• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Single-precision vector cos function.
3  *
4  * Copyright (c) 2019-2022, Arm Limited.
5  * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
6  */
7 
8 #include "mathlib.h"
9 #include "v_math.h"
10 #if V_SUPPORTED
11 
12 static const float Poly[] = {
13   /* 1.886 ulp error */
14   0x1.5b2e76p-19f,
15   -0x1.9f42eap-13f,
16   0x1.110df4p-7f,
17   -0x1.555548p-3f,
18 };
19 #define Pi1 v_f32 (0x1.921fb6p+1f)
20 #define Pi2 v_f32 (-0x1.777a5cp-24f)
21 #define Pi3 v_f32 (-0x1.ee59dap-49f)
22 #define A3 v_f32 (Poly[3])
23 #define A5 v_f32 (Poly[2])
24 #define A7 v_f32 (Poly[1])
25 #define A9 v_f32 (Poly[0])
26 #define RangeVal v_f32 (0x1p20f)
27 #define InvPi v_f32 (0x1.45f306p-2f)
28 #define Shift v_f32 (0x1.8p+23f)
29 #define AbsMask v_u32 (0x7fffffff)
30 #define HalfPi v_f32 (0x1.921fb6p0f)
31 
32 VPCS_ATTR
33 static v_f32_t
specialcase(v_f32_t x,v_f32_t y,v_u32_t cmp)34 specialcase (v_f32_t x, v_f32_t y, v_u32_t cmp)
35 {
36   /* Fall back to scalar code.  */
37   return v_call_f32 (cosf, x, y, cmp);
38 }
39 
40 VPCS_ATTR
41 v_f32_t
V_NAME(cosf)42 V_NAME(cosf) (v_f32_t x)
43 {
44   v_f32_t n, r, r2, y;
45   v_u32_t odd, cmp;
46 
47   r = v_as_f32_u32 (v_as_u32_f32 (x) & AbsMask);
48   cmp = v_cond_u32 (v_as_u32_f32 (r) >= v_as_u32_f32 (RangeVal));
49 
50 #if WANT_SIMD_EXCEPT
51   if (unlikely (v_any_u32 (cmp)))
52     /* If fenv exceptions are to be triggered correctly, set any special lanes
53        to 1 (which is neutral w.r.t. fenv). These lanes will be fixed by
54        specialcase later.  */
55     r = v_sel_f32 (cmp, v_f32 (1.0f), r);
56 #endif
57 
58   /* n = rint((|x|+pi/2)/pi) - 0.5 */
59   n = v_fma_f32 (InvPi, r + HalfPi, Shift);
60   odd = v_as_u32_f32 (n) << 31;
61   n -= Shift;
62   n -= v_f32 (0.5f);
63 
64   /* r = |x| - n*pi  (range reduction into -pi/2 .. pi/2) */
65   r = v_fma_f32 (-Pi1, n, r);
66   r = v_fma_f32 (-Pi2, n, r);
67   r = v_fma_f32 (-Pi3, n, r);
68 
69   /* y = sin(r) */
70   r2 = r * r;
71   y = v_fma_f32 (A9, r2, A7);
72   y = v_fma_f32 (y, r2, A5);
73   y = v_fma_f32 (y, r2, A3);
74   y = v_fma_f32 (y * r2, r, r);
75 
76   /* sign fix */
77   y = v_as_f32_u32 (v_as_u32_f32 (y) ^ odd);
78 
79   if (unlikely (v_any_u32 (cmp)))
80     return specialcase (x, y, cmp);
81   return y;
82 }
83 VPCS_ALIAS
84 #endif
85