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

Merge a952716d35 into 8c00b7a6ff

Browse source
This commit is contained in:
Qingyou Meng 2023-09-15 12:12:51 -04:00 • committed by GitHub
commit 8e7eca7e70
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 10 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

11
ggml.c
View file

@ -616,6 +616,8 @@ static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
#define MM256_SET_M128I(a, b) _mm256_insertf128_si256(_mm256_castsi128_si256(b), (a), 1)
#if defined(__AVX__) || defined(__AVX2__) || defined(__AVX512F__) || defined(__SSSE3__)
#if !__ARM_NEON && !__AVX2__ && (__AVX__ || __SSSE3__)
// multiply int8_t, add results pairwise twice
static inline __m128i mul_sum_i8_pairs(const __m128i x, const __m128i y) {
// Get absolute values of x vectors
@ -627,6 +629,7 @@ static inline __m128i mul_sum_i8_pairs(const __m128i x, const __m128i y) {
const __m128i ones = _mm_set1_epi16(1);
return _mm_madd_epi16(ones, dot);
}
#endif
#if __AVX__ || __AVX2__ || __AVX512F__
// horizontally add 8 floats
@ -647,6 +650,7 @@ static inline int hsum_i32_8(const __m256i a) {
return _mm_cvtsi128_si32(_mm_add_epi32(sum64, hi32));
}
#if !__AVX2__ && __AVX__
// horizontally add 4 int32_t
static inline int hsum_i32_4(const __m128i a) {
const __m128i hi64 = _mm_unpackhi_epi64(a, a);
@ -654,6 +658,7 @@ static inline int hsum_i32_4(const __m128i a) {
const __m128i hi32 = _mm_shuffle_epi32(sum64, _MM_SHUFFLE(2, 3, 0, 1));
return _mm_cvtsi128_si32(_mm_add_epi32(sum64, hi32));
}
#endif
#if defined(__AVX2__) || defined(__AVX512F__)
// spread 32 bits to 32 bytes { 0x00, 0xFF }
@ -713,6 +718,7 @@ static inline __m256 mul_sum_i8_pairs_float(const __m256i x, const __m256i y) {
#endif
}
/*
static inline __m128i packNibbles( __m256i bytes )
{
// Move bits within 16-bit lanes from 0000_abcd_0000_efgh into 0000_0000_abcd_efgh
@ -733,6 +739,7 @@ static inline __m128i packNibbles( __m256i bytes )
return _mm_packus_epi16( r0, r1 );
#endif
}
*/
#elif defined(__AVX__)
// spread 32 bits to 32 bytes { 0x00, 0xFF }
static inline __m256i bytes_from_bits_32(const uint8_t * x) {
@ -801,6 +808,7 @@ static inline __m256 mul_sum_i8_pairs_float(const __m256i x, const __m256i y) {
return sum_i16_pairs_float(doth, dotl);
}
/*
static inline __m128i packNibbles( __m128i bytes1, __m128i bytes2 )
{
// Move bits within 16-bit lanes from 0000_abcd_0000_efgh into 0000_0000_abcd_efgh
@ -816,6 +824,7 @@ static inline __m128i packNibbles( __m128i bytes1, __m128i bytes2 )
return _mm_packus_epi16( bytes1, bytes2);
}
*/
#endif
#elif defined(__SSSE3__)
// horizontally add 4x4 floats
@ -1956,7 +1965,7 @@ ggml_type_traits_t ggml_internal_get_type_traits(enum ggml_type type) {
const __m128 t0 = _mm_add_ps(_mm256_castps256_ps128(x[0]), \
_mm256_extractf128_ps(x[0], 1)); \
const __m128 t1 = _mm_hadd_ps(t0, t0); \
res = _mm_cvtss_f32(_mm_hadd_ps(t1, t1)); \
res = (ggml_float)_mm_cvtss_f32(_mm_hadd_ps(t1, t1)); \
}
// TODO: is this optimal ?