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

working on matrix mul matrix, rebuilding task model defines in C#

Browse Source
This commit is contained in:
Youmu Liu
2025-12-12 18:00:49 +08:00
parent 9ff438d3b9
commit 2c7277f3f9
12 changed files with 223 additions and 444 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-src/k_panel.c
+6 -6
View File
@@ -54,7 +54,7 @@ DECLARE_BUILD_PANEL(sse, float)
DECLARE_BUILD_PANEL(512, float)
{
const MAT_KERNEL(float) *const zero = (const MAT_KERNEL(float) *)&SN_ZERO_KERNEL;
const MAT_KERNEL(float) *k[4][4];
const MAT_KERNEL(float) * k[4][4];
for (int kc = 0; kc < 4; ++kc) {
for (int kr = 0; kr < 4; ++kr) {
@@ -68,8 +68,8 @@ DECLARE_BUILD_PANEL(512, float)
/* Each panel column corresponds to a kernel column (0..3) within a block col (0..3). */
for (int pc = 0; pc < 16; ++pc) {
const int blk_col = pc >> 2; /* which kernel column block (0..3) */
const int kc_col = pc & 3; /* column inside a 4x4 kernel */
const int blk_col = pc >> 2; /* which kernel column block (0..3) */
const int kc_col = pc & 3; /* column inside a 4x4 kernel */
const __m128 r0 = k[blk_col][0]->xmm[kc_col];
const __m128 r1 = k[blk_col][1]->xmm[kc_col];
@@ -155,7 +155,7 @@ DECLARE_STORE_PANEL(sse, float)
DECLARE_STORE_PANEL(512, float)
{
MAT_KERNEL(float) zero;
MAT_KERNEL(float) *k[4][4];
MAT_KERNEL(float) * k[4][4];
for (int kc = 0; kc < 4; ++kc) {
for (int kr = 0; kr < 4; ++kr) {
@@ -168,8 +168,8 @@ DECLARE_STORE_PANEL(512, float)
}
for (int pc = 0; pc < 16; ++pc) {
const int blk_col = pc >> 2; /* which kernel column block (0..3) */
const int kc_col = pc & 3; /* column inside a 4x4 kernel */
const int blk_col = pc >> 2; /* which kernel column block (0..3) */
const int kc_col = pc & 3; /* column inside a 4x4 kernel */
const __m512 col = PANEL_COL_VEC((*panel), pc, 0);