2 * Single-precision vector sin function.
4 * Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
5 * See https://llvm.org/LICENSE.txt for license information.
6 * SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
13 static const float Poly
[] = {
20 #define Pi1 v_f32 (0x1.921fb6p+1f)
21 #define Pi2 v_f32 (-0x1.777a5cp-24f)
22 #define Pi3 v_f32 (-0x1.ee59dap-49f)
23 #define A3 v_f32 (Poly[3])
24 #define A5 v_f32 (Poly[2])
25 #define A7 v_f32 (Poly[1])
26 #define A9 v_f32 (Poly[0])
27 #define RangeVal v_f32 (0x1p20f)
28 #define InvPi v_f32 (0x1.45f306p-2f)
29 #define Shift v_f32 (0x1.8p+23f)
30 #define AbsMask v_u32 (0x7fffffff)
34 specialcase (v_f32_t x
, v_f32_t y
, v_u32_t cmp
)
36 /* Fall back to scalar code. */
37 return v_call_f32 (sinf
, x
, y
, cmp
);
42 V_NAME(sinf
) (v_f32_t x
)
45 v_u32_t sign
, odd
, cmp
;
47 r
= v_as_f32_u32 (v_as_u32_f32 (x
) & AbsMask
);
48 sign
= v_as_u32_f32 (x
) & ~AbsMask
;
49 cmp
= v_cond_u32 (v_as_u32_f32 (r
) >= v_as_u32_f32 (RangeVal
));
51 /* n = rint(|x|/pi) */
52 n
= v_fma_f32 (InvPi
, r
, Shift
);
53 odd
= v_as_u32_f32 (n
) << 31;
56 /* r = |x| - n*pi (range reduction into -pi/2 .. pi/2) */
57 r
= v_fma_f32 (-Pi1
, n
, r
);
58 r
= v_fma_f32 (-Pi2
, n
, r
);
59 r
= v_fma_f32 (-Pi3
, n
, r
);
63 y
= v_fma_f32 (A9
, r2
, A7
);
64 y
= v_fma_f32 (y
, r2
, A5
);
65 y
= v_fma_f32 (y
, r2
, A3
);
66 y
= v_fma_f32 (y
* r2
, r
, r
);
69 y
= v_as_f32_u32 (v_as_u32_f32 (y
) ^ sign
^ odd
);
71 if (unlikely (v_any_u32 (cmp
)))
72 return specialcase (x
, y
, cmp
);