]> git.ipfire.org Git - thirdparty/glibc.git/blame - sysdeps/powerpc/powerpc32/power5/fpu/w_sqrtf.S
Fix reallocation bug in last nss_files change.
[thirdparty/glibc.git] / sysdeps / powerpc / powerpc32 / power5 / fpu / w_sqrtf.S
CommitLineData
aece054b
UD
1/* sqrtf function. PowerPC32 version.
2 Copyright (C) 2007 Free Software Foundation, Inc.
3 This file is part of the GNU C Library.
4
5 The GNU C Library is free software; you can redistribute it and/or
6 modify it under the terms of the GNU Lesser General Public
7 License as published by the Free Software Foundation; either
8 version 2.1 of the License, or (at your option) any later version.
9
10 The GNU C Library is distributed in the hope that it will be useful,
11 but WITHOUT ANY WARRANTY; without even the implied warranty of
12 MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 Lesser General Public License for more details.
14
15 You should have received a copy of the GNU Lesser General Public
16 License along with the GNU C Library; if not, write to the Free
17 Software Foundation, Inc., 1 Franklin Street, Fifth Floor, Boston MA
18 02110-1301 USA. */
19
20#include <sysdep.h>
21#include <math_ldbl_opt.h>
22
23/* float [fp1] sqrts (float x [fp1])
24 Power4 (ISA V2.0) and above implement sqrt in hardware (not optional).
25 The fsqrts instruction generates the correct value for all inputs and
26 sets the appropriate floating point exceptions. Extented checking is
27 only needed to set errno (via __kernel_standard) if the input value
28 is negative.
29
30 So compare the input value against the absolute value of itself.
31 This will compare equal unless the value is negative (EDOM) or a NAN,
32 in which case we branch to the extend wrapper. If equal we can return
33 the result directly.
34
35 This part of the function looks like a leaf routine, so no need to
36 stack a frame or execute prologue/epilogue code. It is safe to
37 branch directly to w_sqrt as long as the input value (f1) is
38 preserved. Putting the the sqrt result into f2 (float parameter 2)
39 allows passing both the input value and sqrt result into the extended
40 wrapper so there is no need to recompute.
41
42 This tactic avoids the overhead of stacking a frame for the normal
43 (non-error) case. Until gcc supports prologue shrink-wrapping
44 this is the best we can do. */
45
46 .section ".text"
47 .machine power4
48EALIGN (__sqrtf, 5, 0)
49 fabs fp0,fp1
50 fsqrts fp2,fp1
51 fcmpu cr1,fp0,fp1
52 bne- cr1,.Lw_sqrtf
53 fmr fp1,fp2
54 blr
55 .align 4
56.Lw_sqrtf:
57 mflr r0
58 stwu r1,-16(r1)
59 cfi_adjust_cfa_offset(16)
60 fmr fp12,fp2
61 stw r0,20(r1)
62 stw r30,8(r1)
3f241d75
AS
63 cfi_offset(lr,20-16)
64 cfi_offset(r30,8-16)
aece054b 65#ifdef SHARED
aece054b
UD
66 bcl 20,31,.LCF1
67.LCF1:
68 mflr r30
69 addis r30,r30,_GLOBAL_OFFSET_TABLE_-.LCF1@ha
70 addi r30,r30,_GLOBAL_OFFSET_TABLE_-.LCF1@l
71 lwz r9,_LIB_VERSION@got(30)
72 lwz r0,0(r9)
aece054b
UD
73#else
74 lis r9,_LIB_VERSION@ha
75 lwz r0,_LIB_VERSION@l(r9)
76#endif
77/* if (_LIB_VERSION == _IEEE_) return z; */
78 cmpwi cr7,r0,-1
79 beq- cr7,.L4
80/* if (x != x, 0) return z; !isnan */
81 fcmpu cr7,fp1,fp1
82 bne- cr7,.L4
83/* if (x < 0.0)
84 return __kernel_standard (x, x, 126) */
85 fmr fp2,fp1
86 li r3,126
87 bne- cr1,.L11
88.L4:
89 lwz r0,20(r1)
90 fmr fp1,fp12
91 lwz r30,8(r1)
92 addi r1,r1,16
93 mtlr r0
94 blr
95.L11:
96 bl __kernel_standard@plt
97 fmr fp12,fp1
98 b .L4
99 END (__sqrtf)
100
101weak_alias (__sqrtf, sqrtf)
102