Created attachment 1714068 [details]
patch against git.centos.org gcc repo

Description of problem:

Pytorch can not be built on AArch64 in CentOS 8. Build fails with:

/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h:235:7: error: ‘vst1q_f32_x2’ was not declared in this scope

I filled bug against pytorch [1] and with help of Sebastian Pop did a fix [2]. But proper solution would be fixing it in gcc by backporting two patches.

1. https://github.com/pytorch/pytorch/issues/44198
2. https://github.com/pytorch/pytorch/pull/44199


Version-Release number of selected component (if applicable):

8.3.1-5.el8.0.2

How reproducible:

always

Steps to Reproduce:
1. Clone https://github.com/pytorch/pytorch
2. cd pytorch
3. USE_CUDA=0 BUILD_CAFFE2_OPS=0 USE_DISTRIBUTED=0 USE_QNNPACK=0 USE_XNNPACK=0 python3 setup.py install

Actual results:

In file included from /root/pytorch/aten/src/ATen/cpu/vec256/vec256.h:10,
                 from /root/pytorch/aten/src/ATen/native/cpu/Loops.h:35,
                 from /root/pytorch/aten/src/ATen/native/Normalization.cpp:10:
/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h:262:3: warning: type qualifiers ignored on function return type [-Wignored-qualifiers]
   const float operator[](int idx) const {
   ^~~~~
/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h:267:3: warning: type qualifiers ignored on function return type [-Wignored-qualifiers]
   const float operator[](int idx) {
   ^~~~~
/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h: In member function ‘void at::vec256::{anonymous}::Vec256<float>::store(void*, int64_t) const’:
/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h:235:7: error: ‘vst1q_f32_x2’ was not declared in this scope
       vst1q_f32_x2(reinterpret_cast<float*>(ptr), values);
       ^~~~~~~~~~~~
/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h:235:7: note: suggested alternative: ‘vld1q_f32_x2’
       vst1q_f32_x2(reinterpret_cast<float*>(ptr), values);
       ^~~~~~~~~~~~
       vld1q_f32_x2
/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h:242:7: error: ‘vst1q_f32_x2’ was not declared in this scope
       vst1q_f32_x2(reinterpret_cast<float*>(tmp_values), values);
       ^~~~~~~~~~~~
/root/pytorch/aten/src/ATen/cpu/vec256/vec256_float_neon.h:242:7: note: suggested alternative: ‘vld1q_f32_x2’
       vst1q_f32_x2(reinterpret_cast<float*>(tmp_values), values);
       ^~~~~~~~~~~~
       vld1q_f32_x2
gmake[2]: *** [caffe2/CMakeFiles/torch_cpu.dir/build.make:2014: caffe2/CMakeFiles/torch_cpu.dir/__/aten/src/ATen/native/Normalization.cpp.o] Error 1
gmake[1]: *** [CMakeFiles/Makefile2:6058: caffe2/CMakeFiles/torch_cpu.dir/all] Error 2
gmake: *** [Makefile:141: all] Error 2

Expected results:

pytorch builds

Additional info: