]>
Commit | Line | Data |
---|---|---|
d5efd131 MF |
1 | .file "modf.s" |
2 | ||
3 | ||
4 | // Copyright (c) 2000 - 2003, Intel Corporation | |
5 | // All rights reserved. | |
6 | // | |
7 | // Contributed 2000 by the Intel Numerics Group, Intel Corporation | |
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. | |
23 | ||
0347518d MF |
24 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
25 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
d5efd131 | 26 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
0347518d | 27 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS |
d5efd131 | 28 | // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
0347518d MF |
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 | |
d5efd131 | 32 | // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING |
0347518d MF |
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 | // | |
d5efd131 | 36 | // Intel Corporation is the author of this code, and requests that all |
0347518d | 37 | // problem reports or change requests be submitted to it directly at |
d5efd131 MF |
38 | // http://www.intel.com/software/products/opensource/libraries/num.htm. |
39 | // | |
40 | // History | |
41 | //============================================================== | |
42 | // 02/02/00 Initial version | |
43 | // 04/04/00 Improved speed, corrected result for NaN input | |
0347518d | 44 | // 12/22/00 Fixed so inexact flag is never set, and invalid is not set for |
d5efd131 MF |
45 | // qnans nor for inputs larger than 2^63. |
46 | // 05/20/02 Cleaned up namespace and sf0 syntax | |
47 | // 02/10/03 Reordered header: .section, .global, .proc, .align | |
48 | // | |
49 | // API | |
50 | //============================================================== | |
51 | // double modf(double x, double *iptr) | |
52 | // break a floating point x number into fraction and an exponent | |
53 | // | |
54 | // input floating point f8, address in r33 | |
55 | // output floating point f8 (x fraction), and *iptr (x integral part) | |
56 | // | |
57 | // OVERVIEW | |
58 | //============================================================== | |
59 | // | |
60 | // NO FRACTIONAL PART: HUGE | |
61 | // If | |
62 | // for double-extended | |
63 | // If the true exponent is greater than or equal 63 | |
64 | // 1003e ==> 1003e -ffff = 3f = 63(dec) | |
65 | // for double | |
66 | // If the true exponent is greater than or equal 52 | |
67 | // 10033 -ffff = 34 = 52(dec) | |
68 | // for single | |
69 | // If the true exponent is greater than or equal 23 | |
70 | // 10016 -ffff = 17 = 23(dec) | |
71 | // then | |
72 | // we are already an integer (p9 true) | |
73 | ||
74 | // NO INTEGER PART: SMALL | |
75 | // Is f8 exponent less than register bias (that is, is it | |
76 | // less than 1). If it is, get the right sign of | |
77 | // zero and store this in iptr. | |
78 | ||
79 | // CALCULATION: NOT HUGE, NOT SMALL | |
80 | // To get the integer part | |
0347518d | 81 | // Take the floating-point input and truncate |
d5efd131 MF |
82 | // then convert this integer to fp Call it MODF_INTEGER_PART |
83 | ||
84 | // Subtract MODF_INTEGER_PART from MODF_NORM_F8 to get fraction part | |
0347518d | 85 | // Then put fraction part in f8 |
d5efd131 MF |
86 | // put integer part MODF_INTEGER_PART into *iptr |
87 | ||
88 | // Registers used | |
89 | //============================================================== | |
90 | ||
0347518d | 91 | // predicate registers used: |
d5efd131 MF |
92 | // p6 - p13 |
93 | ||
94 | // 0xFFFF 0x10033 | |
95 | // -----------------------+-----------------+------------- | |
96 | // SMALL | NORMAL | HUGE | |
97 | // p11 --------------->|<----- p12 ----->| <-------------- p9 | |
98 | // p10 --------------------------------->| | |
99 | // p13 --------------------------------------------------->| | |
100 | // | |
101 | ||
0347518d | 102 | // floating-point registers used: |
d5efd131 MF |
103 | MODF_NORM_F8 = f9 |
104 | MODF_FRACTION_PART = f10 | |
105 | MODF_INTEGER_PART = f11 | |
106 | MODF_INT_INTEGER_PART = f12 | |
107 | ||
108 | ||
0347518d | 109 | // general registers used |
d5efd131 MF |
110 | modf_signexp = r14 |
111 | modf_GR_no_frac = r15 | |
112 | modf_GR_FFFF = r16 | |
0347518d | 113 | modf_17_ones = r17 |
d5efd131 MF |
114 | modf_exp = r18 |
115 | // r33 = iptr | |
0347518d | 116 | |
d5efd131 MF |
117 | |
118 | .section .text | |
119 | GLOBAL_LIBM_ENTRY(modf) | |
120 | ||
121 | // Main path is p9, p11, p8 FALSE and p12 TRUE | |
122 | ||
123 | // Assume input is normalized and get signexp | |
124 | // Normalize input just in case | |
0347518d | 125 | // Form exponent bias |
d5efd131 MF |
126 | { .mfi |
127 | getf.exp modf_signexp = f8 | |
128 | fnorm.s0 MODF_NORM_F8 = f8 | |
129 | addl modf_GR_FFFF = 0xffff, r0 | |
130 | } | |
131 | // Get integer part of input | |
132 | // Form exponent mask | |
133 | { .mfi | |
134 | nop.m 999 | |
135 | fcvt.fx.trunc.s1 MODF_INT_INTEGER_PART = f8 | |
136 | mov modf_17_ones = 0x1ffff ;; | |
137 | } | |
138 | ||
139 | // Is x nan or inf? | |
140 | // qnan snan inf norm unorm 0 -+ | |
141 | // 1 1 1 0 0 0 11 = 0xe3 NAN_INF | |
142 | // Form biased exponent where input only has an integer part | |
143 | { .mfi | |
144 | nop.m 999 | |
145 | fclass.m.unc p6,p13 = f8, 0xe3 | |
146 | addl modf_GR_no_frac = 0x10033, r0 ;; | |
147 | } | |
148 | ||
149 | // Mask to get exponent | |
150 | // Is x unnorm? | |
151 | // qnan snan inf norm unorm 0 -+ | |
152 | // 0 0 0 0 1 0 11 = 0x0b UNORM | |
0347518d | 153 | // Set p13 to indicate calculation path, else p6 if nan or inf |
d5efd131 | 154 | { .mfi |
0347518d | 155 | and modf_exp = modf_17_ones, modf_signexp |
d5efd131 MF |
156 | fclass.m.unc p8,p0 = f8, 0x0b |
157 | nop.i 999 ;; | |
158 | } | |
159 | ||
160 | // p11 <== SMALL, no integer part, fraction is everyting | |
161 | // p9 <== HUGE, no fraction part, integer is everything | |
162 | // p12 <== NORMAL, fraction part and integer part | |
163 | { .mii | |
164 | (p13) cmp.lt.unc p11,p10 = modf_exp, modf_GR_FFFF | |
165 | nop.i 999 | |
166 | nop.i 999 ;; | |
167 | } | |
168 | ||
169 | // Is x inf? p6 if inf, p7 if nan | |
170 | { .mfb | |
171 | (p10) cmp.ge.unc p9,p12 = modf_exp, modf_GR_no_frac | |
172 | (p6) fclass.m.unc p6,p7 = f8, 0x23 | |
173 | (p8) br.cond.spnt MODF_DENORM ;; | |
174 | } | |
175 | ||
176 | MODF_COMMON: | |
177 | // For HUGE set fraction to signed 0 | |
178 | { .mfi | |
179 | nop.m 999 | |
180 | (p9) fmerge.s f8 = f8,f0 | |
181 | nop.i 999 | |
182 | } | |
183 | // For HUGE set integer part to normalized input | |
184 | { .mfi | |
185 | nop.m 999 | |
186 | (p9) fnorm.d.s0 MODF_INTEGER_PART = MODF_NORM_F8 | |
187 | nop.i 999 ;; | |
188 | } | |
189 | ||
190 | // For SMALL set fraction to normalized input, integer part to signed 0 | |
191 | { .mfi | |
192 | nop.m 999 | |
193 | (p11) fmerge.s MODF_INTEGER_PART = f8,f0 | |
194 | nop.i 999 | |
195 | } | |
196 | { .mfi | |
197 | nop.m 999 | |
198 | (p11) fnorm.d.s0 f8 = MODF_NORM_F8 | |
199 | nop.i 999 ;; | |
200 | } | |
201 | ||
202 | // For NORMAL float the integer part | |
203 | { .mfi | |
204 | nop.m 999 | |
205 | (p12) fcvt.xf MODF_INTEGER_PART = MODF_INT_INTEGER_PART | |
206 | nop.i 999 ;; | |
207 | } | |
208 | ||
209 | // If x inf set integer part to INF, fraction to signed 0 | |
210 | { .mfi | |
211 | (p6) stfd [r33] = MODF_NORM_F8 | |
212 | (p6) fmerge.s f8 = f8,f0 | |
213 | nop.i 999 ;; | |
214 | } | |
215 | ||
216 | // If x nan set integer and fraction parts to NaN (quietized) | |
217 | { .mfi | |
218 | (p7) stfd [r33] = MODF_NORM_F8 | |
219 | (p7) fmerge.s f8 = MODF_NORM_F8, MODF_NORM_F8 | |
220 | nop.i 999 ;; | |
221 | } | |
222 | ||
223 | { .mmi | |
224 | (p9) stfd [r33] = MODF_INTEGER_PART | |
225 | nop.m 999 | |
226 | nop.i 999 ;; | |
227 | } | |
228 | ||
229 | // For NORMAL compute fraction part | |
230 | { .mfi | |
231 | (p11) stfd [r33] = MODF_INTEGER_PART | |
232 | (p12) fms.d.s0 f8 = MODF_NORM_F8,f1, MODF_INTEGER_PART | |
233 | nop.i 999 ;; | |
234 | } | |
235 | ||
236 | // For NORMAL test if fraction part is zero; if so append correct sign | |
237 | { .mfi | |
238 | nop.m 999 | |
239 | (p12) fcmp.eq.unc.s0 p7,p0 = MODF_NORM_F8, MODF_INTEGER_PART | |
240 | nop.i 999 ;; | |
241 | } | |
242 | ||
243 | { .mfi | |
244 | (p12) stfd [r33] = MODF_INTEGER_PART | |
245 | nop.f 999 | |
246 | nop.i 999 ;; | |
247 | } | |
248 | ||
249 | // For NORMAL if fraction part is zero append sign of input | |
250 | { .mfb | |
251 | nop.m 999 | |
252 | (p7) fmerge.s f8 = MODF_NORM_F8, f0 | |
253 | br.ret.sptk b0 ;; | |
254 | } | |
255 | ||
256 | MODF_DENORM: | |
257 | // If x unorm get signexp from normalized input | |
258 | // If x unorm get integer part from normalized input | |
259 | { .mfi | |
260 | getf.exp modf_signexp = MODF_NORM_F8 | |
261 | fcvt.fx.trunc.s1 MODF_INT_INTEGER_PART = MODF_NORM_F8 | |
262 | nop.i 999 ;; | |
263 | } | |
264 | ||
265 | // If x unorm mask to get exponent | |
266 | { .mmi | |
267 | and modf_exp = modf_17_ones, modf_signexp ;; | |
268 | cmp.lt.unc p11,p10 = modf_exp, modf_GR_FFFF | |
269 | nop.i 999 ;; | |
270 | } | |
271 | ||
272 | { .mfb | |
273 | (p10) cmp.ge.unc p9,p12 = modf_exp, modf_GR_no_frac | |
274 | nop.f 999 | |
275 | br.cond.spnt MODF_COMMON ;; | |
276 | } | |
277 | ||
278 | GLOBAL_LIBM_END(modf) |