1 /* { dg-do run } */
2 /* { dg-options "-O3 --save-temps -ffast-math" } */
3
4 #pragma GCC target "+nosve"
5
6 #include <arm_neon.h>
7
8 extern void abort (void);
9 extern float fabsf (float);
10 extern double fabs (double);
11
12 #define NUM_TESTS 16
13 #define DELTA 0.000001
14
15 int8_t input_int8[] = {1, 56, 2, -9, -90, 23, 54, 76,
16 -4, 34, 110, -110, 6, 4, 75, -34};
17 int16_t input_int16[] = {1, 56, 2, -9, -90, 23, 54, 76,
18 -4, 34, 110, -110, 6, 4, 75, -34};
19 int32_t input_int32[] = {1, 56, 2, -9, -90, 23, 54, 76,
20 -4, 34, 110, -110, 6, 4, 75, -34};
21 int64_t input_int64[] = {1, 56, 2, -9, -90, 23, 54, 76,
22 -4, 34, 110, -110, 6, 4, 75, -34};
23
24 uint8_t input_uint8[] = {1, 56, 2, 9, 90, 23, 54, 76,
25 4, 34, 110, 110, 6, 4, 75, 34};
26 uint16_t input_uint16[] = {1, 56, 2, 9, 90, 23, 54, 76,
27 4, 34, 110, 110, 6, 4, 75, 34};
28 uint32_t input_uint32[] = {1, 56, 2, 9, 90, 23, 54, 76,
29 4, 34, 110, 110, 6, 4, 75, 34};
30
31 uint64_t input_uint64[] = {1, 56, 2, 9, 90, 23, 54, 76,
32 4, 34, 110, 110, 6, 4, 75, 34};
33
34 float input_float32[] = {0.1f, -0.1f, 0.4f, 10.3f,
35 200.0f, -800.0f, -13.0f, -0.5f,
36 7.9f, -870.0f, 10.4f, 310.11f,
37 0.0f, -865.0f, -2213.0f, -1.5f};
38
39 double input_float64[] = {0.1, -0.1, 0.4, 10.3,
40 200.0, -800.0, -13.0, -0.5,
41 7.9, -870.0, 10.4, 310.11,
42 0.0, -865.0, -2213.0, -1.5};
43
44 #define EQUALF(a, b) (fabsf (a - b) < DELTA)
45 #define EQUALD(a, b) (fabs (a - b) < DELTA)
46 #define EQUALL(a, b) (a == b)
47
48 #define TEST(SUFFIX, Q, TYPE, LANES, FLOAT) \
49 int \
50 test_vaddv##SUFFIX##_##TYPE##x##LANES##_t (void) \
51 { \
52 int i, j; \
53 int moves = (NUM_TESTS - LANES) + 1; \
54 TYPE##_t out_l[NUM_TESTS]; \
55 TYPE##_t out_v[NUM_TESTS]; \
56 \
57 /* Calculate linearly. */ \
58 for (i = 0; i < moves; i++) \
59 { \
60 asm ("" : "=r" (out_l[i]) : "0" (0)); \
61 for (j = 0; j < LANES; j++) \
62 out_l[i] += input_##TYPE[i + j]; \
63 } \
64 \
65 /* Calculate using vector reduction intrinsics. */ \
66 for (i = 0; i < moves; i++) \
67 { \
68 TYPE##x##LANES##_t t1 = vld1##Q##_##SUFFIX (input_##TYPE + i); \
69 out_v[i] = vaddv##Q##_##SUFFIX (t1); \
70 } \
71 \
72 /* Compare. */ \
73 for (i = 0; i < moves; i++) \
74 { \
75 if (!EQUAL##FLOAT (out_v[i], out_l[i])) \
76 return 0; \
77 } \
78 return 1; \
79 }
80
81 #define BUILD_VARIANTS(TYPE, STYPE, W32, W64, F) \
82 TEST (STYPE, , TYPE, W32, F) \
83 TEST (STYPE, q, TYPE, W64, F) \
84
85 BUILD_VARIANTS (int8, s8, 8, 16, L)
86 BUILD_VARIANTS (uint8, u8, 8, 16, L)
87 /* { dg-final { scan-assembler "addv\\tb\[0-9\]+, v\[0-9\]+\.8b" } } */
88 /* { dg-final { scan-assembler "addv\\tb\[0-9\]+, v\[0-9\]+\.16b" } } */
89 BUILD_VARIANTS (int16, s16, 4, 8, L)
90 BUILD_VARIANTS (uint16, u16, 4, 8, L)
91 /* { dg-final { scan-assembler "addv\\th\[0-9\]+, v\[0-9\]+\.4h" } } */
92 /* { dg-final { scan-assembler "addv\\th\[0-9\]+, v\[0-9\]+\.8h" } } */
93 BUILD_VARIANTS (int32, s32, 2, 4, L)
94 BUILD_VARIANTS (uint32, u32, 2, 4, L)
95 /* { dg-final { scan-assembler "addp\\tv\[0-9\]+\.2s, v\[0-9\]+\.2s, v\[0-9\]+\.2s" } } */
96 /* { dg-final { scan-assembler "addv\\ts\[0-9\]+, v\[0-9\]+\.4s" } } */
97 TEST (s64, q, int64, 2, D)
98 TEST (u64, q, uint64, 2, D)
99 /* { dg-final { scan-assembler "addp\\td\[0-9\]+\, v\[0-9\]+\.2d" } } */
100
101 BUILD_VARIANTS (float32, f32, 2, 4, F)
102 /* { dg-final { scan-assembler "faddp\\ts\[0-9\]+, v\[0-9\]+\.2s" } } */
103 /* { dg-final { scan-assembler "faddp\\tv\[0-9\]+\.4s, v\[0-9\]+\.4s, v\[0-9\]+\.4s" } } */
104 TEST (f64, q, float64, 2, D)
105 /* { dg-final { scan-assembler "faddp\\td\[0-9\]+\, v\[0-9\]+\.2d" } } */
106
107 #undef TEST
108 #define TEST(SUFFIX, Q, TYPE, LANES, FLOAT) \
109 { \
110 if (!test_vaddv##SUFFIX##_##TYPE##x##LANES##_t ()) \
111 abort (); \
112 }
113
114 int
115 main (int argc, char **argv)
116 {
117 BUILD_VARIANTS (int8, s8, 8, 16, L)
118 BUILD_VARIANTS (uint8, u8, 8, 16, L)
119 BUILD_VARIANTS (int16, s16, 4, 8, L)
120 BUILD_VARIANTS (uint16, u16, 4, 8, L)
121 BUILD_VARIANTS (int32, s32, 2, 4, L)
122 BUILD_VARIANTS (uint32, u32, 2, 4, L)
123
124 BUILD_VARIANTS (float32, f32, 2, 4, F)
125 TEST (f64, q, float64, 2, D)
126
127 return 0;
128 }
129