EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

CUDA: fix --split-mode row for MMQ (#13323)

Browse Source
This commit is contained in:
Johannes Gäßler
2025-05-06 08:36:46 +02:00
committed by GitHub
parent a7366faa5b
commit 15a28ec8c7
2 changed files with 30 additions and 30 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
ggml/src/ggml-cuda/mmq.cu
View File

@@ -128,7 +128,7 @@ void ggml_cuda_mul_mat_q(
const mmq_args args = {
src0_d, src0->type, (const int *) src1_q8_1.ptr, nullptr, nullptr, dst_d,
ne00, ne01, ne1, s01, s1,
ne00, ne01, ne1, s01, ne11, s1,
ne02, ne12, s02, s12, s2,
ne03, ne13, s03, s13, s3,
use_stream_k};
@@ -212,7 +212,7 @@ void ggml_cuda_mul_mat_q(
// Note that ne02 is used instead of ne12 because the number of y channels determines the z dimension of the CUDA grid.
const mmq_args args = {
src0_d, src0->type, (const int *) src1_q8_1.ptr, ids_dst_dev, expert_bounds_dev, dst_d,
ne00, ne01, ne_get_rows, s01, s1,
ne00, ne01, ne_get_rows, s01, ne_get_rows, s1,
ne02, ne02, s02, s12, s2,
ne03, ne13, s03, s13, s3,
use_stream_k};
@@ -251,7 +251,7 @@ void ggml_cuda_op_mul_mat_q(
ggml_cuda_highest_compiled_arch(cc) >= GGML_CUDA_CC_VOLTA && src1_ncols == ne11;
const mmq_args args = {
src0_dd_i, src0->type, (const int *) src1_ddq_i, nullptr, nullptr, dst_dd_i,
ne00, row_diff, src1_ncols, stride01, nrows_dst,
ne00, row_diff, src1_ncols, stride01, ne11, nrows_dst,
1, 1, 0, 0, 0,
1, 1, 0, 0, 0,
use_stream_k};