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
|
.file "roundl.s"
// Copyright (C) 2000, 2001, Intel Corporation
// All rights reserved.
//
// Contributed 10/25/2000 by John Harrison, Cristina Iordache, Ted Kubaska,
// Bob Norin, Tom Rowan, Shane Story, and Ping Tak Peter Tang of the
// Computational Software Lab, Intel Corporation.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
// * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//
// * Redistributions in binary form must reproduce the above copyright
// notice, this list of conditions and the following disclaimer in the
// documentation and/or other materials provided with the distribution.
//
// * The name of Intel Corporation may not be used to endorse or promote
// products derived from this software without specific prior written
// permission.
//
// 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
//==============================================================
// 10/25/2000: Created
//==============================================================
//
// API
//==============================================================
// long double roundl(long double x)
//
#include "libm_support.h"
// general input registers:
//
roundl_GR_half = r14
roundl_GR_big = r15
roundl_GR_expmask = r16
roundl_GR_signexp = r17
roundl_GR_exp = r18
roundl_GR_expdiff = r19
// predicate registers used:
// p6 - p10
// floating-point registers used:
ROUNDL_NORM_f8 = f9
ROUNDL_TRUNC_f8 = f10
ROUNDL_RINT_f8 = f11
ROUNDL_FLOAT_TRUNC_f8 = f12
ROUNDL_FLOAT_RINT_f8 = f13
ROUNDL_REMAINDER = f14
ROUNDL_HALF = f15
// Overview of operation
//==============================================================
// long double roundl(long double x)
// Return an integer value (represented as a long double) that is x
// rounded to nearest integer, halfway cases rounded away from
// zero.
// if x>0 result = trunc(x+0.5)
// if x<0 result = trunc(x-0.5)
// *******************************************************************************
// Set denormal flag for denormal input and
// and take denormal fault if necessary.
// If x is NAN, ZERO, INFINITY, or >= 2^63 then return
// qnan snan inf norm unorm 0 -+
// 1 1 1 0 0 1 11 0xe7
.align 32
.global roundl#
.section .text
.proc roundl#
.align 32
roundl:
// Get exponent for +0.5
// Truncate x to integer
{ .mfi
addl roundl_GR_half = 0x0fffe, r0
fcvt.fx.trunc.s1 ROUNDL_TRUNC_f8 = f8
nop.i 999
}
// Get signexp of x
// Normalize input
// Form exponent mask
{ .mfi
getf.exp roundl_GR_signexp = f8
fnorm ROUNDL_NORM_f8 = f8
addl roundl_GR_expmask = 0x1ffff, r0 ;;
}
// Form +0.5
// Round x to integer
{ .mfi
setf.exp ROUNDL_HALF = roundl_GR_half
fcvt.fx.s1 ROUNDL_RINT_f8 = f8
nop.i 999 ;;
}
// Get exp of x
// Test for NAN, INF, ZERO
// Get exponent at which input has no fractional part
{ .mfi
and roundl_GR_exp = roundl_GR_expmask, roundl_GR_signexp
fclass.m p8,p9 = f8,0xe7
addl roundl_GR_big = 0x1003e, r0 ;;
}
// Get exp-bigexp
// If exp is so big there is no fractional part, then turn on p8, off p9
{ .mmi
sub roundl_GR_expdiff = roundl_GR_exp, roundl_GR_big ;;
#ifdef _LIBC
(p9) cmp.lt.or.andcm p8,p9 = r0, roundl_GR_expdiff
#else
(p9) cmp.ge.or.andcm p8,p9 = roundl_GR_expdiff, r0
#endif
nop.i 999 ;;
}
// Set p6 if x<0, else set p7
{ .mfi
nop.m 999
(p9) fcmp.lt.unc p6,p7 = f8,f0
nop.i 999
}
// If NAN, INF, ZERO, or no fractional part, result is just normalized input
{ .mfi
nop.m 999
(p8) fnorm.s0 f8 = f8
nop.i 999 ;;
}
// Float the truncated integer
{ .mfi
nop.m 999
(p9) fcvt.xf ROUNDL_FLOAT_TRUNC_f8 = ROUNDL_TRUNC_f8
nop.i 999 ;;
}
// Float the rounded integer to get preliminary result
{ .mfi
nop.m 999
(p9) fcvt.xf ROUNDL_FLOAT_RINT_f8 = ROUNDL_RINT_f8
nop.i 999 ;;
}
// If x<0 and the difference of the truncated input minus the input is 0.5
// then result = truncated input - 1.0
// Else if x>0 and the difference of the input minus truncated input is 0.5
// then result = truncated input + 1.0
// Else
// result = rounded input
// Endif
{ .mfi
nop.m 999
(p6) fsub.s1 ROUNDL_REMAINDER = ROUNDL_FLOAT_TRUNC_f8, ROUNDL_NORM_f8
nop.i 999
}
{ .mfi
nop.m 999
(p7) fsub.s1 ROUNDL_REMAINDER = ROUNDL_NORM_f8, ROUNDL_FLOAT_TRUNC_f8
nop.i 999 ;;
}
// Assume preliminary result is rounded integer
{ .mfi
nop.m 999
(p9) fnorm.s0 f8 = ROUNDL_FLOAT_RINT_f8
nop.i 999
}
// If x<0, test if result=0
{ .mfi
nop.m 999
(p6) fcmp.eq.unc p10,p0 = ROUNDL_FLOAT_RINT_f8,f0
nop.i 999 ;;
}
// If x<0 and result=0, set result=-0
{ .mfi
nop.m 999
(p10) fmerge.ns f8 = f1,f8
nop.i 999
}
// If x<0, test if remainder=0.5
{ .mfi
nop.m 999
(p6) fcmp.eq.unc p6,p0 = ROUNDL_REMAINDER, ROUNDL_HALF
nop.i 999 ;;
}
// If x>0, test if remainder=0.5
{ .mfi
nop.m 999
(p7) fcmp.eq.unc p7,p0 = ROUNDL_REMAINDER, ROUNDL_HALF
nop.i 999 ;;
}
// If x<0 and remainder=0.5, result=truncated-1.0
// If x>0 and remainder=0.5, result=truncated+1.0
// Exit
.pred.rel "mutex",p6,p7
{ .mfi
nop.m 999
(p6) fsub.s0 f8 = ROUNDL_FLOAT_TRUNC_f8,f1
nop.i 999
}
{ .mfb
nop.m 999
(p7) fadd.s0 f8 = ROUNDL_FLOAT_TRUNC_f8,f1
br.ret.sptk b0 ;;
}
.endp roundl
ASM_SIZE_DIRECTIVE(roundl)
|