ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-03 21:09:39 +00:00
Code Issues Packages Projects Releases Wiki Activity

iqk_mul_mat: better srategy when nrc_y not divisible by ny

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-10-01 08:12:29 +03:00
parent fd20638bbc
commit 5f3f3bb09e
1 changed files with 31 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

39
ggml/src/iqk/iqk_mul_mat.cpp
View File

@@ -107,16 +107,39 @@ struct MulMat {
while (!funcs[ny-1] && ny > 0) --ny;
int n_step = (nrc_y - info.cur_y)/ny;
if (n_step > 0) {
for (int ix = 0; ix < nrc_x; ix += k_x_step) {
auto this_info = info;
this_info.s += ix;
int this_nrc_x = ix + k_x_step <= nrc_x ? k_x_step : nrc_x - ix;
for (int iy = 0; iy < n_step; ++iy) {
funcs[ny-1](n, (const void *)((const char *)vx + ix*bx), bx, this_info, this_nrc_x);
this_info.cur_y += ny;
if (n_step*ny != nrc_y) {
++n_step;
int ny1 = nrc_y/n_step;
int ny2 = ny1 + 1;
int my1 = n_step*ny2 - nrc_y;
int my2 = n_step - my1;
for (int ix = 0; ix < nrc_x; ix += k_x_step) {
auto this_info = info;
this_info.s += ix;
int this_nrc_x = ix + k_x_step <= nrc_x ? k_x_step : nrc_x - ix;
for (int iy = 0; iy < my1; ++iy) {
funcs[ny1-1](n, (const void *)((const char *)vx + ix*bx), bx, this_info, this_nrc_x);
this_info.cur_y += ny1;
}
for (int iy = 0; iy < my2; ++iy) {
funcs[ny2-1](n, (const void *)((const char *)vx + ix*bx), bx, this_info, this_nrc_x);
this_info.cur_y += ny2;
}
}
info.cur_y += nrc_y;
}
else {
for (int ix = 0; ix < nrc_x; ix += k_x_step) {
auto this_info = info;
this_info.s += ix;
int this_nrc_x = ix + k_x_step <= nrc_x ? k_x_step : nrc_x - ix;
for (int iy = 0; iy < n_step; ++iy) {
funcs[ny-1](n, (const void *)((const char *)vx + ix*bx), bx, this_info, this_nrc_x);
this_info.cur_y += ny;
}
}
info.cur_y += ny * n_step;
}
info.cur_y += ny * n_step;
}
int n_left = nrc_y - info.cur_y;
if (n_left > 0) {