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

ggml : fixes for AVXVNNI instruction set with MSVC and Clang (#11027)

Browse source 
* Fixes for clang AVX VNNI

* enable AVX VNNI and alder lake build for MSVC

* Apply suggestions from code review

---------

Co-authored-by: slaren <slarengh@gmail.com>
This commit is contained in:
Srihari-mcw 2024-12-31 19:53:33 +05:30 • committed by GitHub
parent 45095a61bf
commit 0827b2c1da
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
5 changed files with 15 additions and 7 deletions
Download patch file Download diff file Expand all files Collapse all files

6
ggml/src/ggml-cpu/ggml-cpu-quants.c
View file

@ -103,10 +103,14 @@ static inline __m256 sum_i16_pairs_float(const __m256i x) {
}
static inline __m256 mul_sum_us8_pairs_float(const __m256i ax, const __m256i sy) {
#if defined(__AVXVNNI__) || (defined(__AVX512VNNI__) && defined(__AVX512VL__))
#if defined(__AVX512VNNI__) && defined(__AVX512VL__)
const __m256i zero = _mm256_setzero_si256();
const __m256i summed_pairs = _mm256_dpbusd_epi32(zero, ax, sy);
return _mm256_cvtepi32_ps(summed_pairs);
#elif defined(__AVXVNNI__)
const __m256i zero = _mm256_setzero_si256();
const __m256i summed_pairs = _mm256_dpbusd_avx_epi32(zero, ax, sy);
return _mm256_cvtepi32_ps(summed_pairs);
#else
// Perform multiplication and create 16-bit values
const __m256i dot = _mm256_maddubs_epi16(ax, sy);