ARM NEON: fix arm64 gcc11 build excess elements in vector failure #1216
Conversation
There was a problem hiding this comment.
Thank you @Qingwu-Li ; can you address the test failures? https://github.com/simd-everywhere/simde/actions/runs/10774988233/job/29878397301?pr=1216#step:10:1023
mr-c
force-pushed
the
arm64-gcc11-builtin-shuffle
branch
from
1097884 to
5377444
Compare
|
I can confirm this PR does fix some errors: https://github.com/simd-everywhere/simde/actions/runs/10775769592/job/29880998011?pr=1216#step:9:1271 (log of a gcc-11 aarch64 build without these fixes) |
Qingwu-Li
force-pushed
the
arm64-gcc11-builtin-shuffle
branch
from
5377444 to
53e13c9
Compare
This is another issue related to gcc 11. While I'm new to SIMD and not fully confident in the fix, it provides a temporary solution to the problem. |
Thanks for the fix! Is there a bug about |
Qingwu-Li
force-pushed
the
arm64-gcc11-builtin-shuffle
branch
from
53e13c9 to
84d614c
Compare
|
Yay, it worked! It would be nice to update the reference to include a specific GCC bug number before merging |
mr-c
force-pushed
the
arm64-gcc11-builtin-shuffle
branch
from
84d614c to
63db247
Compare
Didn't found the related bug number , maybe fixed in this commit in releases/gcc-11.5.0，but not verifyed. |
Okay, that seems to come from https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114521 |
mr-c
force-pushed
the
arm64-gcc11-builtin-shuffle
branch
from
63db247 to
84d5515
Compare
mr-c
force-pushed
the
arm64-gcc11-builtin-shuffle
branch
from
84d5515 to
eeba351
Compare
Resolved a build failure on ARM64 with GCC 11 caused by excess elements in a vector initializer. The issue stemmed from the `__builtin_shuffle` function, where the number of elements in the argument vector(s) and the mask vector must match. see [1]. [1]: simd-everywhere/simde#1216 Signed-off-by: LI Qingwu <[email protected]>
Resolved a build failure on ARM64 with GCC 11 caused by excess elements in a vector initializer. The issue stemmed from the `__builtin_shuffle` function, where the number of elements in the argument vector(s) and the mask vector must match. see [1]. [1]: simd-everywhere/simde#1216 Signed-off-by: LI Qingwu <[email protected]>
Resolved a build failure on ARM64 with GCC 11 caused by excess elements in a vector initializer. The issue stemmed from the `__builtin_shuffle` function, where the number of elements in the argument vector(s) and the mask vector must match. For more details, refer to issue simd-everywhere#1211. Signed-off-by: LI Qingwu <[email protected]>
The vst1_*_x4 built-in functions in GCC 11 produce incorrect results,
as following exanple, this patch circumvents the issue by avoiding these
functions in GCC 11 and earlier versions.
float32x2x4_t val = {{
{0.10f, 0.20f},
{0.30f, 0.40f},
{0.50f, 0.60f},
{0.70f, 0.80f}
}};
float result[16];
vst1_f32_x4(result, val);
get result = {0.100000 0.200000 0.500000 0.600000 0.000000 0.000000 0.000000 0.000000}
Signed-off-by: LI Qingwu <[email protected]>
mr-c
force-pushed
the
arm64-gcc11-builtin-shuffle
branch
from
eeba351 to
78182f2
Compare
|
Thank you @Qingwu-Li ! |
Resolved a build failure on ARM64 with GCC 11 caused by excess elements in a vector initializer. The issue stemmed from the
__builtin_shufflefunction, where the number of elements in the argument vector(s) and the mask vector must match.For more details, refer to issue #1211.