Use gather loads for strided accesses
[official-gcc.git] / gcc / testsuite / gcc.target / i386 / sse4_1-dppd-1.c
blobb8e58d47aad3212b9cbc7bab37ed7d87a29af1ad
1 /* { dg-do run } */
2 /* { dg-require-effective-target sse4 } */
3 /* { dg-options "-O2 -msse4.1" } */
5 #ifndef CHECK_H
6 #define CHECK_H "sse4_1-check.h"
7 #endif
9 #ifndef TEST
10 #define TEST sse4_1_test
11 #endif
13 #include CHECK_H
15 #include <smmintrin.h>
17 #define lmskN 0x00
18 #define lmsk0 0x01
19 #define lmsk1 0x02
20 #define lmsk01 0x03
22 #define hmskA 0x30
23 #define hmsk0 0x10
24 #define hmsk1 0x20
25 #define hmsk01 0x30
26 #define hmskN 0x00
28 #ifndef HIMASK
29 #define HIMASK hmskA
30 #endif
32 static void
33 TEST (void)
35 union
37 __m128d x;
38 double d[2];
39 } val1, val2, res[4];
40 int masks[4];
41 int i, j;
43 val1.d[0] = 2.;
44 val1.d[1] = 3.;
46 val2.d[0] = 10.;
47 val2.d[1] = 100.;
49 res[0].x = _mm_dp_pd (val1.x, val2.x, HIMASK | lmskN);
50 res[1].x = _mm_dp_pd (val1.x, val2.x, HIMASK | lmsk0);
51 res[2].x = _mm_dp_pd (val1.x, val2.x, HIMASK | lmsk1);
52 res[3].x = _mm_dp_pd (val1.x, val2.x, HIMASK | lmsk01);
54 masks[0] = HIMASK | lmskN;
55 masks[1] = HIMASK | lmsk0;
56 masks[2] = HIMASK | lmsk1;
57 masks[3] = HIMASK | lmsk01;
59 for (i = 0; i < 4; i++)
61 double tmp = 0.;
63 for (j = 0; j < 2; j++)
64 if (HIMASK & (0x10 << j))
65 tmp = tmp + (val1.d[j] * val2.d[j]);
67 for (j = 0; j < 2; j++)
68 if ((masks[i] & (1 << j)) && res[i].d[j] != tmp)
69 abort ();