blob: b3bac15b9efb8c8b4c099bb152a20e81f5ca974b (
plain) (
blame)
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
|
// This file is dual licensed under the MIT and the University of Illinois Open
// Source Licenses. See LICENSE.TXT for details.
#include "../assembly.h"
// long double __floatundixf(du_int a);
#ifdef __x86_64__
CONST_SECTION
.balign 16
twop64:
.quad 0x43f0000000000000
#define REL_ADDR(_a) (_a)(%rip)
.text
.balign 4
DEFINE_COMPILERRT_FUNCTION(__floatundixf)
movq %rdi, -8(%rsp)
fildq -8(%rsp)
test %rdi, %rdi
js 1f
ret
1: faddl REL_ADDR(twop64)
ret
END_COMPILERRT_FUNCTION(__floatundixf)
#endif // __x86_64__
/* Branch-free implementation is ever so slightly slower, but more beautiful.
It is likely superior for inlining, so I kept it around for future reference.
#ifdef __x86_64__
CONST_SECTION
.balign 4
twop52:
.quad 0x4330000000000000
twop84_plus_twop52_neg:
.quad 0xc530000000100000
twop84:
.quad 0x4530000000000000
#define REL_ADDR(_a) (_a)(%rip)
.text
.balign 4
DEFINE_COMPILERRT_FUNCTION(__floatundixf)
movl %edi, %esi // low 32 bits of input
shrq $32, %rdi // hi 32 bits of input
orq REL_ADDR(twop84), %rdi // 2^84 + hi (as a double)
orq REL_ADDR(twop52), %rsi // 2^52 + lo (as a double)
movq %rdi, -8(%rsp)
movq %rsi, -16(%rsp)
fldl REL_ADDR(twop84_plus_twop52_neg)
faddl -8(%rsp) // hi - 2^52 (as double extended, no rounding occurs)
faddl -16(%rsp) // hi + lo (as double extended)
ret
END_COMPILERRT_FUNCTION(__floatundixf)
#endif // __x86_64__
*/
|
