]>
Commit | Line | Data |
---|---|---|
8da2915d UD |
1 | .file "floor.s" |
2 | ||
a334319f | 3 | // Copyright (C) 2000, 2001, Intel Corporation |
8da2915d | 4 | // All rights reserved. |
a334319f UD |
5 | // |
6 | // Contributed 2/2/2000 by John Harrison, Ted Kubaska, Bob Norin, Shane Story, | |
7 | // and Ping Tak Peter Tang of the Computational Software Lab, Intel Corporation. | |
aeb25823 AJ |
8 | // |
9 | // Redistribution and use in source and binary forms, with or without | |
10 | // modification, are permitted provided that the following conditions are | |
11 | // met: | |
12 | // | |
13 | // * Redistributions of source code must retain the above copyright | |
14 | // notice, this list of conditions and the following disclaimer. | |
15 | // | |
16 | // * Redistributions in binary form must reproduce the above copyright | |
17 | // notice, this list of conditions and the following disclaimer in the | |
18 | // documentation and/or other materials provided with the distribution. | |
19 | // | |
20 | // * The name of Intel Corporation may not be used to endorse or promote | |
21 | // products derived from this software without specific prior written | |
22 | // permission. | |
a334319f UD |
23 | // |
24 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
25 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
8da2915d | 26 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
a334319f | 27 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS |
8da2915d | 28 | // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
a334319f UD |
29 | // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
30 | // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR | |
31 | // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY | |
8da2915d | 32 | // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING |
a334319f UD |
33 | // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
34 | // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
35 | // | |
8da2915d | 36 | // Intel Corporation is the author of this code, and requests that all |
a334319f UD |
37 | // problem reports or change requests be submitted to it directly at |
38 | // http://developer.intel.com/opensource. | |
8da2915d | 39 | // |
a334319f UD |
40 | .align 32 |
41 | .global floor# | |
42 | ||
43 | .section .text | |
44 | .proc floor# | |
45 | .align 32 | |
46 | ||
8da2915d UD |
47 | // History |
48 | //============================================================== | |
a334319f UD |
49 | // 2/02/00: Initial version |
50 | // 3/22/00: Updated to improve performance | |
51 | // 6/13/00: Improved speed, fixed setting of inexact flag | |
52 | // 6/27/00: Eliminated incorrect invalid flag setting | |
53 | // 2/07/01: Corrected sign of zero result in round to -inf mode | |
8da2915d UD |
54 | |
55 | // API | |
56 | //============================================================== | |
57 | // double floor(double x) | |
58 | ||
a334319f UD |
59 | // general input registers: |
60 | ||
61 | floor_GR_FFFF = r14 | |
62 | floor_GR_signexp = r15 | |
63 | floor_GR_exponent = r16 | |
64 | floor_GR_expmask = r17 | |
65 | floor_GR_bigexp = r18 | |
66 | ||
8da2915d | 67 | |
a334319f | 68 | // predicate registers used: |
8da2915d | 69 | |
a334319f UD |
70 | // p6 ==> Input is NaN, infinity, zero |
71 | // p7 ==> Input is denormal | |
72 | // p8 ==> Input is <0 | |
73 | // p9 ==> Input is >=0 | |
74 | // p10 ==> Input is already an integer (bigger than largest integer) | |
75 | // p11 ==> Input is not a large integer | |
76 | // p12 ==> Input is a smaller integer | |
77 | // p13 ==> Input is not an even integer, so inexact must be set | |
8da2915d | 78 | |
8da2915d | 79 | |
a334319f UD |
80 | // floating-point registers used: |
81 | ||
82 | FLOOR_NORM_f8 = f9 | |
83 | FLOOR_FFFF = f10 | |
84 | FLOOR_INEXACT = f11 | |
85 | FLOOR_FLOAT_INT_f8 = f12 | |
86 | FLOOR_INT_f8 = f13 | |
87 | FLOOR_adj = f14 | |
8da2915d UD |
88 | |
89 | // Overview of operation | |
90 | //============================================================== | |
a334319f | 91 | |
8da2915d | 92 | // double floor(double x) |
a334319f | 93 | // Return an integer value (represented as a double) that is the largest |
8da2915d UD |
94 | // value not greater than x |
95 | // This is x rounded toward -infinity to an integral value. | |
96 | // Inexact is set if x != floor(x) | |
a334319f UD |
97 | // ************************************************************************** |
98 | ||
99 | // Set denormal flag for denormal input and | |
100 | // and take denormal fault if necessary. | |
101 | ||
102 | // Is the input an integer value already? | |
8da2915d UD |
103 | |
104 | // double_extended | |
105 | // if the exponent is > 1003e => 3F(true) = 63(decimal) | |
106 | // we have a significand of 64 bits 1.63-bits. | |
107 | // If we multiply by 2^63, we no longer have a fractional part | |
108 | // So input is an integer value already. | |
109 | ||
110 | // double | |
111 | // if the exponent is >= 10033 => 34(true) = 52(decimal) | |
112 | // 34 + 3ff = 433 | |
113 | // we have a significand of 53 bits 1.52-bits. (implicit 1) | |
114 | // If we multiply by 2^52, we no longer have a fractional part | |
115 | // So input is an integer value already. | |
116 | ||
117 | // single | |
118 | // if the exponent is > 10016 => 17(true) = 23(decimal) | |
119 | // we have a significand of 24 bits 1.23-bits. (implicit 1) | |
120 | // If we multiply by 2^23, we no longer have a fractional part | |
121 | // So input is an integer value already. | |
122 | ||
a334319f | 123 | // If x is NAN, ZERO, or INFINITY, then return |
8da2915d | 124 | |
a334319f UD |
125 | // qnan snan inf norm unorm 0 -+ |
126 | // 1 1 1 0 0 1 11 0xe7 | |
127 | ||
128 | #include "libm_support.h" | |
129 | ||
130 | floor: | |
131 | #ifdef _LIBC | |
132 | .global __floor | |
133 | __floor: | |
134 | #endif | |
8da2915d UD |
135 | |
136 | { .mfi | |
a334319f UD |
137 | getf.exp floor_GR_signexp = f8 |
138 | fcvt.fx.trunc.s1 FLOOR_INT_f8 = f8 | |
139 | addl floor_GR_bigexp = 0x10033, r0 | |
8da2915d UD |
140 | } |
141 | { .mfi | |
a334319f UD |
142 | addl floor_GR_FFFF = -1,r0 |
143 | fcmp.lt.s1 p8,p9 = f8,f0 | |
144 | mov floor_GR_expmask = 0x1FFFF ;; | |
8da2915d UD |
145 | } |
146 | ||
a334319f | 147 | // p7 ==> denorm |
8da2915d | 148 | { .mfi |
a334319f UD |
149 | setf.sig FLOOR_FFFF = floor_GR_FFFF |
150 | fclass.m p7,p0 = f8, 0x0b | |
151 | nop.i 999 | |
8da2915d | 152 | } |
a334319f UD |
153 | { .mfi |
154 | nop.m 999 | |
155 | fnorm.s1 FLOOR_NORM_f8 = f8 | |
156 | nop.i 999 ;; | |
8da2915d UD |
157 | } |
158 | ||
a334319f UD |
159 | // p6 ==> NAN, INF, ZERO |
160 | { .mfb | |
161 | nop.m 999 | |
162 | fclass.m p6,p10 = f8, 0xe7 | |
163 | (p7) br.cond.spnt L(FLOOR_DENORM) ;; | |
8da2915d UD |
164 | } |
165 | ||
a334319f | 166 | L(FLOOR_COMMON): |
8da2915d | 167 | .pred.rel "mutex",p8,p9 |
a334319f UD |
168 | // Set adjustment to subtract from trunc(x) for result |
169 | // If x<0, adjustment is -1.0 | |
170 | // If x>=0, adjustment is 0.0 | |
8da2915d | 171 | { .mfi |
a334319f UD |
172 | and floor_GR_exponent = floor_GR_signexp, floor_GR_expmask |
173 | (p8) fnma.s1 FLOOR_adj = f1,f1,f0 | |
174 | nop.i 999 | |
8da2915d UD |
175 | } |
176 | { .mfi | |
a334319f UD |
177 | nop.m 999 |
178 | (p9) fadd.s1 FLOOR_adj = f0,f0 | |
179 | nop.i 999 ;; | |
8da2915d UD |
180 | } |
181 | ||
182 | { .mfi | |
a334319f UD |
183 | nop.m 999 |
184 | fcmp.eq.s0 p12,p0 = f8,f0 // Dummy op to set denormal and invalid flag | |
185 | nop.i 999 | |
8da2915d | 186 | } |
a334319f UD |
187 | { .mfi |
188 | (p10) cmp.ge.unc p10,p11 = floor_GR_exponent, floor_GR_bigexp | |
189 | (p6) fnorm.d f8 = f8 | |
190 | nop.i 999 ;; | |
8da2915d UD |
191 | } |
192 | ||
a334319f UD |
193 | { .mfi |
194 | nop.m 999 | |
195 | (p11) fcvt.xf FLOOR_FLOAT_INT_f8 = FLOOR_INT_f8 | |
196 | nop.i 999 ;; | |
8da2915d UD |
197 | } |
198 | ||
199 | { .mfi | |
a334319f UD |
200 | nop.m 999 |
201 | (p10) fnorm.d f8 = FLOOR_NORM_f8 | |
202 | nop.i 999 ;; | |
8da2915d | 203 | } |
a334319f UD |
204 | |
205 | ||
8da2915d | 206 | { .mfi |
a334319f UD |
207 | nop.m 999 |
208 | (p11) fadd.d f8 = FLOOR_FLOAT_INT_f8,FLOOR_adj | |
209 | nop.i 999 ;; | |
8da2915d UD |
210 | } |
211 | ||
212 | { .mfi | |
a334319f UD |
213 | nop.m 999 |
214 | (p11) fcmp.eq.unc.s1 p12,p13 = FLOOR_FLOAT_INT_f8, FLOOR_NORM_f8 | |
215 | nop.i 999 ;; | |
8da2915d UD |
216 | } |
217 | ||
a334319f | 218 | // Set inexact if result not equal to input |
8da2915d | 219 | { .mfi |
a334319f UD |
220 | nop.m 999 |
221 | (p13) fmpy.s0 FLOOR_INEXACT = FLOOR_FFFF,FLOOR_FFFF | |
222 | nop.i 999 | |
8da2915d | 223 | } |
a334319f | 224 | // Set result to input if integer |
8da2915d | 225 | { .mfb |
a334319f UD |
226 | nop.m 999 |
227 | (p12) fnorm.d f8 = FLOOR_NORM_f8 | |
228 | br.ret.sptk b0 ;; | |
8da2915d UD |
229 | } |
230 | ||
a334319f UD |
231 | // Here if input denorm |
232 | L(FLOOR_DENORM): | |
8da2915d | 233 | { .mfb |
a334319f UD |
234 | getf.exp floor_GR_signexp = FLOOR_NORM_f8 |
235 | fcvt.fx.trunc.s1 FLOOR_INT_f8 = FLOOR_NORM_f8 | |
236 | br.cond.sptk L(FLOOR_COMMON) ;; | |
8da2915d UD |
237 | } |
238 | ||
a334319f UD |
239 | .endp floor |
240 | ASM_SIZE_DIRECTIVE(floor) |