https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121310
Bug ID: 121310
Summary: [16 regression] highway miscompiled on arm32
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: wrong-code
Severity: normal
Priority: P3
Component: target
Assignee: unassigned at gcc dot gnu.org
Reporter: sjames at gcc dot gnu.org
Target Milestone: ---
Created attachment 62002
--> https://gcc.gnu.org/bugzilla/attachment.cgi?id=62002&action=edit
reduced.cxx
This is the arm issue I mentioned at
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=121218#c2.
Building highway on armv7a with `-O3 -march=armv7-a -mfpu=vfpv3-d16
-mfloat-abi=hard` results in its
HwyConvertTestGroup/HwyConvertTest.TestAllPromoteOddEvenTo/EMU128 test failing.
I am not sure if the reduction is fully(?) representative because this one only
fails on trunk but the original fails at least with 15, maybe further back too
(need to check).
```
$ g++ -O3 -march=armv7-a -mfpu=vfpv3-d16 -mfloat-abi=hard reduced.cxx -o
reduced
reduced.cxx:176:3: warning: access declarations are deprecated in favor of
using-declarations; suggestion: add the ‘using’ keyword [-Wdeprecated]
176 | __uniq_ptr_impl::__uniq_ptr_impl;
| ^~~~~~~~~~~~~~~
reduced.cxx:157:1: warning: ‘always_inline’ function might not be inlinable
unless also declared ‘inline’ [-Wattributes]
157 | WidenMulAccumulate(VFromD<Simd<signed char, 16, 0>> mul) {
| ^~~~~~~~~~~~~~~~~~
$ ./reduced
Aborted (core dumped) ./reduced
```
```
Using built-in specs.
COLLECT_GCC=gcc
COLLECT_LTO_WRAPPER=/usr/libexec/gcc/armv7a-unknown-linux-gnueabihf/16/lto-wrapper
Target: armv7a-unknown-linux-gnueabihf
Configured with:
/var/tmp/portage/sys-devel/gcc-16.0.9999/work/gcc-16.0.9999/configure
--host=armv7a-unknown-linux-gnueabihf --build=armv7a-unknown-linux-gnueabihf
--prefix=/usr --bindi'
Thread model: posix
Supported LTO compression algorithms: zlib zstd
gcc version 16.0.0 20250719 (experimental)
e42799be2bb2966487c27897294426a03a99f56d (Gentoo 16.0.9999 p, commit
37758405f27f9ebf31a0e59044da0c0aa9eda679)
```
