[Bug target/111366] New: error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[Bug target/111366] New: error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

            Bug ID: 111366
           Summary: error: inlining failed in call to 'always_inline'
                    'hwy::PreventElision<int&>(int&)void': target specific
                    option mismatch
           Product: gcc
           Version: 13.2.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: target
          Assignee: unassigned at gcc dot gnu.org
          Reporter: malat at debian dot org
  Target Milestone: ---

I cannot compile highway on ppc64el (ubuntu setup). It fails with:

In file included from /<<PKGBUILDDIR>>/hwy/examples/skeleton.h:24,
                 from /<<PKGBUILDDIR>>/hwy/examples/skeleton_test.cc:18:
/<<PKGBUILDDIR>>/hwy/base.h: In member function
‘skeleton::N_PPC10::TestFloorLog2::operator()<float, hwy::N_PPC10::Simd<float,
4ul, 0> >(float, hwy::N_PPC10::Simd<float, 4ul, 0>)void’:
/<<PKGBUILDDIR>>/hwy/base.h:1511:14: error: inlining failed in call to
‘always_inline’ ‘hwy::PreventElision<int&>(int&)void’: target specific option
mismatch
 1511 | HWY_API void PreventElision(T&& output) {
      |              ^~~~~~~~~~~~~~
In file included from /<<PKGBUILDDIR>>/hwy/foreach_target.h:266,
                 from /<<PKGBUILDDIR>>/hwy/examples/skeleton_test.cc:24:
/<<PKGBUILDDIR>>/hwy/examples/skeleton_test.cc:73:24: note: called from here
   73 |     hwy::PreventElision(sum);
      |     ~~~~~~~~~~~~~~~~~~~^~~~~
/<<PKGBUILDDIR>>/hwy/base.h:1511:14: error: inlining failed in call to
‘always_inline’ ‘hwy::PreventElision<int&>(int&)void’: target specific option
mismatch
 1511 | HWY_API void PreventElision(T&& output) {
      |              ^~~~~~~~~~~~~~
/<<PKGBUILDDIR>>/hwy/examples/skeleton_test.cc:73:24: note: called from here
   73 |     hwy::PreventElision(sum);
      |     ~~~~~~~~~~~~~~~~~~~^~~~~

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #1 from Mathieu Malaterre <malat at debian dot org> ---
Full log:

*
https://launchpadlibrarian.net/684588621/buildlog_ubuntu-mantic-ppc64el.highway_1.0.7-1_BUILDING.txt.gz

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #2 from Mathieu Malaterre <malat at debian dot org> ---
Typical setup:

* https://gcc.godbolt.org/z/5W45a6j8x

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #3 from Mathieu Malaterre <malat at debian dot org> ---
compilation flags required are: `-O1 -mcpu=power9 -flto=auto`

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #4 from Mathieu Malaterre <malat at debian dot org> ---
reduction started:

 % cvise check.sh skeleton_test.cc
00:00:02 INFO ===< 4165591 >===
00:00:02 INFO running 4 interestingness tests in parallel
00:00:02 INFO INITIAL PASSES

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[jan.wassenberg at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

Jan Wassenberg <jan.wassenberg at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |jan.wassenberg at gmail dot com

--- Comment #5 from Jan Wassenberg <jan.wassenberg at gmail dot com> ---
Thanks for reporting this. One might think this is caused by -mcpu=power9
clashing with our #pragma target altivec,vsx,power8-vector.

However, what makes this seem like an actual bug is that -flto=auto is
additionally required to trigger the bug.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[jan.wassenberg at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #6 from Jan Wassenberg <jan.wassenberg at gmail dot com> ---
Thinking about this more, the LTO means more opportunity for inlining and thus
for the compiler to hit the legit "don't want to inline POWER9 into POWER8"
error.

Interestingly this does not happen on x86 - even if we have -march=haswell,
#pragma target -mssse3 is not a problem. Perhaps POWER9 vs POWER8 has a
larger/wider impact on codegen than Haswell vs SSE.

I suppose we will have to work around this on the library level: if we detect
any POWER9 flag, then do not attempt to also generate POWER8 codepaths.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #7 from Mathieu Malaterre <malat at debian dot org> ---
In case this helps resolve this, here is the cvise-reduced code:

% cat skeleton_test.cc
#include <stddef.h>
#define HWY_PRAGMA(tokens) _Pragma(#tokens)
namespace hwy {
#define HWY_PUSH_ATTRIBUTES(targets_str) HWY_PRAGMA(GCC target targets_str)
inline __attribute__((always_inline)) void PreventElision(int output) {
  asm("" : "+r"(output) : : "memory");
}
} // namespace hwy
#if HWY_TARGET == HWY_PPC10
#define HWY_BEFORE_NAMESPACE() HWY_PUSH_ATTRIBUTES(",cpu=power10")
#endif
HWY_BEFORE_NAMESPACE() namespace detail {
  template <typename, size_t, int> struct CappedTagChecker {};
}
template <typename T, size_t kLimit, int kPow2 = 0>
using CappedTag = detail::CappedTagChecker<T, kLimit, kPow2>;
template <class D> size_t Lanes(D);
template <typename, size_t, size_t kMinArg, class Test> struct ForeachCappedR {
  static void Do(size_t min_lanes, size_t max_lanes) {
    CappedTag<int, kMinArg> d;
    Test()(int(), d);
    Do(min_lanes, max_lanes);
  }
};
template <class Test> struct ForPartialVectors {
  template <typename T> void operator()(T t) {
    (void)t;
    ForeachCappedR<T, 1, 1, Test>::Do(1, 1);
  }
};
struct TestFloorLog2 {
  template <class T, class DF> void operator()(T, DF df) {
    size_t count = Lanes(df);
    int sum;
    for (size_t i; count; ++i)
      hwy::PreventElision(sum);
  }
};
void TestAllFloorLog2() { ForPartialVectors<TestFloorLog2>()(float()); }


Compilation line is:

% /usr/bin/c++  -O1 -mcpu=power9 -flto=auto -c skeleton_test.cc

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #8 from Mathieu Malaterre <malat at debian dot org> ---
Further reduced:

#include <stddef.h>
#define HWY_PRAGMA(tokens) _Pragma(#tokens)
#define HWY_PUSH_ATTRIBUTES(targets_str) HWY_PRAGMA(GCC target targets_str)
__attribute__((always_inline)) void PreventElision() { asm(""); }
#define HWY_BEFORE_NAMESPACE() HWY_PUSH_ATTRIBUTES(",cpu=power10")
HWY_BEFORE_NAMESPACE() namespace detail {
  template <typename, size_t, int> struct CappedTagChecker {};
}
template <typename T, size_t kLimit, int kPow2 = 0>
using CappedTag = detail::CappedTagChecker<T, kLimit, kPow2>;
template <typename, size_t, size_t kMinArg, class Test> struct ForeachCappedR {
  static void Do(size_t, size_t) {
    CappedTag<int, kMinArg> d;
    Test()(int(), d);
  }
};
template <class Test> struct ForPartialVectors {
  template <typename T> void operator()(T) {
    ForeachCappedR<T, 1, 1, Test>::Do(1, 1);
  }
};
struct TestFloorLog2 {
  template <class T, class DF> void operator()(T, DF) { PreventElision(); }
};
void TestAllFloorLog2() { ForPartialVectors<TestFloorLog2>()(float()); }

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

Kewen Lin <linkw at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|UNCONFIRMED                 |RESOLVED
                 CC|                            |linkw at gcc dot gnu.org

--- Comment #9 from Kewen Lin <linkw at gcc dot gnu.org> ---
It looks to me that the error message is expected, because the source code
forces the function as always_inline, users would like it to be inlined always,
it's compiled with -mcpu=power10 as pragma function target specifies, but the
caller is compiled with -mcpu=power9 as command line specifies, the caller
doesn't have the full ISA capability over the callee, it's unable to make
inlining happen.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #10 from Kewen Lin <linkw at gcc dot gnu.org> ---
(In reply to Jan Wassenberg from comment #5)
> Thanks for reporting this. One might think this is caused by -mcpu=power9
> clashing with our #pragma target altivec,vsx,power8-vector.
> 
> However, what makes this seem like an actual bug is that -flto=auto is
> additionally required to trigger the bug.

This reminds me there is one bug in the current function rs6000_can_inline_p,
see https://gcc.gnu.org/bugzilla/show_bug.cgi?id=102059#c17

-flto=auto makes the callee have a constructed DECL_FUNCTION_SPECIFIC_TARGET,
so callee_tree is not NULL, it would further check ISA flags. Without it, the
current implementation doesn't adopt the default option node and just ignore it
unexpectedly. Filed PR111380.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #11 from Mathieu Malaterre <malat at debian dot org> ---
(In reply to Kewen Lin from comment #9)
> It looks to me that the error message is expected, because the source code
> forces the function as always_inline, users would like it to be inlined
> always, it's compiled with -mcpu=power10 as pragma function target
> specifies, but the caller is compiled with -mcpu=power9 as command line
> specifies, the caller doesn't have the full ISA capability over the callee,
> it's unable to make inlining happen.

Ah right. I missed the fact that `-O1 -mcpu=power8 -flto=auto` also does
trigger the exact same error message. Sorry for the noise.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

Kewen Lin <linkw at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
     Ever confirmed|0                           |1
         Resolution|INVALID                     |---
   Last reconfirmed|                            |2023-09-12
             Status|RESOLVED                    |REOPENED

--- Comment #12 from Kewen Lin <linkw at gcc dot gnu.org> ---
(In reply to Kewen Lin from comment #9)
> It looks to me that the error message is expected, because the source code
> forces the function as always_inline, users would like it to be inlined
> always, it's compiled with -mcpu=power10 as pragma function target
> specifies, but the caller is compiled with -mcpu=power9 as command line
> specifies, the caller doesn't have the full ISA capability over the callee,
> it's unable to make inlining happen.

Sorry, I just realized that I misread the function which the pragma is applied
for (inverted the caller and the callee). It seems fine to make a caller with
-mcpu=power10 to inline a callee with -mcpu=power9, but unfortunately Power10
drops HTM capability, it causes the mismatch. The issue should be gone with an
extra -mno-htm.

This issue is similar to PR102059, whose fix already takes care of some cases
without actual htm feature uses. But it leaves inline asm alone, unfortunately
the reduced test case here is using inline asm. One idea is to enhance the fix
of PR102059 to further parse inline asm mnemonics.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

Kewen Lin <linkw at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|REOPENED                    |NEW

--- Comment #13 from Kewen Lin <linkw at gcc dot gnu.org> ---
I'd assume that most of applications don't adopt HTM features from Power8 and
Power9 at all, adding an extra option -mno-htm can be considered for them for
this special case of HTM removal in a later CPU (Power10).

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

Kewen Lin <linkw at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
           Assignee|unassigned at gcc dot gnu.org      |linkw at gcc dot gnu.org
             Status|NEW                         |ASSIGNED

--- Comment #14 from Kewen Lin <linkw at gcc dot gnu.org> ---
Both the reduced test case and the upstream code
(https://github.com/google/highway/blob/master/hwy/base.h) have empty inline
asm string, one quick simple fix can treat empty asm string specially:

diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc
index efe9adce1f8..6ef75164d08 100644
--- a/gcc/config/rs6000/rs6000.cc
+++ b/gcc/config/rs6000/rs6000.cc
@@ -25475,9 +25475,11 @@ rs6000_update_ipa_fn_target_info (unsigned int &info,
const gimple *stmt)
   /* Assume inline asm can use any instruction features.  */
   if (gimple_code (stmt) == GIMPLE_ASM)
     {
-      /* Should set any bits we concerned, for now OPTION_MASK_HTM is
-         the only bit we care about.  */
-      info |= RS6000_FN_TARGET_INFO_HTM;
+      const char *asm_str = gimple_asm_string (as_a<const gasm *> (stmt));
+      if (strlen (asm_str) > 0)
+        /* Should set any bits we concerned, for now OPTION_MASK_HTM is
+           the only bit we care about.  */
+        info |= RS6000_FN_TARGET_INFO_HTM;
       return false;
     }
   else if (gimple_code (stmt) == GIMPLE_CALL)

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #15 from Mathieu Malaterre <malat at debian dot org> ---
For some reason the no-htm flag does not seems to work in my case:

% /usr/bin/c++  -O1 -mcpu=power8 -flto=auto -c skeleton_test.cc
skeleton_test.cc: In member function 'TestFloorLog2::operator()<int,
detail::CappedTagChecker<int, 1u, 0> >(int, detail::CappedTagChecker<int, 1u,
0>)void':
skeleton_test.cc:5:44: error: inlining failed in call to 'always_inline'
'hwy::PreventElision(int)': target specific option mismatch
    5 | inline __attribute__((always_inline)) void PreventElision(int output) {
      |                                            ^~~~~~~~~~~~~~
skeleton_test.cc:36:26: note: called from here
   36 |       hwy::PreventElision(sum);
      |       ~~~~~~~~~~~~~~~~~~~^~~~~

with:

% cat skeleton_test.cc
#include <stddef.h>
#define HWY_PRAGMA(tokens) _Pragma(#tokens)
namespace hwy {
#define HWY_PUSH_ATTRIBUTES(targets_str) HWY_PRAGMA(GCC target targets_str)
inline __attribute__((always_inline)) void PreventElision(int output) {
  asm("" : "+r"(output) : : "memory");
}
} // namespace hwy
#if HWY_TARGET == HWY_PPC10
#define HWY_BEFORE_NAMESPACE() HWY_PUSH_ATTRIBUTES(",cpu=power10,no-htm")
#endif
HWY_BEFORE_NAMESPACE() namespace detail {
  template <typename, size_t, int> struct CappedTagChecker {};
}
template <typename T, size_t kLimit, int kPow2 = 0>
using CappedTag = detail::CappedTagChecker<T, kLimit, kPow2>;
template <class D> size_t Lanes(D);
template <typename, size_t, size_t kMinArg, class Test> struct ForeachCappedR {
  static void Do(size_t min_lanes, size_t max_lanes) {
    CappedTag<int, kMinArg> d;
    Test()(int(), d);
    Do(min_lanes, max_lanes);
  }
};
template <class Test> struct ForPartialVectors {
  template <typename T> void operator()(T t) {
    (void)t;
    ForeachCappedR<T, 1, 1, Test>::Do(1, 1);
  }
};
struct TestFloorLog2 {
  template <class T, class DF> void operator()(T, DF df) {
    size_t count = Lanes(df);
    int sum;
    for (size_t i; count; ++i)
      hwy::PreventElision(sum);
  }
};
void TestAllFloorLog2() { ForPartialVectors<TestFloorLog2>()(float()); }

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[malat at debian dot org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #16 from Mathieu Malaterre <malat at debian dot org> ---
Interesting, the following works for me:

% /usr/bin/c++  -O1 -mcpu=power8 -mno-htm -flto=auto -c skeleton_test.cc

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #17 from Kewen Lin <linkw at gcc dot gnu.org> ---
(In reply to Mathieu Malaterre from comment #16)
> Interesting, the following works for me:
> 
> % /usr/bin/c++  -O1 -mcpu=power8 -mno-htm -flto=auto -c skeleton_test.cc

Yeah, the suggestion on an extra option -mno-htm is for command line options,
since they are used for compiling the callee PreventElision.

It makes that:
  - the callee(PreventElision): -mcpu=power8 -mno-htm
  - caller: -mcpu=power10

caller doesn't have the htm capability (as it's compiled with p10 which drops
htm), but the callee claims it doesn't need htm with -mno-htm, so it's safe to
do the inlining.

If you want to adjust the pragma

#define HWY_BEFORE_NAMESPACE() HWY_PUSH_ATTRIBUTES(",cpu=power10")

, it should be:

#define HWY_BEFORE_NAMESPACE() HWY_PUSH_ATTRIBUTES(",cpu=power10,htm")

it pretends that the caller has the htm support, then it's also fine.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[jan.wassenberg at gmail dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #18 from Jan Wassenberg <jan.wassenberg at gmail dot com> ---
Ah, got it. We'll change the pragma to ",htm" as you suggest. Thank you :)

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #19 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The master branch has been updated by Kewen Lin <linkw@gcc.gnu.org>:

https://gcc.gnu.org/g:a65b38e361320e0aa45adbc969c704385ab1f45b

commit r14-4247-ga65b38e361320e0aa45adbc969c704385ab1f45b
Author: Kewen Lin <linkw@linux.ibm.com>
Date:   Mon Sep 25 00:28:19 2023 -0500

    rs6000: Skip empty inline asm in rs6000_update_ipa_fn_target_info
[PR111366]

    PR111366 exposes one thing that can be improved in function
    rs6000_update_ipa_fn_target_info is to skip the given empty
    inline asm string, since it's impossible to adopt any
    hardware features (so far HTM).

    Since this rs6000_update_ipa_fn_target_info related approach
    exists in GCC12 and later, the affected project highway has
    updated its target pragma with ",htm", see the link:
    https://github.com/google/highway/commit/15e63d61eb535f478bc
    I'd not bother to consider an inline asm parser for now but
    will file a separated PR for further enhancement.

            PR target/111366

    gcc/ChangeLog:

            * config/rs6000/rs6000.cc (rs6000_update_ipa_fn_target_info): Skip
            empty inline asm.

    gcc/testsuite/ChangeLog:

            * g++.target/powerpc/pr111366.C: New test.

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #20 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-13 branch has been updated by Kewen Lin <linkw@gcc.gnu.org>:

https://gcc.gnu.org/g:4d22de945430e6f8c33ef98c90a4fad8de0ef41e

commit r13-7953-g4d22de945430e6f8c33ef98c90a4fad8de0ef41e
Author: Kewen Lin <linkw@linux.ibm.com>
Date:   Sun Oct 15 21:05:40 2023 -0500

    rs6000: Skip empty inline asm in rs6000_update_ipa_fn_target_info
[PR111366]

    PR111366 exposes one thing that can be improved in function
    rs6000_update_ipa_fn_target_info is to skip the given empty
    inline asm string, since it's impossible to adopt any
    hardware features (so far HTM).

    Since this rs6000_update_ipa_fn_target_info related approach
    exists in GCC12 and later, the affected project highway has
    updated its target pragma with ",htm", see the link:
    https://github.com/google/highway/commit/15e63d61eb535f478bc
    I'd not bother to consider an inline asm parser for now but
    will file a separated PR for further enhancement.

            PR target/111366

    gcc/ChangeLog:

            * config/rs6000/rs6000.cc (rs6000_update_ipa_fn_target_info): Skip
            empty inline asm.

    gcc/testsuite/ChangeLog:

            * g++.target/powerpc/pr111366.C: New test.

    (cherry picked from commit a65b38e361320e0aa45adbc969c704385ab1f45b)

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[cvs-commit at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

--- Comment #21 from CVS Commits <cvs-commit at gcc dot gnu.org> ---
The releases/gcc-12 branch has been updated by Kewen Lin <linkw@gcc.gnu.org>:

https://gcc.gnu.org/g:2b50ddce27be4738b494f5c4dd7083099b3de68c

commit r12-9925-g2b50ddce27be4738b494f5c4dd7083099b3de68c
Author: Kewen Lin <linkw@linux.ibm.com>
Date:   Sun Oct 15 21:08:26 2023 -0500

    rs6000: Skip empty inline asm in rs6000_update_ipa_fn_target_info
[PR111366]

    PR111366 exposes one thing that can be improved in function
    rs6000_update_ipa_fn_target_info is to skip the given empty
    inline asm string, since it's impossible to adopt any
    hardware features (so far HTM).

    Since this rs6000_update_ipa_fn_target_info related approach
    exists in GCC12 and later, the affected project highway has
    updated its target pragma with ",htm", see the link:
    https://github.com/google/highway/commit/15e63d61eb535f478bc
    I'd not bother to consider an inline asm parser for now but
    will file a separated PR for further enhancement.

            PR target/111366

    gcc/ChangeLog:

            * config/rs6000/rs6000.cc (rs6000_update_ipa_fn_target_info): Skip
            empty inline asm.

    gcc/testsuite/ChangeLog:

            * g++.target/powerpc/pr111366.C: New test.

    (cherry picked from commit a65b38e361320e0aa45adbc969c704385ab1f45b)

[Bug target/111366] error: inlining failed in call to 'always_inline' 'hwy::PreventElision<int&>(int&)void': target specific option mismatch

[linkw at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=111366

Kewen Lin <linkw at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|ASSIGNED                    |RESOLVED
         Resolution|---                         |FIXED

--- Comment #22 from Kewen Lin <linkw at gcc dot gnu.org> ---
Should be fixed on trunk and gcc-{12,13}.