EngineX-Ascend/enginex-ascend-910-llama.cpp
10
0
Fork 0
Code Issues Pull Requests Actions 4 Projects Releases Wiki Activity

metal : various optimizations + refactoring (#16446)

Browse Source 
* metal : ssm_scan minor opts

* metal : get_rows optimize

* metal : cpy optimize

* metal : ssm_conv opt

* metal : ssm_scan simplify

* metal : ssm_Scan opt
This commit is contained in:
Georgi Gerganov
2025-10-07 08:21:40 +03:00
committed by GitHub
parent 3df2244df4
commit 8ae32dc9ec
5 changed files with 258 additions and 452 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
ggml/src/ggml-metal/ggml-metal-impl.h
View File

@@ -178,6 +178,7 @@ typedef struct {
} ggml_metal_kargs_clamp;
typedef struct {
int64_t nk0;
int64_t ne00;
int64_t ne01;
int64_t ne02;
@@ -572,32 +573,45 @@ typedef struct {
int64_t n_seq_tokens;
int64_t n_seqs;
uint64_t s_off;
uint64_t nb00;
uint64_t nb01;
uint64_t nb02;
uint64_t nb03;
uint64_t nb10;
uint64_t nb11;
uint64_t nb12;
uint64_t ns12;
uint64_t nb13;
uint64_t nb20;
uint64_t nb21;
uint64_t ns21;
uint64_t nb22;
int64_t ne30;
uint64_t nb31;
uint64_t nb41;
uint64_t nb42;
uint64_t ns42;
uint64_t nb43;
uint64_t nb51;
uint64_t nb52;
uint64_t ns52;
uint64_t nb53;
uint64_t nb0;
} ggml_metal_kargs_ssm_scan;
typedef struct {
int64_t ne00;
int32_t ne00t;
int32_t ne00;
uint64_t nb01;
uint64_t nb02;
int64_t ne10;
uint64_t nb03;
int32_t ne10;
uint64_t nb10;
uint64_t nb11;
uint64_t nb12;
uint64_t nb1;
uint64_t nb2;
uint64_t nb3;
} ggml_metal_kargs_get_rows;
typedef struct {