1 /*
2 * Single-precision vector tanpi(x) function.
3 *
4 * Copyright (c) 2024, Arm Limited.
5 * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
6 */
7
8 #include "v_math.h"
9 #include "test_sig.h"
10 #include "test_defs.h"
11
12 const static struct v_tanpif_data
13 {
14 float32x4_t c0, c2, c4, c6;
15 float c1, c3, c5, c7;
16 } tanpif_data = {
17 /* Coefficents for tan(pi * x). */
18 .c0 = V4 (0x1.921fb4p1f), .c1 = 0x1.4abbcep3f, .c2 = V4 (0x1.466b8p5f),
19 .c3 = 0x1.461c72p7f, .c4 = V4 (0x1.42e9d4p9f), .c5 = 0x1.69e2c4p11f,
20 .c6 = V4 (0x1.e85558p11f), .c7 = 0x1.a52e08p16f,
21 };
22
23 /* Approximation for single-precision vector tanpi(x)
24 The maximum error is 3.34 ULP:
25 _ZGVnN4v_tanpif(0x1.d6c09ap-2) got 0x1.f70aacp+2
26 want 0x1.f70aa6p+2. */
V_NAME_F1(tanpi)27 float32x4_t VPCS_ATTR NOINLINE V_NAME_F1 (tanpi) (float32x4_t x)
28 {
29 const struct v_tanpif_data *d = ptr_barrier (&tanpif_data);
30
31 float32x4_t n = vrndnq_f32 (x);
32
33 /* inf produces nan that propagates. */
34 float32x4_t xr = vsubq_f32 (x, n);
35 float32x4_t ar = vabdq_f32 (x, n);
36 uint32x4_t flip = vcgtq_f32 (ar, v_f32 (0.25f));
37 float32x4_t r = vbslq_f32 (flip, vsubq_f32 (v_f32 (0.5f), ar), ar);
38
39 /* Order-7 pairwise Horner polynomial evaluation scheme. */
40 float32x4_t r2 = vmulq_f32 (r, r);
41 float32x4_t r4 = vmulq_f32 (r2, r2);
42
43 float32x4_t odd_coeffs = vld1q_f32 (&d->c1);
44 float32x4_t p01 = vfmaq_laneq_f32 (d->c0, r2, odd_coeffs, 0);
45 float32x4_t p23 = vfmaq_laneq_f32 (d->c2, r2, odd_coeffs, 1);
46 float32x4_t p45 = vfmaq_laneq_f32 (d->c4, r2, odd_coeffs, 2);
47 float32x4_t p67 = vfmaq_laneq_f32 (d->c6, r2, odd_coeffs, 3);
48 float32x4_t p = vfmaq_f32 (p45, r4, p67);
49 p = vfmaq_f32 (p23, r4, p);
50 p = vfmaq_f32 (p01, r4, p);
51
52 p = vmulq_f32 (r, p);
53 float32x4_t p_recip = vdivq_f32 (v_f32 (1.0f), p);
54 float32x4_t y = vbslq_f32 (flip, p_recip, p);
55
56 uint32x4_t sign
57 = veorq_u32 (vreinterpretq_u32_f32 (xr), vreinterpretq_u32_f32 (ar));
58 return vreinterpretq_f32_u32 (vorrq_u32 (vreinterpretq_u32_f32 (y), sign));
59 }
60
61 HALF_WIDTH_ALIAS_F1 (tanpi)
62
63 #if WANT_TRIGPI_TESTS
64 TEST_DISABLE_FENV (V_NAME_F1 (tanpi))
65 TEST_ULP (V_NAME_F1 (tanpi), 2.84)
66 TEST_SYM_INTERVAL (V_NAME_F1 (tanpi), 0, 0x1p-31, 50000)
67 TEST_SYM_INTERVAL (V_NAME_F1 (tanpi), 0x1p-31, 0.5, 100000)
68 TEST_SYM_INTERVAL (V_NAME_F1 (tanpi), 0.5, 0x1p23f, 100000)
69 TEST_SYM_INTERVAL (V_NAME_F1 (tanpi), 0x1p23f, inf, 100000)
70 #endif
71