@@ -133,24 +133,26 @@ endif
133133
134134# it is recommended to use the CMAKE file to build for cublas if you can - will likely work better
135135ifdef LLAMA_CUBLAS
136- CFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH ) /targets/x86_64-linux/include
137- CXXFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH ) /targets/x86_64-linux/include
138- LDFLAGS += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH ) /targets/x86_64-linux/lib
139- OBJS += ggml-cuda.o
140- NVCC = nvcc
141- NVCCFLAGS = --forward-unknown-to-host-compiler -arch=native
136+ CFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
137+ CXXFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
138+ LDFLAGS += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH)/targets/x86_64-linux/lib
139+ OBJS += ggml-cuda.o ggml_v2 -cuda.o
140+ NVCC = nvcc
141+ NVCCFLAGS = --forward-unknown-to-host-compiler -arch=native
142142ifdef LLAMA_CUDA_DMMV_X
143- NVCCFLAGS += -DGGML_CUDA_DMMV_X=$(LLAMA_CUDA_DMMV_X )
143+ NVCCFLAGS += -DGGML_CUDA_DMMV_X=$(LLAMA_CUDA_DMMV_X)
144144else
145- NVCCFLAGS += -DGGML_CUDA_DMMV_X=32
145+ NVCCFLAGS += -DGGML_CUDA_DMMV_X=32
146146endif # LLAMA_CUDA_DMMV_X
147147ifdef LLAMA_CUDA_DMMV_Y
148- NVCCFLAGS += -DGGML_CUDA_DMMV_Y=$(LLAMA_CUDA_DMMV_Y )
148+ NVCCFLAGS += -DGGML_CUDA_DMMV_Y=$(LLAMA_CUDA_DMMV_Y)
149149else
150- NVCCFLAGS += -DGGML_CUDA_DMMV_Y=1
150+ NVCCFLAGS += -DGGML_CUDA_DMMV_Y=1
151151endif # LLAMA_CUDA_DMMV_Y
152152ggml-cuda.o : ggml-cuda.cu ggml-cuda.h
153- $(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -Wno-pedantic -c $< -o $@
153+ $(NVCC ) $(NVCCFLAGS ) $(CXXFLAGS ) $(CUBLAS_CXXFLAGS ) -Wno-pedantic -c $< -o $@
154+ ggml_v2-cuda.o : otherarch/ggml_v2-cuda.cu otherarch/ggml_v2-cuda.h
155+ $(NVCC ) $(NVCCFLAGS ) $(CXXFLAGS ) $(CUBLAS_CXXFLAGS ) -Wno-pedantic -c $< -o $@
154156endif # LLAMA_CUBLAS
155157
156158ifdef LLAMA_METAL
0 commit comments