add sm_90 to CUDA11.8 builds (pytorch#1278)

ptrblck · web-flow · commit e40ab7c9e3cd · 2023-01-23T20:23:58.000-08:00
diff --git a/common/install_cuda.sh b/common/install_cuda.sh
@@ -155,8 +155,8 @@ function prune_118 {
     export NVPRUNE="/usr/local/cuda-11.8/bin/nvprune"
     export CUDA_LIB_DIR="/usr/local/cuda-11.8/lib64"
 
-    export GENCODE="-gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86"
-    export GENCODE_CUDNN="-gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86"
+    export GENCODE="-gencode arch=compute_35,code=sm_35 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
+    export GENCODE_CUDNN="-gencode arch=compute_35,code=sm_35 -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_61,code=sm_61 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_75,code=sm_75 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -gencode arch=compute_90,code=sm_90"
 
     if [[ -n "$OVERRIDE_GENCODE" ]]; then
         export GENCODE=$OVERRIDE_GENCODE
diff --git a/conda/pytorch-nightly/bld.bat b/conda/pytorch-nightly/bld.bat
@@ -20,21 +20,18 @@ if "%build_with_cuda%" == "" goto cuda_flags_end
 set CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%desired_cuda%
 set CUDA_BIN_PATH=%CUDA_PATH%\bin
 set TORCH_NVCC_FLAGS=-Xfatbin -compress-all
-set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0
+set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6
 if "%desired_cuda%" == "11.5" (
-    set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6
     set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2
 )
 if "%desired_cuda%" == "11.6" (
-    set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6
     set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2
 )
 if "%desired_cuda%" == "11.7" (
-    set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6
     set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2
 )
 if "%desired_cuda%" == "11.8" (
-    set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6
+    set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;9.0
     set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2
 )
 
diff --git a/conda/pytorch-nightly/build.sh b/conda/pytorch-nightly/build.sh
@@ -70,7 +70,7 @@ if [[ -n "$build_with_cuda" ]]; then
         #for cuda 11.7 include all dynamic loading libraries
         DEPS_LIST=(/usr/local/cuda/lib64/libcudnn*.so.8 /usr/local/cuda-11.7/extras/CUPTI/lib64/libcupti.so.11.7)
     elif [[ $CUDA_VERSION == 11.8* ]]; then
-	export TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;6.0;6.1;7.0;7.5;8.0;8.6"
+	export TORCH_CUDA_ARCH_LIST="$TORCH_CUDA_ARCH_LIST;6.0;6.1;7.0;7.5;8.0;8.6;9.0"
 	#for cuda 11.8 we use cudnn 8.5
 	#which does not have single static libcudnn_static.a deliverable to link with
 	export USE_STATIC_CUDNN=0
diff --git a/magma/Makefile b/magma/Makefile
@@ -2,7 +2,7 @@ SHELL=/usr/bin/env bash
 
 DESIRED_CUDA ?= 11.6
 PACKAGE_NAME ?= magma-cuda116
-CUDA_ARCH_LIST ?= -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70
+CUDA_ARCH_LIST ?= -gencode arch=compute_37,code=sm_37 -gencode arch=compute_50,code=sm_50 -gencode arch=compute_60,code=sm_60 -gencode arch=compute_70,code=sm_70 -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86
 
 DOCKER_RUN = set -eou pipefail; docker run --rm -i \
 	-v $(shell git rev-parse --show-toplevel):/builder \
@@ -26,20 +26,18 @@ clean:
 .PHONY: magma-cuda118
 magma-cuda118: DESIRED_CUDA := 11.8
 magma-cuda118: PACKAGE_NAME := magma-cuda118
-magma-cuda118: CUDA_ARCH_LIST += -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86
+magma-cuda118: CUDA_ARCH_LIST += -gencode arch=compute_90,code=sm_90
 magma-cuda118:
 	$(DOCKER_RUN)
 
 .PHONY: magma-cuda117
 magma-cuda117: DESIRED_CUDA := 11.7
 magma-cuda117: PACKAGE_NAME := magma-cuda117
-magma-cuda117: CUDA_ARCH_LIST += -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86
 magma-cuda117:
 	$(DOCKER_RUN)
 
 .PHONY: magma-cuda116
 magma-cuda116: DESIRED_CUDA := 11.6
 magma-cuda116: PACKAGE_NAME := magma-cuda116
-magma-cuda116: CUDA_ARCH_LIST += -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86
 magma-cuda116:
 	$(DOCKER_RUN)
diff --git a/manywheel/build_cuda.sh b/manywheel/build_cuda.sh
@@ -58,12 +58,12 @@ cuda_version_nodot=$(echo $CUDA_VERSION | tr -d '.')
 
 TORCH_CUDA_ARCH_LIST="3.7;5.0;6.0;7.0"
 case ${CUDA_VERSION} in
-    11.[678])
-        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6"
+    11.8)
+        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6;9.0"
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
-    10.*)
-        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST}"
+    11.[67])
+        TORCH_CUDA_ARCH_LIST="${TORCH_CUDA_ARCH_LIST};7.5;8.0;8.6"
         EXTRA_CAFFE2_CMAKE_FLAGS+=("-DATEN_NO_TEST=ON")
         ;;
     *)
diff --git a/windows/cuda118.bat b/windows/cuda118.bat
@@ -37,10 +37,10 @@ IF "%CUDA_PATH_V118%"=="" (
 )
 
 IF "%BUILD_VISION%" == "" (
-    set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6
+    set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0
     set TORCH_NVCC_FLAGS=-Xfatbin -compress-all
 ) ELSE (
-    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86
+    set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90,code=compute_90
 )
 
 set "CUDA_PATH=%CUDA_PATH_V118%"

Original file line number	Diff line number	Diff line change
`@@ -20,21 +20,18 @@ if "%build_with_cuda%" == "" goto cuda_flags_end`
`20`	`20`	`set CUDA_PATH=C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v%desired_cuda%`
`21`	`21`	`set CUDA_BIN_PATH=%CUDA_PATH%\bin`
`22`	`22`	`set TORCH_NVCC_FLAGS=-Xfatbin -compress-all`
`23`		`-set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0`
	`23`	`+set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6`
`24`	`24`	`if "%desired_cuda%" == "11.5" (`
`25`		`- set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6`
`26`	`25`	`set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2`
`27`	`26`	`)`
`28`	`27`	`if "%desired_cuda%" == "11.6" (`
`29`		`- set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6`
`30`	`28`	`set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2`
`31`	`29`	`)`
`32`	`30`	`if "%desired_cuda%" == "11.7" (`
`33`		`- set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6`
`34`	`31`	`set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2`
`35`	`32`	`)`
`36`	`33`	`if "%desired_cuda%" == "11.8" (`
`37`		`- set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;6.0;6.1;7.0;7.5;8.0;8.6`
	`34`	`+ set TORCH_CUDA_ARCH_LIST=%TORCH_CUDA_ARCH_LIST%;9.0`
`38`	`35`	`set TORCH_NVCC_FLAGS=-Xfatbin -compress-all --threads 2`
`39`	`36`	`)`
`40`	`37`
Original file line number	Diff line number	Diff line change
`@@ -37,10 +37,10 @@ IF "%CUDA_PATH_V118%"=="" (`
`37`	`37`	`)`
`38`	`38`
`39`	`39`	`IF "%BUILD_VISION%" == "" (`
`40`		`- set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6`
	`40`	`+ set TORCH_CUDA_ARCH_LIST=3.7+PTX;5.0;6.0;6.1;7.0;7.5;8.0;8.6;9.0`
`41`	`41`	`set TORCH_NVCC_FLAGS=-Xfatbin -compress-all`
`42`	`42`	`) ELSE (`
`43`		`- set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86`
	`43`	`+ set NVCC_FLAGS=-D__CUDA_NO_HALF_OPERATORS__ --expt-relaxed-constexpr -gencode=arch=compute_35,code=sm_35 -gencode=arch=compute_50,code=sm_50 -gencode=arch=compute_60,code=sm_60 -gencode=arch=compute_70,code=sm_70 -gencode=arch=compute_75,code=sm_75 -gencode=arch=compute_80,code=compute_80 -gencode=arch=compute_86,code=compute_86 -gencode=arch=compute_90,code=compute_90`
`44`	`44`	`)`
`45`	`45`
`46`	`46`	`set "CUDA_PATH=%CUDA_PATH_V118%"`