Skip to content

Small (and unreliable) AVX2 ggml_vec_dot_q4_K_q8_K improvement - remove instruction dependency #2819

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

Merged
merged 1 commit into from
Aug 28, 2023
Merged

Small (and unreliable) AVX2 ggml_vec_dot_q4_K_q8_K improvement - remove instruction dependency #2819

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
4 changes: 2 additions & 2 deletions k_quants.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2694,13 +2694,13 @@ void ggml_vec_dot_q4_K_q8_K(const int n, float * restrict s, const void * restri
const __m256i q8l = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
__m256i p16l = _mm256_maddubs_epi16(q4l, q8l);
p16l = _mm256_madd_epi16(scale_l, p16l);
sumi = _mm256_add_epi32(sumi, p16l);

const __m256i q8h = _mm256_loadu_si256((const __m256i*)q8); q8 += 32;
__m256i p16h = _mm256_maddubs_epi16(q4h, q8h);
p16h = _mm256_madd_epi16(scale_h, p16h);
sumi = _mm256_add_epi32(sumi, p16h);
const __m256i sumj = _mm256_add_epi32(p16l, p16h);

sumi = _mm256_add_epi32(sumi, sumj);
}

__m256 vd = _mm256_set1_ps(d);
Expand Down