xref: /freebsd/contrib/arm-optimized-routines/pl/math/v_acoshf_3u1.c (revision e1c4c8dd8d2d10b6104f06856a77bd5b4813a801)
1 /*
2  * Single-precision vector acosh(x) function.
3  * Copyright (c) 2023, Arm Limited.
4  * SPDX-License-Identifier: MIT OR Apache-2.0 WITH LLVM-exception
5  */
6 
7 #include "v_math.h"
8 #include "pl_sig.h"
9 #include "pl_test.h"
10 #include "v_log1pf_inline.h"
11 
12 const static struct data
13 {
14   struct v_log1pf_data log1pf_consts;
15   uint32x4_t one;
16   uint16x4_t thresh;
17 } data = {
18   .log1pf_consts = V_LOG1PF_CONSTANTS_TABLE,
19   .one = V4 (0x3f800000),
20   .thresh = V4 (0x2000) /* asuint(0x1p64) - asuint(1).  */
21 };
22 
23 #define SignMask 0x80000000
24 
25 static float32x4_t NOINLINE VPCS_ATTR
26 special_case (float32x4_t x, float32x4_t y, uint16x4_t special,
27 	      const struct v_log1pf_data d)
28 {
29   return v_call_f32 (acoshf, x, log1pf_inline (y, d), vmovl_u16 (special));
30 }
31 
32 /* Vector approximation for single-precision acosh, based on log1p. Maximum
33    error depends on WANT_SIMD_EXCEPT. With SIMD fp exceptions enabled, it
34    is 2.78 ULP:
35    __v_acoshf(0x1.07887p+0) got 0x1.ef9e9cp-3
36 			   want 0x1.ef9ea2p-3.
37    With exceptions disabled, we can compute u with a shorter dependency chain,
38    which gives maximum error of 3.07 ULP:
39   __v_acoshf(0x1.01f83ep+0) got 0x1.fbc7fap-4
40 			   want 0x1.fbc7f4p-4.  */
41 
42 VPCS_ATTR float32x4_t V_NAME_F1 (acosh) (float32x4_t x)
43 {
44   const struct data *d = ptr_barrier (&data);
45   uint32x4_t ix = vreinterpretq_u32_f32 (x);
46   uint16x4_t special = vcge_u16 (vsubhn_u32 (ix, d->one), d->thresh);
47 
48 #if WANT_SIMD_EXCEPT
49   /* Mask special lanes with 1 to side-step spurious invalid or overflow. Use
50      only xm1 to calculate u, as operating on x will trigger invalid for NaN.
51      Widening sign-extend special predicate in order to mask with it.  */
52   uint32x4_t p
53       = vreinterpretq_u32_s32 (vmovl_s16 (vreinterpret_s16_u16 (special)));
54   float32x4_t xm1 = v_zerofy_f32 (vsubq_f32 (x, v_f32 (1)), p);
55   float32x4_t u = vfmaq_f32 (vaddq_f32 (xm1, xm1), xm1, xm1);
56 #else
57   float32x4_t xm1 = vsubq_f32 (x, v_f32 (1));
58   float32x4_t u = vmulq_f32 (xm1, vaddq_f32 (x, v_f32 (1.0f)));
59 #endif
60 
61   float32x4_t y = vaddq_f32 (xm1, vsqrtq_f32 (u));
62 
63   if (unlikely (v_any_u16h (special)))
64     return special_case (x, y, special, d->log1pf_consts);
65   return log1pf_inline (y, d->log1pf_consts);
66 }
67 
68 PL_SIG (V, F, 1, acosh, 1.0, 10.0)
69 #if WANT_SIMD_EXCEPT
70 PL_TEST_ULP (V_NAME_F1 (acosh), 2.29)
71 #else
72 PL_TEST_ULP (V_NAME_F1 (acosh), 2.58)
73 #endif
74 PL_TEST_EXPECT_FENV (V_NAME_F1 (acosh), WANT_SIMD_EXCEPT)
75 PL_TEST_INTERVAL (V_NAME_F1 (acosh), 0, 1, 500)
76 PL_TEST_INTERVAL (V_NAME_F1 (acosh), 1, SquareLim, 100000)
77 PL_TEST_INTERVAL (V_NAME_F1 (acosh), SquareLim, inf, 1000)
78 PL_TEST_INTERVAL (V_NAME_F1 (acosh), -0, -inf, 1000)
79