diff --git a/src/Makefile b/src/Makefile index 1191d61a5..9c93508e4 100644 --- a/src/Makefile +++ b/src/Makefile @@ -12,7 +12,7 @@ else all: pannotia rodinia_2.0-ft proxy-apps microbench rodinia-3.1 ispass-2009 polybench parboil shoc custom_apps endif endif -ci: rodinia_2.0-ft rodinia-3.1 GPU_Microbenchmark cutlass cuda_samples +ci: rodinia_2.0-ft rodinia-3.1 GPU_Microbenchmark cutlass_mini cuda_samples accelwattch: accelwattch_validation accelwattch_hw_power accelwattch_ubench accelwattch_validation: rodinia-3.1_accelwattch_validation parboil_accelwattch_validation cutlass cuda_samples-11.0_accelwattch_validation accelwattch_hw_power: rodinia-3.1_hw_power parboil_hw_power cuda_samples-11.0_hw_power @@ -430,15 +430,27 @@ clean_accelwattch_ubench: cutlass: mkdir -p $(BINDIR)/$(BINSUBDIR)/ - echo $(BINDIR) git submodule init && git submodule update - $(SETENV) mkdir -p cuda/cutlass-bench/build && cd cuda/cutlass-bench/build && cmake .. -DUSE_GPGPUSIM=1 -DCUTLASS_NVCC_ARCHS=80 -DCMAKE_CUDA_ARCHITECTURES=80 -DCMAKE_CUDA_COMPILER=$(CUDA_INSTALL_PATH)/bin/nvcc && $(MAKE) cutlass_profiler && $(MAKE) cutlass_examples - echo $(BINDIR) - cp cuda/cutlass-bench/build/tools/profiler/cutlass_profiler $(BINDIR)/$(BINSUBDIR)/cutlass_profiler + mkdir -p cuda/cutlass/build + $(SETENV) cmake -S cuda/cutlass-bench -B cuda/cutlass-bench/build -DCUTLASS_NVCC_ARCHS="80;90;90a" -DCMAKE_CUDA_ARCHITECTURES="80;90;90a" -DCMAKE_CUDA_COMPILER=$(CUDA_INSTALL_PATH)/bin/nvcc + $(SETENV) $(MAKE) -C cuda/cutlass-bench/build/ cutlass_examples cutlass_profiler + find cuda/cutlass-bench/build/examples -type f -executable -exec sh -c 'cp "$$1" "$(BINDIR)/$(BINSUBDIR)/cutlass_$$(basename "$$1")"' sh {} \; + +cutlass_mini: + mkdir -p $(BINDIR)/$(BINSUBDIR)/ + git submodule init && git submodule update + mkdir -p cuda/cutlass-bench/build + $(SETENV) cmake -S cuda/cutlass-bench -B cuda/cutlass-bench/build -DCUTLASS_NVCC_ARCHS="80;90;90a" -DCMAKE_CUDA_ARCHITECTURES="80;90;90a" -DCMAKE_CUDA_COMPILER=$(CUDA_INSTALL_PATH)/bin/nvcc +# add any other cutlass examples you want to build here + $(SETENV) $(MAKE) -C cuda/cutlass-bench/build/ 00_basic_gemm 48_hopper_warp_specialized_gemm 52_hopper_gather_scatter_fusion 57_hopper_grouped_gemm + find cuda/cutlass-bench/build/examples -type f -executable -exec sh -c 'cp "$$1" "$(BINDIR)/$(BINSUBDIR)/cutlass_$$(basename "$$1")"' sh {} \; + + + cutlass_examples_turing: mkdir -p $(BINDIR)/$(BINSUBDIR)/ git submodule init && git submodule update diff --git a/src/setup_environment b/src/setup_environment index 2e0a5894e..d8b0caff5 100755 --- a/src/setup_environment +++ b/src/setup_environment @@ -97,6 +97,7 @@ if [ $CUDA_VERSION_MAJOR -eq 12 ]; then export CUDA_CPPFLAGS="$CUDA_CPPFLAGS -gencode=arch=compute_86,code=compute_86" if [ $CUDA_VERSION_MINOR -ge 8 ]; then export CUDA_CPPFLAGS="$CUDA_CPPFLAGS -gencode=arch=compute_90,code=compute_90" + export CUDA_CPPFLAGS="$CUDA_CPPFLAGS -gencode=arch=compute_90a,code=compute_90a" export CUDA_CPPFLAGS="$CUDA_CPPFLAGS -gencode=arch=compute_100,code=compute_100" fi export CUDA_CPPFLAGS="$CUDA_CPPFLAGS -std=c++17"