Added bwd data v3r1 v4r1, tweaking v1 (#10)

* Added bwd data v3r1: breaking down compute into a series of load balanced GEMM, and launch in a single kernel
* Added bwd data v4r1: like v3r1, but launch GEMMs in multiple kernels
* Tweaked v1r1  and v1r2 (atomic) on AMD GPU

script/cmake-cuda.sh

@@ -14,13 +14,11 @@ cmake
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                           \
 -D DEVICE_BACKEND=NVIDIA                                                                    \
 -D CUDA_COMMON_INCLUDE_DIR="/package/install/cuda/10.1/NVIDIA_CUDA-10.1_Samples/common/inc" \
--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 -maxrregcount=128" \
 ${MY_PROJECT_SOURCE}
 
 #-D BOOST_ROOT="/package/install/boost_1.67.0"                                               \
 
 #-D CMAKE_CUDA_COMPILER="/package/install/cuda_10.0/bin/nvcc"                                \
 #-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" \
+#-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 -maxrregcount=128" \

script/cmake-cuda_docker.sh

@@ -15,11 +15,9 @@ cmake
 -D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                           \
 -D DEVICE_BACKEND=NVIDIA                                                                    \
 -D CUDA_COMMON_INCLUDE_DIR="/root/NVIDIA_CUDA-10.1_Samples/common/inc"                      \
--D CMAKE_CUDA_FLAGS="-ccbin clang++-6.0 -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep  -Xptxas -v -gencode=arch=compute_60,code=sm_60 -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 -maxrregcount=128" \
 ${MY_PROJECT_SOURCE}
 
 
 #-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" \
+#-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 -maxrregcount=128" \

script/extract_asm-cuda.sh

-cuobjdump -xelf sm_60 ./driver/driver && nvdisasm --print-code -g driver.sm_60.cubin > driver.sm_60.asm
-cuobjdump -xelf sm_61 ./driver/driver && nvdisasm --print-code -g driver.sm_61.cubin > driver.sm_61.asm
-cuobjdump -xelf sm_70 ./driver/driver && nvdisasm --print-code -g driver.sm_70.cubin > driver.sm_70.asm
+DRIVER=$1
+ARCH=$2
+cuobjdump -xelf $ARCH ./driver/$DRIVER && nvdisasm --print-code -g $DRIVER.$ARCH.cubin > $DRIVER.$ARCH.asm