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

Merge c2c6790f9b into f72f8f22c9

Browse source
This commit is contained in:
Salvador E. Tropea 2023-10-04 09:34:34 +02:00 • committed by GitHub
commit 86c18669f0
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
1 changed files with 6 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

14
ggml.c
View file

@ -1121,12 +1121,11 @@ static void quantize_row_q8_0_reference(const float * restrict x, block_q8_0 * r
static void quantize_row_q8_0(const float * restrict x, void * restrict vy, int k) {
assert(QK8_0 == 32);
assert(k % QK8_0 == 0);
const int nb = k / QK8_0;
block_q8_0 * restrict y = vy;
#if defined(__ARM_NEON)
for (int i = 0; i < nb; i++) {
for (int i = 0; i < k / QK8_0; i++) {
float32x4_t srcv [8];
float32x4_t asrcv[8];
float32x4_t amaxv[8];
@ -1156,7 +1155,7 @@ static void quantize_row_q8_0(const float * restrict x, void * restrict vy, int
}
}
#elif defined(__wasm_simd128__)
for (int i = 0; i < nb; i++) {
for (int i = 0; i < k / QK8_0; i++) {
v128_t srcv [8];
v128_t asrcv[8];
v128_t amaxv[8];
@ -1189,7 +1188,7 @@ static void quantize_row_q8_0(const float * restrict x, void * restrict vy, int
}
}
#elif defined(__AVX2__) || defined(__AVX__)
for (int i = 0; i < nb; i++) {
for (int i = 0; i < k / QK8_0; i++) {
// Load elements into 4 AVX vectors
__m256 v0 = _mm256_loadu_ps( x );
__m256 v1 = _mm256_loadu_ps( x + 8 );
@ -1343,12 +1342,11 @@ static void quantize_row_q8_1_reference(const float * restrict x, block_q8_1 * r
static void quantize_row_q8_1(const float * restrict x, void * restrict vy, int k) {
assert(k % QK8_1 == 0);
const int nb = k / QK8_1;
block_q8_1 * restrict y = vy;
#if defined(__ARM_NEON)
for (int i = 0; i < nb; i++) {
for (int i = 0; i < k / QK8_1; i++) {
float32x4_t srcv [8];
float32x4_t asrcv[8];
float32x4_t amaxv[8];
@ -1384,7 +1382,7 @@ static void quantize_row_q8_1(const float * restrict x, void * restrict vy, int
y[i].s = d * vaddvq_s32(accv);
}
#elif defined(__wasm_simd128__)
for (int i = 0; i < nb; i++) {
for (int i = 0; i < k / QK8_1; i++) {
v128_t srcv [8];
v128_t asrcv[8];
v128_t amaxv[8];
@ -1426,7 +1424,7 @@ static void quantize_row_q8_1(const float * restrict x, void * restrict vy, int
wasm_i32x4_extract_lane(accv, 3));
}
#elif defined(__AVX2__) || defined(__AVX__)
for (int i = 0; i < nb; i++) {
for (int i = 0; i < k / QK8_1; i++) {
// Load elements into 4 AVX vectors
__m256 v0 = _mm256_loadu_ps( x );
__m256 v1 = _mm256_loadu_ps( x + 8 );