vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

ggml backend interface wip

Browse source 
refactor ggml-cuda
This commit is contained in:
slaren 2023-07-10 17:32:06 +02:00
parent 6e7cca4047
commit 0d2b66c638
15 changed files with 4480 additions and 4782 deletions
Download patch file Download diff file Expand all files Collapse all files

29
Makefile
View file

@ -55,6 +55,12 @@ else
CXXFLAGS += -DNDEBUG
endif
ifdef LLAMA_SANITIZE
CFLAGS += -g -fsanitize=$(LLAMA_SANITIZE) -fno-omit-frame-pointer
CXXFLAGS += -g -fsanitize=$(LLAMA_SANITIZE) -fno-omit-frame-pointer
LDFLAGS += -g -fsanitize=$(LLAMA_SANITIZE)
endif
ifdef LLAMA_SERVER_VERBOSE
CXXFLAGS += -DSERVER_VERBOSE=$(LLAMA_SERVER_VERBOSE)
endif
@ -163,13 +169,16 @@ ifdef LLAMA_BLIS
LDFLAGS += -lblis -L/usr/local/lib
endif # LLAMA_BLIS
ifdef LLAMA_CUBLAS
CFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
CXXFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
ifdef LLAMA_CUDA
CFLAGS += -DGGML_USE_CUDA -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
CXXFLAGS += -DGGML_USE_CUDA -I/usr/local/cuda/include -I/opt/cuda/include -I$(CUDA_PATH)/targets/x86_64-linux/include
LDFLAGS += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64 -L/opt/cuda/lib64 -L$(CUDA_PATH)/targets/x86_64-linux/lib
OBJS += ggml-cuda.o
NVCC = nvcc
NVCCFLAGS = --forward-unknown-to-host-compiler
ifdef LLAMA_DEBUG
NVCCFLAGS += -lineinfo
endif # LLAMA_DEBUG
ifdef CUDA_DOCKER_ARCH
NVCCFLAGS += -Wno-deprecated-gpu-targets -arch=$(CUDA_DOCKER_ARCH)
else
@ -198,10 +207,9 @@ ifdef LLAMA_CUDA_KQUANTS_ITER
else
NVCCFLAGS += -DK_QUANTS_PER_ITERATION=2
endif
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml-cuda-kern.h ggml-cuda-quant.h
$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -Wno-pedantic -c $< -o $@
endif # LLAMA_CUBLAS
endif # LLAMA_CUDA
ifdef LLAMA_CLBLAST
CFLAGS += -DGGML_USE_CLBLAST
@ -275,6 +283,9 @@ $(info I CXXFLAGS: $(CXXFLAGS))
$(info I LDFLAGS: $(LDFLAGS))
$(info I CC: $(CCV))
$(info I CXX: $(CXXV))
ifdef LLAMA_CUDA
$(info I NVCC: $(NVCCV))
endif # LLAMA_CUDA
$(info )
#
@ -284,6 +295,12 @@ $(info )
ggml.o: ggml.c ggml.h ggml-cuda.h
$(CC) $(CFLAGS) -c $< -o $@
# temporary, probably will be added to ggml.c
ggml-backend.o: ggml-backend.c ggml-backend.h ggml.h
$(CC) $(CFLAGS) -c $< -o $@
OBJS += ggml-backend.o
llama.o: llama.cpp ggml.h ggml-cuda.h ggml-metal.h llama.h llama-util.h
$(CXX) $(CXXFLAGS) -c $< -o $@