ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-02-28 17:14:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

Fix q5_0_r4 and q6_0_r4 also on Zen4

Browse Source
This commit is contained in:
Iwan Kawrakow
2025-11-13 12:49:34 +02:00
parent 64126305da
commit 2396c41ef8
1 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
ggml/src/iqk/iqk_gemm_legacy_quants.cpp
View File

@@ -1324,12 +1324,12 @@ static void mul_mat_q5_0_r4_q8_2(int n, const void * vx, size_t bx, const DataIn
for (int ib = 4*(nb/4); ib < nb; ++ib) {
auto scales = prepare(iq5l[ib], iq5h[ib]);
for (int iy = 0; iy < nrc_y; ++iy) {
auto qy = (const block_q8_1 *)q8.y[iy];
auto qy = (const block_q8_2 *)q8.y[iy];
auto sumi = dot(_mm256_loadu_si256((const __m256i*)qy[ib].qs));
ggml_bf16_t d{qy[ib].d}, s{qy[ib].s};
auto dy = _mm512_set1_ps(GGML_BF16_TO_FP32(d));
auto [d8, m8] = ScaleHelperQ8_2::prepare1(qy + ib);
auto dy = _mm512_set1_ps(d8);
acc[2*iy+0] = _mm512_fmadd_ps(_mm512_mul_ps(scales, dy), _mm512_cvtepi32_ps(sumi), acc[2*iy+0]);
acc[2*iy+1] = _mm512_fmadd_ps(scales, _mm512_set1_ps(GGML_BF16_TO_FP32(s)), acc[2*iy+1]);
acc[2*iy+1] = _mm512_fmadd_ps(scales, _mm512_set1_ps(m8), acc[2*iy+1]);
}
}
for (int iy = 0; iy < nrc_y; ++iy) {
@@ -1494,12 +1494,12 @@ static void mul_mat_q6_0_r4_q8_2(int n, const void * vx, size_t bx, const DataIn
for (int ib = 4*(nb/4); ib < nb; ++ib) {
auto scales = prepare(iq6l[ib], iq6h[ib]);
for (int iy = 0; iy < nrc_y; ++iy) {
auto qy = (const block_q8_1 *)q8.y[iy];
auto qy = (const block_q8_2 *)q8.y[iy];
auto sumi = dot(_mm256_loadu_si256((const __m256i*)qy[ib].qs));
ggml_bf16_t d{qy[ib].d}, s{qy[ib].s};
auto dy = _mm512_set1_ps(GGML_BF16_TO_FP32(d));
auto [d8, m8] = ScaleHelperQ8_2::prepare1(qy + ib);
auto dy = _mm512_set1_ps(d8);
acc[2*iy+0] = _mm512_fmadd_ps(_mm512_mul_ps(scales, dy), _mm512_cvtepi32_ps(sumi), acc[2*iy+0]);
acc[2*iy+1] = _mm512_fmadd_ps(scales, _mm512_set1_ps(GGML_BF16_TO_FP32(s)), acc[2*iy+1]);
acc[2*iy+1] = _mm512_fmadd_ps(scales, _mm512_set1_ps(m8), acc[2*iy+1]);
}
}
for (int iy = 0; iy < nrc_y; ++iy) {