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
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
|
#![allow(unreachable_code)]
use float::Float;
use int::Int;
#[derive(Clone, Copy)]
enum Result {
Less,
Equal,
Greater,
Unordered,
}
impl Result {
fn to_le_abi(self) -> i32 {
match self {
Result::Less => -1,
Result::Equal => 0,
Result::Greater => 1,
Result::Unordered => 1,
}
}
fn to_ge_abi(self) -> i32 {
match self {
Result::Less => -1,
Result::Equal => 0,
Result::Greater => 1,
Result::Unordered => -1,
}
}
}
fn cmp<F: Float>(a: F, b: F) -> Result {
let one = F::Int::ONE;
let zero = F::Int::ZERO;
let szero = F::SignedInt::ZERO;
let sign_bit = F::SIGN_MASK as F::Int;
let abs_mask = sign_bit - one;
let exponent_mask = F::EXPONENT_MASK;
let inf_rep = exponent_mask;
let a_rep = a.repr();
let b_rep = b.repr();
let a_abs = a_rep & abs_mask;
let b_abs = b_rep & abs_mask;
// If either a or b is NaN, they are unordered.
if a_abs > inf_rep || b_abs > inf_rep {
return Result::Unordered;
}
// If a and b are both zeros, they are equal.
if a_abs | b_abs == zero {
return Result::Equal;
}
let a_srep = a.signed_repr();
let b_srep = b.signed_repr();
// If at least one of a and b is positive, we get the same result comparing
// a and b as signed integers as we would with a fp_ting-point compare.
if a_srep & b_srep >= szero {
if a_srep < b_srep {
Result::Less
} else if a_srep == b_srep {
Result::Equal
} else {
Result::Greater
}
// Otherwise, both are negative, so we need to flip the sense of the
// comparison to get the correct result. (This assumes a twos- or ones-
// complement integer representation; if integers are represented in a
// sign-magnitude representation, then this flip is incorrect).
} else if a_srep > b_srep {
Result::Less
} else if a_srep == b_srep {
Result::Equal
} else {
Result::Greater
}
}
fn unord<F: Float>(a: F, b: F) -> bool {
let one = F::Int::ONE;
let sign_bit = F::SIGN_MASK as F::Int;
let abs_mask = sign_bit - one;
let exponent_mask = F::EXPONENT_MASK;
let inf_rep = exponent_mask;
let a_rep = a.repr();
let b_rep = b.repr();
let a_abs = a_rep & abs_mask;
let b_abs = b_rep & abs_mask;
a_abs > inf_rep || b_abs > inf_rep
}
intrinsics! {
pub extern "C" fn __lesf2(a: f32, b: f32) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __gesf2(a: f32, b: f32) -> i32 {
cmp(a, b).to_ge_abi()
}
#[arm_aeabi_alias = __aeabi_fcmpun]
pub extern "C" fn __unordsf2(a: f32, b: f32) -> i32 {
unord(a, b) as i32
}
pub extern "C" fn __eqsf2(a: f32, b: f32) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __ltsf2(a: f32, b: f32) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __nesf2(a: f32, b: f32) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __gtsf2(a: f32, b: f32) -> i32 {
cmp(a, b).to_ge_abi()
}
pub extern "C" fn __ledf2(a: f64, b: f64) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __gedf2(a: f64, b: f64) -> i32 {
cmp(a, b).to_ge_abi()
}
#[arm_aeabi_alias = __aeabi_dcmpun]
pub extern "C" fn __unorddf2(a: f64, b: f64) -> i32 {
unord(a, b) as i32
}
pub extern "C" fn __eqdf2(a: f64, b: f64) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __ltdf2(a: f64, b: f64) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __nedf2(a: f64, b: f64) -> i32 {
cmp(a, b).to_le_abi()
}
pub extern "C" fn __gtdf2(a: f64, b: f64) -> i32 {
cmp(a, b).to_ge_abi()
}
}
#[cfg(target_arch = "arm")]
intrinsics! {
pub extern "aapcs" fn __aeabi_fcmple(a: f32, b: f32) -> i32 {
(__lesf2(a, b) <= 0) as i32
}
pub extern "aapcs" fn __aeabi_fcmpge(a: f32, b: f32) -> i32 {
(__gesf2(a, b) >= 0) as i32
}
pub extern "aapcs" fn __aeabi_fcmpeq(a: f32, b: f32) -> i32 {
(__eqsf2(a, b) == 0) as i32
}
pub extern "aapcs" fn __aeabi_fcmplt(a: f32, b: f32) -> i32 {
(__ltsf2(a, b) < 0) as i32
}
pub extern "aapcs" fn __aeabi_fcmpgt(a: f32, b: f32) -> i32 {
(__gtsf2(a, b) > 0) as i32
}
pub extern "aapcs" fn __aeabi_dcmple(a: f64, b: f64) -> i32 {
(__ledf2(a, b) <= 0) as i32
}
pub extern "aapcs" fn __aeabi_dcmpge(a: f64, b: f64) -> i32 {
(__gedf2(a, b) >= 0) as i32
}
pub extern "aapcs" fn __aeabi_dcmpeq(a: f64, b: f64) -> i32 {
(__eqdf2(a, b) == 0) as i32
}
pub extern "aapcs" fn __aeabi_dcmplt(a: f64, b: f64) -> i32 {
(__ltdf2(a, b) < 0) as i32
}
pub extern "aapcs" fn __aeabi_dcmpgt(a: f64, b: f64) -> i32 {
(__gtdf2(a, b) > 0) as i32
}
// On hard-float targets LLVM will use native instructions
// for all VFP intrinsics below
pub extern "C" fn __gesf2vfp(a: f32, b: f32) -> i32 {
(a >= b) as i32
}
pub extern "C" fn __gedf2vfp(a: f64, b: f64) -> i32 {
(a >= b) as i32
}
pub extern "C" fn __gtsf2vfp(a: f32, b: f32) -> i32 {
(a > b) as i32
}
pub extern "C" fn __gtdf2vfp(a: f64, b: f64) -> i32 {
(a > b) as i32
}
pub extern "C" fn __ltsf2vfp(a: f32, b: f32) -> i32 {
(a < b) as i32
}
pub extern "C" fn __ltdf2vfp(a: f64, b: f64) -> i32 {
(a < b) as i32
}
pub extern "C" fn __lesf2vfp(a: f32, b: f32) -> i32 {
(a <= b) as i32
}
pub extern "C" fn __ledf2vfp(a: f64, b: f64) -> i32 {
(a <= b) as i32
}
pub extern "C" fn __nesf2vfp(a: f32, b: f32) -> i32 {
(a != b) as i32
}
pub extern "C" fn __nedf2vfp(a: f64, b: f64) -> i32 {
(a != b) as i32
}
pub extern "C" fn __eqsf2vfp(a: f32, b: f32) -> i32 {
(a == b) as i32
}
pub extern "C" fn __eqdf2vfp(a: f64, b: f64) -> i32 {
(a == b) as i32
}
}
|
