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

arrange of proj structure

Browse Source
This commit is contained in:
Youmu Liu
2026-02-27 17:09:13 +08:00
parent 15434ed053
commit a798ab07f1
173 changed files with 32304 additions and 33955 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-src/m.fma_tile.x86.c
+5 -5
View File
@@ -277,7 +277,7 @@ DECLARE_KERTILE_PROC_LEFT_RIGHT_ANS(float, 512, , fma)
b = _mm_shuffle_pd(b_hi, b_hi, 0x3); \
c##i##_lo = _mm_add_pd(c##i##_lo, _mm_mul_pd(a_lo, b)); \
c##i##_hi = _mm_add_pd(c##i##_hi, _mm_mul_pd(a_hi, b)); \
} while (0);
} while (0)
DECLARE_KERTILE_PROC_LEFT_RIGHT_ANS(double, sse, , fma)
{
@@ -300,13 +300,13 @@ DECLARE_KERTILE_PROC_LEFT_RIGHT_ANS(double, sse, , fma)
const *left_cur = left + GET_INDEX(ansHeight, i, k),
const *right_cur = right + GET_INDEX(commonK, j, k); // GetKernelAddressUnsafe
DOUBLE_KERNEL_FMA_CYCLE(0)
DOUBLE_KERNEL_FMA_CYCLE(0);
DOUBLE_KERNEL_FMA_CYCLE(1)
DOUBLE_KERNEL_FMA_CYCLE(1);
DOUBLE_KERNEL_FMA_CYCLE(2)
DOUBLE_KERNEL_FMA_CYCLE(2);
DOUBLE_KERNEL_FMA_CYCLE(3)
DOUBLE_KERNEL_FMA_CYCLE(3);
}
_mm_store_pd(ans->f64_x + (0 * 2 + 0), c0_lo);