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

add simd id and cpu id

Browse Source
This commit is contained in:
Youmu Liu
2026-01-06 18:04:17 +08:00
parent 88fa653323
commit 3d488abd2e
17 changed files with 372 additions and 615 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-src/m.basic.x86.c
+3 -3
View File
@@ -55,7 +55,7 @@ DECLARE_PNL_PROC_ANS(double, 512, , clear)
}
// Matrix transpose implementations
DECLARE_KER_PROC_RIGHT_ANS(float, sse, transpose)
DECLARE_KER_PROC_RIGHT_ANS(float, sse, ,transpose)
{
register __m128 col0 = right->f32_x[0];
register __m128 col1 = right->f32_x[1];
@@ -72,7 +72,7 @@ DECLARE_KER_PROC_RIGHT_ANS(float, sse, transpose)
ans->f32_x[3] = _mm_movehl_ps(tmp3, tmp1);
}
DECLARE_KER_PROC_RIGHT_ANS(float, avx, transpose)
DECLARE_KER_PROC_RIGHT_ANS(float, avx, ,transpose)
{
register __m256 col_low = right->f32_y[0];
register __m256 col_high = right->f32_y[1];
@@ -87,7 +87,7 @@ DECLARE_KER_PROC_RIGHT_ANS(float, avx, transpose)
ans->f32_y[1] = _mm256_shuffle_ps(col_high, col_low, _MM_SHUFFLE(3, 1, 3, 1));
}
DECLARE_KER_PROC_RIGHT_ANS(float, 512, transpose)
DECLARE_KER_PROC_RIGHT_ANS(float, 512, ,transpose)
{
register __m512 matrix = right->f32_z[0];