vc 1.4.3 build failure

[chenrui333]

👋 trying to build the latest release, but run into some build issue. The error log is as below:

build error

  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/mask.h:37:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/detail.h:1182:64: error: always_inline function '_mm256_testz_ps' requires target feature 'avx', but would be inlined into function 'testz' that is compiled without support for 'avx'
  Vc_INTRINSIC Vc_CONST int testz(__m256  a, __m256  b) { return _mm256_testz_ps(a, b); }
                                                                 ^
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/detail.h:1182:64: error: AVX vector argument of type '__m256' (vector of 8 'float' values) without 'avx' enabled changes the ABI
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/detail.h:651:55: error: always_inline function '_mm256_and_ps' requires target feature 'avx', but would be inlined into function 'and_' that is compiled without support for 'avx'
  Vc_INTRINSIC __m256 and_(__m256 a, __m256 b) { return _mm256_and_ps(a, b); }
                                                        ^
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/detail.h:651:55: error: AVX vector argument of type '__m256' (vector of 8 'float' values) without 'avx' enabled changes the ABI
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/trigonometric_AVX2+FMA+BMI2.cpp:31:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/vector.h:35:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../sse/vector.h:32:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../sse/intrinsics.h:37:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/../common/storage.h:35:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../sse/../avx/intrinsics.h:122:65: error: always_inline function '_mm256_broadcast_ss' requires target feature 'avx', but would be inlined into function 'setabsmask_ps' that is compiled without support for 'avx'
      static Vc_INTRINSIC m256  Vc_CONST setabsmask_ps() { return _mm256_broadcast_ss(reinterpret_cast<const float *>(&c_general::absMaskFloat[1])); }
                                                                  ^
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../sse/../avx/intrinsics.h:122:65: error: AVX vector return of type '__m256' (vector of 8 'float' values) without 'avx' enabled changes the ABI
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/trigonometric_AVX2+FMA+BMI2.cpp:31:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/vector.h:38:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:542:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/vector.tcc:310:14: error: always_inline function '_mm256_or_ps' requires target feature 'avx', but would be inlined into function 'setQnan' that is compiled without support for 'avx'
      data() = _mm256_or_ps(data(), k.dataF());
               ^
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/vector.tcc:310:14: error: AVX vector argument of type '__m256' (vector of 8 'float' values) without 'avx' enabled changes the ABI
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/trigonometric_AVX2+FMA+BMI2.cpp:31:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/vector.h:38:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:36:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/mask.h:37:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/scalar/../common/../avx/detail.h:652:58: error: always_inline function '_mm256_and_pd' requires target feature 'avx', but would be inlined into function 'and_' that is compiled without support for 'avx'
  Vc_INTRINSIC __m256d and_(__m256d a, __m256d b) { return _mm256_and_pd(a, b); }
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:31:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/intrinsics.h:143:89: error: '__builtin_ia32_vinsertf128_ps256' needs target feature avx
      template <int offset> Vc_INTRINSIC __m256  insert128(__m256  a, __m128  b) { return _mm256_insertf128_ps(a, b, offset); }
                                                                                          ^
  /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/lib/clang/13.1.6/include/avxintrin.h:4595:11: note: expanded from macro '_mm256_insertf128_ps'
    (__m256)__builtin_ia32_vinsertf128_ps256((__v8sf)(__m256)(V1), \
            ^
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/avx_sorthelper_AVX2+FMA+BMI2.cpp:28:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:32:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/casts.h:69:66: error: always_inline function '_mm256_castps128_ps256' requires target feature 'avx', but would be inlined into function 'avx_cast' that is compiled without support for 'avx'
      template<> Vc_INTRINSIC __m256  avx_cast(__m128  v) { return _mm256_castps128_ps256(v); }
                                                                   ^
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/casts.h:69:66: error: AVX vector return of type '__m256' (vector of 8 'float' values) without 'avx' enabled changes the ABI
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/casts.h:98:66: error: always_inline function '_mm256_castpd256_pd128' requires target feature 'avx', but would be inlined into function 'avx_cast' that is compiled without support for 'avx'
      template<> Vc_INTRINSIC __m128d avx_cast(__m256d v) { return _mm256_castpd256_pd128(v); }
                                                                   ^
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/casts.h:98:66: error: AVX vector argument of type '__m256d' (vector of 4 'double' values) without 'avx' enabled changes the ABI
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/avx_sorthelper_AVX2+FMA+BMI2.cpp:28:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:31:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/intrinsics.h:154:79: error: '__builtin_ia32_vextractf128_pd256' needs target feature avx
      template <int offset> Vc_INTRINSIC __m128d extract128(__m256d a) { return _mm256_extractf128_pd(a, offset); }
                                                                                ^
  /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/lib/clang/13.1.6/include/avxintrin.h:4725:12: note: expanded from macro '_mm256_extractf128_pd'
    (__m128d)__builtin_ia32_vextractf128_pd256((__v4df)(__m256d)(V), (int)(M))
             ^
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/avx_sorthelper_AVX2+FMA+BMI2.cpp:28:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:31:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/intrinsics.h:144:89: error: '__builtin_ia32_vinsertf128_pd256' needs target feature avx
      template <int offset> Vc_INTRINSIC __m256d insert128(__m256d a, __m128d b) { return _mm256_insertf128_pd(a, b, offset); }
                                                                                          ^
  /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/lib/clang/13.1.6/include/avxintrin.h:4633:12: note: expanded from macro '_mm256_insertf128_pd'
    (__m256d)__builtin_ia32_vinsertf128_pd256((__v4df)(__m256d)(V1), \
             ^
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/avx_sorthelper_AVX2+FMA+BMI2.cpp:28:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:32:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/casts.h:77:66: error: always_inline function '_mm256_castpd128_pd256' requires target feature 'avx', but would be inlined into function 'avx_cast' that is compiled without support for 'avx'
      template<> Vc_INTRINSIC __m256d avx_cast(__m128d v) { return _mm256_castpd128_pd256(v); }
                                                                   ^
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/casts.h:77:66: error: AVX vector return of type '__m256d' (vector of 4 'double' values) without 'avx' enabled changes the ABI
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/build/avx_sorthelper_AVX2+FMA+BMI2.cpp:28:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/vector.h:32:
  In file included from /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/../traits/../common/../sse/../avx/casts.h:34:
  /tmp/vc-20220529-32959-1ggh4jf/Vc-1.4.3/Vc/avx/shuffle.h:293:20: error: '__builtin_ia32_vpermilps' needs target feature avx
              return _mm_permute_ps(x, Dst0 + Dst1 * 4 + Dst2 * 16 + Dst3 * 64);
                     ^
  /Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/lib/clang/13.1.6/include/avxintrin.h:1088:11: note: expanded from macro '_mm_permute_ps'
    (__m128)__builtin_ia32_vpermilps((__v4sf)(__m128)(A), (int)(C))
            ^
  fatal error: too many errors emitted, stopping now [-ferror-limit=]
  20 errors generated.

full build log, https://github.com/Homebrew/homebrew-core/runs/6645127976
relates to Homebrew/homebrew-core#102079

[bernhardmgruber]

Hi! Vc is now in maintenance mode and no longer actively developed. However, if you can provide us with a pull requests we an try to get it in.

[chenrui333]

Got it, thanks!

[dennisklein]

I just encountered the same issue. Turns out brew's compiler wrapper script filters out any -march flags now, unless the formula opts in to not remove it. See Homebrew/brew@0404da7.

Adding this to the beginning of the install step could successfully compile Vc 1.4.3 (FairRootGroup/homebrew-fairsoft@718bf92):

def install
  ENV.runtime_cpu_detection
  #...
end

[chenrui333]

I just encountered the same issue. Turns out brew's compiler wrapper script filters out any -march flags now, unless the formula opts in to not remove it. See Homebrew/brew@0404da7.

Adding this to the beginning of the install step could successfully compile Vc 1.4.3 (FairRootGroup/homebrew-fairsoft@718bf92):

def install
  ENV.runtime_cpu_detection
  #...
end

Gonna give ENV.runtime_cpu_detection a shot

[chenrui333]

Got some audit failure, but the build is successful

  vc:
    * No `cpuid` instruction detected. vc should not use `ENV.runtime_cpu_detection`.

[dennisklein]

Got some audit failure, but the build is successful

  vc:
    * No `cpuid` instruction detected. vc should not use `ENV.runtime_cpu_detection`.

That seems to be a shortcoming of brew's Formula linter currently, which only considers dylibs and executables. However, Vc only emits a libVc.a file.

[dennisklein]

I guess, if you teach this code to also include libVc.a in the set of checked files, the audit will succeed: https://github.com/Homebrew/brew/blob/36d14ae49cc4cd410b0b77e7d4f01e1836fd3be8/Library/Homebrew/formula_cellar_checks.rb#L312-L316

[chenrui333]

I see 😄 I will look into more this weekend