1 /* { dg-do run } */
2 /* { dg-require-effective-target ssse3 } */
3 /* { dg-options "-O2 -fno-strict-aliasing -mssse3" } */
4
5 #ifndef CHECK_H
6 #define CHECK_H "ssse3-check.h"
7 #endif
8
9 #ifndef TEST
10 #define TEST ssse3_test
11 #endif
12
13 #include CHECK_H
14
15 #include "ssse3-vals.h"
16
17 #include <tmmintrin.h>
18
19 #ifndef __AVX__
20 /* Test the 64-bit form */
21 static void
ssse3_test_pmaddubsw(int * i1,int * i2,int * r)22 ssse3_test_pmaddubsw (int *i1, int *i2, int *r)
23 {
24 __m64 t1 = *(__m64 *) i1;
25 __m64 t2 = *(__m64 *) i2;
26 *(__m64 *) r = _mm_maddubs_pi16 (t1, t2);
27 _mm_empty ();
28 }
29 #endif
30
31 /* Test the 128-bit form */
32 static void
ssse3_test_pmaddubsw128(int * i1,int * i2,int * r)33 ssse3_test_pmaddubsw128 (int *i1, int *i2, int *r)
34 {
35 /* Assumes incoming pointers are 16-byte aligned */
36 __m128i t1 = *(__m128i *) i1;
37 __m128i t2 = *(__m128i *) i2;
38 *(__m128i *) r = _mm_maddubs_epi16 (t1, t2);
39 }
40
41 static short
signed_saturate_to_word(int x)42 signed_saturate_to_word(int x)
43 {
44 if (x > (int) 0x7fff)
45 return 0x7fff;
46
47 if (x < (int) 0xffff8000)
48 return 0x8000;
49
50 return (short) x;
51 }
52
53 /* Routine to manually compute the results */
54 static void
compute_correct_result(int * i1,int * i2,int * r)55 compute_correct_result (int *i1, int *i2, int *r)
56 {
57 unsigned char *ub1 = (unsigned char *) i1;
58 char *sb2 = (char *) i2;
59 short *sout = (short *) r;
60 int t0;
61 int i;
62
63 for (i = 0; i < 8; i++)
64 {
65 t0 = ((int) ub1[2 * i] * (int) sb2[2 * i] +
66 (int) ub1[2 * i + 1] * (int) sb2[2 * i + 1]);
67 sout[i] = signed_saturate_to_word (t0);
68 }
69 }
70
71 static void
TEST(void)72 TEST (void)
73 {
74 int i;
75 int r [4] __attribute__ ((aligned(16)));
76 int ck [4];
77 int fail = 0;
78
79 for (i = 0; i < 256; i += 8)
80 {
81 /* Manually compute the result */
82 compute_correct_result (&vals[i + 0], &vals[i + 4], ck);
83
84 #ifndef __AVX__
85 /* Run the 64-bit tests */
86 ssse3_test_pmaddubsw (&vals[i + 0], &vals[i + 4], &r[0]);
87 ssse3_test_pmaddubsw (&vals[i + 2], &vals[i + 6], &r[2]);
88 fail += chk_128 (ck, r);
89 #endif
90
91 /* Run the 128-bit tests */
92 ssse3_test_pmaddubsw128 (&vals[i + 0], &vals[i + 4], r);
93 fail += chk_128 (ck, r);
94 }
95
96 if (fail != 0)
97 abort ();
98 }
99