]>
Commit | Line | Data |
---|---|---|
d5efd131 MF |
1 | .file "modf.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 | ||
0347518d MF |
23 | // THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
24 | // "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT | |
d5efd131 | 25 | // LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
0347518d | 26 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS |
d5efd131 | 27 | // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
0347518d MF |
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 | |
d5efd131 | 31 | // OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY OR TORT (INCLUDING |
0347518d MF |
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 | // | |
d5efd131 | 35 | // Intel Corporation is the author of this code, and requests that all |
0347518d | 36 | // problem reports or change requests be submitted to it directly at |
d5efd131 MF |
37 | // http://www.intel.com/software/products/opensource/libraries/num.htm. |
38 | // | |
39 | // History | |
40 | //============================================================== | |
41 | // 02/02/00 Initial version | |
42 | // 04/04/00 Improved speed, corrected result for NaN input | |
0347518d | 43 | // 12/22/00 Fixed so inexact flag is never set, and invalid is not set for |
d5efd131 MF |
44 | // qnans nor for inputs larger than 2^63. |
45 | // 05/20/02 Cleaned up namespace and sf0 syntax | |
46 | // 02/10/03 Reordered header: .section, .global, .proc, .align | |
47 | // | |
48 | // API | |
49 | //============================================================== | |
50 | // double modf(double x, double *iptr) | |
51 | // break a floating point x number into fraction and an exponent | |
52 | // | |
53 | // input floating point f8, address in r33 | |
54 | // output floating point f8 (x fraction), and *iptr (x integral part) | |
55 | // | |
56 | // OVERVIEW | |
57 | //============================================================== | |
58 | // | |
59 | // NO FRACTIONAL PART: HUGE | |
60 | // If | |
61 | // for double-extended | |
62 | // If the true exponent is greater than or equal 63 | |
63 | // 1003e ==> 1003e -ffff = 3f = 63(dec) | |
64 | // for double | |
65 | // If the true exponent is greater than or equal 52 | |
66 | // 10033 -ffff = 34 = 52(dec) | |
67 | // for single | |
68 | // If the true exponent is greater than or equal 23 | |
69 | // 10016 -ffff = 17 = 23(dec) | |
70 | // then | |
71 | // we are already an integer (p9 true) | |
72 | ||
73 | // NO INTEGER PART: SMALL | |
74 | // Is f8 exponent less than register bias (that is, is it | |
75 | // less than 1). If it is, get the right sign of | |
76 | // zero and store this in iptr. | |
77 | ||
78 | // CALCULATION: NOT HUGE, NOT SMALL | |
79 | // To get the integer part | |
0347518d | 80 | // Take the floating-point input and truncate |
d5efd131 MF |
81 | // then convert this integer to fp Call it MODF_INTEGER_PART |
82 | ||
83 | // Subtract MODF_INTEGER_PART from MODF_NORM_F8 to get fraction part | |
0347518d | 84 | // Then put fraction part in f8 |
d5efd131 MF |
85 | // put integer part MODF_INTEGER_PART into *iptr |
86 | ||
87 | // Registers used | |
88 | //============================================================== | |
89 | ||
0347518d | 90 | // predicate registers used: |
d5efd131 MF |
91 | // p6 - p13 |
92 | ||
93 | // 0xFFFF 0x10033 | |
94 | // -----------------------+-----------------+------------- | |
95 | // SMALL | NORMAL | HUGE | |
96 | // p11 --------------->|<----- p12 ----->| <-------------- p9 | |
97 | // p10 --------------------------------->| | |
98 | // p13 --------------------------------------------------->| | |
99 | // | |
100 | ||
0347518d | 101 | // floating-point registers used: |
d5efd131 MF |
102 | MODF_NORM_F8 = f9 |
103 | MODF_FRACTION_PART = f10 | |
104 | MODF_INTEGER_PART = f11 | |
105 | MODF_INT_INTEGER_PART = f12 | |
106 | ||
107 | ||
0347518d | 108 | // general registers used |
d5efd131 MF |
109 | modf_signexp = r14 |
110 | modf_GR_no_frac = r15 | |
111 | modf_GR_FFFF = r16 | |
0347518d | 112 | modf_17_ones = r17 |
d5efd131 MF |
113 | modf_exp = r18 |
114 | // r33 = iptr | |
0347518d | 115 | |
d5efd131 MF |
116 | |
117 | .section .text | |
118 | GLOBAL_LIBM_ENTRY(modf) | |
119 | ||
120 | // Main path is p9, p11, p8 FALSE and p12 TRUE | |
121 | ||
122 | // Assume input is normalized and get signexp | |
123 | // Normalize input just in case | |
0347518d | 124 | // Form exponent bias |
d5efd131 MF |
125 | { .mfi |
126 | getf.exp modf_signexp = f8 | |
127 | fnorm.s0 MODF_NORM_F8 = f8 | |
128 | addl modf_GR_FFFF = 0xffff, r0 | |
129 | } | |
130 | // Get integer part of input | |
131 | // Form exponent mask | |
132 | { .mfi | |
133 | nop.m 999 | |
134 | fcvt.fx.trunc.s1 MODF_INT_INTEGER_PART = f8 | |
135 | mov modf_17_ones = 0x1ffff ;; | |
136 | } | |
137 | ||
138 | // Is x nan or inf? | |
139 | // qnan snan inf norm unorm 0 -+ | |
140 | // 1 1 1 0 0 0 11 = 0xe3 NAN_INF | |
141 | // Form biased exponent where input only has an integer part | |
142 | { .mfi | |
143 | nop.m 999 | |
144 | fclass.m.unc p6,p13 = f8, 0xe3 | |
145 | addl modf_GR_no_frac = 0x10033, r0 ;; | |
146 | } | |
147 | ||
148 | // Mask to get exponent | |
149 | // Is x unnorm? | |
150 | // qnan snan inf norm unorm 0 -+ | |
151 | // 0 0 0 0 1 0 11 = 0x0b UNORM | |
0347518d | 152 | // Set p13 to indicate calculation path, else p6 if nan or inf |
d5efd131 | 153 | { .mfi |
0347518d | 154 | and modf_exp = modf_17_ones, modf_signexp |
d5efd131 MF |
155 | fclass.m.unc p8,p0 = f8, 0x0b |
156 | nop.i 999 ;; | |
157 | } | |
158 | ||
65cc53fe | 159 | // p11 <== SMALL, no integer part, fraction is everything |
d5efd131 MF |
160 | // p9 <== HUGE, no fraction part, integer is everything |
161 | // p12 <== NORMAL, fraction part and integer part | |
162 | { .mii | |
163 | (p13) cmp.lt.unc p11,p10 = modf_exp, modf_GR_FFFF | |
164 | nop.i 999 | |
165 | nop.i 999 ;; | |
166 | } | |
167 | ||
168 | // Is x inf? p6 if inf, p7 if nan | |
169 | { .mfb | |
170 | (p10) cmp.ge.unc p9,p12 = modf_exp, modf_GR_no_frac | |
171 | (p6) fclass.m.unc p6,p7 = f8, 0x23 | |
172 | (p8) br.cond.spnt MODF_DENORM ;; | |
173 | } | |
174 | ||
175 | MODF_COMMON: | |
176 | // For HUGE set fraction to signed 0 | |
177 | { .mfi | |
178 | nop.m 999 | |
179 | (p9) fmerge.s f8 = f8,f0 | |
180 | nop.i 999 | |
181 | } | |
182 | // For HUGE set integer part to normalized input | |
183 | { .mfi | |
184 | nop.m 999 | |
185 | (p9) fnorm.d.s0 MODF_INTEGER_PART = MODF_NORM_F8 | |
186 | nop.i 999 ;; | |
187 | } | |
188 | ||
189 | // For SMALL set fraction to normalized input, integer part to signed 0 | |
190 | { .mfi | |
191 | nop.m 999 | |
192 | (p11) fmerge.s MODF_INTEGER_PART = f8,f0 | |
193 | nop.i 999 | |
194 | } | |
195 | { .mfi | |
196 | nop.m 999 | |
197 | (p11) fnorm.d.s0 f8 = MODF_NORM_F8 | |
198 | nop.i 999 ;; | |
199 | } | |
200 | ||
201 | // For NORMAL float the integer part | |
202 | { .mfi | |
203 | nop.m 999 | |
204 | (p12) fcvt.xf MODF_INTEGER_PART = MODF_INT_INTEGER_PART | |
205 | nop.i 999 ;; | |
206 | } | |
207 | ||
208 | // If x inf set integer part to INF, fraction to signed 0 | |
209 | { .mfi | |
210 | (p6) stfd [r33] = MODF_NORM_F8 | |
211 | (p6) fmerge.s f8 = f8,f0 | |
212 | nop.i 999 ;; | |
213 | } | |
214 | ||
215 | // If x nan set integer and fraction parts to NaN (quietized) | |
216 | { .mfi | |
217 | (p7) stfd [r33] = MODF_NORM_F8 | |
218 | (p7) fmerge.s f8 = MODF_NORM_F8, MODF_NORM_F8 | |
219 | nop.i 999 ;; | |
220 | } | |
221 | ||
222 | { .mmi | |
223 | (p9) stfd [r33] = MODF_INTEGER_PART | |
224 | nop.m 999 | |
225 | nop.i 999 ;; | |
226 | } | |
227 | ||
228 | // For NORMAL compute fraction part | |
229 | { .mfi | |
230 | (p11) stfd [r33] = MODF_INTEGER_PART | |
231 | (p12) fms.d.s0 f8 = MODF_NORM_F8,f1, MODF_INTEGER_PART | |
232 | nop.i 999 ;; | |
233 | } | |
234 | ||
235 | // For NORMAL test if fraction part is zero; if so append correct sign | |
236 | { .mfi | |
237 | nop.m 999 | |
238 | (p12) fcmp.eq.unc.s0 p7,p0 = MODF_NORM_F8, MODF_INTEGER_PART | |
239 | nop.i 999 ;; | |
240 | } | |
241 | ||
242 | { .mfi | |
243 | (p12) stfd [r33] = MODF_INTEGER_PART | |
244 | nop.f 999 | |
245 | nop.i 999 ;; | |
246 | } | |
247 | ||
248 | // For NORMAL if fraction part is zero append sign of input | |
249 | { .mfb | |
250 | nop.m 999 | |
251 | (p7) fmerge.s f8 = MODF_NORM_F8, f0 | |
252 | br.ret.sptk b0 ;; | |
253 | } | |
254 | ||
255 | MODF_DENORM: | |
256 | // If x unorm get signexp from normalized input | |
257 | // If x unorm get integer part from normalized input | |
258 | { .mfi | |
259 | getf.exp modf_signexp = MODF_NORM_F8 | |
260 | fcvt.fx.trunc.s1 MODF_INT_INTEGER_PART = MODF_NORM_F8 | |
261 | nop.i 999 ;; | |
262 | } | |
263 | ||
264 | // If x unorm mask to get exponent | |
265 | { .mmi | |
266 | and modf_exp = modf_17_ones, modf_signexp ;; | |
267 | cmp.lt.unc p11,p10 = modf_exp, modf_GR_FFFF | |
268 | nop.i 999 ;; | |
269 | } | |
270 | ||
271 | { .mfb | |
272 | (p10) cmp.ge.unc p9,p12 = modf_exp, modf_GR_no_frac | |
273 | nop.f 999 | |
274 | br.cond.spnt MODF_COMMON ;; | |
275 | } | |
276 | ||
277 | GLOBAL_LIBM_END(modf) | |
0609ec0a | 278 | libm_alias_double_other (modf, modf) |