ExternalVendorCode/whisper.cpp
1
0
Fork 0
mirror of https://github.com/ggerganov/whisper.cpp.git synced 2024-12-20 05:07:52 +00:00
Code Issues Actions Packages Projects Releases Wiki Activity

ggml : optimize llamafile cpu matrix multiplication for ppc64le (llama/10156)

Browse Source 
This change upstreams llamafile's cpu matrix
multiplication kernels for ppc64le using MMA
builtins for FP32 datatype.

This change results in a consistent 90%
improvement in input processing time, and 20%
to 80% improvement in output processing time,
across various batch sizes.

The patch is tested with Meta-Lllama-3-8B,
Mistral-7B, Llama-2-7B-chat-hf models on a
IBM POWER10 machine.

Signed-off-by: Amrita H S <amritahs@linux.vnet.ibm.com>
This commit is contained in:
amritahs-ibm 2024-11-09 12:47:50 +05:30 committed by Georgi Gerganov
parent 9f67aab211
commit b7b38f7d68
1 changed files with 7 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

9
ggml/src/CMakeLists.txt
View File

@ -1265,8 +1265,13 @@ elseif (CMAKE_OSX_ARCHITECTURES STREQUAL "x86_64" OR CMAKE_GENERATOR_PLATFORM_LW
endif()
elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "ppc64")
message(STATUS "PowerPC detected")
if (${CMAKE_SYSTEM_PROCESSOR} MATCHES "ppc64le")
list(APPEND ARCH_FLAGS -mcpu=powerpc64le)
execute_process(COMMAND bash -c "grep POWER10 /proc/cpuinfo | head -n 1"
OUTPUT_VARIABLE POWER10_M)
string(FIND ${POWER10_M} "POWER10" substring_index)
if(${substring_index} GREATER_EQUAL 0)
list(APPEND ARCH_FLAGS -mcpu=power10)
elseif (${CMAKE_SYSTEM_PROCESSOR} MATCHES "ppc64le")
list(APPEND ARCH_FLAGS -mcpu=powerpc64le)
else()
list(APPEND ARCH_FLAGS -mcpu=native -mtune=native)
#TODO: Add targets for Power8/Power9 (Altivec/VSX) and Power10(MMA) and query for big endian systems (ppc64/le/be)