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

CUDA: refactor mmq, dmmv, mmvq (#7716)

Browse source 
* CUDA: refactor mmq, dmmv, mmvq

* fix out-of-bounds write

* struct for qk, qr, qi

* fix cmake build

* mmq_type_traits
This commit is contained in:
Johannes Gäßler 2024-06-05 16:53:00 +02:00 • committed by GitHub
parent 2b3389677a
commit 7d1a378b8f
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
112 changed files with 1783 additions and 1767 deletions
Download patch file Download diff file Expand all files Collapse all files

1
Makefile
View file

@ -444,6 +444,7 @@ ifdef LLAMA_CUBLAS
endif
OBJS_CUDA_TEMP_INST = $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-wmma*.cu))
OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/mmq*.cu))
ifdef LLAMA_CUDA_FA_ALL_QUANTS
OBJS_CUDA_TEMP_INST += $(patsubst %.cu,%.o,$(wildcard ggml-cuda/template-instances/fattn-vec*.cu))
else