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
|
.file "floorl.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.
//
.align 32
.global floorl#
.section .text
.proc floorl#
.align 32
// History
//==============================================================
// 2/02/00: Initial version
// 6/13/00: Improved speed
// 6/27/00: Eliminated incorrect invalid flag setting
// 2/07/01: Corrected sign of zero result in round to -inf mode
// API
//==============================================================
// long double floorl(long double x)
// general input registers:
floor_GR_FFFF = r14
floor_GR_signexp = r15
floor_GR_exponent = r16
floor_GR_expmask = r17
floor_GR_bigexp = r18
// predicate registers used:
// p6 ==> Input is NaN, infinity, zero
// p7 ==> Input is denormal
// p8 ==> Input is <0
// p9 ==> Input is >=0
// p10 ==> Input is already an integer (bigger than largest integer)
// p11 ==> Input is not a large integer
// p12 ==> Input is a smaller integer
// p13 ==> Input is not an even integer, so inexact must be set
// floating-point registers used:
FLOOR_NORM_f8 = f9
FLOOR_FFFF = f10
FLOOR_INEXACT = f11
FLOOR_FLOAT_INT_f8 = f12
FLOOR_INT_f8 = f13
FLOOR_adj = f14
// Overview of operation
//==============================================================
// long double floorl(long double x)
// Return an integer value (represented as a long double) that is the largest
// value not greater than x
// This is x rounded toward -infinity to an integral value.
// Inexact is set if x != floorl(x)
// **************************************************************************
// Set denormal flag for denormal input and
// and take denormal fault if necessary.
// Is the input an integer value already?
// double_extended
// if the exponent is > 1003e => 3F(true) = 63(decimal)
// we have a significand of 64 bits 1.63-bits.
// If we multiply by 2^63, we no longer have a fractional part
// So input is an integer value already.
// double
// if the exponent is >= 10033 => 34(true) = 52(decimal)
// 34 + 3ff = 433
// we have a significand of 53 bits 1.52-bits. (implicit 1)
// If we multiply by 2^52, we no longer have a fractional part
// So input is an integer value already.
// single
// if the exponent is > 10016 => 17(true) = 23(decimal)
// we have a significand of 24 bits 1.23-bits. (implicit 1)
// If we multiply by 2^23, we no longer have a fractional part
// So input is an integer value already.
// If x is NAN, ZERO, or INFINITY, then return
// qnan snan inf norm unorm 0 -+
// 1 1 1 0 0 1 11 0xe7
#include "libm_support.h"
floorl:
#ifdef _LIBC
.global __floorl
__floorl:
#endif
{ .mfi
getf.exp floor_GR_signexp = f8
fcvt.fx.trunc.s1 FLOOR_INT_f8 = f8
addl floor_GR_bigexp = 0x1003e, r0
}
{ .mfi
addl floor_GR_FFFF = -1,r0
fcmp.lt.s1 p8,p9 = f8,f0
mov floor_GR_expmask = 0x1FFFF ;;
}
// p7 ==> denorm
{ .mfi
setf.sig FLOOR_FFFF = floor_GR_FFFF
fclass.m p7,p0 = f8, 0x0b
nop.i 999
}
{ .mfi
nop.m 999
fnorm.s1 FLOOR_NORM_f8 = f8
nop.i 999 ;;
}
// p6 ==> NAN, INF, ZERO
{ .mfb
nop.m 999
fclass.m p6,p10 = f8, 0xe7
(p7) br.cond.spnt L(FLOOR_DENORM) ;;
}
L(FLOOR_COMMON):
.pred.rel "mutex",p8,p9
// Set adjustment to subtract from trunc(x) for result
// If x<0, adjustment is -1.0
// If x>=0, adjustment is 0.0
{ .mfi
and floor_GR_exponent = floor_GR_signexp, floor_GR_expmask
(p8) fnma.s1 FLOOR_adj = f1,f1,f0
nop.i 999
}
{ .mfi
nop.m 999
(p9) fadd.s1 FLOOR_adj = f0,f0
nop.i 999 ;;
}
{ .mfi
nop.m 999
fcmp.eq.s0 p12,p0 = f8,f0 // Dummy op to set denormal and invalid flag
nop.i 999
}
{ .mfi
(p10) cmp.ge.unc p10,p11 = floor_GR_exponent, floor_GR_bigexp
(p6) fnorm f8 = f8
nop.i 999 ;;
}
{ .mfi
nop.m 999
(p11) fcvt.xf FLOOR_FLOAT_INT_f8 = FLOOR_INT_f8
nop.i 999 ;;
}
{ .mfi
nop.m 999
(p10) fnorm f8 = FLOOR_NORM_f8
nop.i 999 ;;
}
{ .mfi
nop.m 999
(p11) fadd f8 = FLOOR_FLOAT_INT_f8,FLOOR_adj
nop.i 999 ;;
}
{ .mfi
nop.m 999
(p11) fcmp.eq.unc.s1 p12,p13 = FLOOR_FLOAT_INT_f8, FLOOR_NORM_f8
nop.i 999 ;;
}
// Set inexact if result not equal to input
{ .mfi
nop.m 999
(p13) fmpy.s0 FLOOR_INEXACT = FLOOR_FFFF,FLOOR_FFFF
nop.i 999
}
// Set result to input if integer
{ .mfb
nop.m 999
(p12) fnorm f8 = FLOOR_NORM_f8
br.ret.sptk b0 ;;
}
// Here if input denorm
L(FLOOR_DENORM):
{ .mfb
getf.exp floor_GR_signexp = FLOOR_NORM_f8
fcvt.fx.trunc.s1 FLOOR_INT_f8 = FLOOR_NORM_f8
br.cond.sptk L(FLOOR_COMMON) ;;
}
.endp floorl
ASM_SIZE_DIRECTIVE(floorl)
|
