1 /* { dg-require-effective-target scalar_all_fma } */
2 /* { dg-additional-options "-fdump-tree-optimized -ffp-contract=fast" } */
6 #define N (VECTOR_BITS * 11 / 64 + 3)
9 void __attribute__ ((noipa)) \
10 f_##INV (double *restrict a, double *restrict b, \
11 double *restrict c, double *restrict d) \
13 for (int i = 0; i < N; ++i) \
15 double mb = (INV & 1 ? -b[i] : b[i]); \
17 double md = (INV & 2 ? -d[i] : d[i]); \
18 a[i] = b[i] < 10 ? mb * mc + md : 10.0; \
24 f_##INV (a, b, c, d); \
25 _Pragma("GCC novector") \
26 for (int i = 0; i < N; ++i) \
28 double mb = (INV & 1 ? -b[i] : b[i]); \
30 double md = (INV & 2 ? -d[i] : d[i]); \
31 double fma = __builtin_fma (mb, mc, md); \
32 if (a[i] != (i % 17 < 10 ? fma : 10.0)) \
34 asm volatile ("" ::: "memory"); \
38 #define FOR_EACH_INV(T) \
39 T (0) T (1) T (2) T (3)
46 double a
[N
], b
[N
], c
[N
], d
[N
];
47 for (int i
= 0; i
< N
; ++i
)
52 asm volatile ("" ::: "memory");
58 /* { dg-final { scan-tree-dump-times { = \.COND_FMA } 1 "optimized" { target vect_double_cond_arith } } } */
59 /* { dg-final { scan-tree-dump-times { = \.COND_FMS } 1 "optimized" { target vect_double_cond_arith } } } */
60 /* { dg-final { scan-tree-dump-times { = \.COND_FNMA } 1 "optimized" { target vect_double_cond_arith } } } */
61 /* { dg-final { scan-tree-dump-times { = \.COND_FNMS } 1 "optimized" { target vect_double_cond_arith } } } */