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

Add run_pipeline option of rechunk.

Browse source
This commit is contained in:
Qingtao Li 2024-10-30 10:46:16 +08:00
parent e86c69df8b
commit 080d2ecc56
2 changed files with 6 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

4
ggml/src/CMakeLists.txt
View file

@ -897,6 +897,10 @@ if (GGML_TMAC)
set(GGML_EXTRA_LIBS_PRIVATE ${GGML_EXTRA_LIBS_PRIVATE} t_mac_no_tvm) set(GGML_EXTRA_LIBS_PRIVATE ${GGML_EXTRA_LIBS_PRIVATE} t_mac_no_tvm)
set(GGML_SOURCES_TMAC ${GGML_SOURCES_TMAC} ${TMAC_KERNELS_SOURCE}) set(GGML_SOURCES_TMAC ${GGML_SOURCES_TMAC} ${TMAC_KERNELS_SOURCE})
endif() endif()
if (GGML_TMAC_RECHUNK)
add_compile_definitions(TMAC_RECHUNK)
endif()
else() else()
message(WARNING "TMAC not found") message(WARNING "TMAC not found")
endif() endif()

3
ggml/src/ggml-cpu.c
View file

@ -7577,13 +7577,14 @@ UseGgmlGemm1:;
int64_t dr0 = chunk_size0; int64_t dr0 = chunk_size0;
int64_t dr1 = chunk_size1; int64_t dr1 = chunk_size1;
#if defined(TMAC_RECHUNK)
// Rechunk // Rechunk
if ((nchunk1 == 1) && (nchunk0 > nth * 4)) { if ((nchunk1 == 1) && (nchunk0 > nth * 4)) {
// dr0 should be divisible by chunk_size0 // dr0 should be divisible by chunk_size0
dr0 = (ne0 / (nth * 4) / chunk_size0) * chunk_size0; dr0 = (ne0 / (nth * 4) / chunk_size0) * chunk_size0;
nchunk0 = (nr0 + dr0 - 1) / dr0; nchunk0 = (nr0 + dr0 - 1) / dr0;
} }
#endif
int current_chunk = ith; int current_chunk = ith;