
__bit_ceil argument to unsigned.
* src/c++17/floating_from_chars.cc (__floating_from_chars_hex):
Cast __bit_ceil argument to unsigned.
* src/c++17/memory_resource.cc (big_block): Cast __bit_width
argument to unsigned.
Jonathan Wakely [Fri, 14 Feb 2025 15:28:32 +0000 (15:28 +0000)]
libstdc++: Remove workaround for reserved init_priority warnings
Since r15-7511-g4e7f74225116e7 we can disable the warnings for using a
reserved priority using a diagnostic pragma. That means we no longer
need to put globals using that attribute into separate files that get
included.
This replaces the two uses of such separate files by moving the variable
definition into the source file and adding the diagnostic pragma.
libstdc++-v3/ChangeLog:
* src/c++17/memory_resource.cc (default_res): Define here
instead of including default_resource.h.
* src/c++98/globals_io.cc (__ioinit): Define here instead of
including ios_base_init.h.
* src/c++17/default_resource.h: Removed.
* src/c++98/ios_base_init.h: Removed.
Jonathan Wakely [Wed, 12 Feb 2025 17:29:59 +0000 (17:29 +0000)]
libstdc++: Use init_priority attribute for tzdb globals [PR118811]
When linking statically to libstdc++.a (or to libstdc++_nonshared.a in
the RHEL devtoolset compiler) there's a static initialization order
problem where user code might be constructed before the
std::chrono::tzdb_list globals, and so might try to use them after
they've already been destroyed.
Use the init_priority attribute on those globals so that they are
initialized early. Since r15-7511-g4e7f74225116e7 we can disable the
warnings for using a reserved priority using a diagnostic pragma.
libstdc++-v3/ChangeLog:
PR libstdc++/118811
* src/c++20/tzdb.cc (tzdb_list::_Node): Use init_priority
attribute on static data members.
* testsuite/std/time/tzdb_list/pr118811.cc: New test.
Add the last missing coarray data manipulation routine using remote
accessors.
gcc/fortran/ChangeLog:
PR fortran/107635
* coarray.cc (rewrite_caf_send): Rewrite to
transfer_between_remotes when both sides of the assignment have
a coarray.
(coindexed_code_callback): Prevent duplicate rewrite.
* gfortran.texi: Add documentation for transfer_between_remotes.
* intrinsic.cc (add_subroutines): Add intrinsic symbol for
caf_sendget to allow easy rewrite to transfer_between_remotes.
* trans-decl.cc (gfc_build_builtin_function_decls): Add
prototype for transfer_between_remotes.
* trans-intrinsic.cc (conv_caf_vector_subscript_elem): Mark as
deprecated.
(conv_caf_vector_subscript): Same.
(compute_component_offset): Same.
(conv_expr_ref_to_caf_ref): Same.
(conv_stat_and_team): Extract stat and team from expr.
(gfc_conv_intrinsic_caf_get): Use conv_stat_and_team.
(conv_caf_send_to_remote): Same.
(has_ref_after_cafref): Mark as deprecated.
(conv_caf_sendget): Translate to transfer_between_remotes.
* trans.h: Add prototype for transfer_between_remotes.
libgfortran/ChangeLog:
* caf/libcaf.h: Add prototype for transfer_between_remotes.
* caf/single.c: Implement transfer_between_remotes.
gcc/testsuite/ChangeLog:
* gfortran.dg/coarray_lib_comm_1.f90: Fix up scan_trees.
Refactor to use send_to_remote instead of the slow send_by_ref.
gcc/fortran/ChangeLog:
PR fortran/107635
* coarray.cc (move_coarray_ref): Move the coarray reference out
of the given one. Especially when there is a regular array ref.
(fixup_comp_refs): Move components refs to a derived type where
the codim has been removed, aka a new type.
(split_expr_at_caf_ref): Correctly split the reference chain.
(remove_caf_ref): Simplify.
(create_get_callback): Fix some deficiencies.
(create_allocated_callback): Adapt to new signature of split.
(create_send_callback): New function.
(rewrite_caf_send): Rewrite a call to caf_send to
caf_send_to_remote.
(coindexed_code_callback): Treat caf_send and caf_sendget
correctly.
* gfortran.h (enum gfc_isym_id): Add SENDGET-isym.
* gfortran.texi: Add documentation for send_to_remote.
* resolve.cc (gfc_resolve_code): No longer generate send_by_ref
when allocatable coarray (component) is on the lhs.
* trans-decl.cc (gfc_build_builtin_function_decls): Add
caf_send_to_remote decl.
* trans-intrinsic.cc (conv_caf_func_index): Ensure the static
variables created are not in a block-scope.
(conv_caf_send_to_remote): Translate caf_send_to_remote calls.
(conv_caf_send): Renamed to conv_caf_sendget.
(conv_caf_sendget): Renamed from conv_caf_send.
(gfc_conv_intrinsic_subroutine): Branch correctly for
conv_caf_send and sendget.
* trans.h: Correct decl.
libgfortran/ChangeLog:
* caf/libcaf.h: Add/Correct prototypes for caf_get_from_remote,
caf_send_to_remote.
* caf/single.c (struct accessor_hash_t): Rename accessor_t to
getter_t.
(_gfortran_caf_register_accessor): Use new name of getter_t.
(_gfortran_caf_send_to_remote): New function for sending data to
coarray on a remote image.
gcc/testsuite/ChangeLog:
* gfortran.dg/coarray/send_char_array_1.f90: Extend test to
catch more cases.
* gfortran.dg/coarray_42.f90: Invert tests use, because no
longer a send is needed when local memory in a coarray is
allocated.
Replace caf_is_present by caf_is_present_on_remote which is using a
dedicated callback for each object to test on the remote image.
gcc/fortran/ChangeLog:
PR fortran/107635
* coarray.cc (create_allocated_callback): Add creating remote
side procedure for checking allocation status of coarray.
(rewrite_caf_allocated): Rewrite ALLOCATED on coarray to use caf
routine.
(coindexed_expr_callback): Exempt caf_is_present_on_remote from
being rewritten again.
* gfortran.h (enum gfc_isym_id): Add caf_is_present_on_remote
id.
* gfortran.texi: Add documentation for caf_is_present_on_remote.
* intrinsic.cc (add_functions): Add caf_is_present_on_remote
symbol.
* trans-decl.cc (gfc_build_builtin_function_decls): Define
interface of caf_is_present_on_remote.
* trans-intrinsic.cc (gfc_conv_intrinsic_caf_is_present_remote):
Translate caf_is_present_on_remote.
(trans_caf_is_present): Remove.
(caf_this_image_ref): Remove.
(gfc_conv_allocated): Take out coarray treatment, because that
is rewritten to caf_is_present_on_remote now.
(gfc_conv_intrinsic_function): Handle caf_is_present_on_remote
calls.
* trans.h: Add symbol for caf_is_present_on_remote and remove
old one.
libgfortran/ChangeLog:
* caf/libcaf.h (_gfortran_caf_is_present_on_remote): Add new
function.
(_gfortran_caf_is_present): Remove deprecated one.
* caf/single.c (struct accessor_hash_t): Add function ptr access
for remote side call.
(_gfortran_caf_is_present_on_remote): Added.
(_gfortran_caf_is_present): Removed.
gcc/testsuite/ChangeLog:
* gfortran.dg/coarray/coarray_allocated.f90: Adapt to new method
of checking on remote image.
* gfortran.dg/coarray_lib_alloc_4.f90: Same.
Fortran: Allow to use non-pure/non-elemental functions in coarray indexes [PR107635]
Extract calls to non-pure or non-elemental functions from index
expressions on a coarray.
gcc/fortran/ChangeLog:
PR fortran/107635
* coarray.cc (get_arrayspec_from_expr): Treat array result of
function calls correctly.
(remove_coarray_from_derived_type): Prevent memory loss.
(add_caf_get_from_remote): Correct locus.
(find_comp): New function to find or create a new component in a
derived type.
(check_add_new_comp_handle_array): Handle allocatable arrays or
non-pure/non-elemental functions in indexes of coarrays.
(check_add_new_component): Use above function.
(create_get_parameter_type): Rename to
create_caf_add_data_parameter_type.
(create_caf_add_data_parameter_type): Renaming of variable and
make the additional data a coarray.
(remove_caf_ref): Factor out to reuse in other caf-functions.
(create_get_callback): Use function factored out, set locus
correctly and ensure a kind is set for parameters.
(add_caf_get_intrinsic): Rename to add_caf_get_from_remote and
rename some variables.
(coindexed_expr_callback): Skip over function created by the
rewriter.
(coindexed_code_callback): Filter some intrinsics not to
process.
(gfc_coarray_rewrite): Rewrite also contained functions.
* trans-intrinsic.cc (gfc_conv_intrinsic_caf_get): Reflect
changed order on caf_get_from_remote ().
libgfortran/ChangeLog:
* caf/libcaf.h (_gfortran_caf_register_accessor): Reflect
changed parameter order.
* caf/single.c (struct accessor_hash_t): Same.
(_gfortran_caf_register_accessor): Call accessor using a token
for accessing arrays with a descriptor on the source side.
gcc/testsuite/ChangeLog:
* gfortran.dg/coarray_lib_comm_1.f90: Adapt scan expression.
* gfortran.dg/coarray/get_with_fn_parameter.f90: New test.
* gfortran.dg/coarray/get_with_scalar_fn.f90: New test.
Factor out generation of code to get remote function index and to
create the additional data structure. Rename caf_get_by_ct to
caf_get_from_remote.
gcc/fortran/ChangeLog:
PR fortran/107635
* gfortran.texi: Rename caf_get_by_ct to caf_get_from_remote.
* trans-decl.cc (gfc_build_builtin_function_decls): Rename
intrinsic.
* trans-intrinsic.cc (conv_caf_func_index): Factor out
functionality to be reused by other caf-functions.
(conv_caf_add_call_data): Same.
(gfc_conv_intrinsic_caf_get): Use functions factored out.
* trans.h: Rename intrinsic symbol.
libgfortran/ChangeLog:
* caf/libcaf.h (_gfortran_caf_get_by_ref): Remove from ABI.
This function is replaced by caf_get_from_remote ().
(_gfortran_caf_get_remote_function_index): Use better name.
* caf/single.c (_gfortran_caf_finalize): Free internal data.
(_gfortran_caf_get_by_ref): Remove from public interface, but
keep it, because it is still used by sendget ().
gcc/testsuite/ChangeLog:
* gfortran.dg/coarray_lib_comm_1.f90: Adapt to renamed ABI
function.
* gfortran.dg/coarray_stat_function.f90: Same.
* gfortran.dg/coindexed_1.f90: Same.
Fortran: Move caf_get-rewrite to coarray.cc [PR107635]
Add a rewriter to keep all expression tree that is not optimization
together. At the moment this is just a move from resolve.cc, but will
be extended to handle more cases where rewriting the expression tree may
be easier. The first use case is to extract accessors for coarray
remote image data access.
gcc/fortran/ChangeLog:
PR fortran/107635
* Make-lang.in: Add coarray.cc.
* coarray.cc: New file.
* gfortran.h (gfc_coarray_rewrite): New procedure.
* parse.cc (rewrite_expr_tree): Add entrypoint for rewriting
expression trees.
* resolve.cc (gfc_resolve_ref): Remove caf_lhs handling.
(get_arrayspec_from_expr): Moved to rewrite.cc.
(remove_coarray_from_derived_type): Same.
(convert_coarray_class_to_derived_type): Same.
(split_expr_at_caf_ref): Same.
(check_add_new_component): Same.
(create_get_parameter_type): Same.
(create_get_callback): Same.
(add_caf_get_intrinsic): Same.
(resolve_variable): Remove caf_lhs handling.
Richard Biener [Wed, 12 Feb 2025 10:20:10 +0000 (11:20 +0100)]
tree-optimization/86270 - improve SSA coalescing for loop exit test
The PR indicates a very specific issue with regard to SSA coalescing
failures because there's a pre IV increment loop exit test. While
IVOPTs created the desired IL we later simplify the exit test into
the undesirable form again. The following fixes this up during RTL
expansion where we try to improve coalescing of IVs. That seems
easier that trying to avoid the simplification with some weird
heuristics (it could also have been written this way).
PR tree-optimization/86270
* tree-outof-ssa.cc (insert_backedge_copies): Pattern
match a single conflict in a loop condition and adjust
that avoiding the conflict if possible.
* gcc.target/i386/pr86270.c: Adjust to check for no reg-reg
copies as well.
Patrick Palka [Thu, 20 Feb 2025 04:05:40 +0000 (23:05 -0500)]
libstdc++: Rename concat_view::iterator to ::_Iterator
Even though 'iterator' is a reserved macro name, we can't use it as the
name of this implementation detail type since it could introduce name
lookup ambiguity in valid code, e.g.
struct A { using iterator = void; }
struct B : concat_view<...>, A { using type = iterator; };
libstdc++-v3/ChangeLog:
* include/std/ranges (concat_view::iterator): Rename to ...
(concat_view::_Iterator): ... this throughout.
Patrick Palka [Thu, 20 Feb 2025 03:59:49 +0000 (22:59 -0500)]
libstdc++: Sync concat_view with final P2542 revision [PR115209]
Our concat_view implementation is accidentally based off of an older
revision of the paper, P2542R7 instead of R8. As far as I can tell the
only semantic change in the final revision is the relaxed constraints on
the iterator's iter/sent operator- overloads, which this patch updates.
This patch also simplifies the concat_view::end wording via C++26 pack
indexing as per the final revision. In turn we make the availability of
this library feature conditional on __cpp_pack_indexing. (Note pack
indexing is implemented in GCC 15 and Clang 19).
PR libstdc++/115209
libstdc++-v3/ChangeLog:
* include/bits/version.def (ranges_concat): Depend on
__cpp_pack_indexing.
* include/bits/version.h: Regenerate.
* include/std/ranges (__detail::__last_is_common): Remove.
(__detail::__all_but_first_sized): New.
(concat_view::end): Use C++26 pack indexing instead of
__last_is_common as per R8 of P2542.
(concat_view::iterator::operator-): Update constraints on
iter/sent overloads as per R8 of P2542.
Thomas Schwinge [Wed, 19 Feb 2025 08:30:45 +0000 (09:30 +0100)]
GCN, nvptx: Support '--enable-languages=all'
..., where "support" means that the build doesn't fail, but it doesn't mean
that all target libraries get built and we get pretty test results for the
additional languages.
Andrew Pinski [Wed, 19 Feb 2025 03:02:40 +0000 (19:02 -0800)]
aarch64: Fix testcase pr112105.c
This testcase started to fail with r15-268-g9dbff9c05520a7.
When late_combine was added, it was turned on for -O2+ only,
so this testcase still failed.
This changes the option to be -O2 instead of -O and the testcase
started to pass again.
tested for aarch64-linux-gnu.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/pr112105.c: Change to be -O2 rather
than -O1.
Signed-off-by: Andrew Pinski <quic_apinski@quicinc.com>
David Malcolm [Wed, 19 Feb 2025 14:46:43 +0000 (09:46 -0500)]
input: give file_cache_slot its own copy of the file path [PR118919]
input.cc's file_cache was borrowing copies of the file name.
This could lead to use-after-free when writing out sarif output
from Fortran, which frees its filenames before the sarif output
is fully written out.
Fix by taking a copy in file_cache_slot.
gcc/ChangeLog:
PR other/118919
* input.cc (file_cache_slot::m_file_path): Make non-const.
(file_cache_slot::evict): Free m_file_path.
(file_cache_slot::create): Store a copy of file_path if non-null.
(file_cache_slot::~file_cache_slot): Free m_file_path.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
David Malcolm [Wed, 19 Feb 2025 14:44:46 +0000 (09:44 -0500)]
analyzer: handle more IFN_UBSAN_* as no-ops [PR118300]
Previously the analyzer treated IFN_UBSAN_BOUNDS as a no-op, but
the other IFN_UBSAN_* were unrecognized and conservatively treated
as having arbitrary behavior.
Treat IFN_UBSAN_NULL and IFN_UBSAN_PTR also as no-ops, which should
make -fanalyzer behave better with -fsanitize=undefined.
gcc/analyzer/ChangeLog:
PR analyzer/118300
* kf.cc (class kf_ubsan_bounds): Replace this with...
(class kf_ubsan_noop): ...this.
(register_sanitizer_builtins): Use it to handle IFN_UBSAN_NULL,
IFN_UBSAN_BOUNDS, and IFN_UBSAN_PTR as nop-ops.
(register_known_functions): Drop handling of IFN_UBSAN_BOUNDS
here, as it's now handled by register_sanitizer_builtins above.
gcc/testsuite/ChangeLog:
PR analyzer/118300
* gcc.dg/analyzer/ubsan-pr118300.c: New test.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
Pan Li [Wed, 19 Feb 2025 01:37:51 +0000 (09:37 +0800)]
Vect: Fix ICE when vect_verify_loop_lens acts on relevant mode [PR116351]
This patch would like to fix the ICE similar as below, assump we have
sample code:
1 │ int a, b, c;
2 │ short d, e, f;
3 │ long g (long h) { return h; }
4 │
5 │ void i () {
6 │ for (; b; ++b) {
7 │ f = 5 >> a ? d : d << a;
8 │ e &= c | g(f);
9 │ }
10 │ }
It will ice when compile with -O3 -march=rv64gc_zve64f -mrvv-vector-bits=zvl
During vectorization the override_widen pattern matched and then will get DImode
as vector_mode in loop_info. After that the loop_vinfo will step in vect_analyze_xx
with below flow:
vect_analyze_loop_2
|- vect_pattern_recog // over-widening and set loop_vinfo->vector_mode to DImode
|- ...
|- vect_analyze_loop_operations
|- stmt_info->def_type == vect_reduction_def
|- stmt_info->slp_type == pure_slp
|- vectorizable_lc_phi // Not Hit
|- vectorizable_induction // Not Hit
|- vectorizable_reduction // Not Hit
|- vectorizable_recurr // Not Hit
|- vectorizable_live_operation // Not Hit
|- vect_analyze_stmt
|- stmt_info->relevant == vect_unused_in_scope
|- stmt_info->live == false
|- p pattern_stmt_info == (stmt_vec_info) 0x0
|- return opt_result::success ();
OR
|- PURE_SLP_STMT (stmt_info) && !node then dump "handled only by SLP analysis\n"
|- Early return opt_result::success ();
|- vectorizable_load/store/call_convert/... // Not Hit
|- LOOP_VINFO_CAN_USE_PARTIAL_VECTORS_P && !LOOP_VINFO_MASKS(loop_vinfo).is_empty ()
|- vect_verify_loop_lens (loop_vinfo)
|- assert (VECTOR_MODE_P (loop_vinfo->vector_mode); // Hit assert result in ICE
Finally, the DImode in loop_vinfo will hit the assert (VECTOR_MODE_P (mode))
in vect_verify_loop_lens. This patch would like to return false
directly if the loop_vinfo has relevant mode like DImode for the ICE
fix, but still may have mis-optimization for similar cases. We will try
to cover that in separated patches.
The below test suites are passed for this patch.
* The rv64gcv fully regression test.
* The x86 bootstrap test.
* The x86 fully regression test.
PR middle-end/116351
gcc/ChangeLog:
* tree-vect-loop.cc (vect_verify_loop_lens): Return false if the
loop_vinfo has relevant mode such as DImode.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/base/pr116351-1.c: New test.
* gcc.target/riscv/rvv/base/pr116351-2.c: New test.
* gcc.target/riscv/rvv/base/pr116351.h: New test.
Xi Ruoyao [Mon, 20 Jan 2025 15:13:19 +0000 (23:13 +0800)]
LoongArch: Implement [su]dot_prod* for LSX and LASX modes
Despite it's just a special case of "a widening product of which the
result used for reduction," having these standard names allows to
recognize the dot product pattern earlier and it may be beneficial to
optimization. Also fix some test failures with the test cases:
Xi Ruoyao [Mon, 20 Jan 2025 12:43:07 +0000 (20:43 +0800)]
LoongArch: Implement vec_widen_mult_{even,odd}_* for LSX and LASX modes
Since PR116142 has been fixed, now we can add the standard names so the
compiler will generate better code if the result of a widening
production is reduced.
gcc/ChangeLog:
* config/loongarch/simd.md (even_odd): New define_int_attr.
(vec_widen_<su>mult_<even_odd>_<mode>): New define_expand.
gcc/testsuite/ChangeLog:
* gcc.target/loongarch/wide-mul-reduc-1.c: New test.
* gcc.target/loongarch/wide-mul-reduc-2.c: New test.
Xi Ruoyao [Tue, 11 Feb 2025 18:42:58 +0000 (02:42 +0800)]
LoongArch: Simplify lsx_vpick description
Like what we've done for {lsx_,lasx_x}v{add,sub,mul}l{ev,od}, use
special predicates instead of hard-coded const vectors.
This is not suitable for LASX where lasx_xvpick has a different
semantic.
gcc/ChangeLog:
* config/loongarch/simd.md (LVEC): New define_mode_attr.
(simdfmt_as_i): Make it same as simdfmt for integer vector
modes.
(_f): New define_mode_attr.
* config/loongarch/lsx.md (lsx_vpickev_b): Remove.
(lsx_vpickev_h): Remove.
(lsx_vpickev_w): Remove.
(lsx_vpickev_w_f): Remove.
(lsx_vpickod_b): Remove.
(lsx_vpickod_h): Remove.
(lsx_vpickod_w): Remove.
(lsx_vpickev_w_f): Remove.
(lsx_pick_evod_<mode>): New define_insn.
(lsx_<x>vpick<ev_od>_<simdfmt_as_i><_f>): New
define_expand.
These pattern definitions are tediously long, invoking 32 UNSPECs and
many hard-coded long const vectors. To simplify them, at first we use
the TImode vector operations instead of the UNSPECs, then we adopt an
approach in AArch64: using a special predicate to match the const
vectors for odd/even indices for define_insn's, and generate those
vectors in define_expand's.
For "backward compatibilty" we need to provide a "punned" version for
the operations invoking TImode vectors as the intrinsics still expect
DImode vectors.
Xi Ruoyao [Fri, 24 Jan 2025 00:33:39 +0000 (08:33 +0800)]
LoongArch: Allow moving TImode vectors
We have some vector instructions for operations on 128-bit integer, i.e.
TImode, vectors. Previously they had been modeled with unspecs, but
it's more natural to just model them with TImode vector RTL expressions.
For the preparation, allow moving V1TImode and V2TImode vectors in LSX
and LASX registers so we won't get a reload failure when we start to
save TImode vectors in these registers.
This implicitly depends on the vrepli optimization: without it we'd try
"vrepli.q" which does not really exist and trigger an ICE.
gcc/ChangeLog:
* config/loongarch/lsx.md (mov<LSX:mode>): Remove.
(movmisalign<LSX:mode>): Remove.
(mov<LSX:mode>_lsx): Remove.
* config/loongarch/lasx.md (mov<LASX:mode>): Remove.
(movmisalign<LASX:mode>): Remove.
(mov<LASX:mode>_lasx): Remove.
* config/loongarch/loongarch-modes.def (V1TI): Add.
(V2TI): Mention in the comment.
* config/loongarch/loongarch.md (mode): Add V1TI and V2TI.
* config/loongarch/simd.md (ALLVEC_TI): New mode iterator.
(mov<ALLVEC_TI:mode): New define_expand.
(movmisalign<ALLVEC_TI:mode>): Likewise.
(mov<ALLVEC_TI:mode>_simd): New define_insn_and_split.
Xi Ruoyao [Fri, 24 Jan 2025 00:32:18 +0000 (08:32 +0800)]
LoongArch: Try harder using vrepli instructions to materialize const vectors
For
a = (v4si){0xdddddddd, 0xdddddddd, 0xdddddddd, 0xdddddddd}
we just want
vrepli.b $vr0, 0xdd
but the compiler actually produces a load:
la.local $r14,.LC0
vld $vr0,$r14,0
It's because we only tried vrepli.d which wouldn't work. Try all vrepli
instructions for const int vector materializing to fix it.
gcc/ChangeLog:
* config/loongarch/loongarch-protos.h
(loongarch_const_vector_vrepli): New function prototype.
* config/loongarch/loongarch.cc (loongarch_const_vector_vrepli):
Implement.
(loongarch_const_insns): Call loongarch_const_vector_vrepli
instead of loongarch_const_vector_same_int_p.
(loongarch_split_vector_move_p): Likewise.
(loongarch_output_move): Use loongarch_const_vector_vrepli to
pun operend[1] into a better mode if it's a const int vector,
and decide the suffix of [x]vrepli with the new mode.
* config/loongarch/constraints.md (YI): Call
loongarch_const_vector_vrepli instead of
loongarch_const_vector_same_int_p.
Xi Ruoyao [Mon, 10 Feb 2025 15:39:24 +0000 (23:39 +0800)]
LoongArch: Accept ADD, IOR or XOR when combining objects with no bits in common [PR115478]
Since r15-1120, multi-word shifts/rotates produces PLUS instead of IOR.
It's generally a good thing (allowing to use our alsl instruction or
similar instrunction on other architectures), but it's preventing us
from using bytepick. For example, if we shift a __int128 by 16 bits,
the higher word can be produced via a single bytepick.d instruction with
immediate 2, but we got:
srli.d $r12,$r4,48
slli.d $r5,$r5,16
slli.d $r4,$r4,16
add.d $r5,$r12,$r5
jr $r1
This wasn't work with GCC 14, but after r15-6490 it's supposed to work
if IOR was used instead of PLUS.
To fix this, add a code iterator to match IOR, XOR, and PLUS and use it
instead of just IOR if we know the operands have no overlapping bits.
gcc/ChangeLog:
PR target/115478
* config/loongarch/loongarch.md (any_or_plus): New
define_code_iterator.
(bstrins_<mode>_for_ior_mask): Use any_or_plus instead of ior.
(bytepick_w_<bytepick_imm>): Likewise.
(bytepick_d_<bytepick_imm>): Likewise.
(bytepick_d_<bytepick_imm>_rev): Likewise.
gcc/testsuite/ChangeLog:
PR target/115478
* gcc.target/loongarch/bytepick_shift_128.c: New test.
Jeff Law [Wed, 19 Feb 2025 02:45:29 +0000 (19:45 -0700)]
[PR middle-end/113525] Drop obsolete options from documentation
The sibling and unshare passes were dropped as distinct passes 10+ years ago.
Docs weren't ever updated. This just removes them; given their age I don't
think we need to keep them around any longer.
PR middle-end/113525
gcc/
* doc/invoke.texi (dump-rtl-sibling): Drop documentation for pass
removed long ago.
(dump-rtl-unshare): Likewise.
David Malcolm [Tue, 18 Feb 2025 21:54:57 +0000 (16:54 -0500)]
analyzer: add more properties to sarif output
Add some more properties to the analyzer's sarif output, to
help with debugging -fanalyzer.
gcc/analyzer/ChangeLog:
* diagnostic-manager.cc
(saved_diagnostic::maybe_add_sarif_properties): Add various
properties for debugging, for m_stmt, m_var, and m_duplicates.
Remove stray 'if' statement. Capture the kind of the
pending_diagnostic.
* region-model.cc
(poisoned_value_diagnostic::maybe_add_sarif_properties): New.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
David Malcolm [Tue, 18 Feb 2025 21:54:52 +0000 (16:54 -0500)]
sarif output: fix alphabetization in sarif_scheme_handler::make_sink
No functional change intended.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
gcc/ChangeLog:
* opts-diagnostic.cc (sarif_scheme_handler::make_sink): Put
properties in alphabetical order.
Signed-off-by: David Malcolm <dmalcolm@redhat.com>
Roman Kagan [Thu, 2 Jan 2025 15:32:17 +0000 (16:32 +0100)]
libgcc: i386/linux-unwind.h: always rely on sys/ucontext.h
When gcc is built for x86_64-linux-musl target, stack unwinding from
within signal handler stops at the innermost signal frame. The reason
for this behaviro is that the signal trampoline is not accompanied with
appropiate CFI directives, and the fallback path in libgcc to recognize
it by the code sequence is only enabled for glibc except 2.0. The
latter is motivated by the lack of sys/ucontext.h in that glibc version.
Given that all relevant libc-s ship sys/ucontext.h for over a decade,
and that other arches aren't shy of unconditionally using it, follow
suit and remove the preprocessor condition, too.
libgcc/ChangeLog:
* config/i386/linux-unwind.h: Remove preprocessor
condition to enable fallback path for all libc-s.
The result is vsetvl zero, zero, e64, mf2, ta, ma. The previous vsetvl
set vl = 4 but here we wrongly set it to vl = 2. As all the following
vsetvls only ever change the ratio we never recover.
The issue is quite difficult to trigger because we can often
deduce the value of d at runtime. Then very check for the value of
d will be optimized away.
The last known bad commit is r15-3458-g5326306e7d9d36. With that commit
the output is wrong but -fno-schedule-insns makes it correct. From the
next commit on the issue is latent. I still added the PR's test as scan
and run check even if they don't trigger right now. Not sure if the
run test will ever fail but well. I verified that the
patch fixes the issue when applied on top of r15-3458-g5326306e7d9d36.
PR target/115703
gcc/ChangeLog:
* config/riscv/riscv-vsetvl.cc: Use max_sew for calculating the
new LMUL.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/pr115703-run.c: New test.
* gcc.target/riscv/rvv/autovec/pr115703.c: New test.
Tobias Burnus [Tue, 18 Feb 2025 14:48:39 +0000 (15:48 +0100)]
gfortran.dg/gomp/metadirective-3.f90: xfail on offload_nvptx
Currently, 'target' with a nested metadirective creating a 'teams' will
fail with a bogus error ("‘target’ construct with nested ‘teams’ construct
contains directives outside of the ‘teams’ construct").
That's tracked at PR118694 - and, hence, expected.
However, the testcase metadirective-3.f90 triggers this when compiling for
'target offload_nvptx' (otherwise, the code is optimized away). Use xfail to
silence the error as it is known and there is a tracking PR.
gcc/testsuite/ChangeLog:
* gfortran.dg/gomp/metadirective-3.f90: Add xfail when
compiling for offload_nvptx.
late-combine: Tighten register class check [PR108840]
gcc.target/aarch64/pr108840.c has failed since r15-268-g9dbff9c05520
(which means that I really ought to have looked at it earlier).
The test wants us to fold an SImode AND into all shifts that use it.
This is something that late-combine is supposed to do, but:
(1) the pre-RA pass chickened out because of a register pressure check
(2) the post-RA pass can't handle it, because the shift uses are in
QImode and the sets are in SImode
Both are things that would be good to fix. But (1) is particularly
silly. The constraints on the AND have "rk" for the destination
(so allowing the stack pointer) and "r" for the first source.
Including the stack pointer made the destination seem more permissive
than the source.
The intention was instead to check whether there are any
*allocatable* registers in the destination class that aren't
present in the source.
That's enough for all tests but the last one. The last one still
fails because combine merges the final shift with the move into
the hard return register, giving an arithmetic instruction with
a hard register destination. Pre-RA late-combine currently punts
on those, again due to register pressure concerns. That too is
something I'd like to relax, but not for GCC 15. In the interim,
the best thing seems to be to disable combine for the test.
gcc/
PR rtl-optimization/108840
* late-combine.cc (late_combine::check_register_pressure):
Take only allocatable registers into account when checking
the permissiveness of register classes.
gcc/testsuite/
PR rtl-optimization/108840
* gcc.target/aarch64/pr108840.c: Run at -O2 but disable combine.
Alex Coplan [Tue, 18 Feb 2025 10:48:50 +0000 (10:48 +0000)]
pair-fusion: Tweak wording in dump message [PR118320]
As discussed in
https://gcc.gnu.org/pipermail/gcc-patches/2025-February/675978.html
this tweaks the dump messasge added with the fix for PR118320 since it doesn't
just apply to load pairs.
gcc/ChangeLog:
PR rtl-optimization/118320
* pair-fusion.cc (pair_fusion_bb_info::fuse_pair): Tweak wording in dump
message when punting on invalid use arrays.
Richard Biener [Mon, 17 Feb 2025 14:53:11 +0000 (15:53 +0100)]
tree-optimization/98845 - ICE with tail-merging and DCE/DSE disabled
The following shows that tail-merging will make dead SSA defs live
in paths where it wasn't before, possibly introducing UB or as
in this case, uses of abnormals that eventually fail coalescing
later. The fix is to register such defs for stmt comparison.
PR tree-optimization/98845
* tree-ssa-tail-merge.cc (stmt_local_def): Consider a
def with no uses not local.
* gcc.dg/pr98845.c: New testcase.
* gcc.dg/pr81192.c: Adjust.
projects / thirdparty / gcc.git / log
