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

Working on L4, now on swiss. FMA_TILE is re arranged.

Browse Source
This commit is contained in:
Youmu Liu
2026-03-03 18:02:31 +08:00
parent ae1a49d6fb
commit 91e136f81f
12 changed files with 1083 additions and 477 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-src/load_intrinsic.x86.c
+8 -8
View File
@@ -58,7 +58,7 @@ void load_intrinsic_sse(sn_intrinsic *context, SIMDID id)
context->mat[F32].buffer_sub = STD_MK_PROC_NAME(float, sse, , sub);
context->mat[F32].buffer_transpose = STD_MK_PROC_NAME(float, sse, , transpose);
context->mat[F32].ker_fma = STD_MK_PROC_NAME(float, sse, , fma);
context->mat[F32].ker_tile_fma = STD_MK_PROC_NAME(float, sse, , fma);
context->mat[F32].ker_tile_fma = STD_MK_PROC_NAME(float, sse, , tiled_fma);
context->mat[F32].pivot = NULL;
context->mat[F64].buffer_add = STD_MK_PROC_NAME(double, sse, , add);
@@ -67,7 +67,7 @@ void load_intrinsic_sse(sn_intrinsic *context, SIMDID id)
context->mat[F64].buffer_sub = STD_MK_PROC_NAME(double, sse, , sub);
context->mat[F64].buffer_transpose = NULL;
context->mat[F64].ker_fma = STD_MK_PROC_NAME(double, sse, , fma);
context->mat[F64].ker_tile_fma = STD_MK_PROC_NAME(double, sse, , fma);
context->mat[F64].ker_tile_fma = STD_MK_PROC_NAME(double, sse, , tiled_fma);
context->mat[F64].pivot = NULL;
context->vec_f32_normalize_3 = f32_normalize;
@@ -90,8 +90,8 @@ void load_intrinsic_avx(sn_intrinsic *context, SIMDID id)
context->mat[F32].buffer_transpose = STD_MK_PROC_NAME(float, avx, , transpose);
context->mat[F32].ker_fma = is_fma ? STD_MK_PROC_NAME(float, avx, , fma) :
STD_MK_PROC_NAME(float, avx, _fma, fma);
context->mat[F32].ker_tile_fma = is_fma ? STD_MK_PROC_NAME(float, avx, , fma) :
STD_MK_PROC_NAME(float, avx, _fma, fma);
context->mat[F32].ker_tile_fma = is_fma ? STD_MK_PROC_NAME(float, avx, , tiled_fma) :
STD_MK_PROC_NAME(float, avx, _fma, tiled_fma);
context->mat[F32].pivot = NULL;
context->mat[F64].buffer_add = STD_MK_PROC_NAME(double, avx, , add);
@@ -101,8 +101,8 @@ void load_intrinsic_avx(sn_intrinsic *context, SIMDID id)
context->mat[F64].buffer_transpose = NULL;
context->mat[F64].ker_fma = is_fma ? STD_MK_PROC_NAME(double, avx, , fma) :
STD_MK_PROC_NAME(double, avx, _fma, fma);
context->mat[F64].ker_tile_fma = is_fma ? STD_MK_PROC_NAME(double, avx, , fma) :
STD_MK_PROC_NAME(double, avx, _fma, fma);
context->mat[F64].ker_tile_fma = is_fma ? STD_MK_PROC_NAME(double, avx, , tiled_fma) :
STD_MK_PROC_NAME(double, avx, _fma, tiled_fma);
context->mat[F64].pivot = NULL;
context->vec_f32_normalize_3 = f32_normalize;
@@ -120,7 +120,7 @@ void load_intrinsic_512(sn_intrinsic *context, SIMDID id)
context->mat[F32].buffer_sub = STD_MK_PROC_NAME(float, 512, , sub);
context->mat[F32].buffer_transpose = STD_MK_PROC_NAME(float, 512, , transpose);
context->mat[F32].ker_fma = STD_MK_PROC_NAME(float, 512, , fma);
context->mat[F32].ker_tile_fma = STD_MK_PROC_NAME(float, 512, , fma);
context->mat[F32].ker_tile_fma = STD_MK_PROC_NAME(float, 512, , tiled_fma);
context->mat[F32].pivot = NULL;
context->mat[F64].buffer_add = STD_MK_PROC_NAME(double, 512, , add);
@@ -129,7 +129,7 @@ void load_intrinsic_512(sn_intrinsic *context, SIMDID id)
context->mat[F64].buffer_sub = STD_MK_PROC_NAME(double, 512, , sub);
context->mat[F64].buffer_transpose = NULL;
context->mat[F64].ker_fma = STD_MK_PROC_NAME(double, 512, , fma);
context->mat[F64].ker_tile_fma = STD_MK_PROC_NAME(double, 512, , fma);
context->mat[F64].ker_tile_fma = STD_MK_PROC_NAME(double, 512, , tiled_fma);
context->mat[F64].pivot = NULL;
context->vec_f32_normalize_3 = f32_normalize;