1 /* { dg-do compile } */
2 /* { dg-options "-O2 -ftree-vectorize" } */
3
4 #include <stdint.h>
5
6 #define eq(A, B) ((A) == (B))
7 #define ne(A, B) ((A) != (B))
8 #define olt(A, B) ((A) < (B))
9 #define ole(A, B) ((A) <= (B))
10 #define oge(A, B) ((A) >= (B))
11 #define ogt(A, B) ((A) > (B))
12 #define ordered(A, B) (!__builtin_isunordered (A, B))
13 #define unordered(A, B) (__builtin_isunordered (A, B))
14 #define ueq(A, B) (!__builtin_islessgreater (A, B))
15 #define ult(A, B) (__builtin_isless (A, B))
16 #define ule(A, B) (__builtin_islessequal (A, B))
17 #define uge(A, B) (__builtin_isgreaterequal (A, B))
18 #define ugt(A, B) (__builtin_isgreater (A, B))
19 #define nueq(A, B) (__builtin_islessgreater (A, B))
20 #define nult(A, B) (!__builtin_isless (A, B))
21 #define nule(A, B) (!__builtin_islessequal (A, B))
22 #define nuge(A, B) (!__builtin_isgreaterequal (A, B))
23 #define nugt(A, B) (!__builtin_isgreater (A, B))
24
25 #define TEST_LOOP(TYPE1, TYPE2, CMP) \
26 void __attribute__ ((noinline, noclone)) \
27 test_##TYPE1##_##TYPE2##_##CMP##_zero (TYPE1 *restrict dest, \
28 TYPE1 *restrict src, \
29 TYPE1 fallback, \
30 TYPE2 *restrict a, \
31 int count) \
32 { \
33 for (int i = 0; i < count; ++i) \
34 dest[i] = CMP (a[i], 0) ? src[i] : fallback; \
35 }
36
37 #define TEST_CMP(CMP) \
38 TEST_LOOP (int32_t, float, CMP) \
39 TEST_LOOP (uint32_t, float, CMP) \
40 TEST_LOOP (float, float, CMP) \
41 TEST_LOOP (int64_t, double, CMP) \
42 TEST_LOOP (uint64_t, double, CMP) \
43 TEST_LOOP (double, double, CMP)
44
45 TEST_CMP (eq)
46 TEST_CMP (ne)
47 TEST_CMP (olt)
48 TEST_CMP (ole)
49 TEST_CMP (oge)
50 TEST_CMP (ogt)
51 TEST_CMP (ordered)
52 TEST_CMP (unordered)
53 TEST_CMP (ueq)
54 TEST_CMP (ult)
55 TEST_CMP (ule)
56 TEST_CMP (uge)
57 TEST_CMP (ugt)
58 TEST_CMP (nueq)
59 TEST_CMP (nult)
60 TEST_CMP (nule)
61 TEST_CMP (nuge)
62 TEST_CMP (nugt)
63
64 /* { dg-final { scan-assembler-times {\tfcmeq\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 3 } } */
65 /* { dg-final { scan-assembler-times {\tfcmeq\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 3 } } */
66
67 /* 3 for ne, 3 for ueq and 3 for nueq. */
68 /* { dg-final { scan-assembler-times {\tfcmne\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 9 } } */
69 /* { dg-final { scan-assembler-times {\tfcmne\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 9 } } */
70
71 /* 3 for olt, 3 for ult and 3 for nult. */
72 /* { dg-final { scan-assembler-times {\tfcmlt\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 9 } } */
73 /* { dg-final { scan-assembler-times {\tfcmlt\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 9 } } */
74
75 /* 3 for ole, 3 for ule and 3 for nule. */
76 /* { dg-final { scan-assembler-times {\tfcmle\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 9 } } */
77 /* { dg-final { scan-assembler-times {\tfcmle\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 9 } } */
78
79 /* 3 for ogt, 3 for ugt and 3 for nugt. */
80 /* { dg-final { scan-assembler-times {\tfcmgt\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 9 } } */
81 /* { dg-final { scan-assembler-times {\tfcmgt\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 9 } } */
82
83 /* 3 for oge, 3 for uge and 3 for nuge. */
84 /* { dg-final { scan-assembler-times {\tfcmge\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} 9 } } */
85 /* { dg-final { scan-assembler-times {\tfcmge\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} 9 } } */
86
87 /* { dg-final { scan-assembler-not {\tfcmuo\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, #0\.0\n} } } */
88 /* { dg-final { scan-assembler-not {\tfcmuo\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, #0\.0\n} } } */
89 /* 3 invocations for all 12 unordered comparisons. */
90 /* { dg-final { scan-assembler-times {\tfcmuo\tp[0-9]+\.s, p[0-7]/z, z[0-9]+\.s, z[0-9]+\.s\n} 36 } } */
91 /* { dg-final { scan-assembler-times {\tfcmuo\tp[0-9]+\.d, p[0-7]/z, z[0-9]+\.d, z[0-9]+\.d\n} 36 } } */