deftdawg/bitsandbytes-rocm
1
0
Fork 0
forked from mrq/bitsandbytes-rocm
Code Pull Requests Activity

Added Kepler and fixed V100+CUDA101 support. #4 #5

Browse Source
This commit is contained in:
Tim Dettmers 2021-10-17 21:21:39 -07:00
parent 20294dc0f2
commit 8400b58cbb
1 changed files with 28 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
Makefile
View File

@ -15,11 +15,30 @@ INCLUDE := -I $(CUDA_HOME)/include -I $(ROOT_DIR)/csrc -I $(CONDA_PREFIX)/inclu
LIB := -L $(CUDA_HOME)/lib64 -lcudart -lcuda -lcublas -lcurand -lcusparse -L $(CONDA_PREFIX)/lib LIB := -L $(CUDA_HOME)/lib64 -lcudart -lcuda -lcublas -lcurand -lcusparse -L $(CONDA_PREFIX)/lib
# NVIDIA NVCC compilation flags # NVIDIA NVCC compilation flags
COMPUTE_CAPABILITY := -gencode arch=compute_50,code=sm_50 # Maxwell COMPUTE_CAPABILITY := -gencode arch=compute_35,code=sm_35 # Kepler
COMPUTE_CAPABILITY += -gencode arch=compute_37,code=sm_37 # Kepler
COMPUTE_CAPABILITY += -gencode arch=compute_50,code=sm_50 # Maxwell
COMPUTE_CAPABILITY += -gencode arch=compute_52,code=sm_52 # Maxwell COMPUTE_CAPABILITY += -gencode arch=compute_52,code=sm_52 # Maxwell
COMPUTE_CAPABILITY += -gencode arch=compute_60,code=sm_60 # Pascal
COMPUTE_CAPABILITY += -gencode arch=compute_61,code=sm_61 # Pascal COMPUTE_CAPABILITY += -gencode arch=compute_61,code=sm_61 # Pascal
COMPUTE_CAPABILITY += -gencode arch=compute_70,code=sm_70 # Volta COMPUTE_CAPABILITY += -gencode arch=compute_70,code=sm_70 # Volta
COMPUTE_CAPABILITY += -gencode arch=compute_72,code=sm_72 # Volta COMPUTE_CAPABILITY += -gencode arch=compute_72,code=sm_72 # Volta
COMPUTE_CAPABILITY += -gencode arch=compute_72,code=sm_72 # Volta
# CUDA 9.2 supports CC 3.0, but CUDA >= 11.0 does not
CC_CUDA92 := -gencode arch=compute_30,code=sm_30
# Later versions of CUDA support the new architectures
CC_CUDA10x := -gencode arch=compute_30,code=sm_30
CC_CUDA10x += -gencode arch=compute_75,code=sm_75
CC_CUDA110 := -gencode arch=compute_75,code=sm_75
CC_CUDA110 += -gencode arch=compute_80,code=sm_80
CC_CUDA11x := -gencode arch=compute_75,code=sm_75
CC_CUDA11x += -gencode arch=compute_80,code=sm_80
CC_CUDA11x += -gencode arch=compute_86,code=sm_86
all: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR) all: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) $(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
@ -27,23 +46,23 @@ all: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR)
$(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB) $(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
cuda92: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR) cuda92: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA92) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA92) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o
$(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB) $(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
cuda10x: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR) cuda10x: $(ROOT_DIR)/dependencies/cub $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -gencode arch=compute_75,code=sm_75 -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o
$(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB) $(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
cuda110: $(BUILD_DIR) cuda110: $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -gencode arch=compute_80,code=sm_80 -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o
$(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB) $(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
cuda11x: $(BUILD_DIR) cuda11x: $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -gencode arch=compute_80,code=sm_80 -gencode arch=compute_86,code=sm_86 -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR) $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' --use_fast_math -Xptxas=-v -dc $(FILES_CUDA) $(INCLUDE) $(LIB) --output-directory $(BUILD_DIR)
$(NVCC) $(COMPUTE_CAPABILITY) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o $(NVCC) $(COMPUTE_CAPABILITY) $(CC_CUDA10x) -Xcompiler '-fPIC' -dlink $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o -o $(BUILD_DIR)/link.o
$(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB) $(GPP) -std=c++11 -shared -fPIC $(INCLUDE) $(BUILD_DIR)/ops.o $(BUILD_DIR)/kernels.o $(BUILD_DIR)/link.o $(FILES_CPP) -o ./bitsandbytes/libbitsandbytes.so $(LIB)
$(BUILD_DIR): $(BUILD_DIR):