opencv: Fails to build with GCC 12 (pre-release) on ppc64le

System information (version)

OpenCV => 4.5.5
Operating System / Platform => Fedora Linux Rawhide (development version), ppc64le (64-bit PowerPC, little-endian)
Compiler => GCC 12.0.0 (pre-release)

Detailed description

GCC was just updated to a pre-release of 12.0.0 in Fedora Rawhide (the development version of Fedora Linux), and OpenCV and packages that use it are failing to compile on ppc64le due to a problem with AltiVec/VSX intrinsics.

The downstream bug report is here.

I don’t know if this is an OpenCV bug or a GCC 12 bug.

In file included from /builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/base.hpp:661,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core.hpp:53,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/utility.hpp:56,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/src/precomp.hpp:53,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/src/alloc.cpp:43:
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_double2 vec_cvf(const vec_float4&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:466:22: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  466 | { return fn2(vec_sldw(a, a, 1)); }
      |                      ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:468:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_4_2'
  468 | VSX_IMPL_CONV_EVEN_4_2(vec_double2, vec_float4, vec_cvf,  vec_cvfo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_dword2 vec_ctsl(const vec_float4&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:466:22: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  466 | { return fn2(vec_sldw(a, a, 1)); }
      |                      ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:472:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_4_2'
  472 | VSX_IMPL_CONV_EVEN_4_2(vec_dword2,  vec_float4, vec_ctsl, vec_ctslo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_udword2 vec_ctul(const vec_float4&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:466:22: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  466 | { return fn2(vec_sldw(a, a, 1)); }
      |                      ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:473:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_4_2'
  473 | VSX_IMPL_CONV_EVEN_4_2(vec_udword2, vec_float4, vec_ctul, vec_ctulo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_float4 vec_cvf(const vec_double2&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:479:20: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  479 |     return vec_sldw(v4, v4, 3);                  \
      |                    ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:482:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_2_4'
  482 | VSX_IMPL_CONV_EVEN_2_4(vec_float4, vec_double2, vec_cvf, vec_cvfo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_float4 vec_ctf(const vec_dword2&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:479:20: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  479 |     return vec_sldw(v4, v4, 3);                  \
      |                    ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:483:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_2_4'
  483 | VSX_IMPL_CONV_EVEN_2_4(vec_float4, vec_dword2,  vec_ctf, vec_ctfo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_float4 vec_ctf(const vec_udword2&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:479:20: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  479 |     return vec_sldw(v4, v4, 3);                  \
      |                    ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:484:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_2_4'
  484 | VSX_IMPL_CONV_EVEN_2_4(vec_float4, vec_udword2, vec_ctf, vec_ctfo)
      | ^~~~~~~~~~~~~~~~~~~~~~
In file included from /builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/base.hpp:661,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core.hpp:53,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/utility.hpp:56,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/src/precomp.hpp:53,
                 from /builddir/build/BUILD/opencv-4.5.5/modules/core/src/algorithm.cpp:43:
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_double2 vec_cvf(const vec_float4&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:466:22: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  466 | { return fn2(vec_sldw(a, a, 1)); }
      |                      ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:468:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_4_2'
  468 | VSX_IMPL_CONV_EVEN_4_2(vec_double2, vec_float4, vec_cvf,  vec_cvfo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_dword2 vec_ctsl(const vec_float4&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:466:22: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  466 | { return fn2(vec_sldw(a, a, 1)); }
      |                      ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:472:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_4_2'
  472 | VSX_IMPL_CONV_EVEN_4_2(vec_dword2,  vec_float4, vec_ctsl, vec_ctslo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_udword2 vec_ctul(const vec_float4&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:466:22: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  466 | { return fn2(vec_sldw(a, a, 1)); }
      |                      ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:473:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_4_2'
  473 | VSX_IMPL_CONV_EVEN_4_2(vec_udword2, vec_float4, vec_ctul, vec_ctulo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_float4 vec_cvf(const vec_double2&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:479:20: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  479 |     return vec_sldw(v4, v4, 3);                  \
      |                    ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:482:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_2_4'
  482 | VSX_IMPL_CONV_EVEN_2_4(vec_float4, vec_double2, vec_cvf, vec_cvfo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_float4 vec_ctf(const vec_dword2&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:479:20: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  479 |     return vec_sldw(v4, v4, 3);                  \
      |                    ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:483:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_2_4'
  483 | VSX_IMPL_CONV_EVEN_2_4(vec_float4, vec_dword2,  vec_ctf, vec_ctfo)
      | ^~~~~~~~~~~~~~~~~~~~~~
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp: In function 'vec_float4 vec_ctf(const vec_udword2&)':
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:479:20: error: invalid parameter combination for AltiVec intrinsic '__builtin_vec_sldw'
  479 |     return vec_sldw(v4, v4, 3);                  \
      |                    ^
/builddir/build/BUILD/opencv-4.5.5/modules/core/include/opencv2/core/vsx_utils.hpp:484:1: note: in expansion of macro 'VSX_IMPL_CONV_EVEN_2_4'
  484 | VSX_IMPL_CONV_EVEN_2_4(vec_float4, vec_udword2, vec_ctf, vec_ctfo)
      | ^~~~~~~~~~~~~~~~~~~~~~

Steps to reproduce

Compile with GCC 12 on ppc64le.

Issue submission checklist

I report the issue, it’s not a question
I checked the problem with documentation, FAQ, open issues, forum.opencv.org, Stack Overflow, etc and have not found solution
I updated to latest OpenCV version and the issue is still there
There is reproducer code and related data files: videos, images, onnx, etc N/A

About this issue

Original URL
State: closed
Created 2 years ago
Comments: 18 (13 by maintainers)

Most upvoted comments

with patch applied it built on Fedora Rawhide https://koji.fedoraproject.org/koji/taskinfo?taskID=82845067

for F35 (GCC 11) also built https://koji.fedoraproject.org/koji/taskinfo?taskID=82846659

sergiomb2 on Feb 15, 2022

@alalek Your patch works fine with the Power8 compilation flag and using GCC12. I was able to compile my whole setup.

ChipKerchner on Feb 15, 2022

@alalek,

Could someone try this updated workaround with GCC12?

Lauched some for native-ppc64le: gcc8, gcc11, gcc12 https://copr.fedorainfracloud.org/coprs/rezso/ML/build/3501854
Stopped after a while so it may appear as ‘failed’ due to stop but it passed as described below on ppc64le.

I added PR #21614 to check on CI cross-compilation with GCC-8 (Ubuntu 18.04 x86_64).

It seems on gcc12 (also other gcc) #21614 just passed over src/algorithm.cpp:

https://download.copr.fedorainfracloud.org/results/rezso/ML/fedora-rawhide-ppc64le/03501854-opencv/builder-live.log.gz

+ git clone --depth 1 -n -b next https://github.com/opencv/opencv.git .
Cloning into '.'...
+ git fetch --depth 1 origin 4ba2b05df8d6deb71cef92477989173b2c1ca98a

{...}

+ echo 'Patch #1 (21614.patch):'
Patch #1 (21614.patch):
+ /usr/bin/patch --no-backup-if-mismatch -f -p1 --fuzz=0
patching file modules/core/include/opencv2/core/vsx_utils.hpp

{...}

[  2%] Building CXX object modules/core/CMakeFiles/opencv_core.dir/src/algorithm.cpp.o
[  2%] Building CXX object modules/core/CMakeFiles/opencv_core.dir/src/alloc.cpp.o
[  2%] Building CXX object modules/core/CMakeFiles/opencv_core.dir/src/arithm.cpp.o

cbalint13 on Feb 15, 2022

_ARCH_PWR9 can be used to test if the compiler is invoked to target Power9, so why not change the use of __builtin_altivec_adub to check for the presence of that macro and fall back to the alternate implementation on Power8?

edelsohn on Feb 14, 2022