BLD: Avoid "visibility attribute not supported" warning #16288

embray · 2020-05-18T15:55:43Z

I found that when building the latest master branch on Cygwin, while testing #16246, that thousands of warnings were generated at build time like:

numpy/core/src/npysort/binsearch.c.src: In function ‘binsearch_left_bool’:
numpy/core/src/npysort/binsearch.c.src:82:1: warning: visibility attribute not supported in this configuration; ignored [-Wattributes]

Granted this is just a warning, so I don't think it's a serious issue.

It seems the test that was supposed to check for __attribute__ support was not working as expected. The #pragmas only take effect if I provide a function body--they are ignored for bare declarations. I don't know if that's by intent, or if it's a GCC issue. For reference:

$ gcc --version
gcc (GCC) 7.4.0

embray · 2020-05-18T16:11:11Z

Now getting errors like

numpy/core/src/umath/simd.inc.src: In function ‘fma_scalef_ps’:
  numpy/core/src/umath/simd.inc.src:1519:1: warning: AVX vector return without AVX enabled changes the ABI [-Wpsabi]
   {
   ^
  numpy/core/src/umath/simd.inc.src:1518:1: note: The ABI for passing parameters with 32-byte alignment has changed in GCC 4.6
   fma_scalef_ps(__m256 poly, __m256 quadrant)
   ^~~~~~~~~~~~~
  In file included from /usr/lib/gcc/x86_64-linux-gnu/7/include/immintrin.h:41:0,
                   from numpy/core/src/umath/simd.inc.src:25,
                   from numpy/core/src/umath/loops.c.src:50:
  /usr/lib/gcc/x86_64-linux-gnu/7/include/avxintrin.h:1195:1: error: inlining failed in call to always_inline ‘_mm256_movemask_ps’: target specific option mismatch
   _mm256_movemask_ps (__m256 __A)
   ^~~~~~~~~~~~~~~~~~

I was getting this locally as well but it wasn't clear at first that this was related. Seems maybe it is.

embray · 2020-05-18T16:13:20Z

Oh, I made a stupid bug in this commit. Rebasing again...

anirudh2290 · 2020-05-18T20:38:37Z

numpy/distutils/command/autodist.py

@@ -69,7 +69,10 @@ def check_gcc_function_attribute(cmd, attribute, name):
        #pragma GCC diagnostic error "-Wattributes"
        #pragma clang diagnostic error "-Wattributes"

-        int %s %s(void*);
+        int %s %s(void*)


FWIW I am not able to reproduce this on 7.3.0 or gcc-9 versions. but if this function is not behaving as its supposed to for certain compiler versions, its certainly worth fixing.

that's odd, This build error was perfectly reproducible on bionic with gcc-7.5 and gcc-9.2.

@r-devulap , not sure whats going on. my gcc version for gcc-9 is gcc 9.3.0

I was going to say, it could be Windows-specific. But maybe not.

seberg · 2020-05-19T13:54:35Z

The change looks simple, but there are universal test failures along the lines of. @r-devulap since those failures are AVX related, a ping just in case you know right away what the cause is.

``` numpy/core/src/umath/simd.inc.src: In function ‘fma_scalef_ps’: 4221 numpy/core/src/umath/simd.inc.src:1519:1: warning: AVX vector return without AVX enabled changes the ABI [-Wpsabi] 4222 { 4223 ^ 4224 numpy/core/src/umath/simd.inc.src:1518:1: note: The ABI for passing parameters with 32-byte alignment has changed in GCC 4.6 4225 fma_scalef_ps(__m256 poly, __m256 quadrant) 4226 ^~~~~~~~~~~~~ 4227 In file included from /usr/lib/gcc/x86_64-linux-gnu/7/include/immintrin.h:41:0, 4228 from numpy/core/src/umath/simd.inc.src:25, 4229 from numpy/core/src/umath/loops.c.src:50: 4230 /usr/lib/gcc/x86_64-linux-gnu/7/include/avxintrin.h:1195:1: error: inlining failed in call to always_inline ‘_mm256_movemask_ps’: target specific option mismatch 4231 _mm256_movemask_ps (__m256 __A) 4232 ^~~~~~~~~~~~~~~~~~ 4233 In file included from numpy/core/src/umath/loops.c.src:50:0: 4234 numpy/core/src/umath/simd.inc.src:1533:10: note: called from here 4235 if (_mm256_movemask_ps(denormal_mask) != 0x0000) { 4236 ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 4237 In file included from /usr/lib/gcc/x86_64-linux-gnu/7/include/immintrin.h:41:0, 4238 from numpy/core/src/umath/simd.inc.src:25, 4239 from numpy/core/src/umath/loops.c.src:50: 4240 /usr/lib/gcc/x86_64-linux-gnu/7/include/avxintrin.h:404:1: error: inlining failed in call to always_inline ‘_mm256_cmp_ps’: target specific option mismatch 4241 ```

r-devulap · 2020-05-19T16:41:32Z

You cannot omit parameters names in function definitions in C. This will fix your problem:

-        int %s %s(void*)
+        int %s %s(void* temp)

seberg · 2020-05-19T17:14:02Z

Thanks, shoudl have seen, I guess @embray actually may just have forgotten to push the rebase...

embray · 2020-05-20T10:39:09Z

That's possible. When I said I was rebasing it was to fix the missing parameter name. Previously I had done it properly and it worked, so I think I accidently pushed an older version. I'll fix it in a sec.

One thing I was kinda wondering is if there was a specific reason this test function even needed to take a dummy parameter in the first place. I couldn't find any reference in the commit history as to why that was needed for this test function.

I did also find it odd that my GCC would ignore the unknown attribute on a bare declaration, but not in a function definition. I don't know if that's intentional or a GCC bug, but I couldn't find any reference to that either. I guess it doesn't bother processing the attributes unless there's an actual code module to compile.

embray · 2020-05-20T10:41:10Z

And yes, because of the missing parameter it means all attribute tests fail, and I think that was causing the AVX-related bugs. They go away when I fix it (instead I get #16290, but that's unrelated).

Bumps [hypothesis](https://github.com/HypothesisWorks/hypothesis) from 5.12.0 to 5.14.0. - [Release notes](https://github.com/HypothesisWorks/hypothesis/releases) - [Commits](HypothesisWorks/hypothesis@hypothesis-python-5.12.0...hypothesis-python-5.14.0) Signed-off-by: dependabot-preview[bot] <[email protected]>

on systems that do not support symbol visibility attributes

seberg · 2020-05-20T13:59:35Z

Weird that it shows that dependabot commit, nvm will just squash merge though. I am going to put this in, looks like it just cannot hurt to me. Thanks @embray!

embray · 2020-05-20T15:20:19Z

@seberg Thanks!

I found that when building the latest master branch on Cygwin, while testing numpy#16246, that thousands of warnings were generated at build time like: numpy/core/src/npysort/binsearch.c.src: In function ‘binsearch_left_bool’: numpy/core/src/npysort/binsearch.c.src:82:1: warning: visibility attribute not supported in this configuration; ignored [-Wattributes] Granted this is just a warning, so I don't think it's a serious issue. It seems the test that was supposed to check for __attribute__ support was not working as expected. The #pragmas only take effect if I provide a function body--they are ignored for bare declarations. I don't know if that's by intent, or if it's a GCC issue. For reference: $ gcc --version gcc (GCC) 7.4.0

embray force-pushed the cygwin/visibility-attribute-ignored branch from 08713c6 to 7f7d19d Compare May 18, 2020 15:56

anirudh2290 reviewed May 18, 2020

View reviewed changes

charris changed the title ~~[BUG] avoid "visibility attribute not supported" warning~~ BUG: Avoid "visibility attribute not supported" warning May 18, 2020

charris added 03 - Maintenance component: numpy.distutils labels May 18, 2020

dependabot-preview bot and others added 2 commits May 20, 2020 13:20

BUG: avoid "visibility attribute not supported" warning when building

4ca7de9

on systems that do not support symbol visibility attributes

embray force-pushed the cygwin/visibility-attribute-ignored branch from 7f7d19d to 4ca7de9 Compare May 20, 2020 11:20

seberg merged commit dcf1614 into numpy:master May 20, 2020

seberg changed the title ~~BUG: Avoid "visibility attribute not supported" warning~~ BLD: Avoid "visibility attribute not supported" warning May 20, 2020

embray deleted the cygwin/visibility-attribute-ignored branch May 20, 2020 15:20

charris added the 09 - Backport-Candidate PRs tagged should be backported label May 20, 2020

charris added this to the 1.19.0 release milestone May 20, 2020

charris mentioned this pull request May 22, 2020

BLD: Avoid "visibility attribute not supported" warning. #16343

Merged

charris removed the 09 - Backport-Candidate PRs tagged should be backported label May 22, 2020

charris removed this from the 1.19.0 release milestone May 22, 2020

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

BLD: Avoid "visibility attribute not supported" warning #16288

BLD: Avoid "visibility attribute not supported" warning #16288

Uh oh!

embray commented May 18, 2020

Uh oh!

embray commented May 18, 2020

Uh oh!

embray commented May 18, 2020

Uh oh!

anirudh2290 May 18, 2020

Uh oh!

r-devulap May 19, 2020

Uh oh!

anirudh2290 May 19, 2020

Uh oh!

embray May 28, 2020

Uh oh!

seberg commented May 19, 2020

Uh oh!

r-devulap commented May 19, 2020

Uh oh!

seberg commented May 19, 2020

Uh oh!

embray commented May 20, 2020

Uh oh!

embray commented May 20, 2020

Uh oh!

seberg commented May 20, 2020

Uh oh!

embray commented May 20, 2020

Uh oh!

Uh oh!

Uh oh!

BLD: Avoid "visibility attribute not supported" warning #16288

BLD: Avoid "visibility attribute not supported" warning #16288

Uh oh!

Conversation

embray commented May 18, 2020

Uh oh!

embray commented May 18, 2020

Uh oh!

embray commented May 18, 2020

Uh oh!

anirudh2290 May 18, 2020

Choose a reason for hiding this comment

Uh oh!

r-devulap May 19, 2020

Choose a reason for hiding this comment

Uh oh!

anirudh2290 May 19, 2020

Choose a reason for hiding this comment

Uh oh!

embray May 28, 2020

Choose a reason for hiding this comment

Uh oh!

seberg commented May 19, 2020

Uh oh!

r-devulap commented May 19, 2020

Uh oh!

seberg commented May 19, 2020

Uh oh!

embray commented May 20, 2020

Uh oh!

embray commented May 20, 2020

Uh oh!

seberg commented May 20, 2020

Uh oh!

embray commented May 20, 2020

Uh oh!

Uh oh!