aditya/llama.cpp
1
0
Fork 0
mirror of https://git.adityakumar.xyz/llama.cpp.git synced 2024-11-09 23:29:44 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

add avx2 for dot_q8_0_q8_0, 2x faster than scalar (#1211)

Browse source
This commit is contained in:
Yann Follet 2023-04-28 19:59:48 +08:00 committed by GitHub
parent 0b2da20538
commit 04aaae1d79
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 18 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

18
ggml.c
View file

@ -3626,6 +3626,24 @@ static void ggml_vec_dot_q8_0_q8_0(const int n, float * restrict s, const void *
}
*s = vaddvq_f32(sumv0) + vaddvq_f32(sumv1);
#elif defined(__AVX2__)
// Initialize accumulator with zeros
__m256 acc = _mm256_setzero_ps();
// Main loop
for (int i = 0; i < nb; ++i) {
// Compute combined scale for the block
const __m256 d = _mm256_mul_ps( _mm256_broadcast_ss( &x[i].d ), _mm256_broadcast_ss( &y[i].d ) );
__m256i bx = _mm256_loadu_si256((const __m256i *)x[i].qs);
__m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
const __m256 q = mul_sum_i8_pairs_float(bx, by);
// Multiply q with scale and accumulate
acc = _mm256_fmadd_ps( d, q, acc );
}
*s = hsum_float_8(acc);
#else
// scalar
float sumf = 0.0;