LMU/DKS
5
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

adopted DKS to Cuda12 and GV100. For Tesla K40c code lines are still present but commented by as35.

Browse Source
This commit is contained in:
suter_a
2023-01-30 12:38:52 +01:00
parent efa3311b45
commit 21b4c591b9
3 changed files with 8 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
CMakeLists.txt
View File

@ -148,7 +148,8 @@ IF ( (${CMAKE_C_COMPILER_ID} STREQUAL "GNU" OR ${CMAKE_C_COMPILER_ID} STREQUAL "
MESSAGE (STATUS "cuda version: ${CUDA_VERSION}")
SET(CUDA_PROPAGATE_HOST_FLAGS OFF)
SET (CUDA_NVCC_FLAGS "-arch=sm_35;-DDEBUG;-std=c++11;-D__wsu;-fmad=false")
#as35: Tesla K40c SET (CUDA_NVCC_FLAGS "-arch=sm_35;-DDEBUG;-std=c++11;-D__wsu;-fmad=false")
SET (CUDA_NVCC_FLAGS "-arch=sm_70; -gencode=arch=compute_70,code=sm_70 -DDEBUG;-std=c++11;-D__wsu;-fmad=false")
SET (CUDA_NVCC_FLAGS "${CUDA_NVCC_FLAGS};${OPENCL_KERNELS}")
IF (NOT STATIC_CUDA)

7
src/CUDA/CMakeListsLibcuda.txt
View File

@ -1,8 +1,9 @@
CMAKE_MINIMUM_REQUIRED (VERSION 2.8)
CMAKE_MINIMUM_REQUIRED (VERSION 3.2)
FIND_PACKAGE(CUDA REQUIRED)
SET (CUDA_NVCC_FLAGS "-arch=sm_30")
#as35: Tesla K40c SET (CUDA_NVCC_FLAGS "-arch=sm_30")
SET (CUDA_NVCC_FLAGS "-arch=sm_70")
SET(LIB_TYPE STATIC)
@ -22,4 +23,4 @@ INCLUDE_DIRECTORIES (
${CMAKE_CURRENT_SOURCE_DIR}
)
CUDA_ADD_LIBRARY(cudadks ${DKS_CUDA_SRCS})
CUDA_ADD_LIBRARY(cudadks ${DKS_CUDA_SRCS})

3
src/CUDA/CudaChiSquareRuntime.cu
View File

@ -94,7 +94,8 @@ int CudaChiSquareRuntime::compileProgram(std::string function, bool mlh) {
}
//compile program
const char *opts[] = {"-arch=compute_35", "-fmad=false", ""};
//as35: for Tesla K40c const char *opts[] = {"-arch=compute_35", "-fmad=false", ""};
const char *opts[] = {"-arch=compute_70", "-fmad=false", ""};
int numopts = 2;
if (mlh) {
opts[2] = "-DMLH";