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
2026-01-05 18:10:24 +08:00
parent d0fe4dde1a
commit 88fa653323
11 changed files with 447 additions and 286 deletions
Download Patch File Download Diff File Expand all files Collapse all files
StgSharp.Native/sn-src/m.bufop.x86.c
+25 -24
View File
@@ -3,8 +3,9 @@
#if SN_IS_ARCH(SN_ARCH_X86_64)
#include "sn_intrinsic.h"
#include "sn_intrincontext.x86.h"
DECLARE_BUF_PROC_ANS_SCALAR(float, sse, fill)
DECLARE_BUF_PROC_ANS_SCALAR(float, sse, ,fill)
{
register __m128 vec = _mm_set_ps1(((float *)(scalar->data))[0]);
for (size_t i = 0; i < count; i++) {
@@ -15,7 +16,7 @@ DECLARE_BUF_PROC_ANS_SCALAR(float, sse, fill)
}
}
DECLARE_BUF_PROC_ANS_SCALAR(float, avx, fill)
DECLARE_BUF_PROC_ANS_SCALAR(float, avx, ,fill)
{
register __m256 vec = _mm256_broadcast_ss(((float *)(scalar->data)));
for (size_t i = 0; i < count; i++) {
@@ -24,7 +25,7 @@ DECLARE_BUF_PROC_ANS_SCALAR(float, avx, fill)
}
}
DECLARE_BUF_PROC_ANS_SCALAR(float, 512, fill)
DECLARE_BUF_PROC_ANS_SCALAR(float, 512, ,fill)
{
register __m512 vec = _mm512_broadcastss_ps(((__m128 *)(scalar->data))[0]);
for (size_t i = 0; i < count; i++) {
@@ -32,7 +33,7 @@ DECLARE_BUF_PROC_ANS_SCALAR(float, 512, fill)
}
}
DECLARE_BUF_PROC_ANS_SCALAR(double, sse, fill)
DECLARE_BUF_PROC_ANS_SCALAR(double, sse, ,fill)
{
register __m128d vec = _mm_set1_pd(((double *)(scalar->data))[0]);
for (size_t i = 0; i < count; i++) {
@@ -47,7 +48,7 @@ DECLARE_BUF_PROC_ANS_SCALAR(double, sse, fill)
}
}
DECLARE_BUF_PROC_ANS_SCALAR(double, avx, fill)
DECLARE_BUF_PROC_ANS_SCALAR(double, avx, ,fill)
{
register __m256d vec = _mm256_broadcast_sd(((double *)(scalar->data)));
for (size_t i = 0; i < count; i++) {
@@ -58,7 +59,7 @@ DECLARE_BUF_PROC_ANS_SCALAR(double, avx, fill)
}
}
DECLARE_BUF_PROC_ANS_SCALAR(double, 512, fill)
DECLARE_BUF_PROC_ANS_SCALAR(double, 512, ,fill)
{
register __m512d vec = _mm512_broadcastsd_pd(_mm_loadu_pd(scalar));
for (size_t i = 0; i < count; i++) {
@@ -67,7 +68,7 @@ DECLARE_BUF_PROC_ANS_SCALAR(double, 512, fill)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, add)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, ,add)
{
for (size_t i = 0; i < count; i++) {
register __m128 c0 = _mm_add_ps(left[i].f32_x[0], right[i].f32_x[0]);
@@ -82,7 +83,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, add)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, add)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, ,add)
{
for (size_t i = 0; i < count; i++) {
size_t s = sizeof(MAT_KERNEL(float));
@@ -93,14 +94,14 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, add)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, add)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, ,add)
{
for (size_t i = 0; i < count; i++) {
register __m512 c0 = _mm512_add_ps(left[i].f32_z[0], right[i].f32_z[0]);
ans[i].f32_z[0] = c0;
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, add)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, ,add)
{
for (size_t i = 0; i < count; i++) {
register __m128d c0 = _mm_add_pd(_mm_castps_pd(left[i].f32_x[0]),
@@ -131,7 +132,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, add)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, add)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, ,add)
{
for (size_t i = 0; i < count; i++) {
size_t s = sizeof(MAT_KERNEL(float));
@@ -146,7 +147,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, add)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, add)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, ,add)
{
for (size_t i = 0; i < count; i++) {
register __m512d c0 = _mm512_add_pd(left[i].f64_z[0], right[i].f64_z[0]);
@@ -156,7 +157,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, add)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, sub)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, ,sub)
{
for (size_t i = 0; i < count; i++) {
register __m128 c0 = _mm_sub_ps(left[i].f32_x[0], right[i].f32_x[0]);
@@ -171,7 +172,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, sse, sub)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, sub)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, ,sub)
{
for (size_t i = 0; i < count; i++) {
register __m256 c0 = _mm256_sub_ps(left[i].f32_y[0], right[i].f32_y[0]);
@@ -181,7 +182,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, avx, sub)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, sub)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, ,sub)
{
for (size_t i = 0; i < count; i++) {
register __m512 c0 = _mm512_sub_ps(left[i].f32_z[0], right[i].f32_z[0]);
@@ -189,7 +190,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(float, 512, sub)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, sub)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, ,sub)
{
for (size_t i = 0; i < count; i++) {
register __m128d c0 = _mm_sub_pd(left[i].f64_x[0], right[i].f64_x[0]);
@@ -212,7 +213,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, sse, sub)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, sub)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, ,sub)
{
for (size_t i = 0; i < count; i++) {
register __m256 c0 = _mm256_sub_ps(left[i].f32_y[0], right[i].f32_y[0]);
@@ -222,7 +223,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, avx, sub)
}
}
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, sub)
DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, ,sub)
{
for (size_t i = 0; i < count; i++) {
register __m512 c0 = _mm512_sub_ps(left[i].f32_z[0], right[i].f32_z[0]);
@@ -230,7 +231,7 @@ DECLARE_BUF_PROC_LEFT_RIGHT_ANS(double, 512, sub)
}
}
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, sse, scalar_mul)
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, sse, ,scalar_mul)
{
register __m128 vec = _mm_set_ps1(((float *)(scalar->data))[0]);
for (size_t i = 0; i < count; i++) {
@@ -246,7 +247,7 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, sse, scalar_mul)
}
}
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, avx, scalar_mul)
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, avx, ,scalar_mul)
{
register __m256 vec = _mm256_broadcast_ss(((float *)(scalar->data)));
for (size_t i = 0; i < count; i++) {
@@ -258,7 +259,7 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, avx, scalar_mul)
}
}
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, 512, scalar_mul)
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, 512, ,scalar_mul)
{
register __m512 vec = _mm512_broadcastss_ps(((__m128 *)(scalar->data))[0]);
for (size_t i = 0; i < count; i++) {
@@ -268,7 +269,7 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(float, 512, scalar_mul)
}
}
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, sse, scalar_mul)
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, sse, ,scalar_mul)
{
register __m128d vec = _mm_set1_pd(((double *)(scalar->data))[0]);
for (size_t i = 0; i < count; i++) {
@@ -292,7 +293,7 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, sse, scalar_mul)
}
}
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, avx, scalar_mul)
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, avx, ,scalar_mul)
{
register __m256d vec = _mm256_broadcast_sd(((double *)(scalar->data)));
for (size_t i = 0; i < count; i++) {
@@ -308,7 +309,7 @@ DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, avx, scalar_mul)
}
}
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, 512, scalar_mul)
DECLARE_BUF_PROC_RIGHT_ANS_SCALAR(double, 512, ,scalar_mul)
{
register __m512d vec = _mm512_broadcastsd_pd(_mm_load_pd(scalar));
for (size_t i = 0; i < count; i++) {