Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-20 01:31:42 +02:00
Code Issues Packages Projects Releases Wiki Activity

KCPP Fetches AMD ROCm Memory without a stick, CC_TURING Gets the Boot, koboldcpp_hipblas.dll Talks To The Hand, and hipBLAS Compiler Finds Its Independence! (#517)

Browse Source 
* AMD ROCm memory fetching and max mem setting

* Update .gitignore with koboldcpp_hipblas.dll

* Update CMakeLists.txt remove CC_TURING for AMD

* separate hipBLAS compiler, update MMV_Y, move CXX/CC print

separate hipBLAS compiler, update MMV_Y value, move the section that prints CXX and CC compiler name
This commit is contained in:
YellowRoseCx
2023-11-05 08:23:18 -06:00
committed by GitHub
parent a62468ec4c
commit e2e5fe56a8
4 changed files with 33 additions and 26 deletions
Download Patch File Download Diff File Expand all files Collapse all files
Makefile
+10 -14
View File
@@ -196,16 +196,13 @@ endif # LLAMA_CUBLAS
ifdef LLAMA_HIPBLAS
ROCM_PATH ?= /opt/rocm
CC := $(ROCM_PATH)/llvm/bin/clang
CXX := $(ROCM_PATH)/llvm/bin/clang++
HCC := $(ROCM_PATH)/llvm/bin/clang
HCXX := $(ROCM_PATH)/llvm/bin/clang++
GPU_TARGETS ?= gfx803 gfx900 gfx906 gfx908 gfx90a gfx1030 gfx1100 $(shell $(ROCM_PATH)/llvm/bin/amdgpu-arch)
LLAMA_CUDA_DMMV_X ?= 32
LLAMA_CUDA_MMV_Y ?= 2
LLAMA_CUDA_MMV_Y ?= 1
LLAMA_CUDA_KQUANTS_ITER ?= 2
HIPFLAGS += -DGGML_USE_HIPBLAS -DGGML_USE_CUBLAS $(shell $(ROCM_PATH)/bin/hipconfig -C)
ifdef LLAMA_CUDA_FORCE_DMMV
HIPFLAGS += -DGGML_CUDA_FORCE_DMMV
endif # LLAMA_CUDA_FORCE_DMMV
HIPLDFLAGS += -L$(ROCM_PATH)/lib -Wl,-rpath=$(ROCM_PATH)/lib -lhipblas -lamdhip64 -lrocblas
HIP_OBJS += ggml-cuda.o ggml_v2-cuda.o ggml_v2-cuda-legacy.o
ggml-cuda.o: HIPFLAGS += $(addprefix --offload-arch=,$(GPU_TARGETS)) \
@@ -221,11 +218,11 @@ ggml_v2-cuda-legacy.o: HIPFLAGS += $(addprefix --offload-arch=,$(GPU_TARGETS)) \
-DGGML_CUDA_MMV_Y=$(LLAMA_CUDA_MMV_Y) \
-DK_QUANTS_PER_ITERATION=$(LLAMA_CUDA_KQUANTS_ITER)
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
$(CXX) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
$(HCXX) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
ggml_v2-cuda.o: otherarch/ggml_v2-cuda.cu otherarch/ggml_v2-cuda.h
$(CXX) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
$(HCXX) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
ggml_v2-cuda-legacy.o: otherarch/ggml_v2-cuda-legacy.cu otherarch/ggml_v2-cuda-legacy.h
$(CXX) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
$(HCXX) $(CXXFLAGS) $(HIPFLAGS) -x hip -c -o $@ $<
endif # LLAMA_HIPBLAS
@@ -259,8 +256,6 @@ ifneq ($(filter armv8%,$(UNAME_M)),)
CFLAGS += -mfp16-format=ieee -mno-unaligned-access
endif
CCV := $(shell $(CC) --version | head -n 1)
CXXV := $(shell $(CXX) --version | head -n 1)
DEFAULT_BUILD =
FAILSAFE_BUILD =
@@ -281,7 +276,7 @@ ifeq ($(OS),Windows_NT)
CUBLAS_BUILD = $(CXX) $(CXXFLAGS) $(CUBLAS_FLAGS) $^ -shared -o $@.dll $(CUBLASLD_FLAGS) $(LDFLAGS)
endif
ifdef LLAMA_HIPBLAS
HIPBLAS_BUILD = $(CXX) $(CXXFLAGS) $(HIPFLAGS) $^ -shared -o $@.dll $(HIPLDFLAGS) $(LDFLAGS)
HIPBLAS_BUILD = $(HCXX) $(CXXFLAGS) $(HIPFLAGS) $^ -shared -o $@.dll $(HIPLDFLAGS) $(LDFLAGS)
endif
else
DEFAULT_BUILD = $(CXX) $(CXXFLAGS) $^ -shared -o $@.so $(LDFLAGS)
@@ -300,7 +295,7 @@ else
CUBLAS_BUILD = $(CXX) $(CXXFLAGS) $(CUBLAS_FLAGS) $^ -shared -o $@.so $(CUBLASLD_FLAGS) $(LDFLAGS)
endif
ifdef LLAMA_HIPBLAS
HIPBLAS_BUILD = $(CXX) $(CXXFLAGS) $(HIPFLAGS) $^ -shared -o $@.so $(HIPLDFLAGS) $(LDFLAGS)
HIPBLAS_BUILD = $(HCXX) $(CXXFLAGS) $(HIPFLAGS) $^ -shared -o $@.so $(HIPLDFLAGS) $(LDFLAGS)
endif
ifndef LLAMA_OPENBLAS
@@ -314,7 +309,8 @@ else
endif
endif
CCV := $(shell $(CC) --version | head -n 1)
CXXV := $(shell $(CXX) --version | head -n 1)
#
# Print build information