/* { dg-do run } */
|
/* { dg-do run } */
|
/* { dg-require-effective-target sse4 } */
|
/* { dg-require-effective-target sse4 } */
|
/* { dg-options "-O2 -msse4.2" } */
|
/* { dg-options "-O2 -msse4.2" } */
|
|
|
#ifndef CHECK_H
|
#ifndef CHECK_H
|
#define CHECK_H "sse4_2-check.h"
|
#define CHECK_H "sse4_2-check.h"
|
#endif
|
#endif
|
|
|
#ifndef TEST
|
#ifndef TEST
|
#define TEST sse4_2_test
|
#define TEST sse4_2_test
|
#endif
|
#endif
|
|
|
#include CHECK_H
|
#include CHECK_H
|
|
|
#include "sse4_2-pcmpstr.h"
|
#include "sse4_2-pcmpstr.h"
|
|
|
#define NUM 1024
|
#define NUM 1024
|
|
|
#define IMM_VAL0 \
|
#define IMM_VAL0 \
|
(_SIDD_SBYTE_OPS | _SIDD_CMP_RANGES | _SIDD_MASKED_POSITIVE_POLARITY)
|
(_SIDD_SBYTE_OPS | _SIDD_CMP_RANGES | _SIDD_MASKED_POSITIVE_POLARITY)
|
#define IMM_VAL1 \
|
#define IMM_VAL1 \
|
(_SIDD_UBYTE_OPS | _SIDD_CMP_EQUAL_EACH | _SIDD_NEGATIVE_POLARITY \
|
(_SIDD_UBYTE_OPS | _SIDD_CMP_EQUAL_EACH | _SIDD_NEGATIVE_POLARITY \
|
| _SIDD_BIT_MASK)
|
| _SIDD_BIT_MASK)
|
#define IMM_VAL2 \
|
#define IMM_VAL2 \
|
(_SIDD_UWORD_OPS | _SIDD_CMP_EQUAL_ANY | _SIDD_MASKED_NEGATIVE_POLARITY)
|
(_SIDD_UWORD_OPS | _SIDD_CMP_EQUAL_ANY | _SIDD_MASKED_NEGATIVE_POLARITY)
|
#define IMM_VAL3 \
|
#define IMM_VAL3 \
|
(_SIDD_SWORD_OPS | _SIDD_CMP_EQUAL_ORDERED \
|
(_SIDD_SWORD_OPS | _SIDD_CMP_EQUAL_ORDERED \
|
| _SIDD_POSITIVE_POLARITY | _SIDD_UNIT_MASK)
|
| _SIDD_POSITIVE_POLARITY | _SIDD_UNIT_MASK)
|
|
|
static void
|
static void
|
TEST (void)
|
TEST (void)
|
{
|
{
|
union
|
union
|
{
|
{
|
__m128i x[NUM];
|
__m128i x[NUM];
|
char c[NUM *16];
|
char c[NUM *16];
|
} src1, src2;
|
} src1, src2;
|
__m128i res, correct;
|
__m128i res, correct;
|
int correct_flags;
|
int correct_flags;
|
int flags, cf, zf, sf, of, af;
|
int flags, cf, zf, sf, of, af;
|
int i;
|
int i;
|
|
|
for (i = 0; i < NUM *16; i++)
|
for (i = 0; i < NUM *16; i++)
|
{
|
{
|
src1.c[i] = rand ();
|
src1.c[i] = rand ();
|
src2.c[i] = rand ();
|
src2.c[i] = rand ();
|
}
|
}
|
|
|
for (i = 0; i < NUM; i++)
|
for (i = 0; i < NUM; i++)
|
{
|
{
|
switch ((rand () % 4))
|
switch ((rand () % 4))
|
{
|
{
|
case 0:
|
case 0:
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL0);
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL0);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL0);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL0);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL0);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL0);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL0);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL0);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL0);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL0);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL0);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL0);
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL0,
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL0,
|
&correct_flags);
|
&correct_flags);
|
break;
|
break;
|
|
|
case 1:
|
case 1:
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL1);
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL1);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL1);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL1);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL1);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL1);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL1);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL1);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL1);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL1);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL1);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL1);
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL1,
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL1,
|
&correct_flags);
|
&correct_flags);
|
break;
|
break;
|
|
|
case 2:
|
case 2:
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL2);
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL2);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL2);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL2);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL2);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL2);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL2);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL2);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL2);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL2);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL2);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL2);
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL2,
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL2,
|
&correct_flags);
|
&correct_flags);
|
break;
|
break;
|
|
|
default:
|
default:
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL3);
|
res = _mm_cmpistrm (src1.x[i], src2.x[i], IMM_VAL3);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL3);
|
cf = _mm_cmpistrc (src1.x[i], src2.x[i], IMM_VAL3);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL3);
|
zf = _mm_cmpistrz (src1.x[i], src2.x[i], IMM_VAL3);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL3);
|
sf = _mm_cmpistrs (src1.x[i], src2.x[i], IMM_VAL3);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL3);
|
of = _mm_cmpistro (src1.x[i], src2.x[i], IMM_VAL3);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL3);
|
af = _mm_cmpistra (src1.x[i], src2.x[i], IMM_VAL3);
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL3,
|
correct = cmp_im (&src1.x[i], &src2.x[i], IMM_VAL3,
|
&correct_flags);
|
&correct_flags);
|
break;
|
break;
|
}
|
}
|
|
|
if (memcmp (&correct, &res, sizeof (res)))
|
if (memcmp (&correct, &res, sizeof (res)))
|
abort ();
|
abort ();
|
|
|
flags = 0;
|
flags = 0;
|
if (cf)
|
if (cf)
|
flags |= CFLAG;
|
flags |= CFLAG;
|
if (zf)
|
if (zf)
|
flags |= ZFLAG;
|
flags |= ZFLAG;
|
if (sf)
|
if (sf)
|
flags |= SFLAG;
|
flags |= SFLAG;
|
if (of)
|
if (of)
|
flags |= OFLAG;
|
flags |= OFLAG;
|
|
|
if (flags != correct_flags
|
if (flags != correct_flags
|
|| (af && (cf || zf))
|
|| (af && (cf || zf))
|
|| (!af && !(cf || zf)))
|
|| (!af && !(cf || zf)))
|
abort ();
|
abort ();
|
}
|
}
|
}
|
}
|
|
|
