happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

test: disable l2-block prefetch in polynomail approximation

This commit is contained in:
shouyud 2025-12-11 16:59:06 -05:00
parent 2a787a61d1
commit 72339994d4
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
ggml/src/ggml-hexagon/htp/qhcg_approximation.c
View File

@ -120,7 +120,7 @@ int32_t qhcg_approximation(float *restrict input, float *restrict output, uint32
HVX_Vector sline;
HVX_Vector sline_tmp;
HVX_Vector sout;
int32_t block, l2fetch_block;
int32_t block; //l2fetch_block;
int32_t leftover = size & 31;
int32_t vectors_in_rounddown = size / 32;
int32_t leftover_size = leftover * sizeof(float);
@ -241,12 +241,12 @@ int32_t qhcg_approximation(float *restrict input, float *restrict output, uint32
for (int32_t i = vectors_in_rounddown - 1; i > 0; i -= BLOCK_SIZE)
{
block = Q6_R_min_RR(i, BLOCK_SIZE);
l2fetch_block = Q6_R_min_RR(i - L2FETCH_AHEAD, BLOCK_SIZE);
//l2fetch_block = Q6_R_min_RR(i - L2FETCH_AHEAD, BLOCK_SIZE);
if (l2fetch_block > 0)
{
l2fetch(input_v_ptr + L2FETCH_AHEAD, 128, 128, l2fetch_block, 0);
}
// if (l2fetch_block > 0)
// {
// l2fetch(input_v_ptr + L2FETCH_AHEAD, 128, 128, l2fetch_block, 0);
// }
/* Process one vector at a time */
for (int32_t j = 0; j < block; ++j)