Jakub Jelinek [Thu, 26 Sep 2024 09:55:13 +0000 (11:55 +0200)]
pretty-print: Fix up allocate_object
On Thu, Aug 29, 2024 at 06:58:12PM -0400, David Malcolm wrote:
> The following patch rewrites the internals of pp_format.
> The tokens and token lists are allocated on the chunk_obstack, and so
> there's no additional heap activity required, with the memory reclaimed
> when the chunk_obstack is freed after phase 3 of formatting.
> +static void *
> +allocate_object (size_t sz, obstack &s)
> +{
> + /* We must not be half-way through an object. */
> + gcc_assert (obstack_base (&s) == obstack_next_free (&s));
> +
> + obstack_grow (&s, obstack_base (&s), sz);
> + void *buf = obstack_finish (&s);
> + return buf;
> }
I think this is wrong. I hoped it would be the reason of the
unexpected libstdc++ warnings on certain architectures after
seeing
==4027220== Source and destination overlap in memcpy(0x4627154, 0x4627154, 12)
==4027220== at 0x404B93E: memcpy (vg_replace_strmem.c:1123)
==4027220== by 0xAAD5618: allocate_object(unsigned int, obstack&) (pretty-print.cc:1183)
==4027220== by 0xAAD8C0E: operator new (pretty-print.cc:1210)
==4027220== by 0xAAD8C0E: make (pretty-print-format-impl.h:305)
==4027220== by 0xAAD8C0E: format_phase_1 (pretty-print.cc:1659)
==4027220== by 0xAAD8C0E: pretty_printer::format(text_info&) (pretty-print.cc:1618)
==4027220== by 0xAAA840E: pp_format (pretty-print.h:583)
==4027220== by 0xAAA840E: diagnostic_context::report_diagnostic(diagnostic_info*) (diagnostic.cc:1260)
==4027220== by 0xAAA8703: diagnostic_context::diagnostic_impl(rich_location*, diagnostic_metadata const*, diagnostic_option_id, char const*, char**, diagnostic_t) (diagnostic.cc:1404)
==4027220== by 0xAAB8682: warning(diagnostic_option_id, char const*, ...) (diagnostic-global-context.cc:166)
==4027220== by 0x97725F5: warn_deprecated_use(tree_node*, tree_node*) (tree.cc:12485)
==4027220== by 0x8B6694B: mark_used(tree_node*, int) (decl2.cc:6121)
==4027220== by 0x8C9E25E: tsubst_expr(tree_node*, tree_node*, int, tree_node*) [clone .part.0] (pt.cc:21626)
==4027220== by 0x8C9E5E6: tsubst_expr(tree_node*, tree_node*, int, tree_node*) [clone .part.0] (pt.cc:20935)
==4027220== by 0x8C9E1D7: tsubst_expr(tree_node*, tree_node*, int, tree_node*) [clone .part.0] (pt.cc:20424)
==4027220== by 0x8C9DF2E: tsubst_expr(tree_node*, tree_node*, int, tree_node*) [clone .part.0] (pt.cc:20496)
==4027220==
etc. valgrind warnings, unfortunately it is not, but still
I think this is a bug.
If the obstack has enough space in it, i.e. if obstack_room (&s) >= sz,
then obstack_grow from obstack_base will copy uninitialized bytes
through memcpy (obstack_base (&s), obstack_base (&s), sz);
(which pedantically isn't valid due to the overlap, and so
the reason why valgrind complains, but in reality I think most
implementations can handle it fine, after all, we also use it for
structure assignments which could have full or no overlap but never
partial).
If obstack_room (&s) < sz, then obstack_grow will first
_obstack_newchunk (&s, sz); which will allocate new memory and
copy the existing data of the object (but the above assertion
guartantees it will copy 0 bytes) and then the memcpy copies
sz bytes from the old base to the new (if unlucky, that could crash
as there could be end of page and unmapped next page in between).
I think we should use obstack_blank instead of obstack_grow, which
does everything obstack_grow does, except for the memcpy of the
uninitialized data.
2024-09-25 Jakub Jelinek <jakub@redhat.com>
* pretty-print.cc (allocate_object): Use obstack_blank rather than
obstack_grow.
Levy Hsu [Wed, 11 Sep 2024 04:49:02 +0000 (14:19 +0930)]
x86: Extend AVX512 Vectorization for Popcount in Various Modes
This patch enables vectorization of the popcount operation for V2QI, V4QI,
V8QI, V2HI, V4HI, and V2SI modes.
gcc/ChangeLog:
* config/i386/mmx.md:
(VQI_16_32_64): New mode iterator for 8-byte, 4-byte, and 2-byte QImode.
(popcount<mode>2): New pattern for popcount of V2QI/V4QI/V8QI mode.
(popcount<mode>2): New pattern for popcount of V2HI/V4HI mode.
(popcountv2si2): New pattern for popcount of V2SI mode.
gcc/testsuite/ChangeLog:
* gcc.target/i386/part-vect-popcount-1.c: New test.
r15-3878 exposed a mistake in the testcase, probably from an older
version of the dumping logic.
Apart from the slightly different syntax for the dump line, also check
for importing the type_decl rather than the const_decl (we need the type
anyway and importing the type also brings along the enumerators so it
would be unnecessary to seed an import for them as well).
Pan Li [Sat, 21 Sep 2024 04:51:58 +0000 (12:51 +0800)]
RISC-V: Add testcases for form 3 of signed vector SAT_ADD
Form 3:
#define DEF_VEC_SAT_S_ADD_FMT_3(T, UT, MIN, MAX) \
void __attribute__((noinline)) \
vec_sat_s_add_##T##_fmt_3 (T *out, T *op_1, T *op_2, unsigned limit) \
{ \
unsigned i; \
for (i = 0; i < limit; i++) \
{ \
T x = op_1[i]; \
T y = op_2[i]; \
T sum; \
bool overflow = __builtin_add_overflow (x, y, &sum); \
out[i] = overflow ? x < 0 ? MIN : MAX : sum; \
} \
}
The below test are passed for this patch.
* The rv64gcv fully regression test.
It is test only patch and obvious up to a point, will commit it
directly if no comments in next 48H.
gcc/testsuite/ChangeLog:
* gcc.target/riscv/rvv/autovec/vec_sat_arith.h: Add test helper macros.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-10.c: New test.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-11.c: New test.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-12.c: New test.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-9.c: New test.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-run-10.c: New test.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-run-11.c: New test.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-run-12.c: New test.
* gcc.target/riscv/rvv/autovec/binop/vec_sat_s_add-run-9.c: New test.
Pan Li [Sat, 21 Sep 2024 01:28:39 +0000 (09:28 +0800)]
Match: Support form 3 for vector signed integer .SAT_ADD
This patch would like to support the form 3 of the vector signed
integer .SAT_ADD. Aka below example:
Form 3:
#define DEF_VEC_SAT_S_ADD_FMT_3(T, UT, MIN, MAX) \
void __attribute__((noinline)) \
vec_sat_s_add_##T##_fmt_3 (T *out, T *op_1, T *op_2, unsigned limit) \
{ \
unsigned i; \
for (i = 0; i < limit; i++) \
{ \
T x = op_1[i]; \
T y = op_2[i]; \
T sum; \
bool overflow = __builtin_add_overflow (x, y, &sum); \
out[i] = overflow ? x < 0 ? MIN : MAX : sum; \
} \
}
gfortran testsuite: Remove unit-files in files having open-statements, PR116701
PR testsuite/116701 shows that left-behind files from
unnamed gfortran open statements (named unit.N, where N =
unit number) can interfere with the result of a subsequent
run. While that's unlikely to happen for a "real" fortran
target or a test with a deleting close-statement, test-cases
should not rely on previous test-cases passing and not
execute along different execution paths depending on earlier
runs, even if the difference is benevolent.
Most but not all fortran test-cases go through
gfortran-dg-runtest (gfortran.dg) or fortran-torture-execute
(gfortran.fortran-torture). However, the exceptions, with
more complex framework and call-chains, either don't run or
don't have open-statements, so a more complex solution
doesn't seem worthwhile. If test-cases with open-statements
are added later to those parts of the test-suite, calls to
fortran-delete-unit-files at the right spot may be added or
worst case, "manual" cleanup-calls added, like:
! { dg-final { remote_file target delete "fort.10" } }
Put the new proc in fortran-modules.exp since that's where other
common fortran-testsuite dejagnu-library functions are located.
PR testsuite/116701
* lib/fortran-modules.exp (fortran-delete-unit-files): New proc.
* lib/gfortran-dg.exp (gfortran-dg-runtest): Call
fortran-delete-unit-files after executing test.
* lib/fortran-torture.exp (fortran-torture-execute): Ditto.
Mikael Morin [Wed, 25 Sep 2024 19:23:00 +0000 (21:23 +0200)]
doc: Remove @code wrapping of fortran option names [PR116801]
The documentation of gfortran options uses @code wrappings for arguments
to @opindex. This is superfluous, as 'op' index is a texinfo 'code' index,
that is it already implicitly formats its arguments as if in a @code block.
The superfluous wrapping has the effect of creating a nested
<code class="..."> tag inside the regular automatic <code> tag, in the
option index HTML page, preventing the recognition of the corresponding
option by the option URL generation script.
This change removes those superfluous @code wrappings. Additionally,
variables appearing as separate argument in index are removed, permitting
a few more URL recognition. Finally, the URL files are regenerated with the
new URLs recognized on the updated HTML files.
By the way, a spurious 'option' is removed from the label of the std= option
in the index, without any effect on URL recognition.
PR other/116801
gcc/fortran/ChangeLog:
* invoke.texi: Remove @code wrapping in arguments to @opindex.
(std=): Remove spurious 'option' in index.
(idirafter, imultilib, iprefix, isysroot, iquote, isystem,
fintrinsic-modules-path): Remove variable from index.
* lang.opt.urls: Regenerate.
Jakub Jelinek [Wed, 25 Sep 2024 18:17:11 +0000 (20:17 +0200)]
i386: Add GENERIC and GIMPLE folders of __builtin_ia32_{min,max}* [PR116738]
The following patch adds GENERIC and GIMPLE folders for various
x86 min/max builtins.
As discussed, these builtins have effectively x < y ? x : y
(or x > y ? x : y) behavior.
The GENERIC folding is done if all the (relevant) arguments are
constants (such as VECTOR_CST for vectors) and is done because
the GIMPLE folding can't easily handle masking, rounding and the
ss/sd cases (in a way that it would be pattern recognized back to the
corresponding instructions). The GIMPLE folding is also done just
for TARGET_SSE4 or later when optimizing, otherwise it is apparently
not matched back.
if address override is used to avoid the invalid memory operand like
cmpl %fs:previous_emax@dtpoff(%eax), %r12d
gcc/
PR target/116839
* config/i386/i386.cc (ix86_rewrite_tls_address_1): Make it
static. Return if TLS address is thread register plus an integer
register.
gcc/testsuite/
PR target/116839
* gcc.target/i386/pr116839.c: New file.
In particular, this now allows some harmless diagnostic flags (especially
useful for things like -Werror=odr), more optimization flags, and some
Clang-specific options.
GCC's -flto documentation mentions:
> To use the link-time optimizer, -flto and optimization options should be
> specified at compile time and during the final link. It is recommended
> that you compile all the files participating in the same link with the
> same options and also specify those options at link time.
This allows compliance with that.
* ltmain.sh (func_mode_link): Allow various flags through filter.
Andrew Pinski [Tue, 24 Sep 2024 02:17:42 +0000 (19:17 -0700)]
Add an alternative testcase for PR 70740
While looking into improving phiprop, I noticed that
the current pr70740.c testcase was being optimized almost
all the way before phiprop because the addresses were considered
the same; the arrays were all zero in size.
This adds an alternative testcase which changes the array sizes to be 1
and phiprop can and will act on this testcase now and the fix which was
being tested is actually tested now.
Tested on x86_64-linux-gnu.
PR tree-optimization/70740
gcc/testsuite/ChangeLog:
* gcc.dg/torture/pr70740-1.c: New test.
Signed-off-by: Andrew Pinski <quic_apinski@quicinc.com>
Andrew Pinski [Thu, 19 Sep 2024 20:50:14 +0000 (13:50 -0700)]
match: Fix `a != 0 ? a * b : 0` patterns for things that trap [PR116772]
For generic, `a != 0 ? a * b : 0` would match where `b` would be an expression
which trap (in the case of the testcase, it was an integer division but it could be any).
This adds a new helper function, expr_no_side_effects_p which tests if there is no side effects
and the expression is not trapping which might be used in other locations.
Changes since v1:
* v2: Add move check to helper function instead of inlining it.
PR middle-end/116772
gcc/ChangeLog:
* generic-match-head.cc (expr_no_side_effects_p): New function
* gimple-match-head.cc (expr_no_side_effects_p): New function
* match.pd (`a != 0 ? a / b : 0`): Check expr_no_side_effects_p.
(`a != 0 ? a * b : 0`, `a != 0 ? a & b : 0`): Likewise.
gcc/testsuite/ChangeLog:
* gcc.dg/torture/pr116772-1.c: New test.
Signed-off-by: Andrew Pinski <quic_apinski@quicinc.com>
Jakub Jelinek [Wed, 25 Sep 2024 14:07:50 +0000 (16:07 +0200)]
c++: Add testcase for DR 2836
Seems we already handle it the way the DR clarifies, if double/long double
and std::float64_t have the same mode, foo has long double type (while
x + y would be _Float64 in C23), so this patch just adds a testcase which
verifies that.
match: Fix A || B not optimized to true when !B implies A [PR114326]
In expressions like (a != b || ((a ^ b) & c) == d) and
(a != b || (a ^ b) == c), (a ^ b) is folded to false.
In the equivalent expressions (((a ^ b) & c) == d || a != b) and
((a ^ b) == c || a != b) this is not happening.
This patch adds the following simplifications in match.pd:
((a ^ b) & c) cmp d || a != b --> 0 cmp d || a != b
(a ^ b) cmp c || a != b --> 0 cmp c || a != b
PR tree-optimization/114326
gcc/ChangeLog:
* match.pd: Add two patterns to fold a ^ b to 0, when a == b.
gcc/testsuite/ChangeLog:
* gcc.dg/tree-ssa/fold-xor-and-or.c: New test.
* gcc.dg/tree-ssa/fold-xor-or.c: New test.
Tested-by: Christoph Müllner <christoph.muellner@vrull.eu> Signed-off-by: Philipp Tomsich <philipp.tomsich@vrull.eu> Signed-off-by: Konstantinos Eleftheriou <konstantinos.eleftheriou@vrull.eu>
Richard Biener [Wed, 25 Sep 2024 11:15:42 +0000 (13:15 +0200)]
Speed up wide_int_storage::operator=(wide_int_storage const&)
wide_int_storage shows up high in the profile for the testcase in
PR114855 where the apparent issue is that the conditional jump
on 'precision' after the (inlined) memcpy stalls the pipeline due
to the data dependence and required store-to-load forwarding. We
can add scheduling freedom by instead testing precision as from the
source which speeds up the function by 30%. I've applied the
same logic to the copy CTOR.
* wide-int.h (wide_int_storage::wide_int_storage): Branch
on source precision to avoid data dependence on memcpy
destination.
(wide_int_storage::operator=): Likewise.
match: Change (A * B) + (-C) to (B - C/A) * A, if C multiple of A [PR109393]
The following function:
int foo(int *a, int j)
{
int k = j - 1;
return a[j - 1] == a[k];
}
does not fold to `return 1;` using -O2 or higher. The cause of this is that
the expression `4 * j + (-4)` for the index computation is not folded to
`4 * (j - 1)`. Existing simplifications that handle similar cases are applied
when A == C, which is not the case in this instance.
A previous attempt to address this issue is
https://gcc.gnu.org/pipermail/gcc-patches/2024-April/649896.html
This patch adds the following simplification in match.pd:
(A * B) + (-C) -> (B - C/A) * A, if C a multiple of A
which also handles cases where the index is j - 2, j - 3, etc.
Bootstrapped for all languages and regression tested on x86-64 and aarch64.
PR tree-optimization/109393
gcc/ChangeLog:
* match.pd: (A * B) + (-C) -> (B - C/A) * A, if C a multiple of A.
gcc/testsuite/ChangeLog:
* gcc.dg/pr109393.c: New test.
Tested-by: Christoph Müllner <christoph.muellner@vrull.eu> Signed-off-by: Philipp Tomsich <philipp.tomsich@vrull.eu> Signed-off-by: Konstantinos Eleftheriou <konstantinos.eleftheriou@vrull.eu>
Richard Biener [Wed, 25 Sep 2024 10:46:28 +0000 (12:46 +0200)]
remove dominator recursion from reassoc
The reassoc pass currently walks dominators in a recursive way where
I ran into a stack overflow with. The following replaces it with
worklists following patterns used elsewhere.
* tree-ssa-reassoc.cc (break_up_subtract_bb): Remove recursion.
(reassociate_bb): Likewise.
(do_reassoc): Implement worklist based dominator walks for
both break_up_subtract_bb and reassociate_bb.
Aldy Hernandez [Mon, 26 Aug 2024 05:33:38 +0000 (07:33 +0200)]
Remove recursion in simplify_control_stmt_condition_1 [PR114855].
Remove some ad-hoc simplification code in the forward threader, as the
call into the ranger in m_simplifier->simplify() will handle anything we
can do manually in simplify_control_stmt_condition_1.
In PR114855, DOM time is reduced from 120s to 92s (-23%) and overall
compilation time from 235s to 205s (-12%). The total thread count at -O1 is
unchanged for the testcase.
In our bootstrap .ii benchmark suite, I see we thread 3 threads less
over all files at -O1. At -O2, the backward threader picks up one more,
for no difference over all.
Jason Merrill [Thu, 12 Sep 2024 16:15:51 +0000 (12:15 -0400)]
libstdc++: #ifdef out #pragma GCC system_header
In r15-3714-gd3a7302ec5985a I added -Wsystem-headers to the libstdc++ build
flags to help catch problems in the library. This patch takes a different
approach, of disabling the #pragma system_header unless _GLIBCXX_SYSHDR is
defined. As a result, the testsuites will treat them as non-system-headers
to get better warning coverage during regression testing of both gcc and
libstdc++, not just when building the library.
My rationale for the #ifdef instead of just removing the #pragma is the
three G++ tests that want to test libstdc++ system header behavior, so we
need a way to select it.
This doesn't affect installed libraries, as they get their
system-header status from the lookup path. But testsuite_flags
--build-includes gives -I directives rather than -isystem.
This patch doesn't change the headers in config/ because I'm not compiling
with most of them, so won't see any warnings that need fixing. Adjusting
them could happen later, or we can not bother.
Richard Biener [Wed, 25 Sep 2024 10:52:32 +0000 (12:52 +0200)]
Use tree view for find_always_executed_bbs result
The following makes us use bitmap tree view for the always-executed-BBs
bitmap as computed by IPA utils find_always_executed_bbs and used by
IPA modref (where it shows up in the profile for PR114855.
* ipa-utils.cc (find_always_executed_bbs): Switch result
bitmap to tree view.
OpenMP: Update OMP_REQUIRES_TARGET_USED for declare_target + interop
Older versions of the OpenMP specification were not clear about what counted
as device usage. Newer (like TR13) are rather clear. Hence, this commit adds
GCC's target-used flag also when a 'declare target' or an 'interop' are
encountered. (The latter only to Fortran as C/C++ parsing support is still
missing.) TR13 also lists 'dispatch' as target-used construct (as it has the
device clause) and 'device_safesync' as clause with global requirement
property, but both are not yet supported in GCC.
gcc/c/ChangeLog:
* c-parser.cc (c_parser_omp_declare_target): Set target-used bit
in omp_requires_mask.
gcc/cp/ChangeLog:
* parser.cc (cp_parser_omp_declare_target): Set target-used bit
in omp_requires_mask.
gcc/fortran/ChangeLog:
* parse.cc (decode_omp_directive): Set target-used bit of
omp_requires_mask when encountering the declare_target or interop
directive.
gcc/testsuite/ChangeLog:
* gfortran.dg/gomp/interop-1.f90: Add dg-error for missing
omp requires requirement and declare_variant usage.
* gfortran.dg/gomp/requires-8.f90: Likewise.
Richard Biener [Tue, 24 Sep 2024 12:10:13 +0000 (14:10 +0200)]
rtl-optimization/114855 - slow add_store_equivs in IRA
For the testcase in PR114855 at -O1 add_store_equivs shows up as the
main sink for bitmap_set_bit because it uses a bitmap to mark all
seen insns by UID to make sure the forward walk in memref_used_between_p
will find the insn in question. Given we do have a CFG here the
functions operation is questionable, given memref_used_between_p
together with the walk of all insns is obviously quadratic in the
worst case that whole thing should be re-done ... but, for the
testcase, using a sbitmap of size get_max_uid () + 1 gets
bitmap_set_bit off the profile and improves IRA time from 15.58s (8%)
to 3.46s (2%).
Now, given above quadraticness I wonder whether we should instead
gate add_store_equivs on optimize > 1 or flag_expensive_optimizations.
PR rtl-optimization/114855
* ira.cc (add_store_equivs): Use sbitmap for tracking
visited insns.
Richard Biener [Tue, 24 Sep 2024 09:47:26 +0000 (11:47 +0200)]
tree-optimization/114855 - slow VRP due to equiv oracle queries
For the testcase in PR114855 VRP takes 320.41s (23%) (after mitigating
backwards threader slowness). This is mostly due to the bitmap check
in equiv_oracle::find_equiv_dom. The following turns this bitmap
to tree view, trading the linear search for a O(log N) one which
improves VRP time to 54.54s (5%).
PR tree-optimization/114855
* value-relation.cc (equiv_oracle::equiv_oracle): Switch
m_equiv_set to tree view.
The testcase decare-variant-duplicates.c added in commit 96246bff0bcd9e5cdec9e6cf811ee3db4997f6d4 failed on 32-bit x86
because on that target "i386" is defined as a preprocessor macro
and cannot be used as an identifier. Fixed by rewriting that test
not to do that.
gcc/testsuite/ChangeLog
* c-c++-common/gomp/declare-variant-duplicates.c: Avoid using
"i386" as an identifier.
Thomas Koenig [Tue, 24 Sep 2024 20:57:42 +0000 (22:57 +0200)]
Add random numbers and fix some bugs.
This patch adds random number support for UNSIGNED, plus fixes
two bugs, with array I/O where the type used to be set to BT_INTEGER,
and for division with the divisor being a constant.
gcc/fortran/ChangeLog:
* check.cc (gfc_check_random_number): Adjust for unsigned.
* iresolve.cc (gfc_resolve_random_number): Handle unsigned.
* trans-expr.cc (gfc_conv_expr_op): Handle BT_UNSIGNED for divide.
* trans-types.cc (gfc_get_dtype_rank_type): Handle BT_UNSIGNED.
* gfortran.texi: Add RANDOM_NUMBER for UNSIGNED.
libgfortran/ChangeLog:
* gfortran.map: Add _gfortran_random_m1, _gfortran_random_m2,
_gfortran_random_m4, _gfortran_random_m8 and _gfortran_random_m16.
* intrinsics/random.c (random_m1): New function.
(random_m2): New function.
(random_m4): New function.
(random_m8): New function.
(random_m16): New function.
(arandom_m1): New function.
(arandom_m2): New function.
(arandom_m4): New function.
(arandom_m8): New funciton.
(arandom_m16): New function.
Thomas Koenig [Tue, 24 Sep 2024 20:53:59 +0000 (22:53 +0200)]
Implement IANY, IALL and IPARITY for unsigned.
gcc/fortran/ChangeLog:
* check.cc (gfc_check_transf_bit_intrins): Handle unsigned.
* gfortran.texi: Docment IANY, IALL and IPARITY for unsigned.
* iresolve.cc (gfc_resolve_iall): Set flag to use integer
if type is BT_UNSIGNED.
(gfc_resolve_iany): Likewise.
(gfc_resolve_iparity): Likewise.
* simplify.cc (do_bit_and): Adjust asserts for BT_UNSIGNED.
(do_bit_ior): Likewise.
(do_bit_xor): Likewise
Thomas Koenig [Tue, 24 Sep 2024 19:59:10 +0000 (21:59 +0200)]
Implement SUM and PRODUCT for unsigned.
gcc/fortran/ChangeLog:
* gfortran.texi: Document SUM and PRODUCT.
* iresolve.cc (resolve_transformational): New argument,
use_integer, to translate calls to unsigned to calls to
integer.
(gfc_resolve_product): Use it
(gfc_resolve_sum): Use it.
* simplify.cc (init_result_expr): Handle BT_UNSIGNED.
Thomas Koenig [Tue, 24 Sep 2024 19:51:42 +0000 (21:51 +0200)]
Implement MATMUL and DOT_PRODUCT for unsigned.
gcc/fortran/ChangeLog:
* arith.cc (gfc_arith_uminus): Fix warning.
(gfc_arith_minus): Correctly truncate unsigneds.
* check.cc (gfc_check_dot_product): Handle unsigned arguments.
(gfc_check_matmul): Likewise.
* expr.cc (gfc_get_unsigned_expr): New function.
* gfortran.h (gfc_get_unsigned_expr): Add prototype.
* iresolve.cc (gfc_resolve_matmul): If using UNSIGNED, use the
signed integer version.
* gfortran.texi: Document MATMUL and DOT_PRODUCT for unsigned.
* simplify.cc (compute_dot_product): Handle unsigneds.
libgfortran/ChangeLog:
* m4/iparm.m4: Add UNSIGED if type is m.
* m4/matmul.m4: If type is GFC_INTEGER, use GFC_UINTEGER instead.
Whitespace fixes.
* m4/matmul_internal.m4: Whitespace fixes.
Jakub Jelinek [Tue, 24 Sep 2024 18:19:50 +0000 (20:19 +0200)]
c++: Implement C++23 P2718R0 - Wording for P2644R1 Fix for Range-based for Loop [PR107637]
The following patch implements the C++23 P2718R0 paper
- Wording for P2644R1 Fix for Range-based for Loop.
The patch introduces a new option, -f{,no-}range-for-ext-temps so that
user can control the behavior even in older C++ versions.
The option is on by default in C++23 and later (-fno-range-for-ext-temps
is an error in that case) and in the -std=gnu++11 ... -std=gnu++20 modes
(one can use -fno-range-for-ext-temps to request previous behavior in that
case), and is not enabled by default in -std=c++11 ... -std=c++20 modes
but one can explicitly enable it with -frange-for-ext-temps.
As all the temporaries from __for_range initialization should have life
extended until the end of __for_range scope, this patch disables (for
-frange-for-ext-temps and if !processing_template_decl) CLEANUP_POINT_EXPR wrapping
of the __for_range declaration, also disables -Wdangling-reference warning
as well as the rest of extend_ref_init_temps (we know the __for_range temporary
is not TREE_STATIC and as all the temporaries from the initializer will be life
extended, we shouldn't try to handle temporaries referenced by references any
differently) and adds an extra push_stmt_list/pop_stmt_list before
cp_finish_decl of __for_range and after end of the for body and wraps all
that into CLEANUP_POINT_EXPR.
I had to repeat that also for OpenMP range loops because those are handled
differently.
2024-09-24 Jakub Jelinek <jakub@redhat.com>
PR c++/107637
gcc/
* omp-general.cc (find_combined_omp_for, find_nested_loop_xform):
Handle CLEANUP_POINT_EXPR like TRY_FINALLY_EXPR.
* doc/invoke.texi (frange-for-ext-temps): Document. Add
-fconcepts to the C++ option list.
gcc/c-family/
* c.opt (frange-for-ext-temps): New option.
* c-opts.cc (c_common_post_options): Set flag_range_for_ext_temps
for C++23 or later or for C++11 or later in !flag_iso mode if
the option wasn't set by user.
* c-cppbuiltin.cc (c_cpp_builtins): Change __cpp_range_based_for
value for flag_range_for_ext_temps from 201603L to 202212L in C++17
or later.
* c-omp.cc (c_find_nested_loop_xform_r): Handle CLEANUP_POINT_EXPR
like TRY_FINALLY_EXPR.
gcc/cp/
* cp-tree.h: Implement C++23 P2718R0 - Wording for P2644R1 Fix for
Range-based for Loop.
(cp_convert_omp_range_for): Add bool tmpl_p argument.
(find_range_for_decls): Declare.
* parser.cc (cp_convert_range_for): For flag_range_for_ext_temps call
push_stmt_list () before cp_finish_decl for range_temp and save it
temporarily to FOR_INIT_STMT.
(cp_convert_omp_range_for): Add tmpl_p argument. If set, remember
DECL_NAME of range_temp and for cp_finish_decl call restore it before
clearing it again, if unset, don't adjust DECL_NAME of range_temp at
all.
(cp_parser_omp_loop_nest): For flag_range_for_ext_temps range for add
CLEANUP_POINT_EXPR around sl. Call find_range_for_decls and adjust
DECL_NAMEs for range fors if not processing_template_decl. Adjust
cp_convert_omp_range_for caller. Remove superfluous backslash at the
end of line.
* decl.cc (initialize_local_var): For flag_range_for_ext_temps
temporarily clear stmts_are_full_exprs_p rather than set for
for_range__identifier decls.
* call.cc (extend_ref_init_temps): For flag_range_for_ext_temps return
init early for for_range__identifier decls.
* semantics.cc (find_range_for_decls): New function.
(finish_for_stmt): Use it. For flag_range_for_ext_temps if
cp_convert_range_for set FOR_INIT_STMT, pop_stmt_list it and wrap
into CLEANUP_POINT_EXPR.
* pt.cc (tsubst_omp_for_iterator): Adjust tsubst_omp_for_iterator
caller.
(tsubst_stmt) <case OMP_FOR>: For flag_range_for_ext_temps if there
are any range fors in the loop nest, add push_stmt_list starting
before the initializations, pop_stmt_list it after the body and wrap
into CLEANUP_POINT_EXPR. Change DECL_NAME of range for temps from
NULL to for_range_identifier.
gcc/testsuite/
* g++.dg/cpp23/range-for1.C: New test.
* g++.dg/cpp23/range-for2.C: New test.
* g++.dg/cpp23/range-for3.C: New test.
* g++.dg/cpp23/range-for4.C: New test.
* g++.dg/cpp23/range-for5.C: New test.
* g++.dg/cpp23/range-for6.C: New test.
* g++.dg/cpp23/range-for7.C: New test.
* g++.dg/cpp23/range-for8.C: New test.
* g++.dg/cpp23/feat-cxx2b.C (__cpp_range_based_for): Check for
202212L rather than 201603L.
* g++.dg/cpp26/feat-cxx26.C (__cpp_range_based_for): Likewise.
* g++.dg/warn/Wdangling-reference4.C: Don't expect warning for C++23
or newer. Use dg-additional-options rather than dg-options.
libgomp/
* testsuite/libgomp.c++/range-for-1.C: New test.
* testsuite/libgomp.c++/range-for-2.C: New test.
* testsuite/libgomp.c++/range-for-3.C: New test.
* testsuite/libgomp.c++/range-for-4.C: New test.
* testsuite/libgomp.c++/range-for-5.C: New test.
libgcc, Darwin: Drop the legacy library build for macOS >= 15 [PR116809].
We have been building a legacy libgcc_s.1 DSO to support code that
was built with older compilers.
From macOS 15, the unwinder no longer exports some of the symbols used
in that library which (a) cuases bootstrap fail and (b) means that the
legacy library is no longer useful.
No open branch of GCC emits references to this library - and any already
-built code that depends on the symbols would need rework anyway.
PR target/116809
libgcc/ChangeLog:
* config.host: Build legacy libgcc_s.1 on hosts before macOS 15.
* config/i386/t-darwin: Remove reference to legacy libgcc_s.1
* config/rs6000/t-darwin: Likewise.
* config/t-darwin-libgccs1: New file.
c++/contracts: ICE in build_contract_condition_function [PR116490]
We currently do not expect comdat group of the guarded function to
be set at the time of generating pre and post check function.
However, in the case of an explicit instantiation, the guarded
function has been added to a comdat group before generating contract
check functions, which causes the observed ICE. Current assert
removed and an additional check for comdat group of the guarded
function added. With this change, the pre and post check functions
get added to the same comdat group of the guarded function if the
guarded function is already placed in a comdat group.
PR c++/116490
gcc/cp/ChangeLog:
* contracts.cc (build_contract_condition_function): added
a check for comdat group of the guarded function. If set,
the condition check function is added to the same comdat
group.
libgomp: with USM, init 'link' variables with host address
If requires unified_shared_memory or self_maps is set, make
'declare target link' variables to point initially to the host pointer.
libgomp/ChangeLog:
* target.c (gomp_load_image_to_device): For requires
unified_shared_memory, update 'link' vars to point to the host var.
* testsuite/libgomp.c-c++-common/target-link-3.c: New test.
* testsuite/libgomp.c-c++-common/target-link-4.c: New test.
OpenMP: Check additional restrictions on context selector properties
TR13 (pre-6.0) of the OpenMP spec says:
"Each trait-property may only be specified once in a trait selector
other than those in the construct selector set."
and
"If trait-property any is specified in the kind trait-selector of the
device selector set or the target_device selector sets, no other
trait-property may be specified in the same selector set."
These restrictions (with slightly different wording) date back to
OpenMP 5.1, but were not in 5.0 which was the basis for GCC's
implementation.
This patch adds a diagnostic, adds new testcases, and fixes some older
testcases that include now-invalid selectors.
gcc/ChangeLog
* omp-general.cc (omp_check_context_selector): Reject other
properties in the same selector set with kind(any). Also reject
duplicate name-list properties.
We allowed the operand convert when matching SAT_SUB in match.pd, to support
the zip benchmark SAT_SUB pattern. Aka,
(convert? (minus (convert1? @0) (convert1? @1))) for below sample code.
void test (uint16_t *x, unsigned b, unsigned n)
{
unsigned a = 0;
register uint16_t *p = x;
do {
a = *--p;
*p = (uint16_t)(a >= b ? a - b : 0); // Truncate after .SAT_SUB
} while (--n);
}
The pattern match for SAT_SUB itself may also act on below scalar sample
code too.
unsigned long long GetTimeFromFrames(int);
unsigned long long GetMicroSeconds();
void DequeueEvent(unsigned frame) {
long long frame_time = GetTimeFromFrames(frame);
unsigned long long current_time = GetMicroSeconds();
DequeueEvent(frame_time < current_time ? 0 : frame_time - current_time);
}
Aka:
uint32_t a = (uint32_t)SAT_SUB(uint64_t, uint64_t);
Then there will be a problem when ia32 or -m32 is given when compiling.
Because we only check the lhs (aka uint32_t) type is supported by ifn
instead of the operand (aka uint64_t). Mostly DImode is disabled for
32 bits target like ia32 or rv32gcv, and then trigger ICE when expanding.
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/116814
gcc/ChangeLog:
* tree-ssa-math-opts.cc (build_saturation_binary_arith_call): Make
ifn is_supported type check based on operand instead of lhs.
Richard Biener [Tue, 24 Sep 2024 11:47:04 +0000 (13:47 +0200)]
tree-optimization/116819 - SLP with !STMT_VINFO_RELEVANT representative
Under some circumstances we can end up picking a not relevant stmt
as representative of a SLP node. Instead of skipping stmt analysis
and declaring success we have to either ignore relevancy throughout
the code base or fail SLP operation verification. The following
does the latter.
PR tree-optimization/116819
* tree-vect-stmts.cc (vect_analyze_stmt): When the SLP
representative isn't relevant signal failure instead of
success.
Robin Dapp [Tue, 3 Sep 2024 15:53:34 +0000 (17:53 +0200)]
RISC-V: Add more vector-vector extract cases.
This adds a V16SI -> V4SI and related i.e. "quartering" vector-vector
extract expander for VLS modes. It helps with spills in x264 that may
cause a load-hit-store.
Robin Dapp [Fri, 30 Aug 2024 12:35:08 +0000 (14:35 +0200)]
RISC-V: Fix effective target check.
The return value is inverted in check_effective_target_rvv_zvl256b_ok
and check_effective_target_rvv_zvl512b_ok. Fix this and also just use
the current march.
Jason Merrill [Thu, 19 Sep 2024 19:50:19 +0000 (15:50 -0400)]
build: enable C++11 narrowing warnings
We've been using -Wno-narrowing since gcc 4.7, but at this point narrowing
diagnostics seem like a stable part of C++ and we should adjust.
This patch changes -Wno-narrowing to -Wno-error=narrowing so that narrowing
issues will still not break bootstrap, but we can see them.
The rest of the patch fixes the narrowing warnings I see in an
x86_64-pc-linux-gnu bootstrap. In most of the cases, by adjusting the types
of various declarations so that we store the values in the same types we
compute them in, which seems worthwhile anyway. This also allowed us to
remove a few -Wsign-compare casts.
gcc/ChangeLog:
* configure.ac (CXX_WARNING_OPTS): Change -Wno-narrowing
to -Wno-error=narrowing.
* configure: Regenerate.
* config/i386/i386.h (debugger_register_map)
(debugger64_register_map)
(svr4_debugger_register_map): Make unsigned.
* config/i386/i386.cc: Likewise.
* diagnostic-event-id.h (diagnostic_thread_id_t): Make int.
* vec.h (vec::size): Make unsigned int.
* ipa-modref.cc (escape_point::arg): Make unsigned.
(modref_lattice::add_escape_point): Use eaf_flags_t.
(update_escape_summary_1): Use eaf_flags_t, && for bool.
* pair-fusion.cc (pair_fusion_bb_info::track_access):
Make mem_size unsigned int.
* pretty-print.cc (format_phase_2): Cast va_arg to char.
* tree-ssa-loop-ch.cc (ch_base::copy_headers): Make nheaders
unsigned, remove cast.
* tree-ssa-structalias.cc (bitpos_of_field): Return unsigned.
(push_fields_onto_fieldstack):Make offset unsigned, remove cast.
* tree-vect-slp.cc (vect_prologue_cost_for_slp): Use nelt_limit.
* tree-vect-stmts.cc (vect_truncate_gather_scatter_offset):
Make scale unsigned.
(vectorizable_operation): Make ncopies unsigned.
* rtl-ssa/member-fns.inl: Make num_accesses unsigned int.
Fortran: Assign allocated caf-memory to scalar members [PR84870]
Allocating a coarray required an array-descriptor. For scalars a
temporary descriptor was created. Assigning the allocated memory from
the temporary descriptor back to the scalar is now added.
gcc/fortran/ChangeLog:
PR fortran/84870
* trans-array.cc (duplicate_allocatable_coarray): For scalar
allocatable components the memory allocated is now assigned to
the component's pointer.
gcc/testsuite/ChangeLog:
* gfortran.dg/coarray/alloc_comp_10.f90: New test.
Richard Biener [Tue, 24 Sep 2024 10:53:11 +0000 (12:53 +0200)]
tree-optimization/114855 - more update_ssa speedup
The following tackles another source of slow bitmap operations,
namely populating blocks_to_update. We already have that in
tree view around PHI insertion but also the initial population is
slow. There's unfortunately a conditional inbetween list view
requirement and the bitmap API doesn't allow opportunistic
switching but rejects tree -> tree or list -> list transitions.
So the following patch wraps the early population in a tree view
section with possibly one redundant tree -> list -> tree view
transition.
This cuts tree SSA incremental from 228.25s (21%) to 65.05s (7%).
PR tree-optimization/114855
* tree-into-ssa.cc (update_ssa): Use tree view for the
initial population of blocks_to_update.
OpenMP: Add support for 'self_maps' to the 'require' directive
'self_maps' implies 'unified_shared_memory', except that the latter
also permits that explicit maps copy data to device memory while
self_maps does not. In GCC, currently, both are handled identical.
Richard Biener [Fri, 20 Sep 2024 13:07:24 +0000 (15:07 +0200)]
tree-optimization/115372 - failed store-lanes in some cases
The gcc.target/riscv/rvv/autovec/struct/struct_vect-4.c testcase shows
that we sometimes fail to use store-lanes even though it should be
profitable. We're currently relying on vect_slp_prefer_store_lanes_p
at the point we run into the first SLP discovery mismatch with obviously
limited information. For the case at hand we have 3, 5 or 7 lanes
of VnDImode [2, 2] vectors with the first mismatch at lane 2 so the
new group size is 1. The heuristic says that might be an OK split
given the rest is a multiple of the vector lanes. Now we continue
discovery but in the end mismatches result in uniformly single-lane
SLP instances which we can handle via interleaving but of course are
prime candidates for store-lanes. The following patch re-assesses
with the extra knowledge now just relying on the fact whether the
target supports store-lanes for the given group size.
PR tree-optimization/115372
* tree-vect-slp.cc (vect_build_slp_instance): Compute the
uniform, if, number of lanes of the RHS sub-graphs feeding
the store and if uniformly one, use store-lanes if the target
supports that.
libstdc++: Remove unnecessary 'static' from __is_specialization_of
This makes the declarations internal linkage, which is an ODR issue, and
causes a future modules patch to fail regtest as it now detects attempted
uses of TU-local entities in module CMIs.
Richard Biener [Mon, 23 Sep 2024 13:41:14 +0000 (15:41 +0200)]
tree-optimization/114855 - high update_ssa time
Part of the problem in PR114855 is high update_ssa time. When one fixes
the backward jump threading issue tree SSA incremental is at
439.91s ( 26%), mostly doing bitmap element searches for
blocks_with_phis_to_rewrite. The following turns that bitmap to tree
view noticing the two-dimensional vector of PHIs it guards is excessive
compared to what we actually save with it - walking all PHI nodes
in a block, something we already do once to initialize stmt flags.
So instead of optimizing that walk we use the stmt flag, saving
allocations and global state that lives throughout the whole
compilation.
This reduces the tree SSA incremental time to 203.13 ( 14%)
The array was added in r0-74758-g2ce798794df8e1 when we still possibly
had gazillion virtual operands for PR26830, I checked the testcase
still behaves OK.
PR tree-optimization/114855
* tree-into-ssa.cc (phis_to_rewrite): Remove global var.
(mark_phi_for_rewrite): Simplify.
(rewrite_update_phi_arguments): Walk all PHIs, process
those satisfying rewrite_uses_p.
(delete_update_ssa): Simplify.
(update_ssa): Likewise. Switch blocks_with_phis_to_rewrite
to tree view.
__attribute__((noipa)) int foo () { return 42; }
int bar () __attribute__((alias ("foo")));
int baz () __attribute__((alias ("bar")));
int main ()
{
int n;
#pragma omp target map(from:n)
n = baz ();
return n;
}
gcc emits following ptx for baz:
.visible .func (.param.u32 %value_out) bar;
.alias bar,foo;
.visible .func (.param.u32 %value_out) baz;
.alias baz,bar;
which is incorrect since PTX requires aliasee to be a defined function.
The patch instead uses cgraph_node::get(name)->ultimate_alias_target,
which generates the following PTX:
Gaius Mulley [Mon, 23 Sep 2024 23:28:19 +0000 (00:28 +0100)]
modula2: Add noreturn attribute to m2/gm2-libs/M2RTS.mod
This patch removes a build warning by adding a noreturn attribute
to the M2RTS.mod:HaltC procedure. Also add an infinite loop to
gm2-libs-min/M2RTS.mod.
Marek Polacek [Mon, 23 Sep 2024 16:19:40 +0000 (12:19 -0400)]
c++: diagnose this specifier in requires expr [PR116798]
We don't detect an explicit object parameter in a requires expression.
We can get there by way of requires-expression -> requirement-parameter-list
-> parameter-declaration-clause -> ... -> parameter-declaration with
this[opt]. But [dcl.fct]/5 doesn't allow an explicit object parameter
in this context. So let's fix it like r14-9033 and not like r14-8832.
PR c++/116798
gcc/cp/ChangeLog:
* parser.cc (cp_parser_parameter_declaration): Detect an explicit
object parameter in a requires expression.
gcc/testsuite/ChangeLog:
* g++.dg/cpp23/explicit-obj-diagnostics12.C: New test.
Saurabh Jha [Wed, 7 Aug 2024 11:34:20 +0000 (12:34 +0100)]
aarch64: Add codegen support for AdvSIMD faminmax
The AArch64 FEAT_FAMINMAX extension is optional from Armv9.2-a and
mandatory from Armv9.5-a. It introduces instructions for computing the
floating point absolute maximum and minimum of the two vectors
element-wise.
This patch adds code generation support for famax and famin in terms of
existing RTL operators.
famax/famin is equivalent to first taking abs of the operands and then
taking smax/smin on the results of abs.
famax/famin (a, b) = smax/smin (abs (a), abs (b))
This fusion of operators is only possible when -march=armv9-a+faminmax
flags are passed. We also need to pass -ffast-math flag; if we don't,
then a statement like
c[i] = __builtin_fmaxf16 (a[i], b[i]);
is RTL expanded to UNSPEC_FMAXNM instead of smax (likewise for smin).
This code generation is only available on -O2 or -O3 as that is when
auto-vectorization is enabled.
gcc/ChangeLog:
* config/aarch64/aarch64-simd.md
(*aarch64_faminmax_fused): Instruction pattern for faminmax
codegen.
* config/aarch64/iterators.md: Attribute for faminmax codegen.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/simd/faminmax-codegen-no-flag.c: New test.
* gcc.target/aarch64/simd/faminmax-codegen.c: New test.
* gcc.target/aarch64/simd/faminmax-no-codegen.c: New test.
Saurabh Jha [Tue, 6 Aug 2024 15:34:49 +0000 (16:34 +0100)]
aarch64: Add AdvSIMD faminmax intrinsics
The AArch64 FEAT_FAMINMAX extension is optional from Armv9.2-a and
mandatory from Armv9.5-a. It introduces instructions for computing the
floating point absolute maximum and minimum of the two vectors element-wise.
This patch introduces AdvSIMD faminmax intrinsics. The intrinsics of
this extension are implemented as the following builtin functions:
* vamax_f16
* vamaxq_f16
* vamax_f32
* vamaxq_f32
* vamaxq_f64
* vamin_f16
* vaminq_f16
* vamin_f32
* vaminq_f32
* vaminq_f64
We are defining a new way to add AArch64 AdvSIMD intrinsics by listing
all the intrinsics in a .def file and then using that .def file to
initialise various data structures. This would lead to more concise code
and easier addition of the new AdvSIMD intrinsics in future.
The faminmax intrinsics are defined using the new approach.
gcc/ChangeLog:
* config/aarch64/aarch64-builtins.cc
(ENTRY): Macro to parse the contents of
aarch64-simd-pragma-builtins.def.
(ENTRY_VHSDF): Macro to parse the contents of
aarch64-simd-pragma-builtins.def.
(enum aarch64_builtins): New enum values for faminmax builtins
via aarch64-simd-pragma-builtins.def.
(enum class aarch64_builtin_signatures): Enum class to specify
the number of operands a builtin will take.
(struct aarch64_pragma_builtins_data): Struct to hold data from
aarch64-simd-pragma-builtins.def.
(aarch64_fntype): New function to define function types of
intrinsics given an object of type aarch64_pragma_builtins_data.
(aarch64_init_pragma_builtins): New function to define pragma
builtins.
(aarch64_get_pragma_builtin): New function to get a row of
aarch64_pragma_builtins, given code.
(handle_arm_neon_h): Modify to call
aarch64_init_pragma_builtins.
(aarch64_general_check_builtin_call): Modify to check whether
required flag is being used for pragma builtins.
(aarch64_expand_pragma_builtin): New function to emit
instructions of pragma_builtin.
(aarch64_general_expand_builtin): Modify to call
aarch64_expand_pragma_builtin.
* config/aarch64/aarch64-option-extensions.def
(AARCH64_OPT_EXTENSION): Introduce new flag for this extension.
* config/aarch64/aarch64-simd.md
(@aarch64_<faminmax_uns_op><mode>): Instruction pattern for
faminmax intrinsics.
* config/aarch64/aarch64.h
(TARGET_FAMINMAX): Introduce new flag for this extension.
* config/aarch64/iterators.md: New iterators and unspecs.
* doc/invoke.texi: Document extension in AArch64 Options.
* config/aarch64/aarch64-simd-pragma-builtins.def: New file to
list pragma builtins.
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/simd/faminmax-builtins-no-flag.c: New test.
* gcc.target/aarch64/simd/faminmax-builtins.c: New test.
Jason Merrill [Mon, 9 Sep 2024 15:20:02 +0000 (11:20 -0400)]
libstdc++: operator new/delete are transaction_safe
With the changes to #pragma system_header, g++.dg/tm/pr46270.C was
failing because <new> didn't implement the N4514 change to [new.delete] that
says "The library versions of the global allocation and deallocation
functions are declared transaction_safe (8.3.5 dcl.fct)." We already have
the _GLIBCXX_TXN_SAFE macro, just need to add it.
Matthieu Longo [Mon, 23 Sep 2024 14:35:07 +0000 (15:35 +0100)]
dwarf2: store the RA state in CFI row
On AArch64, the RA state informs the unwinder whether the return address
is mangled and how, or not. This information is encoded in a boolean in
the CFI row. This binary approach prevents from expressing more complex
configuration, as it is the case with PAuth_LR introduced in Armv9.5-A.
This patch addresses this limitation by replacing the boolean by an enum.
gcc/ChangeLog:
* dwarf2cfi.cc
(struct dw_cfi_row): Declare a new enum type to replace ra_mangled.
(cfi_row_equal_p): Use ra_state instead of ra_mangled.
(dwarf2out_frame_debug_cfa_negate_ra_state): Same.
(change_cfi_row): Same.
Matthieu Longo [Mon, 23 Sep 2024 14:34:57 +0000 (15:34 +0100)]
dwarf2: add hooks for architecture-specific CFIs
Architecture-specific CFI directives are currently declared an processed
among others architecture-independent CFI directives in gcc/dwarf2* files.
This approach creates confusion, specifically in the case of DWARF
instructions in the vendor space and using the same instruction code.
Such a clash currently happen between DW_CFA_GNU_window_save (used on
SPARC) and DW_CFA_AARCH64_negate_ra_state (used on AArch64), and both
having the same instruction code 0x2d.
Then AArch64 compilers generates a SPARC CFI directive (.cfi_window_save)
instead of .cfi_negate_ra_state, contrarilly to what is expected in
[DWARF for the Arm 64-bit Architecture (AArch64)](https://github.com/
ARM-software/abi-aa/blob/main/aadwarf64/aadwarf64.rst).
This refactoring does not solve completely the problem, but improve the
situation by moving some of the processing of those directives (more
specifically their output in the assembly) to the backend via 2 target
hooks:
- DW_CFI_OPRND1_DESC: parse the first operand of the directive (if any).
- OUTPUT_CFI_DIRECTIVE: output the CFI directive as a string.
Additionally, this patch also contains a renaming of an enum used for
return address mangling on AArch64.
gcc/ChangeLog:
* config/aarch64/aarch64.cc
(aarch64_output_cfi_directive): New hook for CFI directives.
(aarch64_dw_cfi_oprnd1_desc): Same.
(TARGET_OUTPUT_CFI_DIRECTIVE): Hook for output_cfi_directive.
(TARGET_DW_CFI_OPRND1_DESC): Hook for dw_cfi_oprnd1_desc.
* config/sparc/sparc.cc
(sparc_output_cfi_directive): New hook for CFI directives.
(sparc_dw_cfi_oprnd1_desc): Same.
(TARGET_OUTPUT_CFI_DIRECTIVE): Hook for output_cfi_directive.
(TARGET_DW_CFI_OPRND1_DESC): Hook for dw_cfi_oprnd1_desc.
* coretypes.h
(struct dw_cfi_node): Forward declaration of CFI type from
gcc/dwarf2out.h.
(enum dw_cfi_oprnd_type): Same.
(enum dwarf_call_frame_info): Same.
* doc/tm.texi: Regenerated from doc/tm.texi.in.
* doc/tm.texi.in: Add doc for new target hooks.
type of enum to allow forward declaration.
* dwarf2cfi.cc
(struct dw_cfi_row): Update the description for window_save
and ra_mangled.
(dwarf2out_frame_debug_cfa_negate_ra_state): Use AArch64 CFI
directive instead of the SPARC one.
(change_cfi_row): Use the right CFI directive's name for RA
mangling.
(output_cfi): Remove explicit architecture-specific CFI
directive DW_CFA_GNU_window_save that falls into default case.
(output_cfi_directive): Use target hook as default.
* dwarf2out.cc (dw_cfi_oprnd1_desc): Use target hook as default.
* dwarf2out.h (enum dw_cfi_oprnd_type): specify underlying type
of enum to allow forward declaration.
(dw_cfi_oprnd1_desc): Call target hook.
(output_cfi_directive): Use dw_cfi_ref instead of struct
dw_cfi_node *.
* hooks.cc
(hook_bool_dwcfi_dwcfioprndtyperef_false): New.
(hook_bool_FILEptr_dwcfiptr_false): New.
* hooks.h
(hook_bool_dwcfi_dwcfioprndtyperef_false): New.
(hook_bool_FILEptr_dwcfiptr_false): New.
* target.def: Documentation for new hooks.
Matthieu Longo [Mon, 23 Sep 2024 14:31:18 +0000 (15:31 +0100)]
Rename REG_CFA_TOGGLE_RA_MANGLE to REG_CFA_NEGATE_RA_STATE
The current name REG_CFA_TOGGLE_RA_MANGLE is not representative of what
it really is, i.e. a register to represent several states, not only a
binary one. Same for dwarf2out_frame_debug_cfa_toggle_ra_mangle.
Matthieu Longo [Mon, 23 Sep 2024 14:03:37 +0000 (15:03 +0100)]
libgcc: hide CIE and FDE data for DWARF architecture extensions behind a handler.
This patch provides a new handler MD_ARCH_FRAME_STATE_T to hide an
architecture-specific structure containing CIE and FDE data related
to DWARF architecture extensions.
Hiding the architecture-specific attributes behind a handler has the
following benefits:
1. isolating those data from the generic ones in _Unwind_FrameState
2. avoiding casts to custom types.
3. preserving typing information when debugging with GDB, and so
facilitating their printing.
This approach required to add a new header md-unwind-def.h included at
the top of libgcc/unwind-dw2.h, and redirecting to the corresponding
architecture header via a symbolic link.
An obvious drawback is the increase in complexity with macros, and
headers. It also caused a split of architecture definitions between
md-unwind-def.h (types definitions used in unwind-dw2.h) and
md-unwind.h (local types definitions and handlers implementations).
The naming of md-unwind.h with .h extension is a bit misleading as
the file is only included in the middle of unwind-dw2.c. Changing
this naming would require modification of others backends, which I
prefered to abstain from. Overall the benefits are worth the added
complexity from my perspective.
libgcc/ChangeLog:
* Makefile.in: New target for symbolic link to md-unwind-def.h
* config.host: New parameter md_unwind_def_header. Set it to
aarch64/aarch64-unwind-def.h for AArch64 targets, or no-unwind.h
by default.
* config/aarch64/aarch64-unwind.h
(aarch64_pointer_auth_key): Move to aarch64-unwind-def.h
(aarch64_cie_aug_handler): Update.
(aarch64_arch_extension_frame_init): Update.
(aarch64_demangle_return_addr): Update.
* configure.ac: New substitute variable md_unwind_def_header.
* unwind-dw2.h (defined): MD_ARCH_FRAME_STATE_T.
* config/aarch64/aarch64-unwind-def.h: New file.
* configure: Regenerate.
* config/no-unwind.h: Updated comment
Matthieu Longo [Mon, 23 Sep 2024 14:03:35 +0000 (15:03 +0100)]
aarch64: skip copy of RA state register into target context
The RA state register is local to a frame, so it should not be copied to
the target frame during the context installation.
This patch adds a new backend handler that check whether a register
needs to be skipped or not before its installation.
libgcc/ChangeLog:
* config/aarch64/aarch64-unwind.h
(MD_FRAME_LOCAL_REGISTER_P): new handler checking whether a register
from the current context needs to be skipped before installation into
the target context.
(aarch64_frame_local_register): Likewise.
* unwind-dw2.c (uw_install_context_1): use MD_FRAME_LOCAL_REGISTER_P.
Matthieu Longo [Mon, 23 Sep 2024 14:03:30 +0000 (15:03 +0100)]
aarch64: store signing key and signing method in DWARF _Unwind_FrameState
This patch is only a refactoring of the existing implementation
of PAuth and returned-address signing. The existing behavior is
preserved.
_Unwind_FrameState already contains several CIE and FDE information
(see the attributes below the comment "The information we care
about from the CIE/FDE" in libgcc/unwind-dw2.h).
The patch aims at moving the information from DWARF CIE (signing
key stored in the augmentation string) and FDE (the used signing
method) into _Unwind_FrameState along the already-stored CIE and
FDE information.
Note: those information have to be saved in frame_state_reg_info
instead of _Unwind_FrameState as they need to be savable by
DW_CFA_remember_state and restorable by DW_CFA_restore_state, that
both rely on the attribute "prev".
Those new information in _Unwind_FrameState simplifies the look-up
of the signing key when the return address is demangled. It also
allows future signing methods to be easily added.
_Unwind_FrameState is not a part of the public API of libunwind,
so the change is backward compatible.
A new architecture-specific handler MD_ARCH_EXTENSION_FRAME_INIT
allows to reset values (if needed) in the frame state and unwind
context before changing the frame state to the caller context.
A new architecture-specific handler MD_ARCH_EXTENSION_CIE_AUG_HANDLER
isolates the architecture-specific augmentation strings in AArch64
backend, and allows others architectures to reuse augmentation
strings that would have clashed with AArch64 DWARF extensions.
aarch64_demangle_return_addr, DW_CFA_AARCH64_negate_ra_state and
DW_CFA_val_expression cases in libgcc/unwind-dw2-execute_cfa.h
were documented to clarify where the value of the RA state register
is stored (FS and CONTEXT respectively).
libgcc/ChangeLog:
* config/aarch64/aarch64-unwind.h
(AARCH64_DWARF_RA_STATE_MASK): The mask for RA state register.
(aarch64_ra_signing_method_t): The diversifiers used to sign a
function's return address.
(aarch64_pointer_auth_key): The key used to sign a function's
return address.
(aarch64_cie_signed_with_b_key): Deleted as the signing key is
available now in _Unwind_FrameState.
(MD_ARCH_EXTENSION_CIE_AUG_HANDLER): New CIE augmentation string
handler for architecture extensions.
(MD_ARCH_EXTENSION_FRAME_INIT): New architecture-extension
initialization routine for DWARF frame state and context before
execution of DWARF instructions.
(aarch64_context_ra_state_get): Read RA state register from CONTEXT.
(aarch64_ra_state_get): Read RA state register from FS.
(aarch64_ra_state_set): Write RA state register into FS.
(aarch64_ra_state_toggle): Toggle RA state register in FS.
(aarch64_cie_aug_handler): Handler AArch64 augmentation strings.
(aarch64_arch_extension_frame_init): Initialize defaults for the
signing key (PAUTH_KEY_A), and RA state register (RA_no_signing).
(aarch64_demangle_return_addr): Rely on the frame registers and
the signing_key attribute in _Unwind_FrameState.
* unwind-dw2-execute_cfa.h:
Use the right alias DW_CFA_AARCH64_negate_ra_state for __aarch64__
instead of DW_CFA_GNU_window_save.
(DW_CFA_AARCH64_negate_ra_state): Save the signing method in RA
state register. Toggle RA state register without resetting 'how'
to REG_UNSAVED.
* unwind-dw2.c:
(extract_cie_info): Save the signing key in the current
_Unwind_FrameState while parsing the augmentation data.
(uw_frame_state_for): Reset some attributes related to architecture
extensions in _Unwind_FrameState.
(uw_update_context): Move authentication code to AArch64 unwinding.
* unwind-dw2.h (enum register_rule): Give a name to the existing
enum for the register rules, and replace 'unsigned char' by 'enum
register_rule' to facilitate debugging in GDB.
(_Unwind_FrameState): Add a new architecture-extension attribute
to store the signing key.
OpenMP: Fix omp_get_device_from_uid, minor cleanup
In Fortran, omp_get_device_from_uid can also accept substrings, which are
then not NUL terminated. Fixed by introducing a fortran.c wrapper function.
Additionally, in case of a fail the plugin functions now return NULL instead
of failing fatally such that a fall-back UID is generated.
gcc/ChangeLog:
* omp-general.cc (omp_runtime_api_procname): Strip "omp_" from
string; move get_device_from_uid as now a '_' suffix exists.
libgomp/ChangeLog:
* fortran.c (omp_get_device_from_uid_): New function.
* libgomp.map (GOMP_6.0): Add it.
* oacc-host.c (host_dispatch): Init '.uid' and '.get_uid_func'.
* omp_lib.f90.in: Make it used by removing bind(C).
* omp_lib.h.in: Likewise.
* target.c (omp_get_device_from_uid): Ensure the device is initialized.
* plugin/plugin-gcn.c (GOMP_OFFLOAD_get_uid): Add function comment;
return NULL in case of an error.
* plugin/plugin-nvptx.c (GOMP_OFFLOAD_get_uid): Likewise.
* testsuite/libgomp.fortran/device_uid.f90: Update to test substrings.
The target dependent mlra option was designed to be able to quickly
switch between LRA and reload. The reload register allocator step is
scheduled for retirement, thus, remove the functionality of mlra,
keeping it for backward compatibility.
PR target/113954
gcc/ChangeLog:
* config/arc/arc.cc (TARGET_LRA_P): Always return true.
(arc_lra_p): Remove.
* config/arc/arc.h (TARGET_LRA): Remove.
* config/arc/arc.opt (mlra): Change it to do nothing.
* doc/invoke.texi (mlra): Update option description.
Simon Martin [Mon, 16 Sep 2024 11:45:32 +0000 (13:45 +0200)]
c++: Don't crash when mangling member with anonymous union or template type [PR100632, PR109790]
We currently crash upon mangling members that have an anonymous union or
a template operator type.
The problem is that before calling write_unqualified_name,
write_member_name asserts that it has a declaration whose DECL_NAME is
an identifier node that is not that of an operator. This is wrong:
- In PR100632, it's an anonymous union declaration, hence a 0 DECL_NAME
- In PR109790, it's a legitimate template declaration for an operator
(this was accepted up to GCC 10)
This assert was added via r11-6301, to be sure that we do write the "on"
marker for operator members.
This patch removes that assert and instead
- Lets members with an anonymous union type go through
- For operators, adds the missing "on" marker for ABI versions greater
than the highest usable with GCC 10
PR c++/109790
PR c++/100632
gcc/cp/ChangeLog:
* mangle.cc (write_member_name): Handle members whose type is an
anonymous union member. Write missing "on" marker for operators
when ABI version is at least 16.
gcc/testsuite/ChangeLog:
* g++.dg/cpp0x/decltype83.C: New test.
* g++.dg/cpp0x/decltype83a.C: New test.
* g++.dg/cpp1y/lambda-ice3.C: New test.
* g++.dg/cpp1y/lambda-ice3a.C: New test.
* g++.dg/cpp2a/nontype-class67.C: New test.
Simon Martin [Wed, 18 Sep 2024 10:35:27 +0000 (12:35 +0200)]
c++: Don't ICE due to artificial constructor parameters [PR116722]
The following code triggers an ICE
=== cut here ===
class base {};
class derived : virtual public base {
public:
template<typename Arg> constexpr derived(Arg) {}
};
int main() {
derived obj(1.);
}
=== cut here ===
The problem is that cxx_bind_parameters_in_call ends up attempting to
convert a REAL_CST (the first non artificial parameter) to INTEGER_TYPE
(the type of the __in_chrg parameter), which ICEs.
This patch changes cxx_bind_parameters_in_call to return early if it's
called with a *structor that has an __in_chrg or __vtt_parm parameter
since the expression won't be a constant expression.
Note that in the test case, the constructor is not constexpr-suitable,
however it's OK since it's a template according to my read of paragraph
(3) of [dcl.constexpr].
PR c++/116722
gcc/cp/ChangeLog:
* constexpr.cc (cxx_bind_parameters_in_call): Leave early for
{con,de}structors of classes with virtual bases.
Richard Biener [Mon, 23 Sep 2024 08:30:32 +0000 (10:30 +0200)]
tree-optimization/116810 - out-of-bound access to matches[]
The following makes sure to apply forced splitting of groups for
firced single-lane SLP only when the group being analyzed has more
than one lane. This avoids an out-of-bound access to matches[].
PR tree-optimization/116810
* tree-vect-slp.cc (vect_build_slp_instance): Onlu force
splitting for group_size > 1.
Richard Biener [Mon, 23 Sep 2024 09:05:37 +0000 (11:05 +0200)]
tree-optimization/116796 - virtual LC SSA broken after unrolling
When the unroller unloops loops it tracks whether it changes any
nesting relationship of remaining loops but when scanning a loops
preheader it fails to pass down the LC-SSA-invalidated bitmap, losing
the fact that an unrolled formerly inner loop can now be placed on
an exit of its outer loop. The following fixes that.
PR tree-optimization/116796
* cfgloopmanip.cc (fix_loop_placements): Get LC-SSA-invalidated
bitmap and pass it on.
(remove_path): Pass LC-SSA-invalidated to fix_loop_placements.
projects / thirdparty / gcc.git / log
