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

tile fma added

Browse Source
This commit is contained in:
Youmu Liu
2026-01-09 18:13:46 +08:00
parent 4ff1d856a7
commit 6827259891
12 changed files with 736 additions and 265 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-src/load_intrinsic.x86.c
+16 -8
View File
@@ -52,7 +52,8 @@ void load_intrinsic_sse(sn_intrinsic *context, SIMDID id)
context->mat[F32].buffer_transpose = KER_PROC(float, sse, , transpose);
context->mat[F32].build_panel = NULL;
context->mat[F32].clear_panel = PNL_PROC(float, sse, , clear);
context->mat[F32].panel_fma = KER_PROC(float, sse, , fma);
context->mat[F32].ker_fma = KER_PROC(float, sse, , fma);
context->mat[F32].ker_tile_fma = KERTILE_PROC(float, sse, , fma);
context->mat[F32].pivot = pivot_float;
context->mat[F32].store_panel = NULL;
@@ -63,7 +64,8 @@ void load_intrinsic_sse(sn_intrinsic *context, SIMDID id)
context->mat[F64].buffer_transpose = NULL;
context->mat[F64].build_panel = NULL;
context->mat[F64].clear_panel = PNL_PROC(double, sse, , clear);
context->mat[F64].panel_fma = KER_PROC(double, sse, , fma);
context->mat[F64].ker_fma = KER_PROC(double, sse, , fma);
context->mat[F64].ker_tile_fma = KERTILE_PROC(double, sse, , fma);
context->mat[F64].pivot = NULL;
context->mat[F64].store_panel = NULL;
@@ -87,8 +89,10 @@ void load_intrinsic_avx(sn_intrinsic *context, SIMDID id)
context->mat[F32].buffer_transpose = KER_PROC(float, avx, , transpose);
context->mat[F32].build_panel = BUILD_PANEL(avx, float);
context->mat[F32].clear_panel = PNL_PROC(float, avx, , clear);
context->mat[F32].panel_fma = is_fma ? KER_PROC(float, avx, , fma) :
KER_PROC(float, avx, _fma, fma);
context->mat[F32].ker_fma = is_fma ? KER_PROC(float, avx, , fma) :
KER_PROC(float, avx, _fma, fma);
context->mat[F32].ker_tile_fma = is_fma ? KERTILE_PROC(float, avx, , fma) :
KERTILE_PROC(float, avx, _fma, fma);
context->mat[F32].pivot = pivot_float;
context->mat[F32].store_panel = STORE_PANEL(avx, float);
@@ -99,8 +103,10 @@ void load_intrinsic_avx(sn_intrinsic *context, SIMDID id)
context->mat[F64].buffer_transpose = NULL;
context->mat[F64].build_panel = NULL;
context->mat[F64].clear_panel = PNL_PROC(double, avx, , clear);
context->mat[F64].panel_fma = is_fma ? KER_PROC(double, avx, , fma) :
KER_PROC(double, avx, _fma, fma);
context->mat[F64].ker_fma = is_fma ? KER_PROC(double, avx, , fma) :
KER_PROC(double, avx, _fma, fma);
context->mat[F64].ker_tile_fma = is_fma ? KERTILE_PROC(double, avx, , fma) :
KERTILE_PROC(double, avx, _fma, fma);
context->mat[F64].pivot = NULL;
context->mat[F64].store_panel = NULL;
@@ -120,7 +126,8 @@ void load_intrinsic_512(sn_intrinsic *context, SIMDID id)
context->mat[F32].buffer_transpose = KER_PROC(float, 512, , transpose);
context->mat[F32].build_panel = BUILD_PANEL(512, float);
context->mat[F32].clear_panel = PNL_PROC(float, 512, , clear);
context->mat[F32].panel_fma = KER_PROC(float, 512, , fma);
context->mat[F32].ker_fma = KER_PROC(float, 512, , fma);
context->mat[F32].ker_tile_fma = KERTILE_PROC(float, 512, , fma);
context->mat[F32].pivot = pivot_float;
context->mat[F32].store_panel = STORE_PANEL(512, float);
@@ -131,7 +138,8 @@ void load_intrinsic_512(sn_intrinsic *context, SIMDID id)
context->mat[F64].buffer_transpose = NULL;
context->mat[F64].build_panel = BUILD_PANEL(512, double);
context->mat[F64].clear_panel = PNL_PROC(double, 512, , clear);
context->mat[F64].panel_fma = KER_PROC(double, 512, , fma);
context->mat[F64].ker_fma = KER_PROC(double, 512, , fma);
context->mat[F64].ker_tile_fma = KERTILE_PROC(double, 512, , fma);
context->mat[F64].pivot = NULL;
context->mat[F64].store_panel = STORE_PANEL(512, double);