CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-30 08:42:00 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml-cpu: add fp16->fp32 nnpa first

Browse Source 
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo
2025-06-20 18:43:52 +08:00
parent 4a9f60c201
commit 8d4a7987f9
2 changed files with 16 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

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

@@ -62,11 +62,17 @@ struct ggml_compute_params {
#if defined(__s390x__) && defined(__VEC__)
#ifndef __VXE__
#define __VXE__
#endif
#endif // __VXE__
#ifndef __VXE2__
#define __VXE2__
#endif
#endif
#endif // __VXE2__
#endif // __s390x__ && __VEC__
#if defined(__s390x__) && defined(GGML_NNPA)
#ifndef __NNPA__
#define __NNPA__
#endif // __NNPA__
#endif // __s390x__ && GGML_NNPA
#if defined(__ARM_FEATURE_SVE)
#include <sys/prctl.h>

8
ggml/src/ggml-cpu/simd-mappings.h
View File

@@ -962,7 +962,12 @@ static inline void __lsx_f16x4_store(ggml_fp16_t * x, __m128 y) {
#define GGML_F16_STEP GGML_F32_STEP
#define GGML_F16_EPR GGML_F32_EPR
static inline __vector float __lzs_f16cx4_load(const ggml_fp16_t * x) {
static inline float32x4_t __lzs_f16cx4_load(const ggml_fp16_t * x) {
#ifdef __NNPA__
uint16x8_t tmp = vec_xl(0, (const ggml_fp16_t *)x);
uint16x8_t nnpa = vec_convert_from_fp16(tmp, 0);
return vec_extend_to_fp32_hi(nnpa, 0);
#else
float tmp[4];
for (int i = 0; i < 4; i++) {
@@ -972,6 +977,7 @@ static inline __vector float __lzs_f16cx4_load(const ggml_fp16_t * x) {
// note: keep type-cast here to prevent compiler bugs
// see: https://github.com/ggml-org/llama.cpp/issues/12846
return vec_xl(0, (const float *)(tmp));
#endif
}
static inline void __lzs_f16cx4_store(ggml_fp16_t * x, __vector float y) {