]>
Commit | Line | Data |
---|---|---|
d5efd131 MF |
1 | .file "floor.s" |
2 | ||
3 | ||
4 | // Copyright (c) 2000 - 2003, Intel Corporation | |
5 | // All rights reserved. | |
6 | // | |
d5efd131 MF |
7 | // |
8 | // Redistribution and use in source and binary forms, with or without | |
9 | // modification, are permitted provided that the following conditions are | |
10 | // met: | |
11 | // | |
12 | // * Redistributions of source code must retain the above copyright | |
13 | // notice, this list of conditions and the following disclaimer. | |
14 | // | |
15 | // * Redistributions in binary form must reproduce the above copyright | |
16 | // notice, this list of conditions and the following disclaimer in the | |
17 | // documentation and/or other materials provided with the distribution. | |
18 | // | |
19 | // * The name of Intel Corporation may not be used to endorse or promote | |
20 | // products derived from this software without specific prior written | |
21 | // permission. | |
22 | ||
23 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS | |
24 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
25 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR | |
26 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS | |
27 | // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, | |
28 | // EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, | |
29 | // PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR | |
30 | // PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY | |
31 | // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING | |
32 | // NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS | |
33 | // SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. | |
34 | // | |
35 | // Intel Corporation is the author of this code, and requests that all | |
36 | // problem reports or change requests be submitted to it directly at | |
37 | // http://www.intel.com/software/products/opensource/libraries/num.htm. | |
38 | // | |
39 | // History | |
40 | //============================================================== | |
41 | // 02/02/00 Initial version | |
42 | // 03/22/00 Updated to improve performance | |
43 | // 06/13/00 Improved speed, fixed setting of inexact flag | |
44 | // 06/27/00 Eliminated incorrect invalid flag setting | |
45 | // 02/07/01 Corrected sign of zero result in round to -inf mode | |
46 | // 05/20/02 Cleaned up namespace and sf0 syntax | |
47 | // 01/28/03 Improved performance | |
48 | //============================================================== | |
49 | ||
50 | // API | |
51 | //============================================================== | |
52 | // double floor(double x) | |
53 | //============================================================== | |
54 | ||
55 | // general input registers: | |
56 | // r14 - r18 | |
57 | ||
58 | rSignexp = r14 | |
59 | rExp = r15 | |
60 | rExpMask = r16 | |
61 | rBigexp = r17 | |
62 | rM1 = r18 | |
63 | ||
64 | // floating-point registers: | |
65 | // f8 - f13 | |
66 | ||
67 | fXInt = f9 | |
68 | fNormX = f10 | |
69 | fTmp = f11 | |
70 | fAdj = f12 | |
71 | fPreResult = f13 | |
72 | ||
73 | // predicate registers used: | |
74 | // p6 - p9 | |
75 | ||
76 | // Overview of operation | |
77 | //============================================================== | |
78 | // double floor(double x) | |
79 | // Return an integer value (represented as a double) that is the largest | |
80 | // value not greater than x | |
81 | // This is x rounded toward -infinity to an integral value. | |
82 | // Inexact is set if x != floor(x) | |
83 | //============================================================== | |
84 | ||
85 | // double_extended | |
86 | // if the exponent is > 1003e => 3F(true) = 63(decimal) | |
87 | // we have a significand of 64 bits 1.63-bits. | |
88 | // If we multiply by 2^63, we no longer have a fractional part | |
89 | // So input is an integer value already. | |
90 | ||
91 | // double | |
92 | // if the exponent is >= 10033 => 34(true) = 52(decimal) | |
93 | // 34 + 3ff = 433 | |
94 | // we have a significand of 53 bits 1.52-bits. (implicit 1) | |
95 | // If we multiply by 2^52, we no longer have a fractional part | |
96 | // So input is an integer value already. | |
97 | ||
98 | // single | |
99 | // if the exponent is > 10016 => 17(true) = 23(decimal) | |
100 | // we have a significand of 24 bits 1.23-bits. (implicit 1) | |
101 | // If we multiply by 2^23, we no longer have a fractional part | |
102 | // So input is an integer value already. | |
103 | ||
104 | ||
105 | .section .text | |
106 | GLOBAL_IEEE754_ENTRY(floor) | |
107 | ||
108 | { .mfi | |
109 | getf.exp rSignexp = f8 // Get signexp, recompute if unorm | |
110 | fclass.m p7,p0 = f8, 0x0b // Test x unorm | |
111 | addl rBigexp = 0x10033, r0 // Set exponent at which is integer | |
112 | } | |
113 | { .mfi | |
114 | mov rM1 = -1 // Set all ones | |
115 | fcvt.fx.trunc.s1 fXInt = f8 // Convert to int in significand | |
116 | mov rExpMask = 0x1FFFF // Form exponent mask | |
117 | } | |
118 | ;; | |
119 | ||
120 | { .mfi | |
121 | nop.m 0 | |
122 | fcmp.lt.s1 p8,p9 = f8, f0 // Test x < 0 | |
123 | nop.i 0 | |
124 | } | |
125 | { .mfb | |
126 | setf.sig fTmp = rM1 // Make const for setting inexact | |
127 | fnorm.s1 fNormX = f8 // Normalize input | |
128 | (p7) br.cond.spnt FLOOR_UNORM // Branch if x unorm | |
129 | } | |
130 | ;; | |
131 | ||
132 | FLOOR_COMMON: | |
133 | // Return here from FLOOR_UNORM | |
134 | { .mfi | |
135 | nop.m 0 | |
136 | fclass.m p6,p0 = f8, 0x1e7 // Test x natval, nan, inf, 0 | |
137 | nop.i 0 | |
138 | } | |
139 | ;; | |
140 | ||
141 | .pred.rel "mutex",p8,p9 | |
142 | { .mfi | |
143 | nop.m 0 | |
144 | (p8) fnma.s1 fAdj = f1, f1, f0 // If x < 0, adjustment is -1 | |
145 | nop.i 0 | |
146 | } | |
147 | { .mfi | |
148 | nop.m 0 | |
149 | (p9) fma.s1 fAdj = f0, f0, f0 // If x > 0, adjustment is 0 | |
150 | nop.i 0 | |
151 | } | |
152 | ;; | |
153 | ||
154 | { .mfi | |
155 | nop.m 0 | |
156 | fcvt.xf fPreResult = fXInt // trunc(x) | |
157 | nop.i 0 | |
158 | } | |
159 | { .mfb | |
160 | nop.m 0 | |
161 | (p6) fma.d.s0 f8 = f8, f1, f0 // Result if x natval, nan, inf, 0 | |
162 | (p6) br.ret.spnt b0 // Exit if x natval, nan, inf, 0 | |
163 | } | |
164 | ;; | |
165 | ||
166 | { .mmi | |
167 | and rExp = rSignexp, rExpMask // Get biased exponent | |
168 | ;; | |
169 | cmp.ge p7,p6 = rExp, rBigexp // Is |x| >= 2^52? | |
170 | nop.i 0 | |
171 | } | |
172 | ;; | |
173 | ||
174 | { .mfi | |
175 | nop.m 0 | |
176 | (p6) fma.d.s0 f8 = fPreResult, f1, fAdj // Result if !int, |x| < 2^52 | |
177 | nop.i 0 | |
178 | } | |
179 | { .mfi | |
180 | nop.m 0 | |
181 | (p7) fma.d.s0 f8 = fNormX, f1, f0 // Result, if |x| >= 2^52 | |
182 | nop.i 0 | |
183 | } | |
184 | ;; | |
185 | ||
186 | { .mfi | |
187 | nop.m 0 | |
188 | (p6) fcmp.eq.unc.s1 p8, p9 = fPreResult, fNormX // Is trunc(x) = x ? | |
189 | nop.i 0 | |
190 | } | |
191 | ;; | |
192 | ||
193 | { .mfi | |
194 | nop.m 0 | |
195 | (p9) fmpy.s0 fTmp = fTmp, fTmp // Dummy to set inexact | |
196 | nop.i 0 | |
197 | } | |
198 | { .mfb | |
199 | nop.m 0 | |
200 | (p8) fma.d.s0 f8 = fNormX, f1, f0 // If x int, result normalized x | |
201 | br.ret.sptk b0 // Exit main path, 0 < |x| < 2^52 | |
202 | } | |
203 | ;; | |
204 | ||
205 | ||
206 | FLOOR_UNORM: | |
207 | // Here if x unorm | |
208 | { .mfb | |
209 | getf.exp rSignexp = fNormX // Get signexp, recompute if unorm | |
210 | fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag | |
211 | br.cond.sptk FLOOR_COMMON // Return to main path | |
212 | } | |
213 | ;; | |
214 | ||
215 | GLOBAL_IEEE754_END(floor) | |
0609ec0a | 216 | libm_alias_double_other (__floor, floor) |