tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 20:05:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

ggml: move fp16/bf16 conversion optimizations to CPU backend + export conversion APIs (#13107)

Browse Source 
* ggml: dynamic x86_64 feature detection for FP32 <-> FP16/BF16 conversion

* move fp converter to ggml-cpu

* Switch ggml_compute_forward_get_rows_f16/bf16 to new ggml_cpu_fp16/bf16_to_fp32
This commit is contained in:
SXX
2025-04-26 22:05:31 +08:00
committed by GitHub
parent d5fe4e81bd
commit 77d5e9a76a
4 changed files with 101 additions and 50 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
ggml/src/ggml-cpu/ops.cpp
View File

@ -4222,7 +4222,7 @@ static void ggml_compute_forward_get_rows_f16(
GGML_ASSERT(i01 >= 0 && i01 < ne01);
ggml_fp16_to_fp32_row(
ggml_cpu_fp16_to_fp32(
(const ggml_fp16_t*) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03),
(float *) ((char *) dst->data + i10*nb1 + i11*nb2 + i12*nb3), nc);
}
@ -4263,7 +4263,7 @@ static void ggml_compute_forward_get_rows_bf16(
GGML_ASSERT(i01 >= 0 && i01 < ne01);
ggml_bf16_to_fp32_row(
ggml_cpu_bf16_to_fp32(
(const ggml_bf16_t *) ((char *) src0->data + i01*nb01 + i11*nb02 + i12*nb03),
(float *) ((char *) dst->data + i10*nb1 + i11*nb2 + i12*nb3), nc);
}