mat task adapts L4

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
+23
View File
@@ -7,12 +7,35 @@
#include "sn_intrinsic.std.h"
#include "sn_intrinsic.context.x86.h"
#include <stdio.h>
#define _FMA_SELECT(i) _MM_SHUFFLE(i, i, i, i)
#define GET_INDEX(col_len, x, y) (col_len * x + y)
// float
SN_MK_PROC_DECL_STD(float, sse, , fma_pack)
{
__mk_param_std(TILE, RIGHT_ANS, float);
memcpy(ans, right, sizeof(MAT_KERNEL(float)) * common_y * common_z);
}
SN_MK_PROC_DECL_STD(float, avx, , fma_pack)
{
__mk_param_std(TILE, RIGHT_ANS, float);
memcpy(ans, right, sizeof(MAT_KERNEL(float)) * common_y * common_z);
}
SN_MK_PROC_DECL_STD(float, 512, , fma_pack)
{
__mk_param_std(TILE, RIGHT_ANS, float);
memcpy(ans, right, sizeof(MAT_KERNEL(float)) * common_y * common_z);
}
SN_MK_PROC_DECL_STD(float, sse, , fma)
{
__mk_param_std(TILE, LEFT_RIGHT_ANS, float);