add cuda extract_asm script

e6c86f81 · Chao Liu · c2347410 · e6c86f81 · e6c86f81
Commit e6c86f81 authored Apr 02, 2019 by Chao Liu
Hide whitespace changes
Inline Side-by-side

Showing with 6 additions and 4 deletions

script/cmake-cuda.sh script/cmake-cuda.sh +5 -4

script/extract_asm-cuda.sh script/extract_asm-cuda.sh +1 -0

No files found.
--- a/script/cmake-cuda.sh
+++ b/script/cmake-cuda.sh
@@ -15,11 +15,12 @@ cmake
 -D DEVICE_BACKEND=CUDA                                                                      \
 -D BOOST_ROOT="/package/install/boost_1.67.0"                                               \
 -D CUDA_COMMON_INCLUDE_DIR="/home/chao/code/test_feature/cuda_common/cuda_10.0_common/inc"  \
-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -arch=sm_61" \
+-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -gencode=arch=compute_61,code=sm_61 -Xptxas -v -Xptxas -v -maxrregcount=128" \
 ${MY_PROJECT_SOURCE}


 #-D CMAKE_CUDA_COMPILER="/package/install/cuda_10.0/bin/nvcc"                                \
-#-D CMAKE_CUDA_FLAGS="-G -lineinfo --source-in-ptx -keep -Xptxas -v -arch=sm_61"            \
-#-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -arch=sm_61" \
-#-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -arch=sm_61 -Xptxas -v -maxrregcount=128" \
+#-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -gencode=arch=compute_61,code=sm_61" \
+#-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -gencode=arch=compute_61,code=sm_61 -Xptxas -v -Xptxas -v -maxrregcount=128" \
+#-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -gencode=arch=compute_61,code=sm_61 -Xptxas -v -gencode=arch=compute_70,code=sm_70" \
+#-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -gencode=arch=compute_61,code=sm_61 -Xptxas -v -gencode=arch=compute_70,code=sm_70 -Xptxas -v -maxrregcount=128" \
--- a/script/extract_asm-cuda.sh
+++ b/script/extract_asm-cuda.sh
+cuobjdump -xelf all ./driver/driver && nvdisasm --print-code -g driver.sm_61.cubin > driver.sm_61.asm && nvdisasm --print-code -g driver.sm_70.cubin > driver.sm_70.asm