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
|
/* sqrtf function. PowerPC32 version.
Copyright (C) 2007-2013 Free Software Foundation, Inc.
This file is part of the GNU C Library.
The GNU C Library is free software; you can redistribute it and/or
modify it under the terms of the GNU Lesser General Public
License as published by the Free Software Foundation; either
version 2.1 of the License, or (at your option) any later version.
The GNU C Library is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
Lesser General Public License for more details.
You should have received a copy of the GNU Lesser General Public
License along with the GNU C Library; if not, see
<http://www.gnu.org/licenses/>. */
#include <sysdep.h>
#include <math_ldbl_opt.h>
/* float [fp1] sqrts (float x [fp1])
Power4 (ISA V2.0) and above implement sqrt in hardware (not optional).
The fsqrts instruction generates the correct value for all inputs and
sets the appropriate floating point exceptions. Extended checking is
only needed to set errno (via __kernel_standard) if the input value
is negative.
So compare the input value against the absolute value of itself.
This will compare equal unless the value is negative (EDOM) or a NAN,
in which case we branch to the extend wrapper. If equal we can return
the result directly.
This part of the function looks like a leaf routine, so no need to
stack a frame or execute prologue/epilogue code. It is safe to
branch directly to w_sqrt as long as the input value (f1) is
preserved. Putting the sqrt result into f2 (float parameter 2)
allows passing both the input value and sqrt result into the extended
wrapper so there is no need to recompute.
This tactic avoids the overhead of stacking a frame for the normal
(non-error) case. Until gcc supports prologue shrink-wrapping
this is the best we can do. */
.section ".text"
.machine power5
EALIGN (__sqrtf_power5, 5, 0)
fabs fp0,fp1
fsqrts fp2,fp1
fcmpu cr1,fp0,fp1
bne- cr1,.Lw_sqrtf
fmr fp1,fp2
blr
.align 4
.Lw_sqrtf:
mflr r0
stwu r1,-16(r1)
cfi_adjust_cfa_offset(16)
fmr fp12,fp2
stw r0,20(r1)
stw r30,8(r1)
cfi_offset(lr,20-16)
cfi_offset(r30,8-16)
#ifdef SHARED
SETUP_GOT_ACCESS(r30,got_label)
addis r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@ha
addi r30,r30,_GLOBAL_OFFSET_TABLE_-got_label@l
lwz r9,_LIB_VERSION@got(30)
lwz r0,0(r9)
#else
lis r9,_LIB_VERSION@ha
lwz r0,_LIB_VERSION@l(r9)
#endif
/* if (_LIB_VERSION == _IEEE_) return z; */
cmpwi cr7,r0,-1
beq- cr7,.L4
/* if (x != x, 0) return z; !isnan */
fcmpu cr7,fp1,fp1
bne- cr7,.L4
/* if (x < 0.0)
return __kernel_standard (x, x, 126) */
fmr fp2,fp1
li r3,126
bne- cr1,.L11
.L4:
lwz r0,20(r1)
fmr fp1,fp12
lwz r30,8(r1)
addi r1,r1,16
mtlr r0
blr
.L11:
bl __kernel_standard@plt
fmr fp12,fp1
b .L4
END (__sqrtf_power5)
|
