Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
freebsd
GitHub Repository: freebsd/freebsd-src
Path: blob/main/contrib/arm-optimized-routines/math/aarch64/advsimd/expm1.c
48378 views
1
/*
2
* Double-precision vector exp(x) - 1 function.
3
*
4
* Copyright (c) 2022-2024, 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
#include "v_expm1_inline.h"
12
13
static const struct data
14
{
15
struct v_expm1_data d;
16
#if WANT_SIMD_EXCEPT
17
uint64x2_t thresh, tiny_bound;
18
#else
19
float64x2_t oflow_bound;
20
#endif
21
} data = {
22
.d = V_EXPM1_DATA,
23
#if WANT_SIMD_EXCEPT
24
/* asuint64(oflow_bound) - asuint64(0x1p-51), shifted left by 1 for abs
25
compare. */
26
.thresh = V2 (0x78c56fa6d34b552),
27
/* asuint64(0x1p-51) << 1. */
28
.tiny_bound = V2 (0x3cc0000000000000 << 1),
29
#else
30
/* Value above which expm1(x) should overflow. Absolute value of the
31
underflow bound is greater than this, so it catches both cases - there is
32
a small window where fallbacks are triggered unnecessarily. */
33
.oflow_bound = V2 (0x1.62b7d369a5aa9p+9),
34
#endif
35
};
36
37
static float64x2_t VPCS_ATTR NOINLINE
38
special_case (float64x2_t x, uint64x2_t special, const struct data *d)
39
{
40
return v_call_f64 (expm1, x, expm1_inline (v_zerofy_f64 (x, special), &d->d),
41
special);
42
}
43
44
/* Double-precision vector exp(x) - 1 function.
45
The maximum error observed error is 2.05 ULP:
46
_ZGVnN2v_expm1(0x1.6329669eb8c87p-2) got 0x1.a8897eef87b34p-2
47
want 0x1.a8897eef87b32p-2. */
48
float64x2_t VPCS_ATTR V_NAME_D1 (expm1) (float64x2_t x)
49
{
50
const struct data *d = ptr_barrier (&data);
51
52
#if WANT_SIMD_EXCEPT
53
uint64x2_t ix = vreinterpretq_u64_f64 (x);
54
/* If fp exceptions are to be triggered correctly, fall back to scalar for
55
|x| < 2^-51, |x| > oflow_bound, Inf & NaN. Add ix to itself for
56
shift-left by 1, and compare with thresh which was left-shifted offline -
57
this is effectively an absolute compare. */
58
uint64x2_t special
59
= vcgeq_u64 (vsubq_u64 (vaddq_u64 (ix, ix), d->tiny_bound), d->thresh);
60
#else
61
/* Large input, NaNs and Infs. */
62
uint64x2_t special = vcageq_f64 (x, d->oflow_bound);
63
#endif
64
65
if (unlikely (v_any_u64 (special)))
66
return special_case (x, special, d);
67
68
/* expm1(x) ~= p * t + (t - 1). */
69
return expm1_inline (x, &d->d);
70
}
71
72
TEST_SIG (V, D, 1, expm1, -9.9, 9.9)
73
TEST_ULP (V_NAME_D1 (expm1), 1.56)
74
TEST_DISABLE_FENV_IF_NOT (V_NAME_D1 (expm1), WANT_SIMD_EXCEPT)
75
TEST_SYM_INTERVAL (V_NAME_D1 (expm1), 0, 0x1p-51, 1000)
76
TEST_SYM_INTERVAL (V_NAME_D1 (expm1), 0x1p-51, 0x1.62b7d369a5aa9p+9, 100000)
77
TEST_SYM_INTERVAL (V_NAME_D1 (expm1), 0x1.62b7d369a5aa9p+9, inf, 100)
78
79