Merge branch 'master' into inline_asm_v2

script/cmake-cuda.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+
+rm -f CMakeCache.txt
+rm -f *.cmake
+rm -rf CMakeFiles
+
+MY_PROJECT_SOURCE=/home/chao/code/modular_convolution
+MY_PROJECT_INSTALL=../install.dir
+
+cmake                                                                                       \
+-D CMAKE_INSTALL_PREFIX=${MY_PROJECT_INSTALL}                                               \
+-D CMAKE_CXX_COMPILER=clang++                                                               \
+-D CMAKE_BUILD_TYPE=Release                                                                 \
+-D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                           \
+-D DEVICE_BACKEND=CUDA                                                                      \
+-D BOOST_ROOT="/package/install/boost_1.67.0"                                               \
+-D CUDA_COMMON_INCLUDE_DIR="/home/chao/code/test_feature/cuda_common/cuda_10.0_common/inc"  \
+-D CMAKE_CUDA_FLAGS="-ccbin clang++ -m64 -Xcompiler -fopenmp -lineinfo --source-in-ptx -keep -Xptxas -v -gencode=arch=compute_61,code=sm_61 -Xptxas -v -Xptxas -v -maxrregcount=128" \
+${MY_PROJECT_SOURCE}
+
+
+#-D CMAKE_CUDA_COMPILER="/package/install/cuda_10.0/bin/nvcc"                                \
+#-D CMAKE_CUDA_FLAGS="-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" \

script/cmake-hip.sh

@@ -0,0 +1,19 @@
+#!/bin/bash
+
+rm -f CMakeCache.txt
+rm -f *.cmake
+rm -rf CMakeFiles
+
+MY_PROJECT_SOURCE=/home/chao/code/modular_convolution
+MY_PROJECT_INSTALL=../install.dir
+
+cmake                                                                                       \
+-D CMAKE_INSTALL_PREFIX=${MY_PROJECT_INSTALL}                                               \
+-D CMAKE_BUILD_TYPE=Release                                                                 \
+-D DEVICE_BACKEND="HIP"                                                                     \
+-D HIP_HIPCC_FLAGS="${HIP_HIPCC_FLAGS} -gline-tables-only"                                  \
+-D CMAKE_CXX_FLAGS="-gline-tables-only"                                                     \
+-D CMAKE_CXX_COMPILER=/opt/rocm/bin/hipcc                                                   \
+-D CMAKE_PREFIX_PATH="/opt/rocm;/home/package/build/mlopen_dep"                             \
+-D CMAKE_VERBOSE_MAKEFILE:BOOL=ON                                                           \
+${MY_PROJECT_SOURCE}

script/compile-hip.sh

@@ -0,0 +1,6 @@
+#!/bin/bash
+export KMDUMPISA=1
+export KMDUMPLLVM=1
+
+make -j driver
+/opt/rocm/hcc/bin/llvm-objdump -mcpu=gfx906 -source -line-numbers driver/dump-gfx906.isabin > driver/dump-gfx906.isabin.isa

script/extract_asm-cuda.sh

@@ -0,0 +1 @@
+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

script/tracer-hip.sh

@@ -0,0 +1,3 @@
+#!/bin/bash
+
+/root/workspace/rocprofiler_pkg/bin/rpl_run.sh --timestamp on -i /root/workspace/rocprofiler_pkg/input.xml -d ./trace ./driver/driver 0 10