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 #undef MASK
13 #define MASK 0xfe
14
15 static void
init_pblendw(short * src1,short * src2)16 init_pblendw (short *src1, short *src2)
17 {
18 int i, sign = 1;
19
20 for (i = 0; i < NUM * 8; i++)
21 {
22 src1[i] = i * i * sign;
23 src2[i] = (i + 20) * sign;
24 sign = -sign;
25 }
26 }
27
28 static int
check_pblendw(__m128i * dst,short * src1,short * src2)29 check_pblendw (__m128i *dst, short *src1, short *src2)
30 {
31 short tmp[8];
32 int j;
33
34 memcpy (&tmp[0], src1, sizeof (tmp));
35 for (j = 0; j < 8; j++)
36 if ((MASK & (1 << j)))
37 tmp[j] = src2[j];
38
39 return memcmp (dst, &tmp[0], sizeof (tmp));
40 }
41
42 static void
sse4_1_test(void)43 sse4_1_test (void)
44 {
45 __m128i x, y;
46 union
47 {
48 __m128i x[NUM];
49 short s[NUM * 8];
50 } dst, src1, src2;
51 union
52 {
53 __m128i x;
54 short s[8];
55 } src3;
56 int i;
57
58 init_pblendw (src1.s, src2.s);
59
60 /* Check pblendw imm8, m128, xmm */
61 for (i = 0; i < NUM; i++)
62 {
63 dst.x[i] = _mm_blend_epi16 (src1.x[i], src2.x[i], MASK);
64 if (check_pblendw (&dst.x[i], &src1.s[i * 8], &src2.s[i * 8]))
65 abort ();
66 }
67
68 /* Check pblendw imm8, xmm, xmm */
69 src3.x = _mm_setzero_si128 ();
70
71 x = _mm_blend_epi16 (dst.x[2], src3.x, MASK);
72 y = _mm_blend_epi16 (src3.x, dst.x[2], MASK);
73
74 if (check_pblendw (&x, &dst.s[16], &src3.s[0]))
75 abort ();
76
77 if (check_pblendw (&y, &src3.s[0], &dst.s[16]))
78 abort ();
79 }
80