- Put the whole line into defined()

- Use __SSSE3__ instead of __SSE__
This commit is contained in:
3ooabkhxtn 2023-05-12 13:59:20 +00:00
parent 70c2b6c696
commit fc26f54e74

4
ggml.c
View file

@ -472,7 +472,7 @@ static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
// quantization // quantization
// //
#if __AVX__ || __AVX2__ || __AVX512F__ || defined(__SSE3__) #if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__)
// multiply int8_t, add results pairwise twice // multiply int8_t, add results pairwise twice
static inline __m128i mul_sum_i8_pairs(const __m128i x, const __m128i y) { static inline __m128i mul_sum_i8_pairs(const __m128i x, const __m128i y) {
// Get absolute values of x vectors // Get absolute values of x vectors
@ -2142,7 +2142,7 @@ static void ggml_vec_dot_q4_0_q8_0(const int n, float * restrict s, const void *
} }
*s = hsum_float_8(acc); *s = hsum_float_8(acc);
#elif defined(__SSE3__) #elif defined(__SSSE3__)
// set constants // set constants
const __m128i lowMask = _mm_set1_epi8(0xF); const __m128i lowMask = _mm_set1_epi8(0xF);
const __m128i off = _mm_set1_epi8(8); const __m128i off = _mm_set1_epi8(8);