1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
|
// Adapted from https://github.com/Alexhuszagh/rust-lexical.
//! Defines rounding schemes for floating-point numbers.
use super::float::ExtendedFloat;
use super::num::*;
use super::shift::*;
use core::mem;
// MASKS
/// Calculate a scalar factor of 2 above the halfway point.
#[inline]
pub(crate) fn nth_bit(n: u64) -> u64 {
let bits: u64 = mem::size_of::<u64>() as u64 * 8;
debug_assert!(n < bits, "nth_bit() overflow in shl.");
1 << n
}
/// Generate a bitwise mask for the lower `n` bits.
#[inline]
pub(crate) fn lower_n_mask(n: u64) -> u64 {
let bits: u64 = mem::size_of::<u64>() as u64 * 8;
debug_assert!(n <= bits, "lower_n_mask() overflow in shl.");
if n == bits {
u64::max_value()
} else {
(1 << n) - 1
}
}
/// Calculate the halfway point for the lower `n` bits.
#[inline]
pub(crate) fn lower_n_halfway(n: u64) -> u64 {
let bits: u64 = mem::size_of::<u64>() as u64 * 8;
debug_assert!(n <= bits, "lower_n_halfway() overflow in shl.");
if n == 0 {
0
} else {
nth_bit(n - 1)
}
}
/// Calculate a bitwise mask with `n` 1 bits starting at the `bit` position.
#[inline]
pub(crate) fn internal_n_mask(bit: u64, n: u64) -> u64 {
let bits: u64 = mem::size_of::<u64>() as u64 * 8;
debug_assert!(bit <= bits, "internal_n_halfway() overflow in shl.");
debug_assert!(n <= bits, "internal_n_halfway() overflow in shl.");
debug_assert!(bit >= n, "internal_n_halfway() overflow in sub.");
lower_n_mask(bit) ^ lower_n_mask(bit - n)
}
// NEAREST ROUNDING
// Shift right N-bytes and round to the nearest.
//
// Return if we are above halfway and if we are halfway.
#[inline]
pub(crate) fn round_nearest(fp: &mut ExtendedFloat, shift: i32) -> (bool, bool) {
// Extract the truncated bits using mask.
// Calculate if the value of the truncated bits are either above
// the mid-way point, or equal to it.
//
// For example, for 4 truncated bytes, the mask would be b1111
// and the midway point would be b1000.
let mask: u64 = lower_n_mask(shift as u64);
let halfway: u64 = lower_n_halfway(shift as u64);
let truncated_bits = fp.mant & mask;
let is_above = truncated_bits > halfway;
let is_halfway = truncated_bits == halfway;
// Bit shift so the leading bit is in the hidden bit.
overflowing_shr(fp, shift);
(is_above, is_halfway)
}
// Tie rounded floating point to event.
#[inline]
pub(crate) fn tie_even(fp: &mut ExtendedFloat, is_above: bool, is_halfway: bool) {
// Extract the last bit after shifting (and determine if it is odd).
let is_odd = fp.mant & 1 == 1;
// Calculate if we need to roundup.
// We need to roundup if we are above halfway, or if we are odd
// and at half-way (need to tie-to-even).
if is_above || (is_odd && is_halfway) {
fp.mant += 1;
}
}
// Shift right N-bytes and round nearest, tie-to-even.
//
// Floating-point arithmetic uses round to nearest, ties to even,
// which rounds to the nearest value, if the value is halfway in between,
// round to an even value.
#[inline]
pub(crate) fn round_nearest_tie_even(fp: &mut ExtendedFloat, shift: i32) {
let (is_above, is_halfway) = round_nearest(fp, shift);
tie_even(fp, is_above, is_halfway);
}
// DIRECTED ROUNDING
// Shift right N-bytes and round towards a direction.
//
// Return if we have any truncated bytes.
#[inline]
fn round_toward(fp: &mut ExtendedFloat, shift: i32) -> bool {
let mask: u64 = lower_n_mask(shift as u64);
let truncated_bits = fp.mant & mask;
// Bit shift so the leading bit is in the hidden bit.
overflowing_shr(fp, shift);
truncated_bits != 0
}
// Round down.
#[inline]
fn downard(_: &mut ExtendedFloat, _: bool) {}
// Shift right N-bytes and round toward zero.
//
// Floating-point arithmetic defines round toward zero, which rounds
// towards positive zero.
#[inline]
pub(crate) fn round_downward(fp: &mut ExtendedFloat, shift: i32) {
// Bit shift so the leading bit is in the hidden bit.
// No rounding schemes, so we just ignore everything else.
let is_truncated = round_toward(fp, shift);
downard(fp, is_truncated);
}
// ROUND TO FLOAT
// Shift the ExtendedFloat fraction to the fraction bits in a native float.
//
// Floating-point arithmetic uses round to nearest, ties to even,
// which rounds to the nearest value, if the value is halfway in between,
// round to an even value.
#[inline]
pub(crate) fn round_to_float<F, Algorithm>(fp: &mut ExtendedFloat, algorithm: Algorithm)
where
F: Float,
Algorithm: FnOnce(&mut ExtendedFloat, i32),
{
// Calculate the difference to allow a single calculation
// rather than a loop, to minimize the number of ops required.
// This does underflow detection.
let final_exp = fp.exp + F::DEFAULT_SHIFT;
if final_exp < F::DENORMAL_EXPONENT {
// We would end up with a denormal exponent, try to round to more
// digits. Only shift right if we can avoid zeroing out the value,
// which requires the exponent diff to be < M::BITS. The value
// is already normalized, so we shouldn't have any issue zeroing
// out the value.
let diff = F::DENORMAL_EXPONENT - fp.exp;
if diff <= u64::FULL {
// We can avoid underflow, can get a valid representation.
algorithm(fp, diff);
} else {
// Certain underflow, assign literal 0s.
fp.mant = 0;
fp.exp = 0;
}
} else {
algorithm(fp, F::DEFAULT_SHIFT);
}
if fp.mant & F::CARRY_MASK == F::CARRY_MASK {
// Roundup carried over to 1 past the hidden bit.
shr(fp, 1);
}
}
// AVOID OVERFLOW/UNDERFLOW
// Avoid overflow for large values, shift left as needed.
//
// Shift until a 1-bit is in the hidden bit, if the mantissa is not 0.
#[inline]
pub(crate) fn avoid_overflow<F>(fp: &mut ExtendedFloat)
where
F: Float,
{
// Calculate the difference to allow a single calculation
// rather than a loop, minimizing the number of ops required.
if fp.exp >= F::MAX_EXPONENT {
let diff = fp.exp - F::MAX_EXPONENT;
if diff <= F::MANTISSA_SIZE {
// Our overflow mask needs to start at the hidden bit, or at
// `F::MANTISSA_SIZE+1`, and needs to have `diff+1` bits set,
// to see if our value overflows.
let bit = (F::MANTISSA_SIZE + 1) as u64;
let n = (diff + 1) as u64;
let mask = internal_n_mask(bit, n);
if (fp.mant & mask) == 0 {
// If we have no 1-bit in the hidden-bit position,
// which is index 0, we need to shift 1.
let shift = diff + 1;
shl(fp, shift);
}
}
}
}
// ROUND TO NATIVE
// Round an extended-precision float to a native float representation.
#[inline]
pub(crate) fn round_to_native<F, Algorithm>(fp: &mut ExtendedFloat, algorithm: Algorithm)
where
F: Float,
Algorithm: FnOnce(&mut ExtendedFloat, i32),
{
// Shift all the way left, to ensure a consistent representation.
// The following right-shifts do not work for a non-normalized number.
fp.normalize();
// Round so the fraction is in a native mantissa representation,
// and avoid overflow/underflow.
round_to_float::<F, _>(fp, algorithm);
avoid_overflow::<F>(fp);
}
|
