1 /* { dg-do run } */
2 /* { dg-require-effective-target sse4 } */
3 /* { dg-options "-O2 -msse4.1" } */
4
5 #include "sse4_1-check.h"
6
7 #include <smmintrin.h>
8 #include <string.h>
9
10 #define NUM 20
11
12 static void
init_blendvps(float * src1,float * src2,float * mask)13 init_blendvps (float *src1, float *src2, float *mask)
14 {
15 int i, msk, sign = 1;
16
17 msk = -1;
18 for (i = 0; i < NUM * 4; i++)
19 {
20 if((i % 4) == 0)
21 msk++;
22 src1[i] = i* (i + 1) * sign;
23 src2[i] = (i + 20) * sign;
24 mask[i] = (i + 120) * i;
25 if( (msk & (1 << (i % 4))))
26 mask[i] = -mask[i];
27 sign = -sign;
28 }
29 }
30
31 static int
check_blendvps(__m128 * dst,float * src1,float * src2,float * mask)32 check_blendvps (__m128 *dst, float *src1, float *src2,
33 float *mask)
34 {
35 float tmp[4];
36 int j;
37
38 memcpy (&tmp[0], src1, sizeof (tmp));
39 for (j = 0; j < 4; j++)
40 if (mask [j] < 0.0)
41 tmp[j] = src2[j];
42
43 return memcmp (dst, &tmp[0], sizeof (tmp));
44 }
45
46 static void
sse4_1_test(void)47 sse4_1_test (void)
48 {
49 union
50 {
51 __m128 x[NUM];
52 float f[NUM * 4];
53 } dst, src1, src2, mask;
54 int i;
55
56 init_blendvps (src1.f, src2.f, mask.f);
57
58 for (i = 0; i < NUM; i++)
59 {
60 dst.x[i] = _mm_blendv_ps (src1.x[i], src2.x[i], mask.x[i]);
61 if (check_blendvps (&dst.x[i], &src1.f[i * 4], &src2.f[i * 4],
62 &mask.f[i * 4]))
63 abort ();
64 }
65 }
66