For the test case
int32_t foo (svint32_t x)
{
svbool_t pg = svpfalse ();
return svlastb_s32 (pg, x);
}
compiled with -O3 -mcpu=grace -msve-vector-bits=128, GCC produced:
foo:
pfalse p3.b
lastb w0, p3, z0.s
ret
when it could use a Neon lane extract instead:
foo:
This was leftover from work-in-progress state, where only std::queue was
tested.
libstdc++-v3/ChangeLog:
* testsuite/std/format/ranges/adaptors.cc: Updated test.
---
Tested on x86_64-linux.
libstdc++-v3/testsuite/std/format/ranges/adaptors.cc | 2 +-
1 file changed, 1 insertion(+), 1 d
Tested so far on x86_64-pc-linux-gnu (just modules.exp), OK for trunk/15
if full bootstrap+regtest succeeds?
A potentially safer approach that would slightly bloat out the size of
the built modules would be to always stream this variable rather than
having any conditions, but from what I can tell
From: Arthur Cohen
Content was validated using the Nu HTML checker per the contributing doc.
---
htdocs/gcc-15/changes.html | 57 ++
1 file changed, 57 insertions(+)
diff --git a/htdocs/gcc-15/changes.html b/htdocs/gcc-15/changes.html
index 3e3c6655..10b1ce
On 4/19/25 3:29 PM, Denis Chertykov wrote:
Bugfix for PR118591
This bug occurs only with '-Os' option.
The function 'inherit_reload_reg ()' have a wrong condition:
static bool
inherit_reload_reg (bool def_p, int
SVE loads and stores where the predicate is all-true can be optimized to
unpredicated instructions. For example,
svuint8_t foo (uint8_t *x)
{
return svld1 (svptrue_b8 (), x);
}
was compiled to:
foo:
ptrue p3.b, all
ld1bz0.b, p3/z, [x0]
ret
but can be compiled to:
foo
On Fri, 18 Apr 2025, Jason Merrill wrote:
> limit_bad_template_recursion currently avoids immediate instantiation of
> templates from uses in an already ill-formed instantiation, but we still can
> get unnecessary recursive instantiation in pending_templates if the
> instantiation was queued befor
Hello, gentle maintainer.
This is a message from the Translation Project robot.
A revised PO file for textual domain 'gcc' has been submitted
by the French team of translators. The file is available at:
https://translationproject.org/latest/gcc/fr.po
(This file, 'gcc-15.1.0.fr.po', has jus
Pengfei Li writes:
> This patch transforms RTL expressions of the form (subreg (not X) off)
> into (not (subreg X off)) when the subreg is an operand of a bitwise AND
> or OR. This transformation can expose opportunities to combine a NOT
> operation with the bitwise AND/OR.
>
> For example, it imp
This ensures that wstring-inst.o and similar files will be rebuilt when
string-inst.cc changes.
libstdc++-v3/ChangeLog:
* src/c++11/Makefile.am: Add prerequisites for targets that
depend on string-inst.cc.
* src/c++11/Makefile.in: Regenerate.
---
We could make a lot of ot
Andrew Stubbs wrote:
This testcase ensures that the interop HSA support is sufficient to run
a kernel manually on the same device. It reuses an OpenMP kernel in
order to avoid all the complication of compiling a custom kernel in
Dejagnu (although, this does mean matching the OpenMP runtime
envir
Hi!
Here is a patch to handle the PARALLEL case too.
I think we can just use rtx_equal_p there, because it will always use
SImode in the EXPR_LIST REGs in that case.
Bootstrapped/regtested on s390x-linux, ok for trunk and 15.2 (with
CALL_EXPR_MUST_TAIL_CALL (call_expr) && added in that case)?
20
> On 25 Apr 2025, at 12:06, Richard Sandiford wrote:
>
> Kyrylo Tkachov writes:
>> Hi Richard,
>>
>>> On 23 Apr 2025, at 13:47, Richard Sandiford
>>> wrote:
>>>
>>> Thanks for all the feedback. I've tried to address it in the version
>>> below. I'll push later today if there are no furth
On Fri, Apr 25, 2025 at 11:49 AM Jonathan Wakely wrote:
> On 23/04/25 13:56 +0200, Tomasz Kamiński wrote:
> >When width parameter is specified for formatting range, tuple or escaped
> >presentation of string, we used to format characters to temporary string,
> >and write produce sequence padded a
On Fri, Apr 25, 2025 at 12:55 PM Jonathan Wakely wrote:
> This ensures that wstring-inst.o and similar files will be rebuilt when
> string-inst.cc changes.
>
> libstdc++-v3/ChangeLog:
>
> * src/c++11/Makefile.am: Add prerequisites for targets that
> depend on string-inst.cc.
>
On Fri, Apr 25, 2025 at 01:21:46PM +0200, Jakub Jelinek wrote:
> Hi!
>
> Here is a patch to handle the PARALLEL case too.
> I think we can just use rtx_equal_p there, because it will always use
> SImode in the EXPR_LIST REGs in that case.
>
> Bootstrapped/regtested on s390x-linux, ok for trunk an
> On Apr 24, 2025, at 19:56, Kees Cook wrote:
>
>
>
> On April 24, 2025 1:44:23 PM PDT, Qing Zhao wrote:
>>
>>
>>> On Apr 24, 2025, at 15:43, Bill Wendling wrote:
>>>
>>> On Thu, Apr 24, 2025 at 8:15 AM Qing Zhao wrote:
Hi,
Kees reported a segmentation failure when
If -msve-vector-bits=128, SVE loads and stores (LD1 and ST1) with a
ptrue predicate can be replaced by neon instructions (LDR and STR),
thus avoiding the predicate altogether. This also enables formation of
LDP/STP pairs.
For example, the test cases
svfloat64_t
ptrue_load (float64_t *x)
{
svboo
On Fri, 25 Apr 2025 at 15:31, Tomasz Kamiński wrote:
>
> This was leftover from work-in-progress state, where only std::queue was
> tested.
>
> libstdc++-v3/ChangeLog:
>
> * testsuite/std/format/ranges/adaptors.cc: Updated test.
> ---
> Tested on x86_64-linux.
OK for trunk and gcc-15.
>
When the patch,
https://gcc.gnu.org/pipermail/gcc-patches/2024-August/660807.html was rewroked
into r15-3047-g404d947d8ddd3c,
the include for calls.h was still included and missed that it was no longer
needed.
Pushed as obvious.
PR tree-optimization/119811
gcc/ChangeLog:
* tre
Fifty-eight new testcases for COBOL. These cover a range of fundamental
operations and data representation.
>From 591831dcd4bc9cb9c089d952e73ec8bfcb6cb3fb Mon Sep 17 00:00:00 2001
From: Robert Dubner mailto:rdub...@symas.com
Date: Fri, 25 Apr 2025 10:19:35 -0400
Subject: [PATCH] cobol: New testca
On Fri, 25 Apr 2025 at 15:41, Jonathan Wakely wrote:
>
> On Fri, 25 Apr 2025 at 15:31, Tomasz Kamiński wrote:
> >
> > This was leftover from work-in-progress state, where only std::queue was
> > tested.
> >
> > libstdc++-v3/ChangeLog:
> >
> > * testsuite/std/format/ranges/adaptors.cc: Up
On 1/12/25 10:44 PM, Zhijin Zeng wrote:
Compared to the patch v2, I added Zfinx check and Zfh check. Please help to
review again.
Thanks,
Zhijin
From 9ddb402cebe868050ebc2f75e4d87238161411b4 Mon Sep 17 00:00:00 2001
From: Zhijin Zeng
Date: Sat, 11 Jan 2025 12:09:11 +0800
Subject: [PATCH]
Hello, gentle maintainer.
This is a message from the Translation Project robot. (If you have
any questions, send them to .)
A new POT file for textual domain 'gcc' has been made available
to the language teams for translation. It is archived as:
https://translationproject.org/POT-files/gcc
Jennifer Schmitz writes:
> For the test case
> int32_t foo (svint32_t x)
> {
> svbool_t pg = svpfalse ();
> return svlastb_s32 (pg, x);
> }
> compiled with -O3 -mcpu=grace -msve-vector-bits=128, GCC produced:
> foo:
> pfalse p3.b
> lastb w0, p3, z0.s
> ret
> when it could us
Tested x86_64-pc-linux-gnu, applying to trunk.
-- 8< --
The patch for PR87185 fixed the ICE without fixing the underlying problem,
that we were failing to find the declaration of the capture proxy that we
are trying to decide whether to prune. Fixed by looking at the right index
in stmt_list_sta
The idea of -march-map= is to simply and future proof select the
best -march for a certain arch, without requiring that the compiler
has support for it (like having a special multilib for it) - while
-march= sets the actually used '.target' (and the compiler might
actually generate specialized cod
On Fri, Apr 25, 2025 at 12:19 AM Jonathan Wakely wrote:
> In r10-452-ge625ccc21a91f3 I noted that we don't have an accessor for
> invoking _M_impl._M_key_compare in the associative containers. That
> meant that the static assertions to check for valid comparison functions
> were squirrelled away
On Fri, Apr 25, 2025 at 12:20 AM Jonathan Wakely wrote:
> Moving the static_assert that checks is_invocable_r_v into _Task_state
> means it is checked when we instantiate that class template.
>
> Replacing the __create_task_state function with a static member function
> _Task_state::_S_create ens
Hi!
We now have some script which transforms e.g.
GCC 15.1
line in gcc-15/changes.html to
GCC 15.1
This unfortunately breaks the gcc_release script, which looks for
GCC 15.1 appearing in gennews after optional blanks from the start of
the line in the NEWS file, which is no longer the case, there
>
> I am not so sure about this when it come to relatively common
> instructions. Hiding things in unspec prevents combine and other RTL
> passes from doing their job. I would say that it only makes sense for
> siutations where RTL equivalent is very inconvenient.
>
In the direction of using gener
On Fri, 25 Apr 2025 at 09:13, Tomasz Kaminski wrote:
>
>
>
> On Fri, Apr 25, 2025 at 12:19 AM Jonathan Wakely wrote:
>>
>> In r10-452-ge625ccc21a91f3 I noted that we don't have an accessor for
>> invoking _M_impl._M_key_compare in the associative containers. That
>> meant that the static assertio
s/thread:id/thread::id/ in the subject line
On 24/04/25 12:50 +0200, Tomasz Kamiński wrote:
This patch add constrains __formatter::__char to _CharT type parameter
s/constrains/constraint/
of formatter specialization, matching the constrains
same change here
of formatting of integer/point
On Tue, 22 Apr 2025 at 08:13, Tomasz Kaminski wrote:
>
> The test cover constructors introduced in [PATCH v2] libstdc++-v3: Implement
> missing allocator-aware constructors for unordered containers,
> so I will merge that after.
This change is OK for trunk now too.
>
> On Fri, Apr 18, 2025 at 5
On Fri, 18 Apr 2025 at 09:11, Tomasz Kamiński wrote:
>
> This patch implements remainder of LWG2713 (after r15-8293-g64f5c854597759)
> by adding missing allocator aware version of unordered associative containers
> constructors accepting pair of iterators or initializer_list, and
> corresponding
This fixes
FAIL: 22_locale/ctype/is/string/89728_neg.cc -std=gnu++98 (test for errors,
line )
Since r16-133-g8acea9ffa82ed8 we don't keep issuing more errors after
the first one, so this dg-error no longer matches anything.
libstdc++-v3/ChangeLog:
* testsuite/22_locale/ctype/is/string
Rename _M_a to match the name of the exposition-only data member shown
in the standard, i.e. 'counter'.
libstdc++-v3/ChangeLog:
* include/std/latch (latch::_M_a): Rename to _M_counter.
---
IMHO this makes it a little easier to compare the implementation to the
spec.
Tested x86_64-linux.
Thanks Jeff for comments.
> The point is the separation isn't neat and clean in rtx_costs. We've
> kind of let it go as-is with the vector short-cut out, but I don't think
> that's really where we want to be, it was just an expedient decision to
> allow us to focus on more important stuff. N
On Fri, Apr 25, 2025 at 1:51 PM Jonathan Wakely wrote:
> Rename _M_a to match the name of the exposition-only data member shown
> in the standard, i.e. 'counter'.
>
> libstdc++-v3/ChangeLog:
>
> * include/std/latch (latch::_M_a): Rename to _M_counter.
> ---
>
> IMHO this makes it a little
Kyrylo Tkachov writes:
> Hi Richard,
>
>> On 23 Apr 2025, at 13:47, Richard Sandiford
>> wrote:
>>
>> Thanks for all the feedback. I've tried to address it in the version
>> below. I'll push later today if there are no further comments.
>>
>> Richard
>>
>>
>> The list is structured as:
>>
---
Pushed to wwwdocs because the old URL gives a 404 error.
htdocs/robots.txt | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/htdocs/robots.txt b/htdocs/robots.txt
index 01ae43dd..526fa1dd 100644
--- a/htdocs/robots.txt
+++ b/htdocs/robots.txt
@@ -1,4 +1,4 @@
-# See http://ww
---
Pushed to wwwdocs.
htdocs/gcc-15/changes.html | 25 -
1 file changed, 24 insertions(+), 1 deletion(-)
diff --git a/htdocs/gcc-15/changes.html b/htdocs/gcc-15/changes.html
index 3cec4ff4..3e3c6655 100644
--- a/htdocs/gcc-15/changes.html
+++ b/htdocs/gcc-15/changes.html
libgomp/
* testsuite/libgomp.c-c++-common/target-cdtor-1.c: New.
---
.../libgomp.c-c++-common/target-cdtor-1.c | 89 +++
1 file changed, 89 insertions(+)
create mode 100644 libgomp/testsuite/libgomp.c-c++-common/target-cdtor-1.c
diff --git a/libgomp/testsuite/
Hi!
On 2025-04-23T21:49:27+0200, I wrote:
> [...]
With 'libgomp.c++/target-cdtor-2.C' changed to
'dg-require-effective-target init_priority', and
'libgomp.c++/target-cdtor-1.C', 'libgomp.c++/target-cdtor-2.C' changed to
scan for '__cxa_atexit' only for 'target cxa_atexit', I've pushed to
trunk br
This testcase ensures that the interop HSA support is sufficient to run
a kernel manually on the same device. It reuses an OpenMP kernel in
order to avoid all the complication of compiling a custom kernel in
Dejagnu (although, this does mean matching the OpenMP runtime
environment, which might be
On Thu, 24 Apr 2025 at 15:48, Jonathan Wakely wrote:
>
> On Fri, 18 Apr 2025 at 23:08, Jason Merrill wrote:
> >
> > limit_bad_template_recursion currently avoids immediate instantiation of
> > templates from uses in an already ill-formed instantiation, but we still can
> > get unnecessary recursi
On 23/04/25 13:56 +0200, Tomasz Kamiński wrote:
When width parameter is specified for formatting range, tuple or escaped
presentation of string, we used to format characters to temporary string,
and write produce sequence padded according to the spec. However, once the
estimated width of formatte
Hi!
I've committed the following patch to rotate news, chose to move
202{2,3}-ish items to news.html. We'll soon have 14.3/13.4 releases
so the first column will be about the same length as the second.
diff --git a/htdocs/index.html b/htdocs/index.html
index 8bdc4071..a662ab09 100644
--- a/htdoc
Bootstrapped and regtested on x86_64-pc-linux-gnu, OK for trunk/15?
-- >8 --
When importing a CNTTP object, since r15-3031-g0b7904e274fbd6 we
shortcut the processing of the generated NTTP so that we don't attempt
to recursively load pendings. However, due to an oversight we do not
properly set T
On Fri, Apr 25, 2025 at 01:55:58PM +0100, Andre Vieira (lists) wrote:
> Differences from v1:
>
> Made suggested changes to testcase and moved to target-agnostic area.
>
> gcc/ChangeLog:
>
> PR rtl-optimization/116479
> * modulo-sched.cc (doloop_register_get): Reject conditions with
>
Differences from v1:
Made suggested changes to testcase and moved to target-agnostic area.
gcc/ChangeLog:
PR rtl-optimization/116479
* modulo-sched.cc (doloop_register_get): Reject conditions with
decrements that are not 1.
gcc/testsuite/ChangeLog:
* gcc.dg/pr1
On 2025-04-23 10:18, Richard Biener wrote:
On Tue, Apr 22, 2025 at 5:43 PM Andi Kleen wrote:
On 2025-04-22 13:22, Richard Biener wrote:
> On Sat, Apr 12, 2025 at 5:09 PM Andi Kleen wrote:
>>
>> From: Andi Kleen
>>
>> ... that uses -march=native -mtune=native to build a compiler
>> optimized
Tested x86_64-pc-linux-gnu, applying to trunk.
-- 8< --
I'm about to add a bugfix that changes the ABI of noexcept lambdas, so first
let's add the new ABI version. And I think it's time to update the
compatibility version; let's bump to GCC 13, before the addition of concepts
mangling.
gcc/Chan
On Fri, Apr 25, 2025 at 04:56:52PM +, Qing Zhao wrote:
>
>
> > On Apr 24, 2025, at 13:07, Kees Cook wrote:
> >
> > On Thu, Apr 24, 2025 at 04:36:14PM +, Qing Zhao wrote:
> >>
> >>> On Apr 24, 2025, at 11:59, Martin Uecker wrote:
> >>>
> >>> Am Donnerstag, dem 24.04.2025 um 15:15 +000
The _Padding_sink was behaving incorreclty, when the predicated width (based on
code units counts) was higher than _M_maxwidth, but lower than _M_padwidth.
In this case _M_update() returned without calling _M_force_update() and
computing
field width for Unicode encondings, because _M_buffering() r
Hi,
This patch marks m32c*-*-* targets obsolete in GCC 16. The target has
not had a maintainer since GCC 9 (r9-1950), and fails to compile even
the simplest of functions since GCC 8 (r8-777, as reported in PR83670).
OK for trunk?
Regards,
Iain.
---
contrib/ChangeLog:
* config-list.mk:
On 4/25/25 10:30 AM, Nathaniel Shead wrote:
Tested so far on x86_64-pc-linux-gnu (just modules.exp), OK for trunk/15
if full bootstrap+regtest succeeds?
OK.
A potentially safer approach that would slightly bloat out the size of
the built modules would be to always stream this variable rather
Jennifer Schmitz writes:
> If -msve-vector-bits=128, SVE loads and stores (LD1 and ST1) with a
> ptrue predicate can be replaced by neon instructions (LDR and STR),
> thus avoiding the predicate altogether. This also enables formation of
> LDP/STP pairs.
>
> For example, the test cases
>
> svfloat
I noticed the comments and the code don't match.
The correct form is:
'if (0 != 0)': false
and
'if (1 != 0)': true
That is always NE and always 0 as the second operand.
Also there is a spello for statement in the comment in
front of gimple_cond_true_p.
Pushed as obvious.
gcc/ChangeLog:
> On Apr 25, 2025, at 13:18, Kees Cook wrote:
>
> On Fri, Apr 25, 2025 at 04:56:52PM +, Qing Zhao wrote:
>>
>>
>>> On Apr 24, 2025, at 13:07, Kees Cook wrote:
>>>
>>> On Thu, Apr 24, 2025 at 04:36:14PM +, Qing Zhao wrote:
> On Apr 24, 2025, at 11:59, Martin Uecker wrote:
On Thu, Apr 24, 2025 at 4:56 PM Kees Cook wrote:
> On April 24, 2025 1:44:23 PM PDT, Qing Zhao wrote:
> >> On Apr 24, 2025, at 15:43, Bill Wendling wrote:
> >>
> >> On Thu, Apr 24, 2025 at 8:15 AM Qing Zhao wrote:
> >>>
> >>> Hi,
> >>>
> >>> Kees reported a segmentation failure when he used the
On Sat, Apr 19, 2025 at 08:33:48PM -0700, Andrew Pinski wrote:
> > + {
> > + tree other_value = NULL_TREE;
> > + /* If we have a function call that we know the return value is
> > the same
> > +as the argument, try the argument too. */
> > + int flags = gi
During the base register initialization, when we are eliminating the load
instruction, we were calling `emit_move_insn` on registers of the same
size but of different mode in some cases, causing an ICE.
This patch fixes this, by adding a check for the modes to match before
calling `emit_move_insn`
Some backends do not define TARGET_ASM_OUTPUT_MI_THUNK. But the generic
thunk support cannot emit code for calling variadic methods of
multiple-inheritance classes. Example error for pru-unknown-elf:
.../gcc/gcc/testsuite/g++.dg/ipa/pr83549.C:7:24: error: generic thunk code
fails for method 'v
On Fri, Apr 25, 2025 at 11:54 AM Jakub Jelinek wrote:
>
> On Sat, Apr 19, 2025 at 08:33:48PM -0700, Andrew Pinski wrote:
> > > + {
> > > + tree other_value = NULL_TREE;
> > > + /* If we have a function call that we know the return value is
> > > the same
> > > +a
Testcases for musttail call optimization fail on pru-unknown-elf:
FAIL: c-c++-common/musttail14.c -std=gnu++17 (test for excess errors)
Excess errors:
.../gcc/gcc/testsuite/c-c++-common/musttail14.c:37:14: error: cannot
tail-call: caller uses sjlj exceptions
Silence these errors by disabli
On Thu, Apr 24, 2025 at 11:56:52PM +0200, Bernhard Reutner-Fischer wrote:
>
> >>* lib/target-supports.exp
> >>(check_effective_target_variadic_mi_thunk): New function.
> >OK.
> >jeff
> >
>
> Please document new effective_target checks in sourcebuild.texi
>
> thanks
Pushed with added doc
I forgot to push this before going on vacation; sorry.
Pushed to trunk.
---
htdocs/gcc-15/changes.html | 260 +++-
htdocs/gcc-15/diag-color-screenshot.png | Bin 0 -> 33062 bytes
2 files changed, 257 insertions(+), 3 deletions(-)
create mode 100644 htdocs/gcc-15
Hi Jerry,
Am 24.04.25 um 22:56 schrieb Jerry D:
On 4/24/25 12:59 PM, Harald Anlauf wrote:
Dear all,
the attached patch is the result of my attempts to fix an ICE when
compiling gfortran.dg/proc_ptr_52.f90 with -fcheck=all. While
trying to reduce this, I found several oddities with functions
r
Autoconf documents AC_LANG_SAVE, AC_LANG_CPLUSPLUS etc. as deprecated.
libstdc++-v3/ChangeLog:
* acinclude.m4 (GLIBCXX_LANG_PUSH, GLIBCXX_LANG_POP): New
macros to replace uses of AC_LANG_SAVE, AC_LANG_CPLUSPLUS, and
AC_LANG_RESTORE.
* configure: Regenerate.
---
To
> On Apr 24, 2025, at 13:07, Kees Cook wrote:
>
> On Thu, Apr 24, 2025 at 04:36:14PM +, Qing Zhao wrote:
>>
>>> On Apr 24, 2025, at 11:59, Martin Uecker wrote:
>>>
>>> Am Donnerstag, dem 24.04.2025 um 15:15 + schrieb Qing Zhao:
Hi,
Kees reported a segmentation failur
Ping https://gcc.gnu.org/pipermail/gcc-patches/2025-January/672568.html
Missed the version 15 freeze with the last ping, I believe the project
is open for general development again now?
[sorry for hijacking this thread]
Does anyone have any response to Yeoul Na's suggestion to hold off on
implementing __counted_by_expr() until the next C standards committee?
https://discourse.llvm.org/t/rfc-bounds-safety-in-c-syntax-compatibility-with-gcc/85885/17?u=void
I'm fine with it, bu
This fixes:
FAIL: tr1/2_general_utilities/shared_ptr/cons/43820_neg.cc -std=gnu++26 (test
for errors, line 283)
FAIL: tr1/2_general_utilities/shared_ptr/cons/43820_neg.cc -std=gnu++26 (test
for errors, line 305)
This is another consequence of r16-133-g8acea9ffa82ed8 which prevents
the -Wdele
Currently std::addressof calls std::__addressof which uses
__builtin_addressof. This leads to me prefering std::__addressof in some
code, to avoid the extra hop. But it's not as though the implementation
of std::__addressof is complicated and reusing it avoids any code
duplication.
So let's just m
libstdc++-v3/ChangeLog:
* include/bits/ptr_traits.h (to_address): Use markdown for
formatting in Doxygen comments.
---
Pushed to trunk.
libstdc++-v3/include/bits/ptr_traits.h | 6 +++---
1 file changed, 3 insertions(+), 3 deletions(-)
diff --git a/libstdc++-v3/include/bits/ptr_
This will hardly make a dent in the very slow compile times for
but it seems worth doing anyway.
libstdc++-v3/ChangeLog:
* include/bits/regex_compiler.h (_AnyMatcher::operator()):
Replace tag dispatching with 'if constexpr'.
(_AnyMatcher::_M_apply): Remove both overloads.
Hi Jeff,
On Fri, Apr 25, 2025 at 6:04 AM Jeff Law wrote:
>
> On 4/24/25 2:37 AM, Anton Blanchard wrote:
> > Add integer and floating point scheduling models for the Tenstorrent
> > Ascalon 8 wide CPU.
> >
> > gcc/ChangeLog:
> > * config/riscv/riscv-cores.def (RISCV_TUNE): Update.
> >
This patch adds more tests for vector permutes which can now be optimized as
FMOV with the generic PERM change and the aarch64 AND patch.
PR target/100165
gcc/testsuite/ChangeLog:
* gcc.target/aarch64/fmov-3.c: New test.
* gcc.target/aarch64/fmov-4.c: New test.
*
We can optimize AND with certain vector of immediates as FMOV if the result of
the AND is as if the upper lane of the input vector is set to zero and the lower
lane remains unchanged.
For example, at present:
v4hi
f_v4hi (v4hi x)
{
return x & (v4hi){ 0x, 0x, 0, 0 };
}
generates:
f_v4h
Certain permute that blends a vector with zero can be interpreted as an AND of a
mask. This idea was suggested by Richard Sandiford when he was reviewing my
patch which tries to optimizes certain vector permute with the FMOV instruction
for the aarch64 target. Canonicalizing this class of vector pe
> Richard Sandiford writes:
> > I think this would also simplify the evpc detection, since the
> > requirement for using AND is the same for big-endian and
> > little-endian, namely that index I of the result must either come from
> > index I of the nonzero vector or from any element of the zero v
82 matches
Mail list logo
