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

new compute kernel design, less mem read delay

Browse Source
This commit is contained in:
Nitload
2026-04-07 01:04:05 +08:00
parent d702b3224c
commit 12a216d404
47 changed files with 225 additions and 145 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-src/m.bufop.x86.c
+2 -2
View File
@@ -68,7 +68,7 @@ SN_MK_PROC_DECL_STD(double, avx, , fill)
SN_MK_PROC_DECL_STD(double, 512, , fill)
{
__mk_param_std(BUFFER, ANS_SCALAR, double);
register __m512d vec = _mm512_set1_pd(SN_CONST_PTR_CAST(double, scalar->data)[0]);
register __m512d vec = _mm512_set1_pd(((double const *)scalar->data)[0]);
for (size_t i = 0; i < count; i++) {
ans[i].f64_z[0] = vec;
ans[i].f64_z[1] = vec;
@@ -354,7 +354,7 @@ SN_MK_PROC_DECL_STD(double, 512, , scalar_mul)
{
__mk_param_std(BUFFER, RIGHT_ANS_SCALAR, double);
register __m512d vec = _mm512_set1_pd(SN_CONST_PTR_CAST(double, scalar->data)[0]);
register __m512d vec = _mm512_set1_pd(((double const *)scalar->data)[0]);
for (size_t i = 0; i < count; i++) {
register __m512d c0 = _mm512_mul_pd(right[i].f64_z[0], vec);
register __m512d c1 = _mm512_mul_pd(right[i].f64_z[1], vec);