URL
https://opencores.org/ocsvn/openrisc/openrisc/trunk
Go to most recent revision |
Only display areas with differences |
Details |
Blame |
View Log
Rev 318 |
Rev 338 |
/* { dg-do run } */
|
/* { dg-do run } */
|
/* { dg-require-effective-target sse4 } */
|
/* { dg-require-effective-target sse4 } */
|
/* { dg-options "-O2 -msse4.1" } */
|
/* { dg-options "-O2 -msse4.1" } */
|
|
|
#ifndef CHECK_H
|
#ifndef CHECK_H
|
#define CHECK_H "sse4_1-check.h"
|
#define CHECK_H "sse4_1-check.h"
|
#endif
|
#endif
|
|
|
#ifndef TEST
|
#ifndef TEST
|
#define TEST sse4_1_test
|
#define TEST sse4_1_test
|
#endif
|
#endif
|
|
|
#include CHECK_H
|
#include CHECK_H
|
|
|
#include <smmintrin.h>
|
#include <smmintrin.h>
|
|
|
#include <string.h>
|
#include <string.h>
|
|
|
#define lmskN 0x00
|
#define lmskN 0x00
|
#define lmsk0 0x01
|
#define lmsk0 0x01
|
#define lmsk1 0x02
|
#define lmsk1 0x02
|
#define lmsk01 0x03
|
#define lmsk01 0x03
|
|
|
#define hmskA 0x30
|
#define hmskA 0x30
|
#define hmsk0 0x10
|
#define hmsk0 0x10
|
#define hmsk1 0x20
|
#define hmsk1 0x20
|
#define hmsk01 0x30
|
#define hmsk01 0x30
|
#define hmskN 0x00
|
#define hmskN 0x00
|
|
|
#ifndef HIMASK
|
#ifndef HIMASK
|
#define HIMASK hmskA
|
#define HIMASK hmskA
|
#endif
|
#endif
|
|
|
#ifndef LOMASK
|
#ifndef LOMASK
|
#define LOMASK lmsk01
|
#define LOMASK lmsk01
|
#endif
|
#endif
|
|
|
static void
|
static void
|
TEST (void)
|
TEST (void)
|
{
|
{
|
union
|
union
|
{
|
{
|
__m128d x;
|
__m128d x;
|
double d[2];
|
double d[2];
|
} val1[4], val2[4], res[4], chk[4];
|
} val1[4], val2[4], res[4], chk[4];
|
int i, j;
|
int i, j;
|
double tmp;
|
double tmp;
|
|
|
for (i = 0; i < 4; i++)
|
for (i = 0; i < 4; i++)
|
{
|
{
|
val1[i].d [0] = 2.;
|
val1[i].d [0] = 2.;
|
val1[i].d [1] = 3.;
|
val1[i].d [1] = 3.;
|
|
|
val2[i].d [0] = 10.;
|
val2[i].d [0] = 10.;
|
val2[i].d [1] = 100.;
|
val2[i].d [1] = 100.;
|
|
|
tmp = 0.;
|
tmp = 0.;
|
for (j = 0; j < 2; j++)
|
for (j = 0; j < 2; j++)
|
if ((HIMASK & (0x10 << j)))
|
if ((HIMASK & (0x10 << j)))
|
tmp += val1[i].d [j] * val2[i].d [j];
|
tmp += val1[i].d [j] * val2[i].d [j];
|
|
|
for (j = 0; j < 2; j++)
|
for (j = 0; j < 2; j++)
|
if ((LOMASK & (1 << j)))
|
if ((LOMASK & (1 << j)))
|
chk[i].d[j] = tmp;
|
chk[i].d[j] = tmp;
|
}
|
}
|
|
|
for (i = 0; i < 4; i++)
|
for (i = 0; i < 4; i++)
|
{
|
{
|
res[i].x = _mm_dp_pd (val1[i].x, val2[i].x, HIMASK | LOMASK);
|
res[i].x = _mm_dp_pd (val1[i].x, val2[i].x, HIMASK | LOMASK);
|
if (memcmp (&res[i], &chk[i], sizeof (chk[i])))
|
if (memcmp (&res[i], &chk[i], sizeof (chk[i])))
|
abort ();
|
abort ();
|
}
|
}
|
}
|
}
|
|
|
© copyright 1999-2024
OpenCores.org, equivalent to Oliscience, all rights reserved. OpenCores®, registered trademark.