Skip to content

ggml-cpu: fuse the row-max scan into the mask pass in ggml_compute_forward_soft_max_f32 #16

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
codspeed-hq wants to merge 1 commit into
base: master
Choose a base branch
from
Open

ggml-cpu: fuse the row-max scan into the mask pass in ggml_compute_forward_soft_max_f32 #16

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
13 changes: 10 additions & 3 deletions ggml/src/ggml-cpu/ops.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5516,16 +5516,26 @@ static void ggml_compute_forward_soft_max_f32(

ggml_vec_cpy_f32 (ne00, wp, sp);
ggml_vec_scale_f32(ne00, wp, scale);

// The additive mask is applied in a full pass over `wp`, so the
// row maximum (needed next for the numerically-stable soft-max)
// is accumulated in that same pass instead of re-reading the
// whole row afterwards with a separate ggml_vec_max_f32 scan.
float max = -INFINITY;
if (mp_f32) {
if (use_f16) {
for (int i = 0; i < ne00; ++i) {
wp[i] += slope*GGML_CPU_FP16_TO_FP32(mp_f16[i]);
max = MAX(max, wp[i]);
}
} else {
for (int i = 0; i < ne00; ++i) {
wp[i] += slope*mp_f32[i];
max = MAX(max, wp[i]);
}
}
} else {
ggml_vec_max_f32(ne00, &max, wp);
}

#ifndef NDEBUG
Expand All @@ -5535,9 +5545,6 @@ static void ggml_compute_forward_soft_max_f32(
}
#endif // NDEBUG

float max = -INFINITY;
ggml_vec_max_f32(ne00, &max, wp);

// if we have sinks, make a correction as if they were included in the softmax
if (sk) {
max = MAX(max, sk[i02]);
Expand Down
Loading