Added sm_75 support for CI Xenial CUDA 11.1 cuDNN 8 builds (#57320)

Summary:
This PR adds `sm_75` CUDA architecture support for the PR CI build Xenial CUDA 11.1 cuDNN 8, with build name:`pytorch_linux_xenial_cuda11_1_cudnn8_py3_gcc7_build`, so that generated artifacts from these builds can be installed and run on machines with CUDA capability sm_75.

In PR https://github.com/pytorch/pytorch/issues/57207, the Xenial CUDA 10.2 cuDNN 7 build `pytorch_linux_xenial_cuda10_2_cudnn7_py3_gcc7_build` was taken off the list of builds done for PRs to `master`. PR https://github.com/pytorch/pytorch/issues/56619 has added `sm_75` support for this build. This PR removes this support for the Xenial CUDA 10.2 cuDNN7 builds, and adds it for the current PR CI build Xenial CUDA 11.1 cuDNN 8 `pytorch_linux_xenial_cuda11_1_cudnn8_py3_gcc7_build`.

Pull Request resolved: https://github.com/pytorch/pytorch/pull/57320

Reviewed By: astaff

Differential Revision: D28125542

Pulled By: malfet

fbshipit-source-id: f220b8f3279054c98cab9eef1e0d7e37161a946f
This commit is contained in:
Mustafa Bal 2021-04-30 17:50:40 -07:00 committed by Facebook GitHub Bot
parent 2b54cec7e8
commit bd347012ec

View File

@ -184,8 +184,8 @@ fi
# Target only our CI GPU machine's CUDA arch to speed up the build
export TORCH_CUDA_ARCH_LIST="5.2"
# Add sm_75 support for the Linux CUDA 10.2 cuDNN 7 build
if [[ "$BUILD_ENVIRONMENT" == *cuda10.2-cudnn7*build ]]; then
# Add sm_75 support for the Linux CUDA 11.1 cuDNN 8 CircleCI build
if [[ "$BUILD_ENVIRONMENT" == *xenial-cuda11.1-cudnn8*build ]]; then
export TORCH_CUDA_ARCH_LIST=$TORCH_CUDA_ARCH_LIST";7.5"
fi