CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-29 08:41:22 +00:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit 'refs/pull/14417/head' of github.com:ggerganov/llama.cpp into xsn/ggml_scale_bias

Browse Source
This commit is contained in:
Xuan Son Nguyen
2025-07-09 14:13:30 +02:00
parent 60b03ff968 ebbad7796d
commit a67685e0e1
2 changed files with 11 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
ggml/src/ggml-cpu/ops.cpp
View File

@@ -4670,17 +4670,17 @@ static void ggml_compute_forward_scale_f32(
for (int i1 = ir0; i1 < ir1; i1++) { for (int i1 = ir0; i1 < ir1; i1++) {
if (dst->data != src0->data) { if (dst->data != src0->data) {
// src0 is same shape as dst => same indices // src0 is same shape as dst => same indices
// TODO: add x parameter to ggml_vec_scale_f32 and remove this memcpy
memcpy((char *)dst->data + i1*nb1, (char *)src0->data + i1*nb01, nc * sizeof(float)); memcpy((char *)dst->data + i1*nb1, (char *)src0->data + i1*nb01, nc * sizeof(float));
} }
ggml_vec_scale_f32(nc, (float *) ((char *) dst->data + i1*nb1), s); ggml_vec_scale_f32(nc, (float *) ((char *) dst->data + i1*nb1), s);
} }
} else { } else {
for (int i1 = ir0; i1 < ir1; i1++) { for (int i1 = ir0; i1 < ir1; i1++) {
if (dst->data != src0->data) { ggml_vec_mad1_f32(nc,
// src0 is same shape as dst => same indices (float *) ((char *) dst->data + i1*nb1),
memcpy((char *)dst->data + i1*nb1, (char *)src0->data + i1*nb01, nc * sizeof(float)); (float *) ((char *) src0->data + i1*nb1),
} s, b);
ggml_vec_mad1_f32(nc, (float *) ((char *) dst->data + i1*nb1), s, b);
} }
} }
} }

12
ggml/src/ggml-cpu/vec.h
View File

@@ -351,14 +351,14 @@ inline static void ggml_vec_mad_f32_unroll(const int n, const int xs, const int
#endif #endif
} }
inline static void ggml_vec_mad1_f32(const int n, float * y, const float s, const float b) { inline static void ggml_vec_mad1_f32(const int n, float * y, const float * x, const float s, const float b) {
#if defined(GGML_USE_ACCELERATE) #if defined(GGML_USE_ACCELERATE)
vDSP_vsmsa(y, 1, &s, &b, y, 1, n); vDSP_vsmsa(x, 1, &s, &b, y, 1, n);
#elif defined(GGML_SIMD) #elif defined(GGML_SIMD)
#if defined(__ARM_FEATURE_SVE) #if defined(__ARM_FEATURE_SVE)
// scalar ; TODO: Write SVE code // scalar ; TODO: Write SVE code
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
y[i] = y[i]*s + b; y[i] = x[i]*s + b;
} }
#else #else
const int np = (n & ~(GGML_F32_STEP - 1)); const int np = (n & ~(GGML_F32_STEP - 1));
@@ -370,7 +370,7 @@ inline static void ggml_vec_mad1_f32(const int n, float * y, const float s, cons
for (int i = 0; i < np; i += GGML_F32_STEP) { for (int i = 0; i < np; i += GGML_F32_STEP) {
for (int j = 0; j < GGML_F32_ARR; j++) { for (int j = 0; j < GGML_F32_ARR; j++) {
ay[j] = GGML_F32_VEC_LOAD(y + i + j*GGML_F32_EPR); ay[j] = GGML_F32_VEC_LOAD(x + i + j*GGML_F32_EPR);
ay[j] = GGML_F32_VEC_FMA(ay[j], vs, vb); ay[j] = GGML_F32_VEC_FMA(ay[j], vs, vb);
GGML_F32_VEC_STORE(y + i + j*GGML_F32_EPR, ay[j]); GGML_F32_VEC_STORE(y + i + j*GGML_F32_EPR, ay[j]);
@@ -379,13 +379,13 @@ inline static void ggml_vec_mad1_f32(const int n, float * y, const float s, cons
// leftovers // leftovers
for (int i = np; i < n; ++i) { for (int i = np; i < n; ++i) {
y[i] = y[i]*s + b; y[i] = x[i]*s + b;
} }
#endif #endif
#else #else
// scalar // scalar
for (int i = 0; i < n; ++i) { for (int i = 0; i < n; ++i) {
y[i] = y[i]*s + b; y[i] = x[i]*s + b;
} }
#endif #endif
} }