bitmap_set_bit checks the original value of the bit to return it to the
caller and then only writes the new value back if it changes.
Most callers of bitmap_set_bit don't need the return value, but with the conditional store
the CPU still has to predict it correctly since gcc doesn't know how to do
that without APX on x86 (even though CMOV could do it with a dummy target).
Really if-conversion should handle this case, but for now we can fix
it.
This simple patch improves runtime by 15% for the test case in the PR.
Which is more than I expected given it only has ~1.44% of the cycles, but I guess
the mispredicts caused some down stream effects.
cc1plus-bitmap -std=gnu++20 -O2 pr119482.cc -quiet
ran 1.15 ± 0.01 times faster than cc1plus -std=gnu++20 -O2 pr119482.cc -quiet
At least with this test case the total number of branches decreases
drastically. Even though the mispredict rate goes up slightly it is
still a big win.
$ perf stat -e branches,branch-misses,uncore_imc/cas_count_read/,uncore_imc/cas_count_write/ \
-a ../obj-fast/gcc/cc1plus -std=gnu++20 -O2 pr119482.cc -quiet -w
Performance counter stats for 'system wide':
41,932,957,091 branches
686,117,623 branch-misses # 1.64% of all branches
43,690.47 MiB uncore_imc/cas_count_read/
12,362.56 MiB uncore_imc/cas_count_write/
Doc: Cross-reference constructor and init_priority attributes [PR118982]
Per the issue, the discussion of these two attributes needed to be
better integrated. I also did some editing for style and readability,
and clarified that almost all targets support this feature (it is
enabled by default unless the back end disables it), not just "some".
Co-Authored_by: Jonathan Wakely <jwakely@redhat.com>
gcc/ChangeLog
PR c++/118982
* doc/extend.texi (Common Function Attributes): For the
constructor/destructory attribute, be more explicit about the
relationship between the constructor attribute and
the C++ init_priority attribute, and add a cross-reference.
Also document that most targets support this.
(C++ Attributes): Similarly for the init_priority attribute.
Doc: Document enum with underlying type extension [PR117689]
This is a C23/C++11 feature that is supported as an extension with
earlier -std= options too, but was never previously documented. It
interacts with the already-documented forward enum definition extension,
so I have merged discussion of the two extensions into the same section.
gcc/ChangeLog
PR c/117689
* doc/extend.texi (Incomplete Enums): Rename to....
(Enum Extensions): This. Document support for specifying the
underlying type of an enum as an extension in all earlier C
and C++ standards. Document that a forward declaration with
underlying type is not an incomplete type, and which dialects
GCC supports that in.
c++/modules: Forbid exposures of TU-local entities in inline variables [PR119551]
An inline variable has vague linkage, and needs to be conditionally
emitted in TUs that reference it. Unfortunately this clashes with
[basic.link] p14.2, which says that we ignore the initialisers of all
variables (including inline ones), since importers will not have access
to the referenced TU-local entities to write the definition.
This patch makes such exposures be ill-formed. One case that continues
to work is if the exposure is part of the dynamic initialiser of an
inline variable; in such cases, the definition has been built as part of
the module interface unit anyway, and importers don't need to write it
out again, so such exposures are "harmless".
PR c++/119551
gcc/cp/ChangeLog:
* module.cc (trees_out::write_var_def): Only ignore non-inline
variable initializers.
gcc/testsuite/ChangeLog:
* g++.dg/modules/internal-5_a.C: Add cases that should be
ignored.
* g++.dg/modules/internal-5_b.C: Test these new cases, and make
the testcase more robust.
* g++.dg/modules/internal-11.C: New test.
* g++.dg/modules/internal-12_a.C: New test.
* g++.dg/modules/internal-12_b.C: New test.
Jonathan Wakely [Mon, 31 Mar 2025 11:30:44 +0000 (12:30 +0100)]
libstdc++: Fix -Warray-bounds warning in std::vector::resize [PR114945]
This is yet another false positive warning fix. This time the compiler
can't prove that when the vector has sufficient excess capacity to
append new elements, the pointer to the existing storage is not null.
libstdc++-v3/ChangeLog:
PR libstdc++/114945
* include/bits/vector.tcc (vector::_M_default_append): Add
unreachable condition so the compiler knows that _M_finish is
not null.
* testsuite/23_containers/vector/capacity/114945.cc: New test.
Reviewed-by: Tomasz Kamiński <tkaminsk@redhat.com>
Tom Tromey [Wed, 21 Aug 2024 17:46:52 +0000 (11:46 -0600)]
Further use of mod_scope in modified_type_die
I am working on some changes to GNAT to emit hierarchical DWARF --
i.e., where entities will have simple names nested in a DW_TAG_module.
While working on this I found a couple of paths in modified_type_die
where "mod_scope" should be used, but is not. I suspect these cases
are only reachable by Ada code, as in both spots (subrange types and
base types), I believe that other languages don't generally have named
types in a non-top-level scope, and in these other situations,
mod_scope will still be correct.
gcc
* dwarf2out.cc (modified_type_die): Use mod_scope for
ranged types, base types, and array types.
Jakub Jelinek [Tue, 1 Apr 2025 14:47:37 +0000 (16:47 +0200)]
tailc: Improve tail recursion handling [PR119493]
This is a partial step towards fixing that PR.
For musttail recursive calls which have non-is_gimple_reg_type typed
parameters, the only case we've handled was if the exact parameter
was passed through (perhaps modified, but still the same PARM_DECL).
That isn't necessary, we can copy the argument to the parameter as well
(just need to watch for the use of the parameter in later arguments,
say musttail recursive call which swaps 2 structure arguments).
The patch attempts to play safe and punts if any of the parameters are
addressable (like we do for all normal tail calls and tail recursions,
except for musttail in the posted unreviewed patch).
With this patch (at least when early inlining isn't done on not yet
optimized body) inlining should see already tail recursion optimized
body and will not have problems with SRA breaking musttail.
This version of the patch limits this for musttail tail recursions,
with intent to enable for all tail recursions in GCC 16.
2025-04-01 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/119493
* tree-tailcall.cc (find_tail_calls): Don't punt on tail recusion
if some arguments don't have is_gimple_reg_type, only punt if they
have non-POD types, or volatile, or addressable or (for now) it is
not a musttail call. Set tailr_arg_needs_copy in those cases too.
(eliminate_tail_call): Copy call arguments to params if they don't
have is_gimple_reg_type, use temporaries if the argument is used
later.
(tree_optimize_tail_calls_1): Skip !is_gimple_reg_type
tailr_arg_needs_copy parameters. Formatting fix.
Jakub Jelinek [Tue, 1 Apr 2025 14:40:55 +0000 (16:40 +0200)]
combine: Use reg_used_between_p rather than modified_between_p in two spots [PR119291]
The following testcase is miscompiled on x86_64-linux at -O2 by the combiner.
We have from earlier combinations
(insn 22 21 23 4 (set (reg:SI 104 [ _7 ])
(const_int 0 [0])) "pr119291.c":25:15 96 {*movsi_internal}
(nil))
(insn 23 22 24 4 (set (reg/v:SI 117 [ e ])
(reg/v:SI 116 [ e ])) 96 {*movsi_internal}
(expr_list:REG_DEAD (reg/v:SI 116 [ e ])
(nil)))
(note 24 23 25 4 NOTE_INSN_DELETED)
(insn 25 24 26 4 (parallel [
(set (reg:CCZ 17 flags)
(compare:CCZ (neg:SI (reg:SI 104 [ _7 ]))
(const_int 0 [0])))
(set (reg/v:SI 116 [ e ])
(neg:SI (reg:SI 104 [ _7 ])))
]) "pr119291.c":26:13 977 {*negsi_2}
(expr_list:REG_DEAD (reg:SI 104 [ _7 ])
(nil)))
(note 26 25 27 4 NOTE_INSN_DELETED)
(insn 27 26 28 4 (set (reg:DI 128 [ _9 ])
(ne:DI (reg:CCZ 17 flags)
(const_int 0 [0]))) "pr119291.c":26:13 1447 {*setcc_di_1}
(expr_list:REG_DEAD (reg:CCZ 17 flags)
(nil)))
and try_combine is called on i3 25 and i2 22 (second time)
and reach the hunk being patched with simplified i3
(insn 25 24 26 4 (parallel [
(set (pc)
(pc))
(set (reg/v:SI 116 [ e ])
(const_int 0 [0]))
]) "pr119291.c":28:13 977 {*negsi_2}
(expr_list:REG_DEAD (reg:SI 104 [ _7 ])
(nil)))
and
(insn 22 21 23 4 (set (reg:SI 104 [ _7 ])
(const_int 0 [0])) "pr119291.c":27:15 96 {*movsi_internal}
(nil))
Now, the try_combine code there attempts to split two independent
sets in newpat by moving one of them to i2.
And among other tests it checks
!modified_between_p (SET_DEST (set1), i2, i3)
which is certainly needed, if there would be say
(set (reg/v:SI 116 [ e ]) (const_int 42 [0x2a]))
in between i2 and i3, we couldn't do that, as that set would overwrite
the value set by set1 we want to move to the i2 position.
But in this case pseudo 116 isn't set in between i2 and i3, but used
(and additionally there is a REG_DEAD note for it).
This is equally bad for the move, because while the i3 insn
and later will see the pseudo value that we set, the insn in between
which uses the value will see a different value from the one that
it should see.
As we don't check for that, in the end try_combine succeeds and
changes the IL to:
(insn 22 21 23 4 (set (reg/v:SI 116 [ e ])
(const_int 0 [0])) "pr119291.c":27:15 96 {*movsi_internal}
(nil))
(insn 23 22 24 4 (set (reg/v:SI 117 [ e ])
(reg/v:SI 116 [ e ])) 96 {*movsi_internal}
(expr_list:REG_DEAD (reg/v:SI 116 [ e ])
(nil)))
(note 24 23 25 4 NOTE_INSN_DELETED)
(insn 25 24 26 4 (set (pc)
(pc)) "pr119291.c":28:13 2147483647 {NOOP_MOVE}
(nil))
(note 26 25 27 4 NOTE_INSN_DELETED)
(insn 27 26 28 4 (set (reg:DI 128 [ _9 ])
(const_int 0 [0])) "pr119291.c":28:13 95 {*movdi_internal}
(nil))
(note, the i3 got turned into a nop and try_combine also modified insn 27).
The following patch replaces the modified_between_p
tests with reg_used_between_p, my understanding is that
modified_between_p is a subset of reg_used_between_p, so one
doesn't need both.
Looking at this some more today, I think we should special case
set_noop_p because that can be put into i2 (except for the JUMP_P
violations), currently both modified_between_p (pc_rtx, i2, i3)
and reg_used_between_p (pc_rtx, i2, i3) returns false.
I'll post a patch incrementally for that (but that feels like
new optimization, so probably not something that should be backported).
On Tue, Apr 01, 2025 at 11:27:25AM +0200, Richard Biener wrote:
> Can we constrain SET_DEST (set1/set0) to a REG_P in combine? Why
> does the comment talk about memory?
I was worried about making too risky changes this late in stage4
(and especially also for backports). Most of this code is 1992-ish.
I think many of the functions are just misnamed, the reg_ in there doesn't
match what those functions do (bet they initially supported just REGs
and later on support for other kinds of expressions was added, but haven't
done git archeology to prove that).
What we know for sure is:
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != ZERO_EXTRACT
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 0))) != STRICT_LOW_PART
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != ZERO_EXTRACT
&& GET_CODE (SET_DEST (XVECEXP (newpat, 0, 1))) != STRICT_LOW_PART
that is checked earlier in the condition.
Then it calls
&& ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 1)),
XVECEXP (newpat, 0, 0))
&& ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 0)),
XVECEXP (newpat, 0, 1))
While it has reg_* in it, that function mostly calls reg_overlap_mentioned_p
which is also misnamed, that function handles just fine all of
REG, MEM, SUBREG of REG, (SUBREG of MEM not, see below), ZERO_EXTRACT,
STRICT_LOW_PART, PC and even some further cases.
So, IMHO SET_DEST (set0) or SET_DEST (set0) can be certainly a REG, SUBREG
of REG, PC (at least the REG and PC cases are triggered on the testcase)
and quite possibly also MEM (SUBREG of MEM not, see below).
Now, the code uses !modified_between_p (SET_SRC (set{1,0}), i2, i3) where that
function for constants just returns false, for PC returns true, for REG
returns reg_set_between_p, for MEM recurses on the address, for
MEM_READONLY_P otherwise returns false, otherwise checks using alias.cc code
whether the memory could have been modified in between, for all other
rtxes recurses on the subrtxes. This part didn't change in my patch.
I've only changed those
- && !modified_between_p (SET_DEST (set{1,0}), i2, i3)
+ && !reg_used_between_p (SET_DEST (set{1,0}), i2, i3)
where the former has been described above and clearly handles all of
REG, SUBREG of REG, PC, MEM and SUBREG of MEM among other things.
The replacement reg_used_between_p calls reg_overlap_mentioned_p on each
instruction in between i2 and i3. So, there is clearly a difference
in behavior if SET_DEST (set{1,0}) is pc_rtx, in that case modified_between_p
returns unconditionally true even if there are no instructions in between,
but reg_used_between_p if there are no non-debug insns in between returns
false. Sorry for missing that, guess I should check for that (with the
exception of the noop moves which are often (set (pc) (pc)) and handled
by the incremental patch). In fact not just that, reg_used_between_p
will only return true for PC if it is mentioned anywhere in the insns
in between.
Anyway, except for that, for REG it calls refers_to_regno_p
and so should find any occurrences of any of the REG or parts of it for hard
registers, for MEM returns true if it sees any MEMs in insns in between
(conservatively), for SUBREGs apparently it relies on it being SUBREG of REG
(so doesn't handle SUBREG of MEM) and handles SUBREG of REG like the
SUBREG_REG, PC I've already described.
Now, because reg_overlap_mentioned_p doesn't handle SUBREG of MEM, I think
already the initial
&& ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 1)),
XVECEXP (newpat, 0, 0))
&& ! reg_referenced_p (SET_DEST (XVECEXP (newpat, 0, 0)),
XVECEXP (newpat, 0, 1))
calls would have failed --enable-checking=rtl or would have misbehaved, so
I think there is no need to check for it further.
To your question why I don't use reg_referenced_p, that is because
reg_referenced_p is something to call on one insn pattern, while
reg_used_between_p is pretty much that on all insns in between two
instructions (excluding the boundaries).
So, I think it would be safer to add && SET_DEST (set{1,0} != pc_rtx
checks to preserve former behavior, like in the following version.
2025-04-01 Jakub Jelinek <jakub@redhat.com>
PR rtl-optimization/119291
* combine.cc (try_combine): For splitting of PARALLEL with
2 independent SETs into i2 and i3 sets check reg_used_between_p
of the SET_DESTs rather than just modified_between_p.
Kito Cheng [Tue, 1 Apr 2025 01:14:51 +0000 (09:14 +0800)]
RISC-V: Tweak testcase for PIE
Linux toolchain may configured with --enable-default-pie, and that will
cause lots of regression test failures because the function name will
append with @plt suffix (e.g. `call foo` become `call foo@plt`), also
some code generation will different due to the code model like the address
generation for global variable, so we may add -fno-pie to those
testcases to prevent that.
We may consider just drop @plt suffix to prevent that at all, because
it's not difference between w/ and w/o @plt suffix, the linker will pick
the right one to do, however it's late stage of GCC development, so just
tweak the testcase should be the best way to do now.
Changes from v1:
- Add more testcase for PIE (from rvv.exp).
- Tweak the rule for match @plt.
Marek Polacek [Tue, 25 Mar 2025 17:36:24 +0000 (13:36 -0400)]
c++: fix missing lifetime extension [PR119383]
Since r15-8011 cp_build_indirect_ref_1 won't do the *&TARGET_EXPR ->
TARGET_EXPR folding not to change its value category. That fix seems
correct but it made us stop extending the lifetime in this testcase,
causing a wrong-code issue -- extend_ref_init_temps_1 did not see
through the extra *& because it doesn't use a tree walk.
This patch reverts r15-8011 and instead handles the problem in
build_over_call by calling force_lvalue in the is_really_empty_class
case as well as in the general case.
PR c++/119383
gcc/cp/ChangeLog:
* call.cc (build_over_call): Use force_lvalue to ensure op= returns
an lvalue.
* cp-tree.h (force_lvalue): Declare.
* cvt.cc (force_lvalue): New.
* typeck.cc (cp_build_indirect_ref_1): Revert r15-8011.
Jakub Jelinek [Tue, 1 Apr 2025 09:45:16 +0000 (11:45 +0200)]
profile: Another profiling musttail call fix [PR119535]
As the following testcase shows, EDGE_FAKE edges from musttail calls to
EXIT aren't the only edges we should ignore, we need to ignore also
edges created by the splitting of blocks for the EDGE_FAKE creation that
point from the musttail calls to the fallthrough block, which typically does
the return or with PHIs for the return value.
2025-04-01 Jakub Jelinek <jakub@redhat.com>
PR gcov-profile/119535
* profile.cc (branch_prob): Ignore any edges from bbs ending with
musttail call, rather than only EDGE_FAKE edges from those to EXIT.
Jakub Jelinek [Tue, 1 Apr 2025 09:43:16 +0000 (11:43 +0200)]
tailr: Punt on tail recursions that would break musttail [PR119493]
While working on the previous tailc patch, I've noticed the following
problem.
The testcase below fails, because we decide to tail recursion optimize
the call, but tail recursion (as documented in tree-tailcall.cc) needs to
add some result multiplication and/or addition if any tail recursion uses
accumulator, which is added right before the return.
So, if there are musttail non-recurive calls in the function, successful
tail recursion optimization will mean we'll later error on the musttail
calls. musttail recursive calls are ok, those would be tail recursion
optimized.
So, the following patch punts on all tail recursion optimizations if it
needs accumulators (add and/or mult) if there is at least one non-recursive
musttail call.
2025-04-01 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/119493
* tree-tailcall.cc (tree_optimize_tail_calls_1): Ignore tail recursion
candidates which need accumulators if there is at least one musttail
non-recursive call.
Jakub Jelinek [Tue, 1 Apr 2025 09:40:58 +0000 (11:40 +0200)]
gimple-low: Diagnose assume attr expressions defining labels which are used as unary && operands outside of those [PR119537]
The following testcases ICE on invalid code which defines
labels inside of statement expressions and then uses &&label
from code outside of the statement expressions.
The C++ FE diagnoses that with a warning (not specifically for
assume attribute, genericallly about taking address of a label
outside of a statement expression so computed goto could violate
the requirement that statement expression is not entered from
outside of it through a jump into it), the C FE doesn't diagnose
anything.
Normal direct gotos to such labels are diagnosed by both C and C++.
In the assume attribute case it is actually worse than for
addresses of labels in normal statement expressions, in that case
the labels are still in the current function, so invalid program
can still jump to those (and in case of OpenMP/OpenACC where it
is also invalid and stuff is moved to a separate function, such
movement is done post cfg discovery of FORCED_LABELs and worst
case one can run into cases which fail to assemble, but I haven't
succeeded in creating ICE for that).
For assume at -O0 we'd just throw away the assume expression if
it is not a simple condition and so the label is then not defined
anywhere and we ICE during cfg pass.
The gimplify.cc hunks fix that, as we don't have FORCED_LABELs
discovery done yet, it preserves all assume expressions which contain
used user labels.
With that we ICE during IRA, which is upset about an indirect jump
to a label which doesn't exist.
So, the gimple-low.cc hunks add diagnostics of the problem, it gathers
uids of all the user used labels inside of the assume expressions (usually
none) and if it finds any, walks the IL to find uses of those from outside
of those expressions now outlined into separate magic functions.
2025-04-01 Jakub Jelinek <jakub@redhat.com>
PR middle-end/119537
* gimplify.cc (find_used_user_labels): New function.
(gimplify_call_expr): Don't remove complex assume expression at -O0
if it defines any user labels.
* gimple-low.cc: Include diagnostic-core.h.
(assume_labels): New variable.
(diagnose_assume_labels): New function.
(lower_function_body): Call it via walk_gimple_seq if assume_labels
is non-NULL, then BITMAP_FREE assume_labels.
(find_assumption_locals_r): Record in assume_labels uids of user
labels defined in assume attribute expressions.
* c-c++-common/pr119537-1.c: New test.
* c-c++-common/pr119537-2.c: New test.
Thomas Schwinge [Mon, 31 Mar 2025 07:55:14 +0000 (09:55 +0200)]
GCN: Don't emit weak undefined symbols [PR119369]
This resolves all instances of PR119369
"GCN: weak undefined symbols -> execution test FAIL, 'HSA_STATUS_ERROR_VARIABLE_UNDEFINED'";
for all affected test cases, the execution test status progresses FAIL -> PASS.
This however also causes a small number of (expected) regressions, very similar
to GCC/nvptx:
[-PASS:-]{+FAIL:+} g++.dg/abi/pure-virtual1.C -std=c++17 (test for excess errors)
[-PASS:-]{+FAIL:+} g++.dg/abi/pure-virtual1.C -std=c++26 (test for excess errors)
[-PASS:-]{+FAIL:+} g++.dg/abi/pure-virtual1.C -std=c++98 (test for excess errors)
[-PASS:-]{+FAIL:+} gcc.dg/attr-weakref-1.c (test for excess errors)
[-FAIL:-]{+UNRESOLVED:+} gcc.dg/attr-weakref-1.c [-execution test-]{+compilation failed to produce executable+}
This fixes a few hundreds of compilation/linking FAILs (similar to PR69506),
where the GCN/LLVM 'ld' reported:
ld: error: relocation R_AMDGPU_REL32_LO cannot be used against symbol '_ZGTtnam'; recompile with -fPIC
>>> defined in [...]/amdgcn-amdhsa/./libstdc++-v3/src/.libs/libstdc++.a(cow-stdexcept.o)
>>> referenced by cow-stdexcept.cc:259 ([...]/libstdc++-v3/src/c++11/cow-stdexcept.cc:259)
>>> cow-stdexcept.o:(_txnal_cow_string_C1_for_exceptions(void*, char const*, void*)) in archive [...]/amdgcn-amdhsa/./libstdc++-v3/src/.libs/libstdc++.a
ld: error: relocation R_AMDGPU_REL32_HI cannot be used against symbol '_ZGTtnam'; recompile with -fPIC
>>> defined in [...]/amdgcn-amdhsa/./libstdc++-v3/src/.libs/libstdc++.a(cow-stdexcept.o)
>>> referenced by cow-stdexcept.cc:259 ([...]/source-gcc/libstdc++-v3/src/c++11/cow-stdexcept.cc:259)
>>> cow-stdexcept.o:(_txnal_cow_string_C1_for_exceptions(void*, char const*, void*)) in archive [...]/amdgcn-amdhsa/./libstdc++-v3/src/.libs/libstdc++.a
[...]
..., which is:
$ c++filt _ZGTtnam
transaction clone for operator new[](unsigned long)
..., and similarly for other libitm symbols.
However, the affected test cases, if applicable, then run into execution test
FAILs, due to PR119369
"GCN: weak undefined symbols -> execution test FAIL, 'HSA_STATUS_ERROR_VARIABLE_UNDEFINED'".
PR target/119369
libstdc++-v3/
* config/cpu/gcn/cpu_defines.h: New.
* configure.host [GCN] (cpu_defines_dir): Point to it.
Richard Biener [Mon, 31 Mar 2025 12:56:25 +0000 (14:56 +0200)]
target/119549 - fixup handling of -mno-sse4 in target attribute
The following fixes ix86_valid_target_attribute_inner_p to properly
handle target("no-sse4") via OPT_mno_sse4 rather than as unset OPT_msse4.
I've added asserts to ix86_handle_option that RejectNegative is honored
for both.
PR target/119549
* common/config/i386/i386-common.cc (ix86_handle_option):
Assert that both OPT_msse4 and OPT_mno_sse4 are never unset.
* config/i386/i386-options.cc (ix86_valid_target_attribute_inner_p):
Process negated OPT_msse4 as OPT_mno_sse4.
Jakub Jelinek [Tue, 1 Apr 2025 08:05:18 +0000 (10:05 +0200)]
libquadmath: Avoid old-style function definition warnings
I've noticed
../../../libquadmath/printf/gmp-impl.h:104:18: warning: old-style function definition [-Wold-style-definition]
../../../libquadmath/printf/gmp-impl.h:104:18: warning: old-style function definition [-Wold-style-definition]
../../../libquadmath/printf/gmp-impl.h:104:18: warning: old-style function definition [-Wold-style-definition]
../../../libquadmath/strtod/strtod_l.c:456:22: warning: old-style function definition [-Wold-style-definition]
warnings during bootstrap (clearly since the switch to -std=gnu23 by default).
The following patch fixes those in libquadmath, the only other warnings are
in zlib.
Hu, Lin1 [Wed, 26 Mar 2025 08:15:52 +0000 (16:15 +0800)]
i386: Add attr_isa for vaes patterns to sync with attr gpr16. [pr119473]
For vaes patterns with jm constraint and gpr16 attr, it requires "isa"
attr to distinct avx/avx512 alternatives in ix86_memory_address_reg_class.
Also adds missing type and mode attributes for those vaes patterns.
gcc/ChangeLog:
PR target/119473
* config/i386/sse.md
(vaesdec_<mode>): Set attr "isa" as "avx,vaes_avx512vl", "type" as
"sselog1", "mode" as "TI".
(vaesdeclast_<mode>): Ditto.
(vaesenc_<mode>): Ditto.
(vaesenclast_<mode>): Ditto.
gcc/testsuite/ChangeLog:
PR target/119473
* gcc.target/i386/pr119473.c: New test.
Co-authored-by: Hongyu Wang <hongyu.wang@intel.com>
Monk Chiang [Tue, 4 Feb 2025 07:29:17 +0000 (15:29 +0800)]
RISC-V: Fix wrong LMUL when only implict zve32f.
According to Section 3.4.2, Vector Register Grouping, in the RISC-V
Vector Specification, the rule for LMUL is LMUL >= SEW/ELEN
Changes since V2:
- Add check on vector-iterators.md
- Add one more testcase to check the VLS use correct mode.
gcc/ChangeLog:
* config/riscv/riscv-v.cc: Add restrict for insert LMUL.
* config/riscv/riscv-vector-builtins-types.def:
Use RVV_REQUIRE_ELEN_64 to check LMUL number.
* config/riscv/riscv-vector-switch.def: Likewise.
* config/riscv/vector-iterators.md: Check TARGET_VECTOR_ELEN_64
rather than "TARGET_MIN_VLEN > 32" for all iterator.
Jonathan Wakely [Fri, 28 Mar 2025 15:41:41 +0000 (15:41 +0000)]
libstdc++: Fix -Warray-bounds warning in std::vector<bool> [PR110498]
In this case, we need to tell the compiler that the current size is not
larger than the new size so that all the existing elements can be copied
to the new storage. This avoids bogus warnings about overflowing the new
storage when the compiler can't tell that that cannot happen.
We might as well also hoist the loads of begin() and end() before the
allocation too. All callers will have loaded at least begin() before
calling _M_reallocate.
libstdc++-v3/ChangeLog:
PR libstdc++/110498
* include/bits/vector.tcc (vector<bool, A>::_M_reallocate):
Hoist loads of begin() and end() before allocation and use them
to state an unreachable condition.
* testsuite/23_containers/vector/bool/capacity/110498.cc: New
test.
Reviewed-by: Tomasz Kamiński <tkaminsk@redhat.com>
Jonathan Wakely [Fri, 28 Mar 2025 15:41:41 +0000 (15:41 +0000)]
libstdc++: Fix -Wstringop-overread warning in std::vector<bool> [PR114758]
As in r13-4393-gcca06f0d6d76b0 and a few other commits, we can avoid
bogus warnings in std::vector<bool> by hoisting some loads to before the
allocation that calls operator new. This means that the compiler has
enough info to remove the dead branches that trigger bogus warnings.
On trunk this is only needed with -fno-assume-sane-operators-new-delete
but it will help on the branches where that option doesn't exist.
libstdc++-v3/ChangeLog:
PR libstdc++/114758
* include/bits/vector.tcc (vector<bool, A>::_M_fill_insert):
Hoist loads of begin() and end() before allocation.
* testsuite/23_containers/vector/bool/capacity/114758.cc: New
test.
Reviewed-by: Tomasz Kamiński <tkaminsk@redhat.com>
Jonathan Wakely [Mon, 31 Mar 2025 14:07:12 +0000 (15:07 +0100)]
Libstdc++: Fix bootstrap failure for cross without tm.tm_zone [PR119550]
In r15-8491-g778c28c70f8573 I added a use of the Autoconf macro
AC_STRUCT_TIMEZONE, but that requires a link-test for the global tzname
object if tm.tm_zone isn't supported. That link-test isn't allowed for
cross-compilation, so bootstrap fails if tm.tm_zone isn't supported.
Since libstdc++ only cares about tm.tm_zone and won't use tzname anyway,
we don't need the link-test. Replace AC_STRUCT_TIMEZONE with a custom
macro that only checks for tm.tm_zone. We can improve on the Autoconf
macro by checking it's a suitable type, which isn't actually checked by
AC_STRUCT_TIMEZONE.
libstdc++-v3/ChangeLog:
PR libstdc++/119550
* acinclude.m4 (GLIBCXX_STRUCT_TM_TM_ZONE): New macro.
* config.h.in: Regenerate.
* configure: Regenerate.
* configure.ac: Use GLIBCXX_STRUCT_TM_TM_ZONE.
* include/bits/chrono_io.h (__formatter_chrono::_M_c): Check
_GLIBCXX_USE_STRUCT_TM_TM_ZONE instead of
_GLIBCXX_HAVE_STRUCT_TM_TM_ZONE.
Philip Herron [Fri, 28 Mar 2025 18:59:33 +0000 (18:59 +0000)]
gccrs: Fix SEGV when type path resolver fails outright
When we resolve paths we resolve to Types first we walk each segment to
the last module which has no type but then in the event that the child
of a module is not found we have a null root_tyty which needs to be caught
and turned into an ErrorType node.
Philip Herron [Fri, 28 Mar 2025 18:24:57 +0000 (18:24 +0000)]
gccrs: fix ice when setting up regions
num regions is based on the used arguments of regions which can be
less than the substutions requirements. So lets check for that and allow
anon regions to be created for them.
Fixes Rust-GCC#3605
gcc/rust/ChangeLog:
* typecheck/rust-tyty-subst.h: check for min range
gcc/testsuite/ChangeLog:
* rust/compile/issue-3605.rs: New test.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
Philip Herron [Thu, 27 Mar 2025 17:27:56 +0000 (17:27 +0000)]
gccrs: Fix ICE during const expr eval on array expressions
Array expressions are still getting turned into VIEW_CONVERT_EXPR's becuase
TYPE_MAIN_VARIANT is not set so then we might as well reuse the type-hasher
to sort this out.
Fixes Rust-GCC#3588
gcc/rust/ChangeLog:
* backend/rust-compile-context.h: only push named types
* backend/rust-compile-type.cc (TyTyResolveCompile::visit): run the type hasher
gcc/testsuite/ChangeLog:
* rust/compile/issue-3588.rs: New test.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
gccrs: Resolve module final self segment in use decls
Lowercase self suffix with path was not resolved properly, this should
point to the module right before.
gcc/rust/ChangeLog:
* resolve/rust-forever-stack.hxx: Add a new specialized function
to retrieve the last "real" segment depending on the namespace.
* resolve/rust-forever-stack.h: Add new function prototype.
* resolve/rust-early-name-resolver-2.0.cc (Early::finalize_rebind_import):
Set declared name according to the selected segment, if there is a self
suffix in the use declaration then select the previous segment.
Philip Herron [Thu, 27 Mar 2025 15:15:08 +0000 (15:15 +0000)]
gccrs: Give the builtin unit struct an actual locus
This has been a pet peeve of mine for a while because the gimple never
emitted the struct () name properly it was always empty which for record
types they always require a real locus or they dont get a proper name.
gcc/rust/ChangeLog:
* backend/rust-compile-base.cc (HIRCompileBase::unit_expression): pass ctx
* backend/rust-compile-base.h: cant be static
* backend/rust-compile-intrinsic.cc (try_handler_inner): pass ctx
* backend/rust-compile-type.cc
(TyTyResolveCompile::get_unit_type): update to grab the first locus
(TyTyResolveCompile::visit): pass ctx
* backend/rust-compile-type.h: likewise
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
Philip Herron [Thu, 27 Mar 2025 12:04:41 +0000 (12:04 +0000)]
gccrs: Fix ICE when compiling block expressions in array capacity
We need to reuse the existing compile_constant_item helper which handles
the case if this is a simple expression, fn-call or a block expression.
The patch extracts out this helper as a static method so this can be used
in more places.
Philip Herron [Wed, 26 Mar 2025 19:00:41 +0000 (19:00 +0000)]
gccrs: Add check for super traits being implemented by Self
We need to recursively check the super traits of the predicate the Self
type is trying to implement. Otherwise its cannot implement it.
Fixes Rust-GCC#3553
gcc/rust/ChangeLog:
* typecheck/rust-hir-type-check-item.cc (TypeCheckItem::resolve_impl_block_substitutions):
Track the polarity
* typecheck/rust-tyty-bounds.cc (TypeBoundPredicate::validate_type_implements_this):
new validator
* typecheck/rust-tyty.h: new prototypes
gcc/testsuite/ChangeLog:
* rust/compile/issue-3553.rs: New test.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
Philip Herron [Wed, 26 Mar 2025 17:11:36 +0000 (17:11 +0000)]
gccrs: Fix ICE when array elements are not a value
We need to check for error_mark_node when doing adjustments from coercion
sites otherwise we hit assetions as part of the coercion. That fixes the
ICE but the reason for the error_mark_node is because the array element
value.
Fixes Rust-GCC#3567
gcc/rust/ChangeLog:
* backend/rust-compile-expr.cc (CompileExpr::array_value_expr): add value chk for array expr
gcc/testsuite/ChangeLog:
* rust/compile/issue-3567.rs: New test.
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
Philip Herron [Wed, 26 Mar 2025 14:05:03 +0000 (14:05 +0000)]
gccrs: fix unconstrained infer vars on generic associated type
The trick here is that when Bar::test is resolved it resolves to the
trait method:
fn <Bar<i32>, T> (placeholder) -> placeholder
Which is fine so we need to setup the associated types for Bar<i32> which
means looking up the associated impl block then setting up the projection
of A = T so it becomes:
The issue is that the binding of the generics was still T so this caused
inference variables to be injected again but unlinked. A possible tweak
would be that we are substituting again with new infer vars to actually
just unify them enplace so they are all part of the chain. This still
might be needed but lets hold off for now.
So basically when we are Path probing we dont allow GAT's to generate new
inference vars because they wont be bound to this current segment which
just causes confusion.
Fixes Rust-GCC#3242
gcc/rust/ChangeLog:
* typecheck/rust-hir-trait-reference.h: add default infer arg
* typecheck/rust-hir-trait-resolve.cc: dont add new infer vars
* typecheck/rust-hir-type-check-path.cc (TypeCheckExpr::resolve_segments): dont infer
gcc/testsuite/ChangeLog:
* rust/compile/issue-3242.rs: no longer skip the test
Signed-off-by: Philip Herron <herron.philip@googlemail.com>
Owen Avery [Tue, 25 Mar 2025 22:18:21 +0000 (18:18 -0400)]
gccrs: Fix validation of constant items
gcc/rust/ChangeLog:
* checks/errors/rust-ast-validation.cc
(ASTValidation::visit): Allow constant items lacking expressions
if and only if they're associated with a trait definition, not a
trait implementation.
gcc/testsuite/ChangeLog:
* rust/compile/issue-3541-1.rs: New test.
* rust/compile/issue-3541-2.rs: Likewise.
Arthur Cohen [Mon, 24 Mar 2025 14:32:51 +0000 (15:32 +0100)]
rust: Lower minimum supported Rust version to 1.49
gcc/rust/ChangeLog:
* checks/errors/borrowck/ffi-polonius/Cargo.lock: Regenerate.
* checks/errors/borrowck/ffi-polonius/Cargo.toml: Update to use source patching instead of
vendoring, lower edition to 2018.
* checks/errors/borrowck/ffi-polonius/vendor/log/Cargo.toml: Change edition to 2018.
* checks/errors/borrowck/ffi-polonius/vendor/log/src/lib.rs: Remove uses of unstable
feature.
* checks/errors/borrowck/ffi-polonius/.cargo/config.toml: Removed.
libgrust/ChangeLog:
* libformat_parser/Makefile.am: Avoid using --config as it is unsupported by cargo 1.49.
* libformat_parser/Makefile.in: Regenerate.
* libformat_parser/generic_format_parser/src/lib.rs: Use extension trait for missing
features.
* libformat_parser/src/lib.rs: Likewise.
* libformat_parser/.cargo/config: Moved to...
* libformat_parser/.cargo/config.toml: ...here.
Owen Avery [Thu, 20 Mar 2025 16:17:20 +0000 (12:17 -0400)]
gccrs: nr2.0: Fix test const_generics_3.rs
gcc/testsuite/ChangeLog:
* rust/compile/const_generics_3.rs: Modify test to run with name
resolution 2.0 only and to handle the absence of a bogus
resolution error.
* rust/compile/nr2/exclude: Remove const_generics_3.rs.
Those tests are coming from libcore and module inlining was wrong, in
libcore there was a use declaration to import those modules which was
missing here.
Function could not be found and triggered an error message.
gcc/testsuite/ChangeLog:
* rust/compile/feature_rust_attri0.rs: Add extern
function declaration and change name to printf.
* rust/compile/nr2/exclude: Remove now passing test from exclusion
list.
Jørgen Kvalsvik [Mon, 31 Mar 2025 17:03:37 +0000 (19:03 +0200)]
Only write gcov when file output is on [PR119553]
gcov_write_* functions must be guarded so they only are called when
output_to_file is true, like for -fcondition-coverage, otherwise it
triggers an invalid read as detected by valgrind. The gcno file is
mostly written to from profile.cc, so it doesn't make too much sense
to hide it in path-coverage.cc. The find_paths name was a bit
imprecise, and is renamed to instrument_prime_paths.
PR gcov-profile/119553
gcc/ChangeLog:
* path-coverage.cc (find_paths): Return path count, don't
write to gcno, and rename to ...
(instrument_prime_paths): ... this.
* profile.cc (branch_prob): Write path counts to gcno.
Jonathan Wakely [Mon, 31 Mar 2025 16:07:55 +0000 (17:07 +0100)]
libstdc++: Tweak linker script to avoid conflict on Solaris
The new symbols for the _M_construct<bool> function template match an
existing pattern in the GLIBCXX_3.4.21 version, as well as the intended
pattern in the GLIBCXX_3.4.34 version. That causes a linker error on
Solaris.
libstdc++-v3/ChangeLog:
* config/abi/pre/gnu.ver (GLIBCXX_3.4.21): Make
std::basic_string::_M_construct patterns more precise.
Iain Buclaw [Sat, 29 Mar 2025 22:16:25 +0000 (23:16 +0100)]
d: Fix error with -Warray-bounds and -O2 [PR117002]
The record layout of class types in D don't get any tail padding, so it
is possible for the `classInstanceSize' to not be a multiple of the
`classInstanceAlignment'.
Rather than setting the instance alignment on the underlying
RECORD_TYPE, instead give the type an alignment of 1, which will mark it
as TYPE_PACKED. The value of `classInstanceAlignment' is instead
applied to the DECL_ALIGN of both the static `init' symbol, and the
stack allocated variable used when generating `new' for a `scope' class.
PR d/117002
gcc/d/ChangeLog:
* decl.cc (aggregate_initializer_decl): Set explicit decl alignment of
class instance.
* expr.cc (ExprVisitor::visit (NewExp *)): Likewise.
* types.cc (TypeVisitor::visit (TypeClass *)): Mark the record type of
classes as packed.
Marek Polacek [Thu, 27 Mar 2025 19:03:18 +0000 (15:03 -0400)]
c++: fix reporting routines re-entered [PR119303]
We crash while we call warning_at ("inline function used but never defined")
since it invokes dump_template_bindings -> tsubst -> ... -> convert_like ->
... -> c_common_truthvalue_conversion -> warning_at ("enum constant in boolean
context")
cp_truthvalue_conversion correctly gets complain=0 but it calls
c_common_truthvalue_conversion from c-family which doesn't have
a similar parameter.
We can fix this by tweaking diagnostic_context::report_diagnostic to
check for recursion after checking if the diagnostic was enabled.
PR c++/116960
PR c++/119303
gcc/ChangeLog:
* diagnostic.cc (diagnostic_context::report_diagnostic): Check for
non-zero m_lock later, after checking diagnostic_enabled.
gcc/testsuite/ChangeLog:
* g++.dg/cpp2a/lambda-uneval26.C: New test.
* g++.dg/warn/undefined2.C: New test.
aarch64: Remove +sme -> +sve2 feature flag dependency
As per the AArch64 ISA FEAT_SME does not require FEAT_SVE2. However, we don't
support SME without SVE2 and bail out with a 'sorry' if this configuration is
encountered. We may choose to support this in the future.
gcc/ChangeLog:
* config/aarch64/aarch64-option-extensions.def (SME): Remove SVE2 as
prerequisite and add in FCMA and F16FML.
* config/aarch64/aarch64.cc (aarch64_override_options_internal):
Diagnose use of SME without SVE2 and implicitly enable SVE2 when
enabling SME after streaming mode diagnosis.
* doc/invoke.texi (sme): Document that this can only be used with the
sve2 extension.
Jason Merrill [Sat, 29 Mar 2025 18:00:55 +0000 (14:00 -0400)]
c++: lambda in function template signature [PR119401]
Here we instantiate the lambda three times in producing A<0>::f:
1) in tsubst_function_type, substituting the type of A<>::f
2) in tsubst_function_decl, substituting the parameters of A<>::f
3) in regenerate_decl_from_template when instantiating A<>::f
The first one gets thrown away by maybe_rebuild_function_decl_type. Before
r15-7202, we happily built all of them and mangled the result wrongly as
lambda #3. After r15-7202, we try to mangle #3 as #1, which breaks because
#1 is already mangled as #1.
This patch avoids building #3 by suppressing regenerate_decl_from_template
if the template signature includes a lambda, fixing the ICE.
We now mangle the lambda as #2, which is still wrong. Addressing that
should involve not calling tsubst_function_type from tsubst_function_decl,
and building the type from the parms types in the first place rather than
fixing it up in maybe_rebuild_function_decl_type.
PR c++/119401
gcc/cp/ChangeLog:
* pt.cc (regenerate_decl_from_template): Don't regenerate if the
signature involves a lambda.
Richard Earnshaw [Mon, 31 Mar 2025 09:37:11 +0000 (10:37 +0100)]
arm: testsuite: fix vect-fmaxmin.c test
This is another case of a test that was both an executable test
requiring specific hardware and an assembler scan test. The
requirement for the hardware was masking some useful testing that
could be done (by scanning the assembly output) on almost all test
runs. Fixed in a similar manner to fmaxmin{,-2}.c by splitting the
test into two, one that scans the assembler output and one that
executes the compiled code if suitable hardware is available.
The masked issue was that this test was expecting vectorization to
occur that was incorrect given the options passed. For correct
vectorization we need -funsafe-math-optimizations as the vector
version of the single-precision operation will apply a truncation of
denormal values.
gcc/testsuite/ChangeLog:
* gcc.target/arm/vect-fmaxmin-2.c: New compile test. Split from ...
* gcc.target/arm/vect-fmaxmin.c: ... here. Remove scan-assembler
subtests. For both, add -funsafe-math-optimizations.
Tobias Burnus [Mon, 31 Mar 2025 09:44:26 +0000 (11:44 +0200)]
OpenMP: modify_call_for_omp_dispatch - fix invalid memory access after 'error' [PR119541]
OpenMP requires that the number of dispatch 'interop' clauses (ninterop)
is less or equal to the number of declare variant 'append_args' interop
objects (nappend).
While 'nappend < ninterop' was diagnosed as error, the processing continues,
which lead to an invalid out-of-bounds memory access. Solution: only
process the first nappend 'interop' clauses.
gcc/ChangeLog:
PR middle-end/119541
* gimplify.cc (modify_call_for_omp_dispatch): Limit interop claues
processing by the number of append_args arguments.
Kyrylo Tkachov [Mon, 24 Mar 2025 08:53:06 +0000 (01:53 -0700)]
PR middle-end/119442: expr.cc: Fix vec_duplicate into vector boolean modes
In this testcase GCC tries to expand a VNx4BI vector:
vector(4) <signed-boolean:4> _40;
_39 = (<signed-boolean:4>) _24;
_40 = {_39, _39, _39, _39};
This ends up in a scalarised sequence of bitfield insert operations.
This is despite the fact that AArch64 provides a vec_duplicate pattern
specifically for vec_duplicate into VNx4BI.
The store_constructor code is overly conservative when trying vec_duplicate
as it sees a requested VNx4BImode and an element mode of QImode, which I guess
is the storage mode of BImode objects.
The vec_duplicate expander in aarch64-sve.md explicitly allows QImode element
modes so it should be safe to use it. This patch extends that mode check
to allow such expanders.
The testcase is heavily auto-reduced from a real application but in itself is
nonsensical, but it does demonstrate the current problematic codegen.
Tomasz Kamiński [Fri, 28 Mar 2025 15:17:18 +0000 (16:17 +0100)]
libstdc++: Constrain formatters for chrono types [PR119517]
The formatters for chrono types defined the parse/format methods
as accepting unconstrained types, this in combination with lack
of constrain on _CharT lead to them falsely satisfying formattable
requirements for any type used as character.
This patch adjust the fromatter<T, CharT>::parse signature to:
constexpr typename basic_format_parse_context<_CharT>::iterator
parse(basic_format_parse_context<_CharT>& __pc);
And formatter<T, CharT>::format to:
template<typename _Out>
typename basic_format_context<_Out, _CharT>::iterator
format(const T& __t,
basic_format_context<_Out, _CharT>& __fc) const;
Furthermore we _CharT with __format::__char (char or wchar_t),
PR libstdc++/119517
libstdc++-v3/ChangeLog:
* include/bits/chrono_io.h (formatter):
Add __format::__char for _CharT and adjust parse and format
method signatures.
* testsuite/std/time/format/pr119517.cc: New test.
Reviewed-by: Jonathan Wakely <jwakely@redhat.com> Signed-off-by: Tomasz Kamiński <tkaminsk@redhat.com>
Jakub Jelinek [Mon, 31 Mar 2025 08:13:42 +0000 (10:13 +0200)]
gcc_release: Generate srcdir extras/infos/man pages from all FEs [PR119510]
Enabling cobol explicitly (at least unconditionally) in gcc_release has the
disadvantage that the script no longer works for GCC <= 14, I think it would
be better to keep it working for all still supported release branches.
And as mentioned in the PR, we still don't generate the
--enable-generated-files-in-srcdir extras/infos/man pages for languages
not actually enabled.
Using --enable-languages=all would mean gcc_release takes far longer and
more importantly, various FEs have extra dependencies, Ada requires a
working Ada compiler (furthermore not newer than the gcc release, so if
I run this on a system with say GCC 15 installed, even when I have Ada
installed, I won't be able to gcc_release GCC 14 or 13 etc.), D working D
compiler, Go takes a long time to build libgo.
So, the following patch instead takes similar approach to what
make regenerate-opt-urls
takes, it generates stuff even for non-enabled languages.
For most languages it works just fine and is a matter of say for cobol
make cobol.srcextra cobol.srcinfo cobol.srcman
The only problem is Modula 2, which has some messed up dependencies and
when the FE is not enabled, this will try to build the whole FE as well and
fail. I think it would be useful to fix that but at least before that is
fixed on the trunk and all release branches, the following patch just
conditionally (so that it works even for GCC 12 which doesn't have Modula 2)
enables also m2.
And lastly, libffi seems to be only enabled for Go (and maybe D), I'd prefer
not to enable those languages for the reasons stated above, so if we really
need libffi.info in release tarballs (despite libffi being used only as
implementation detail and not installed), the patch just generates it by
hand.
2025-03-29 Jakub Jelinek <jakub@redhat.com>
PR other/119510
* gcc_release: Use --enable-languages=c,c++,lto and if m2 is available,
with,m2 appended to that. Check for all possible languages and run
make $lang.srcextra $lang.srcinfo $lang.srcman for those. Add
libffi/doc/libffi.info.
They were using ssecvt instead of sseicvt, I've also added handling
for sseicvt2 which was introduced without fixing up automata, and
the relevant instruction uses DFmode. IMO this is a quite messy
area that could need TLC in the machine description itself.
Richard Biener [Thu, 27 Mar 2025 11:40:15 +0000 (12:40 +0100)]
target/119010 - add reservations for integer vector compares to zen4/zen5
The following handles TI, OI and XI mode in the respective EVEX
compare reservations that do not use memory (I've not yet run into
ones with). The znver automata has separate reservations for
integer compares (but only for zen1, for zen2 and zen3 there are
no compare reservations at all), but I don't see why that should
be necessary here.
Richard Biener [Thu, 27 Mar 2025 10:29:21 +0000 (11:29 +0100)]
target/119010 - missing reservations for Zen4/5 and SSE compares
There's the znver4_sse_test reservation which matches the memory-less
SSE compares but currently requires prefix_extra == 1. The old
znver automata in this case sometimes uses znver1-double instead of
znver1-direct, but it's quite a maze. The following simply drops
the prefix_extra requirement, but I have no idea what I'm doing here.
There doesn't seem to be any documentation on the scheduler relevant
attributes used, or at least I cannot find that.
PR target/119010
* config/i386/zn4zn5.md (znver4_sse_test): Drop test of
prefix_extra attribute.
which does not look at the mode at all. The zn4zn5 automaton lacks
this and instead has separated store and load-store reservations
in odd ways. The following renames the store one and introduces
a none variant.
PR target/119010
* config/i386/zn4zn5.md (znver4_sse_log1): Rename to
znver4_sse_log1_store.
(znver5_sse_log1): Rename to znver5_sse_log1_store.
(znver4_sse_log1): New memory-less variant.
Jakub Jelinek [Mon, 31 Mar 2025 05:51:04 +0000 (07:51 +0200)]
c++: Honor noipa attribute for FE nothrow discovery [PR119518]
The following testcase has different code generation in bar depending on
whether foo is defined or just declared.
That is undesirable when it has noipa attribute, that attribute is
documented to be a black box between caller and callee, so the caller
shouldn't know about any implicitly determined properties of the callee
and callee shouldn't know about its callers.
E.g. the ipa-pure-const passes including nothrow discovery in there all
honor noipa attribute, but the FE did not.
2025-03-31 Jakub Jelinek <jakub@redhat.com>
PR c++/119518
* decl.cc (finish_function): Don't set TREE_NOTHROW for
functions with "noipa" attribute even when we can prove
they can't throw.
Jakub Jelinek [Mon, 31 Mar 2025 05:39:53 +0000 (07:39 +0200)]
libstdc++: Fix up string _M_constructor<bool> exports [PR103827]
On Thu, Mar 27, 2025 at 02:04:24PM +0100, Jan Hubicka wrote:
> Seems I missed the approval, sorry. I will push it - I think it would
> be useful to have it in.
Unfortunately the exports in this patch only work on targets where size_t is
unsigned long, not e.g. on ia32 where it is unsigned int, or targets where
it is unsigned long long.
2025-03-31 Jakub Jelinek <jakub@redhat.com>
PR tree-optimization/103827
PR tree-optimization/80331
PR tree-optimization/87502
* config/abi/pre/gnu.ver (GLIBCXX_3.4.34): Use [jmy] rather than m
in pattern for _M_construct<bool>(char const*, size_t).
Jan Hubicka [Sun, 30 Mar 2025 21:49:49 +0000 (23:49 +0200)]
Optimize string constructor
this patch improves code generation on string constructors. We currently have
_M_construct which takes as a parameter two iterators (begin/end pointers to
other string) and produces new string. This patch adds special case of
constructor where instead of begining/end pointers we readily know the string
size and also special case when we know that source is 0 terminated. This
happens commonly when producing stirng copies. Moreover currently ipa-prop is
not able to propagate information that beg-end is known constant (copied string
size) which makes it impossible for inliner to spot the common case where
string size is known to be shorter than 15 bytes and fits in local buffer.
Finally I made new constructor inline. Because it is explicitely instantiated
without C++20 constexpr we do not produce implicit instantiation (as required
by standard) which prevents inlining, ipa-modref and any other IPA analysis to
happen. I think we need to make many of the other functions inline, since
optimization accross string manipulation is quite important. There is PR94960
to track this issue.
* config/abi/pre/gnu.ver: Add version for _M_construct<bool>
* include/bits/basic_string.h: (basic_string::_M_construct<bool>): Declare.
(basic_string constructors): Use it.
* include/bits/basic_string.tcc: (basic_string::_M_construct<bool>): New template.
* src/c++11/string-inst.cc: Instantated S::_M_construct<bool>.
gcc/testsuite/ChangeLog:
* g++.dg/tree-ssa/pr80331.C: New test.
* g++.dg/tree-ssa/pr87502.C: New test.
Sandra Loosemore [Wed, 26 Mar 2025 18:25:51 +0000 (18:25 +0000)]
Doc: Clean up New/Delete Builtins manual section
I noticed that the "New/Delete Builtins" section failed to explicitly
name or describe the arguments of the builtin functions it purported
to document, outside of using them in an example. I've fixed that
and cleaned up the whole section.
gcc/ChangeLog
* doc/extend.texi (New/Delete Builtins): Cleanup up the text and
explicitly list the builtins being documented.
projects / thirdparty / gcc.git / log
