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

llamafile : improve sgemm.cpp (#6796)

Browse source 
* llamafile : improve sgemm.cpp

- Re-enable by default
- Fix issue described in #6716
- Make code more abstract, elegant, and maintainable
- Faster handling of weirdly shaped `m` an `n` edge cases

* Address review comments

* Help clang produce fma instructions

* Address review comments
This commit is contained in:
Justine Tunney 2024-04-22 15:00:36 -04:00 • committed by GitHub
parent e931888d50
commit 192090bae4
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 406 additions and 567 deletions
Download patch file Download diff file Expand all files Collapse all files

4
Makefile
View file

@ -384,10 +384,6 @@ ifdef LLAMA_OPENBLAS
MK_LDFLAGS += $(shell pkg-config --libs openblas)
endif # LLAMA_OPENBLAS
# TODO: temporary disable until MoE is fixed
# https://github.com/ggerganov/llama.cpp/pull/6716
LLAMA_NO_LLAMAFILE := 1
ifndef LLAMA_NO_LLAMAFILE
MK_CPPFLAGS += -DGGML_USE_LLAMAFILE
OBJS += sgemm.o