about summary refs log tree commit diff
path: root/sysdeps/ia64/fpu/libm_frexp4f.S
blob: d94ad095a04dde2ae23704ac46867fdef965f9ed (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
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
.file "libm_frexp_4f.s"

// Copyright (c) 2000, 2001, Intel Corporation
// All rights reserved.
// 
// Contributed 2/2/2000 by John Harrison, Ted Kubaska, Bob Norin, Shane Story,
// and Ping Tak Peter Tang of the Computational Software Lab, Intel Corporation.
// 
// WARRANTY DISCLAIMER
// 
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS 
// CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
// EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 
// PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 
// PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY 
// OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING
// NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 
// SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 
// 
// Intel Corporation is the author of this code, and requests that all
// problem reports or change requests be submitted to it directly at 
// http://developer.intel.com/opensource.
//
// History
//==============================================================
// 2/02/00: Initial version
// 3/20/00: Improved speed
// 6/01/00: Fixed bug when x a double-extended denormal
// 12/08/00 Corrected label on .endp
//
// API
//==============================================================
// float frexp(float x, int* y)
// float __libm_frexp_4f(float x, int* y)
//   where int* y is a 32-bit integer
//
// Overview of operation
//==============================================================
// break a floating point x number into fraction and an exponent
// The fraction is returned as a float
// The exponent is returned as an integer pointed to by y
//    This is a true (not a biased exponent) but 0fffe is subtracted
//    as a bias instead of 0xffff. This is because the fraction returned
//    is between 0.5 and 1.0, not the expected IEEE range.
//
// The fraction is 0.5 <= fraction < 1.0
//
// Registers used
//==============================================================

// general registers: 
// r14  exponent bias for x negative
// r15  exponent bias for x positive
// r16  signexp of x
// r17  exponent mask
// r18  exponent of x
// r19  exponent result
// r20  signexp of 2^64
// r32  on input contains the 32-bit IEEE float that is in f8
// r33  on input pointer to 32-bit integer for exponent

// predicate registers:
// p6   set if x is Nan, zero, or infinity
// p7   set if x negative
// p8   set if x positive
// p9   set if x double-extended denormal

// floating-point registers:
// f8  input, output
// f9  normalized x
// f10 signexp for significand result for x positive
// f11 signexp for significand result for x negative
// f12 2^64

#include "libm_support.h"

.align 32
.global __libm_frexp_4f#

.section .text
.proc  __libm_frexp_4f#
.align 32

__libm_frexp_4f: 

// Set signexp for significand result for x>0
// If x is a NaN, zero, or infinity, return it.
// Put 0 in the int pointer.
// x NAN, ZERO, INFINITY?
// Set signexp for significand result for x<0
{ .mfi
(p0)    mov         r15 = 0x0fffe
(p0)    fclass.m.unc p6,p0 = f8, 0xe7
(p0)    mov         r14 = 0x2fffe
}
// Form signexp of 2^64 in case x double-extended denormal
// Save the normalized value of input in f9
// The normalization also sets fault flags and takes faults if necessary
{ .mfi
(p0)    mov         r20 = 0x1003f
(p0)    fnorm       f9 = f8 
        nop.i 999 ;;
}

// Move signexp for significand result for x>0 to FP reg
// Form 2^64 in case x double-extended denormal
{ .mmi
(p0)    setf.exp    f10 = r15
(p0)    setf.exp    f12 = r20
        nop.i 999 ;;
}

// Move signexp for significand result for x<0 to FP reg
// If x NAN, ZERO, INFINITY, set *y=0 as a 32-bit integer, and exit
{ .mmb
(p0)    setf.exp    f11 = r14
(p6)    st4         [r33] = r0
(p6)    br.ret.spnt b0 ;;
}

// Form exponent mask
// p7 if x<0, else p8
{ .mfi
(p0)    mov         r17 = 0x1ffff
(p0)    fcmp.lt.unc p7,p8 = f8,f0
        nop.i 999 ;;
}

// Test for fnorm(x) denormal, means x double-extended denormal
{ .mfi
        nop.m 999
(p0)    fclass.m.unc p9,p0 = f9, 0x0b
        nop.i 999 ;;
}

// If x double-extended denormal add 64 to exponent bias for scaling
// If x double-extended denormal multiply x * 2^64 which is normal
{ .mfi
(p9)    add         r15 = 64, r15
(p9)    fmpy        f9 = f9, f12
        nop.i 999 ;;
}

// true exponent stored to int pointer
// the bias is treated as 0xfffe instead of 
// normal 0xffff because we want the significand
// to be in the range <=0.5 sig < 1.0
// Store the value of the exponent at the pointer in r33

// If x>0 form significand result 
{ .mfi
        nop.m 999
(p8)    fmerge.se   f8 = f10,f9
        nop.i 999  ;;
}

// Get signexp of normalized x
// If x<0 form significand result 
{ .mfi
(p0)    getf.exp    r16 = f9
(p7)    fmerge.se   f8 = f11,f9
        nop.i 999  ;;
}

// Get exp of normalized x
// Subtract off bias to get true exponent of x
{ .mmi
(p0)    and         r18 = r17,r16 ;;
(p0)    sub         r19 = r18,r15
        nop.i 999  ;;
}

// Store int y as a 32-bit integer
// Make the value a float
{ .mfb
(p0)    st4         [r33] = r19
(p0)    fnorm.s     f8 = f8
(p0)    br.ret.sptk b0 ;;
}

.endp __libm_frexp_4f
ASM_SIZE_DIRECTIVE(__libm_frexp_4f)