Skip to content

Commit

Permalink
Merge 'origin/master' into hipblas
Browse files Browse the repository at this point in the history
  • Loading branch information
SlyEcho committed Aug 9, 2023
2 parents 4024f91 + 25d43e0 commit 610ba4c
Show file tree
Hide file tree
Showing 4 changed files with 682 additions and 396 deletions.
1 change: 0 additions & 1 deletion CMakeLists.txt
Original file line number Diff line number Diff line change
Expand Up @@ -375,7 +375,6 @@ if (LLAMA_HIPBLAS)
message(STATUS "HIP and hipBLAS found")
add_compile_definitions(GGML_USE_HIPBLAS GGML_USE_CUBLAS)
add_library(ggml-rocm OBJECT ggml-cuda.cu ggml-cuda.h)
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_MMQ_Y=${LLAMA_CUDA_MMQ_Y})
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_DMMV_X=${LLAMA_CUDA_DMMV_X})
target_compile_definitions(ggml-rocm PRIVATE GGML_CUDA_MMV_Y=${LLAMA_CUDA_MMV_Y})
target_compile_definitions(ggml-rocm PRIVATE K_QUANTS_PER_ITERATION=${LLAMA_CUDA_KQUANTS_ITER})
Expand Down
7 changes: 0 additions & 7 deletions Makefile
Original file line number Diff line number Diff line change
Expand Up @@ -253,11 +253,6 @@ ifdef LLAMA_CUDA_KQUANTS_ITER
else
NVCCFLAGS += -DK_QUANTS_PER_ITERATION=2
endif
ifdef LLAMA_CUDA_MMQ_Y
NVCCFLAGS += -DGGML_CUDA_MMQ_Y=$(LLAMA_CUDA_MMQ_Y)
else
NVCCFLAGS += -DGGML_CUDA_MMQ_Y=64
endif # LLAMA_CUDA_MMQ_Y
#ifdef LLAMA_CUDA_CUBLAS
# NVCCFLAGS += -DGGML_CUDA_CUBLAS
#endif # LLAMA_CUDA_CUBLAS
Expand Down Expand Up @@ -292,7 +287,6 @@ ifdef LLAMA_HIPBLAS
GPU_TARGETS ?= gfx803 gfx900 gfx906 gfx908 gfx90a gfx1030 gfx1100
LLAMA_CUDA_DMMV_X ?= 32
LLAMA_CUDA_MMV_Y ?= 1
LLAMA_CUDA_MMQ_Y ?= 64
LLAMA_CUDA_KQUANTS_ITER ?= 2
CFLAGS += -DGGML_USE_HIPBLAS -DGGML_USE_CUBLAS $(shell $(ROCM_PATH)/bin/hipconfig -C)
CXXFLAGS += -DGGML_USE_HIPBLAS -DGGML_USE_CUBLAS $(shell $(ROCM_PATH)/bin/hipconfig -C)
Expand All @@ -301,7 +295,6 @@ ifdef LLAMA_HIPBLAS
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 += -DGGML_CUDA_MMQ_Y=$(LLAMA_CUDA_MMQ_Y)
ggml-cuda.o: CXXFLAGS += -DK_QUANTS_PER_ITERATION=$(LLAMA_CUDA_KQUANTS_ITER)
ggml-cuda.o: ggml-cuda.cu ggml-cuda.h
$(CXX) $(CXXFLAGS) -x hip -c -o $@ $<
Expand Down
2 changes: 0 additions & 2 deletions README.md
Original file line number Diff line number Diff line change
Expand Up @@ -406,7 +406,6 @@ Building the program with BLAS support may lead to some performance improvements
--->
| Option | Legal values | Default | Description |
|-------------------------|------------------------|---------|-------------|
| LLAMA_CUDA_MMQ_Y | Positive integer >= 32 | 64 | Tile size in y direction when using the custom CUDA kernels for prompt processing. Higher values can be faster depending on the amount of shared memory available. Power of 2 heavily recommended. |
| LLAMA_CUDA_FORCE_DMMV | Boolean | false | Force the use of dequantization + matrix vector multiplication kernels instead of using kernels that do matrix vector multiplication on quantized data. By default the decision is made based on compute capability (MMVQ for 6.1/Pascal/GTX 1000 or higher). Does not affect k-quants. |
| LLAMA_CUDA_DMMV_X | Positive integer >= 32 | 32 | Number of values in x direction processed by the CUDA dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants. |
| LLAMA_CUDA_MMV_Y | Positive integer | 1 | Block size in y direction for the CUDA mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended. Does not affect k-quants. |
Expand Down Expand Up @@ -438,7 +437,6 @@ Building the program with BLAS support may lead to some performance improvements
| Option | Legal values | Default | Description |
|-------------------------|------------------------|---------|-------------|
| LLAMA_CUDA_MMQ_Y | Positive integer >= 32 | 64 | Tile size in y direction when using the custom HIP kernels for prompt processing. Higher values can be faster depending on the amount of shared memory available. Power of 2 heavily recommended. |
| LLAMA_CUDA_DMMV_X | Positive integer >= 32 | 32 | Number of values in x direction processed by the HIP dequantization + matrix vector multiplication kernel per iteration. Increasing this value can improve performance on fast GPUs. Power of 2 heavily recommended. Does not affect k-quants. |
| LLAMA_CUDA_MMV_Y | Positive integer | 1 | Block size in y direction for the HIP mul mat vec kernels. Increasing this value can improve performance on fast GPUs. Power of 2 recommended. Does not affect k-quants. |
| LLAMA_CUDA_KQUANTS_ITER | 1 or 2 | 2 | Number of values processed per iteration and per HIP thread for Q2_K and Q6_K quantization formats. Setting this value to 1 can improve performance for slow GPUs. |
Expand Down
Loading

0 comments on commit 610ba4c

Please sign in to comment.