Jianyu Huang

@mnicely Thanks for checking. We are mostly interested in dense for the 1st step.

Hi @gregrodgers , do you have more details on clang version and repro instruction? We were able to build with clang 9.

Do you have more details on the need of conanfile.py ? We don't have it currently.

Thanks for reporting! Do you have more contexts and repros? I saw the failure is on `dev-libs/FBGEMM-2022.01.13`. Wonder if the issue still exists with the latest source build, with asmjit...

Which compiler version did you use and do you have more detailed repro? Thanks!

https://github.com/pytorch/FBGEMM/issues/1070 is closed, with the latest asmjit upgraded in the trunk. Please feel free to reopen the issue if the asmjit Clang 14 Build failure still exists.

cc @mjanderson09 This specific error is only for the GEMM benchmark not for the library (where we use `((volatile char*)(llc.data()));` to flush cache). Is it possible to comment it out...

Right: FBGEMM_BUILD_BENCHMARKS is disabled for PyTorch build. https://github.com/pytorch/FBGEMM/pull/1202 , https://github.com/pytorch/pytorch/pull/82396, https://github.com/pytorch/pytorch/pull/82676, https://github.com/pytorch/FBGEMM/issues/1227, #1233 all merged.

The line is needed for avoiding the cache effects in the benchmark. It will help flush the cache. We need to check the proper clang14 flags for that. But this...

Closing with https://github.com/pytorch/FBGEMM/pull/1246. Feel free to reopen if the issue still exists.