youmuliu/stgsharp
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

mat task adapts L4

Browse Source
This commit is contained in:
Youmu Liu
2026-06-09 18:13:36 +08:00
parent 413e41db45
commit 0356f5aff8
8 changed files with 286 additions and 47 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-head/sn_intrinsic.h
+25 -4
View File
@@ -46,9 +46,9 @@
#define zero_vec
#define zero_vec_256
#define ALIGN(simdVec)
#define ALIGN256(simdVec)
#define ALIGN512(simdVec)
#define ALIGN(simdVec)
#define ALIGN256(simdVec)
#define ALIGN512(simdVec)
#endif
@@ -79,6 +79,27 @@ typedef struct sn_L4_cache_head {
uint64_t reverse_mask_2;
} sn_L4_cache_head;
typedef struct sn_L4_handle {
void *ptr;
sn_L4_cache_head head;
} sn_L4_handle;
typedef struct matrix_task_pack {
sn_L4_handle buffer_0;
sn_L4_handle buffer_1;
sn_L4_handle buffer_2;
sn_L4_handle buffer_3;
uint64_t reverse_0;
uint64_t reverse_1;
uint64_t reverse_2;
uint64_t reverse_3;
uint64_t reverse_4;
uint64_t reverse_5;
uint64_t reverse_6;
uint64_t reverse_7;
} matrix_task_pack;
typedef struct matrix_task {
void *mat_0;
void *mat_1;
@@ -95,7 +116,7 @@ typedef struct matrix_task {
scalar_pack padding;
} matrix_task;
#define MAT_TASK(T) matrix_task
#define MAT_TASK(T) matrix_task_pack
#if defined(__cplusplus)
static_assert(sizeof(matrix_task) == 128, "matrix_parallel_task_package size mismatch");