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

makefile rewrite

Browse source
This commit is contained in:
Henri Vasserman 2023-08-16 21:28:54 +03:00
parent 3de6a9aed2
commit bbbc0ce717
No known key found for this signature in database
GPG key ID: 2995FC0F58B1A986
1 changed files with 17 additions and 14 deletions
Download patch file Download diff file Expand all files Collapse all files

27
Makefile
View file

@ -282,23 +282,26 @@ endif # LLAMA_CLBLAST
ifdef LLAMA_HIPBLAS ifdef LLAMA_HIPBLAS
ROCM_PATH ?= /opt/rocm ROCM_PATH ?= /opt/rocm
CC := $(ROCM_PATH)/llvm/bin/clang HIPCC ?= $(ROCM_PATH)/bin/hipcc
CXX := $(ROCM_PATH)/llvm/bin/clang++ GPU_TARGETS ?= $(shell $(ROCM_PATH)/llvm/bin/amdgpu-arch)
GPU_TARGETS ?= gfx803 gfx900 gfx906 gfx908 gfx90a gfx1030 gfx1100
LLAMA_CUDA_DMMV_X ?= 32 LLAMA_CUDA_DMMV_X ?= 32
LLAMA_CUDA_MMV_Y ?= 1 LLAMA_CUDA_MMV_Y ?= 1
LLAMA_CUDA_KQUANTS_ITER ?= 2 LLAMA_CUDA_KQUANTS_ITER ?= 2
CFLAGS += -DGGML_USE_HIPBLAS -DGGML_USE_CUBLAS $(shell $(ROCM_PATH)/bin/hipconfig -C) CFLAGS += -DGGML_USE_HIPBLAS -DGGML_USE_CUBLAS
CXXFLAGS += -DGGML_USE_HIPBLAS -DGGML_USE_CUBLAS $(shell $(ROCM_PATH)/bin/hipconfig -C) CXXFLAGS += -DGGML_USE_HIPBLAS -DGGML_USE_CUBLAS
LDFLAGS += -L$(ROCM_PATH)/lib -Wl,-rpath=$(ROCM_PATH)/lib -lhipblas -lamdhip64 -lrocblas LDFLAGS += -L$(ROCM_PATH)/lib -Wl,-rpath=$(ROCM_PATH)/lib -fuse-ld=lld
LDFLAGS += -lhipblas -lamdhip64 -lrocblas
HIPFLAGS += $(addprefix --offload-arch=,$(GPU_TARGETS))
HIPFLAGS += -DGGML_CUDA_DMMV_X=$(LLAMA_CUDA_DMMV_X)
HIPFLAGS += -DGGML_CUDA_MMV_Y=$(LLAMA_CUDA_MMV_Y)
HIPFLAGS += -DK_QUANTS_PER_ITERATION=$(LLAMA_CUDA_KQUANTS_ITER)
HIPFLAGS += -DCC_TURING=1000000000
ifdef LLAMA_CUDA_FORCE_DMMV
HIPFLAGS += -DGGML_CUDA_FORCE_DMMV
endif # LLAMA_CUDA_FORCE_DMMV
OBJS += ggml-cuda.o OBJS += ggml-cuda.o
ggml-cuda.o: CXXFLAGS += $(addprefix --offload-arch=,$(GPU_TARGETS))
ggml-cuda.o: CXXFLAGS += -DGGML_CUDA_DMMV_X=$(LLAMA_CUDA_DMMV_X)
ggml-cuda.o: CXXFLAGS += -DGGML_CUDA_MMV_Y=$(LLAMA_CUDA_MMV_Y)
ggml-cuda.o: CXXFLAGS += -DK_QUANTS_PER_ITERATION=$(LLAMA_CUDA_KQUANTS_ITER)
ggml-cuda.o: CXXFLAGS += -DCC_TURING=1000000000
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
$(CXX) $(CXXFLAGS) -x hip -c -o $@ $< $(HIPCC) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
endif # LLAMA_HIPBLAS endif # LLAMA_HIPBLAS
ifdef LLAMA_METAL ifdef LLAMA_METAL