1 /* { dg-do run } */
2 /* { dg-options "-O3 --save-temps" } */
3
4 #include <arm_neon.h>
5
6 extern void abort (void);
7 extern float fabsf (float);
8 extern double fabs (double);
9
10 extern double trunc (double);
11 extern double round (double);
12 extern double nearbyint (double);
13 extern double floor (double);
14 extern double ceil (double);
15 extern double rint (double);
16
17 extern float truncf (float);
18 extern float roundf (float);
19 extern float nearbyintf (float);
20 extern float floorf (float);
21 extern float ceilf (float);
22 extern float rintf (float);
23
24 #define NUM_TESTS 8
25 #define DELTA 0.000001
26
27 float input_f32[] = {0.1f, -0.1f, 0.4f, 10.3f,
28 200.0f, -800.0f, -13.0f, -0.5f};
29 double input_f64[] = {0.1, -0.1, 0.4, 10.3,
30 200.0, -800.0, -13.0, -0.5};
31
32 #define TEST(SUFFIX, Q, WIDTH, LANES, C_FN, F) \
33 int \
34 test_vrnd##SUFFIX##_float##WIDTH##x##LANES##_t (void) \
35 { \
36 int ret = 1; \
37 int i = 0; \
38 int nlanes = LANES; \
39 float##WIDTH##_t expected_out[NUM_TESTS]; \
40 float##WIDTH##_t actual_out[NUM_TESTS]; \
41 \
42 for (i = 0; i < NUM_TESTS; i++) \
43 { \
44 expected_out[i] = C_FN##F (input_f##WIDTH[i]); \
45 /* Don't vectorize this. */ \
46 asm volatile ("" : : : "memory"); \
47 } \
48 \
49 /* Prevent the compiler from noticing these two loops do the same \
50 thing and optimizing away the comparison. */ \
51 asm volatile ("" : : : "memory"); \
52 \
53 for (i = 0; i < NUM_TESTS; i+=nlanes) \
54 { \
55 float##WIDTH##x##LANES##_t out = \
56 vrnd##SUFFIX##Q##_f##WIDTH \
57 (vld1##Q##_f##WIDTH (input_f##WIDTH + i)); \
58 vst1##Q##_f##WIDTH (actual_out + i, out); \
59 } \
60 \
61 for (i = 0; i < NUM_TESTS; i++) \
62 ret &= fabs##F (expected_out[i] - actual_out[i]) < DELTA; \
63 \
64 return ret; \
65 } \
66
67
68 #define BUILD_VARIANTS(SUFFIX, C_FN) \
69 TEST (SUFFIX, , 32, 2, C_FN, f) \
70 TEST (SUFFIX, q, 32, 4, C_FN, f) \
71 TEST (SUFFIX, q, 64, 2, C_FN, ) \
72
73 BUILD_VARIANTS ( , trunc)
74 /* { dg-final { scan-assembler "frintz\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
75 /* { dg-final { scan-assembler "frintz\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
76 /* { dg-final { scan-assembler "frintz\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
77 BUILD_VARIANTS (a, round)
78 /* { dg-final { scan-assembler "frinta\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
79 /* { dg-final { scan-assembler "frinta\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
80 /* { dg-final { scan-assembler "frinta\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
81 BUILD_VARIANTS (i, nearbyint)
82 /* { dg-final { scan-assembler "frinti\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
83 /* { dg-final { scan-assembler "frinti\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
84 /* { dg-final { scan-assembler "frinti\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
85 BUILD_VARIANTS (m, floor)
86 /* { dg-final { scan-assembler "frintm\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
87 /* { dg-final { scan-assembler "frintm\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
88 /* { dg-final { scan-assembler "frintm\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
89 BUILD_VARIANTS (p, ceil)
90 /* { dg-final { scan-assembler "frintp\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
91 /* { dg-final { scan-assembler "frintp\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
92 /* { dg-final { scan-assembler "frintp\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
93 BUILD_VARIANTS (x, rint)
94 /* { dg-final { scan-assembler "frintx\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
95 /* { dg-final { scan-assembler "frintx\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
96 /* { dg-final { scan-assembler "frintx\\tv\[0-9\]+\.2d, v\[0-9\]+\.2d" } } */
97
98 #undef TEST
99 #define TEST(SUFFIX, Q, WIDTH, LANES, C_FN, F) \
100 { \
101 if (!test_vrnd##SUFFIX##_float##WIDTH##x##LANES##_t ()) \
102 abort (); \
103 }
104
105 int
106 main (int argc, char **argv)
107 {
108 BUILD_VARIANTS ( , trunc)
109 BUILD_VARIANTS (a, round)
110 BUILD_VARIANTS (i, nearbyint)
111 BUILD_VARIANTS (m, floor)
112 BUILD_VARIANTS (p, ceil)
113 BUILD_VARIANTS (x, rint)
114 return 0;
115 }
116