From 338f4624c6a94ae60899bc219c32e7aad86a4d7f Mon Sep 17 00:00:00 2001 From: atalman Date: Mon, 19 Sep 2022 11:21:10 -0700 Subject: [PATCH 1/3] Add manywheel special build Testing Builder change Testing Adding manywheel cuda workflow Simplify Fix expr --- manywheel/build_cuda.sh | 89 +++++++++++++++++++++++++---------------- 1 file changed, 54 insertions(+), 35 deletions(-) diff --git a/manywheel/build_cuda.sh b/manywheel/build_cuda.sh index efea1ae93..f947b49ab 100644 --- a/manywheel/build_cuda.sh +++ b/manywheel/build_cuda.sh @@ -232,42 +232,61 @@ DEPS_SONAME=( "libgomp.so.1" ) elif [[ $CUDA_VERSION == "11.7" ]]; then -export USE_STATIC_CUDNN=0 -DEPS_LIST=( - "/usr/local/cuda/lib64/libcudart.so.11.0" - "/usr/local/cuda/lib64/libnvToolsExt.so.1" - "/usr/local/cuda/lib64/libnvrtc.so.11.2" # this is not a mistake for 11.7, it links to 11.7.50 - "/usr/local/cuda/lib64/libnvrtc-builtins.so.11.7" - "/usr/local/cuda/lib64/libcudnn_adv_infer.so.8" - "/usr/local/cuda/lib64/libcudnn_adv_train.so.8" - "/usr/local/cuda/lib64/libcudnn_cnn_infer.so.8" - "/usr/local/cuda/lib64/libcudnn_cnn_train.so.8" - "/usr/local/cuda/lib64/libcudnn_ops_infer.so.8" - "/usr/local/cuda/lib64/libcudnn_ops_train.so.8" - "/usr/local/cuda/lib64/libcudnn.so.8" - "/usr/local/cuda/lib64/libcublas.so.11" - "/usr/local/cuda/lib64/libcublasLt.so.11" - "$LIBGOMP_PATH" -) -DEPS_SONAME=( - "libcudart.so.11.0" - "libnvToolsExt.so.1" - "libnvrtc.so.11.2" - "libnvrtc-builtins.so.11.7" - "libcudnn_adv_infer.so.8" - "libcudnn_adv_train.so.8" - "libcudnn_cnn_infer.so.8" - "libcudnn_cnn_train.so.8" - "libcudnn_ops_infer.so.8" - "libcudnn_ops_train.so.8" - "libcudnn.so.8" - "libcublas.so.11" - "libcublasLt.so.11" - "libgomp.so.1" -) + export USE_STATIC_CUDNN=0 + if [[ "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" == *"cudnn"* ]]; then + echo "Bundling without cudnn and cublas." + DEPS_LIST=( + "/usr/local/cuda/lib64/libcudart.so.11.0" + "/usr/local/cuda/lib64/libnvToolsExt.so.1" + "/usr/local/cuda/lib64/libnvrtc.so.11.2" # this is not a mistake for 11.7, it links to 11.7.50 + "/usr/local/cuda/lib64/libnvrtc-builtins.so.11.7" + "$LIBGOMP_PATH" + ) + DEPS_SONAME=( + "libcudart.so.11.0" + "libnvToolsExt.so.1" + "libnvrtc.so.11.2" + "libnvrtc-builtins.so.11.7" + "libgomp.so.1" + ) + else + echo "Bundling with cudnn and cublas." + DEPS_LIST=( + "/usr/local/cuda/lib64/libcudart.so.11.0" + "/usr/local/cuda/lib64/libnvToolsExt.so.1" + "/usr/local/cuda/lib64/libnvrtc.so.11.2" # this is not a mistake for 11.7, it links to 11.7.50 + "/usr/local/cuda/lib64/libnvrtc-builtins.so.11.7" + "/usr/local/cuda/lib64/libcudnn_adv_infer.so.8" + "/usr/local/cuda/lib64/libcudnn_adv_train.so.8" + "/usr/local/cuda/lib64/libcudnn_cnn_infer.so.8" + "/usr/local/cuda/lib64/libcudnn_cnn_train.so.8" + "/usr/local/cuda/lib64/libcudnn_ops_infer.so.8" + "/usr/local/cuda/lib64/libcudnn_ops_train.so.8" + "/usr/local/cuda/lib64/libcudnn.so.8" + "/usr/local/cuda/lib64/libcublas.so.11" + "/usr/local/cuda/lib64/libcublasLt.so.11" + "$LIBGOMP_PATH" + ) + DEPS_SONAME=( + "libcudart.so.11.0" + "libnvToolsExt.so.1" + "libnvrtc.so.11.2" + "libnvrtc-builtins.so.11.7" + "libcudnn_adv_infer.so.8" + "libcudnn_adv_train.so.8" + "libcudnn_cnn_infer.so.8" + "libcudnn_cnn_train.so.8" + "libcudnn_ops_infer.so.8" + "libcudnn_ops_train.so.8" + "libcudnn.so.8" + "libcublas.so.11" + "libcublasLt.so.11" + "libgomp.so.1" + ) + fi -# Try parallelizing nvcc as well -export TORCH_NVCC_FLAGS="-Xfatbin -compress-all --threads 2" + # Try parallelizing nvcc as well + export TORCH_NVCC_FLAGS="-Xfatbin -compress-all --threads 2" else echo "Unknown cuda version $CUDA_VERSION" exit 1 From 56bff31e66c867fd8fd5015865f4861d41acf4ea Mon Sep 17 00:00:00 2001 From: atalman Date: Tue, 20 Sep 2022 12:32:48 -0700 Subject: [PATCH 2/3] address comments --- manywheel/build_cuda.sh | 52 ++++++++++++++++------------------------- 1 file changed, 20 insertions(+), 32 deletions(-) diff --git a/manywheel/build_cuda.sh b/manywheel/build_cuda.sh index f947b49ab..bb2a622e1 100644 --- a/manywheel/build_cuda.sh +++ b/manywheel/build_cuda.sh @@ -233,29 +233,26 @@ DEPS_SONAME=( ) elif [[ $CUDA_VERSION == "11.7" ]]; then export USE_STATIC_CUDNN=0 - if [[ "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" == *"cudnn"* ]]; then - echo "Bundling without cudnn and cublas." - DEPS_LIST=( - "/usr/local/cuda/lib64/libcudart.so.11.0" - "/usr/local/cuda/lib64/libnvToolsExt.so.1" - "/usr/local/cuda/lib64/libnvrtc.so.11.2" # this is not a mistake for 11.7, it links to 11.7.50 - "/usr/local/cuda/lib64/libnvrtc-builtins.so.11.7" - "$LIBGOMP_PATH" - ) - DEPS_SONAME=( - "libcudart.so.11.0" - "libnvToolsExt.so.1" - "libnvrtc.so.11.2" - "libnvrtc-builtins.so.11.7" - "libgomp.so.1" - ) - else + # Try parallelizing nvcc as well + export TORCH_NVCC_FLAGS="-Xfatbin -compress-all --threads 2" + DEPS_LIST=( + "/usr/local/cuda/lib64/libcudart.so.11.0" + "/usr/local/cuda/lib64/libnvToolsExt.so.1" + "/usr/local/cuda/lib64/libnvrtc.so.11.2" # this is not a mistake for 11.7, it links to 11.7.50 + "/usr/local/cuda/lib64/libnvrtc-builtins.so.11.7" + "$LIBGOMP_PATH" + ) + DEPS_SONAME=( + "libcudart.so.11.0" + "libnvToolsExt.so.1" + "libnvrtc.so.11.2" + "libnvrtc-builtins.so.11.7" + "libgomp.so.1" + ) + + if [[ "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" != *"cudnn"* ]]; then echo "Bundling with cudnn and cublas." - DEPS_LIST=( - "/usr/local/cuda/lib64/libcudart.so.11.0" - "/usr/local/cuda/lib64/libnvToolsExt.so.1" - "/usr/local/cuda/lib64/libnvrtc.so.11.2" # this is not a mistake for 11.7, it links to 11.7.50 - "/usr/local/cuda/lib64/libnvrtc-builtins.so.11.7" + DEPS_LIST+=( "/usr/local/cuda/lib64/libcudnn_adv_infer.so.8" "/usr/local/cuda/lib64/libcudnn_adv_train.so.8" "/usr/local/cuda/lib64/libcudnn_cnn_infer.so.8" @@ -265,13 +262,8 @@ elif [[ $CUDA_VERSION == "11.7" ]]; then "/usr/local/cuda/lib64/libcudnn.so.8" "/usr/local/cuda/lib64/libcublas.so.11" "/usr/local/cuda/lib64/libcublasLt.so.11" - "$LIBGOMP_PATH" ) - DEPS_SONAME=( - "libcudart.so.11.0" - "libnvToolsExt.so.1" - "libnvrtc.so.11.2" - "libnvrtc-builtins.so.11.7" + DEPS_SONAME+=( "libcudnn_adv_infer.so.8" "libcudnn_adv_train.so.8" "libcudnn_cnn_infer.so.8" @@ -281,12 +273,8 @@ elif [[ $CUDA_VERSION == "11.7" ]]; then "libcudnn.so.8" "libcublas.so.11" "libcublasLt.so.11" - "libgomp.so.1" ) fi - - # Try parallelizing nvcc as well - export TORCH_NVCC_FLAGS="-Xfatbin -compress-all --threads 2" else echo "Unknown cuda version $CUDA_VERSION" exit 1 From b16bff23b7aeb0b5986ed05e7e27231d5c4caf94 Mon Sep 17 00:00:00 2001 From: atalman Date: Tue, 20 Sep 2022 12:47:46 -0700 Subject: [PATCH 3/3] checking for general setting --- manywheel/build_cuda.sh | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/manywheel/build_cuda.sh b/manywheel/build_cuda.sh index bb2a622e1..c7a177fee 100644 --- a/manywheel/build_cuda.sh +++ b/manywheel/build_cuda.sh @@ -250,7 +250,7 @@ elif [[ $CUDA_VERSION == "11.7" ]]; then "libgomp.so.1" ) - if [[ "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" != *"cudnn"* ]]; then + if [[ -z "$PYTORCH_EXTRA_INSTALL_REQUIREMENTS" ]]; then echo "Bundling with cudnn and cublas." DEPS_LIST+=( "/usr/local/cuda/lib64/libcudnn_adv_infer.so.8"