Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

metal : fuse NORM + MUL + ADD, support non-multiples of 4 (#16220)

Browse Source 
* metal : fuse NORM + MUL + ADD

* metal : support norms of non-multiple of 4

* cont : fix comment [no ci]
This commit is contained in:
Georgi Gerganov
2025-09-25 11:30:16 +03:00
committed by GitHub
parent 4ea00794b8
commit dfcd53f7ec
9 changed files with 206 additions and 232 deletions
Download Patch File Download Diff File Expand all files Collapse all files
ggml/src/ggml-metal/ggml-metal-device.h
+1 -2
View File
@@ -123,10 +123,9 @@ ggml_metal_pipeline_t ggml_metal_library_get_pipeline_mul_mv_id (ggml_me
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_argmax (ggml_metal_library_t lib, const struct ggml_tensor * op);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_argsort (ggml_metal_library_t lib, const struct ggml_tensor * op);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_bin (ggml_metal_library_t lib, enum ggml_op op, int32_t n_fuse, bool row);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_rms_norm (ggml_metal_library_t lib, const struct ggml_tensor * op, int32_t n_fuse);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_l2_norm (ggml_metal_library_t lib, const struct ggml_tensor * op);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_group_norm (ggml_metal_library_t lib, const struct ggml_tensor * op);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_norm (ggml_metal_library_t lib, const struct ggml_tensor * op);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_norm (ggml_metal_library_t lib, const struct ggml_tensor * op, int32_t n_fuse);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_rope (ggml_metal_library_t lib, const struct ggml_tensor * op);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_im2col (ggml_metal_library_t lib, const struct ggml_tensor * op);
ggml_metal_pipeline_t ggml_metal_library_get_pipeline_conv_transpose_1d (ggml_metal_library_t lib, const struct ggml_tensor * op);