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

Improve cuBLAS performance by using a memory pool (#1094)

Browse source 
* Improve cuBLAS performance by using a memory pool

* Move cuda specific definitions to ggml-cuda.h/cu

* Add CXX flags to nvcc

* Change memory pool synchronization mechanism to a spin lock
General code cleanup
This commit is contained in:
slaren 2023-04-21 21:59:17 +02:00 • committed by GitHub
parent 25d7abbd1f
commit 50cb666b8a
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
4 changed files with 170 additions and 109 deletions
Download patch file Download diff file Expand all files Collapse all files

10
Makefile
View file

@ -101,11 +101,13 @@ ifdef LLAMA_OPENBLAS
LDFLAGS += -lopenblas
endif
ifdef LLAMA_CUBLAS
CFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include
LDFLAGS += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64
OBJS += ggml-cuda.o
CFLAGS += -DGGML_USE_CUBLAS -I/usr/local/cuda/include
LDFLAGS += -lcublas -lculibos -lcudart -lcublasLt -lpthread -ldl -lrt -L/usr/local/cuda/lib64
OBJS += ggml-cuda.o
NVCC = nvcc
NVCCFLAGS = --forward-unknown-to-host-linker -arch=native
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
nvcc -arch=native -c -o $@ $<
$(NVCC) $(NVCCFLAGS) $(CXXFLAGS) -c $< -o $@
endif
ifdef LLAMA_GPROF
CFLAGS += -pg