tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-07-02 05:15:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml-cpu: nnpa activate ggml_cpu_fp16_to_fp32 for 8 elements

Browse Source 
Signed-off-by: Aaron Teo <aaron.teo1@ibm.com>
This commit is contained in:
Aaron Teo
2025-06-21 16:10:44 +08:00
parent ebc1d19f62
commit 6a25fd8531
1 changed files with 15 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
ggml/src/ggml-cpu/ggml-cpu.c
View File

@ -3170,6 +3170,21 @@ void ggml_cpu_fp16_to_fp32(const ggml_fp16_t * x, float * y, int64_t n) {
#endif
#if defined(__NNPA__)
for (; i + 7 < n; i += 8) {
uint16x8_t v_x = vec_xl(0, (const ggml_fp16_t *)(x + i));
uint16x8_t v_xd = vec_convert_from_fp16(v_x, 0);
float32x4_t v_xdh = vec_extend_to_fp32_hi(v_xd, 0);
float32x4_t v_xdl = vec_extend_to_fp32_lo(v_xd, 0);
y[i + 0] = v_xdh[0];
y[i + 1] = v_xdh[1];
y[i + 2] = v_xdh[2];
y[i + 3] = v_xdh[3];
y[i + 4] = v_xdl[0];
y[i + 5] = v_xdl[1];
y[i + 6] = v_xdl[2];
y[i + 7] = v_xdl[3];
}
for (; i + 3 < n; i += 4) {
uint16x8_t v_x = vec_xl(0, (const ggml_fp16_t *)(x + i));
uint16x8_t nnpa_dfl16 = vec_convert_from_fp16(v_x, 0);