Rostyslav Geyyer

Results 14 comments of Rostyslav Geyyer

Clang 14 Build Failure

Hi @juanpaez22, thanks for sharing the issue! We are working on the asmjit update https://github.com/pytorch/FBGEMM/issues/1070, stay tuned!

Clang 14 Build Failure

Hi @juanpaez22, thanks for reminding! We have updated internal deps to d3fbf7c9bc7c1d1365a94a45614b91c5a3706b81, currently working on FBGEMM API to support it.

Clang 14 Build Failure

Hi @juanpaez22, thanks for double checking! Could you try https://github.com/pytorch/FBGEMM/pull/1202 and test if it works with Clang 14?

Clang 14 Build Failure

Looks like it is something Clang-specific... I have added a compilation flag to the CMakeLists, @juanpaez22 could you re-check https://github.com/pytorch/FBGEMM/pull/1202?

Clang 14 Build Failure

Thanks, @juanpaez22! I won't be able to help at this point, so pinging @jianyuh and @brad-mengchi ;-)

FBGEMM_gpu support for CUDA 11.5

Hi @chantat, thanks for sharing the issue! Looks like you are using a CUDA 11.X version, so you don't have to specify the CUB_DIR explicitly. Could you re-try without running...

Failure in import: undefined symbol error from Python 3.7 + CUDA113

Hi @chongxiaoc, thanks for sharing! It is a known issue we are working on! As a workaround please build fbgemm_gpu from source.

Refactor f8_t and bf8_t as custom types, enable use of custom types

@zjing14 There is no perf drop on example_gemm_xdl_fp8, nice!

Refactor f8_t and bf8_t as custom types, enable use of custom types

Current work is covered in https://github.com/ROCm/composable_kernel/pull/1333

Tags for ROCm releases

Hi @tpkessler, thanks for pinging! We are discussing possible release strategies, but haven't decided yet... let us get back to you in early 2023.