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

left and right use diff pack kernel

Browse Source
This commit is contained in:
Youmu Liu
2026-06-12 18:25:04 +08:00
parent 06a119cc6f
commit 2741b34583
6 changed files with 390 additions and 164 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-head/sn_intrinsic.context.x86.h
+24 -2
View File
@@ -175,17 +175,39 @@ SN_MK_PROC_DECL_STD(float, sse, , quality);
*
* CONVENSIONS:
*
* buffer_1 is pointer to original colum, offset_1 is the offset to the original column, count_1 is the count can be packed
* buffer_1 is pointer to original colum,
* offset_1 is the offset to the original column,
* count_1 is the count can be packed
*
* C# shceduler is responsible for calculating the offset and count based on the actual buffer size and matrix dimension,
* and make sure that count_1 will not exceeds lenght of the column
*
* scalar pack carries column length and offset data:
* [column length, offset, buffer requires, empty]
* [column length, empty, empty, empty]
*/
SN_MK_PROC_DECL_STD(float, sse, , fma_pack_left);
SN_MK_PROC_DECL_STD(float, avx, , fma_pack_left);
SN_MK_PROC_DECL_STD(float, 512, , fma_pack_left);
/*
* pack right matrix data to L4 buffer
*
* CONVENSIONS:
*
* buffer_1 is pointer to original row,
* offset_1 is the offset to the original row,
* count_1 is the count can be packed
*
* C# shceduler is responsible for calculating the offset and count based on the actual buffer size and matrix dimension,
* and make sure that count_1 will not exceeds lenght of the row
*
* scalar pack carries column length and offset data:
* [column length, empty, empty, empty]
*/
SN_MK_PROC_DECL_STD(float, sse, , fma_pack_right);
SN_MK_PROC_DECL_STD(float, avx, , fma_pack_right);
SN_MK_PROC_DECL_STD(float, 512, , fma_pack_right);
SN_MK_PROC_DECL_STD(float, sse, , fma);
SN_MK_PROC_DECL_STD(float, avx, , fma);
SN_MK_PROC_DECL_STD(float, avx, _fma, fma);