vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

iq3_s sllv can be safely replaced with sse multiply

Browse source
This commit is contained in:
netrunnereve 2024-06-17 21:43:24 -04:00
parent b57187fb37
commit 29e2a96d28
1 changed files with 6 additions and 11 deletions
Download patch file Download diff file Expand all files Collapse all files

17
ggml-quants.c
View file

@ -10798,6 +10798,8 @@ void ggml_vec_dot_iq3_s_q8_K (int n, float * restrict s, size_t bs, const void *
const __m128i mask2_0 = _mm_loadu_si128((const __m128i*)k_mask2);
const __m128i mask2_1 = _mm_loadu_si128((const __m128i*)k_mask2 + 1);
const __m128i idx_mul_0 = _mm_set_epi32(32, 64, 128, 256);
const __m128i idx_mul_1 = _mm_set_epi32(2, 4, 8, 16);
const __m128i idx_mask = _mm_set1_epi32(256);
typedef union {
@ -10831,17 +10833,10 @@ void ggml_vec_dot_iq3_s_q8_K (int n, float * restrict s, size_t bs, const void *
idx.vec[2] = _mm_set1_epi32(qh[ib32+1]);
idx.vec[3] = idx.vec[2];
// AVX has no sllv so we have to do this
for (int j = 0; j < 2; ++j) {
for (int k = 0; k < 8; ++k) {
idx.index[j*8+k] <<= 8 - k;
}
}
idx.vec[0] = _mm_and_si128(idx.vec[0], idx_mask);
idx.vec[1] = _mm_and_si128(idx.vec[1], idx_mask);
idx.vec[2] = _mm_and_si128(idx.vec[2], idx_mask);
idx.vec[3] = _mm_and_si128(idx.vec[3], idx_mask);
idx.vec[0] = _mm_and_si128(_mm_mullo_epi32(idx.vec[0], idx_mul_0), idx_mask);
idx.vec[1] = _mm_and_si128(_mm_mullo_epi32(idx.vec[1], idx_mul_1), idx_mask);
idx.vec[2] = _mm_and_si128(_mm_mullo_epi32(idx.vec[2], idx_mul_0), idx_mask);
idx.vec[3] = _mm_and_si128(_mm_mullo_epi32(idx.vec[3], idx_mul_1), idx_mask);
idx.vec[0] = _mm_or_si128(idx.vec[0], _mm_cvtepi16_epi32(idx_l_0));
idx.vec[1] = _mm_or_si128(idx.vec[1], _mm_cvtepi16_epi32(_mm_srli_si128(idx_l_0, 8)));