]>
Commit | Line | Data |
---|---|---|
d5efd131 MF |
1 | .file "truncf.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 | // 07/07/00 Created | |
42 | // 05/20/02 Cleaned up namespace and sf0 syntax | |
43 | // 01/20/03 Improved performance and reduced code size | |
44 | //============================================================== | |
45 | ||
46 | // API | |
47 | //============================================================== | |
48 | // float truncf(float x) | |
49 | //============================================================== | |
50 | ||
51 | // general input registers: | |
52 | // r14 - r18 | |
53 | ||
54 | rExpBias = r14 | |
55 | rSignexp = r15 | |
56 | rExp = r16 | |
57 | rExpMask = r17 | |
58 | rBigexp = r18 | |
59 | ||
60 | // floating-point registers: | |
61 | // f8 - f10 | |
62 | ||
63 | fXtruncInt = f9 | |
64 | fNormX = f10 | |
65 | ||
66 | // predicate registers used: | |
67 | // p6, p7 | |
68 | ||
69 | // Overview of operation | |
70 | //============================================================== | |
71 | // float truncf(float x) | |
72 | // Return an integer value (represented as a float) less than or | |
73 | // equal to x in magnitude. | |
74 | // This is x rounded toward zero to an integral value. | |
75 | //============================================================== | |
76 | ||
77 | // double_extended | |
78 | // if the exponent is > 1003e => 3F(true) = 63(decimal) | |
79 | // we have a significand of 64 bits 1.63-bits. | |
80 | // If we multiply by 2^63, we no longer have a fractional part | |
81 | // So input is an integer value already. | |
82 | ||
83 | // double | |
84 | // if the exponent is >= 10033 => 34(true) = 52(decimal) | |
85 | // 34 + 3ff = 433 | |
86 | // we have a significand of 53 bits 1.52-bits. (implicit 1) | |
87 | // If we multiply by 2^52, we no longer have a fractional part | |
88 | // So input is an integer value already. | |
89 | ||
90 | // single | |
91 | // if the exponent is > 10016 => 17(true) = 23(decimal) | |
92 | // we have a significand of 24 bits 1.23-bits. (implicit 1) | |
93 | // If we multiply by 2^23, we no longer have a fractional part | |
94 | // So input is an integer value already. | |
95 | ||
96 | .section .text | |
97 | GLOBAL_LIBM_ENTRY(truncf) | |
98 | ||
99 | { .mfi | |
100 | getf.exp rSignexp = f8 // Get signexp, recompute if unorm | |
101 | fcvt.fx.trunc.s1 fXtruncInt = f8 // Convert to int in significand | |
102 | addl rBigexp = 0x10016, r0 // Set exponent at which is integer | |
103 | } | |
104 | { .mfi | |
105 | mov rExpBias = 0x0FFFF // Form exponent bias | |
106 | fnorm.s1 fNormX = f8 // Normalize input | |
107 | mov rExpMask = 0x1FFFF // Form exponent mask | |
108 | } | |
109 | ;; | |
110 | ||
111 | { .mfi | |
112 | nop.m 0 | |
113 | fclass.m p7,p0 = f8, 0x0b // Test x unorm | |
114 | nop.i 0 | |
115 | } | |
116 | ;; | |
117 | ||
118 | { .mfb | |
119 | nop.m 0 | |
120 | fclass.m p6,p0 = f8, 0x1e3 // Test x natval, nan, inf | |
121 | (p7) br.cond.spnt TRUNC_UNORM // Branch if x unorm | |
122 | } | |
123 | ;; | |
124 | ||
125 | TRUNC_COMMON: | |
126 | // Return here from TRUNC_UNORM | |
127 | { .mfb | |
128 | and rExp = rSignexp, rExpMask // Get biased exponent | |
129 | (p6) fma.s.s0 f8 = f8, f1, f0 // Result if x natval, nan, inf | |
130 | (p6) br.ret.spnt b0 // Exit if x natval, nan, inf | |
131 | } | |
132 | ;; | |
133 | ||
134 | { .mfi | |
135 | cmp.lt p6,p0 = rExp, rExpBias // Is |x| < 1? | |
136 | fcvt.xf f8 = fXtruncInt // Result, assume 1 <= |x| < 2^23 | |
137 | cmp.ge p7,p0 = rExp, rBigexp // Is |x| >= 2^23? | |
138 | } | |
139 | ;; | |
140 | ||
141 | // We must correct result if |x| < 1, or |x| >= 2^23 | |
142 | .pred.rel "mutex",p6,p7 | |
143 | { .mfi | |
144 | nop.m 0 | |
145 | (p6) fmerge.s f8 = fNormX, f0 // If |x| < 1, result sgn(x)*0 | |
146 | nop.i 0 | |
147 | } | |
148 | { .mfb | |
149 | nop.m 0 | |
150 | (p7) fma.s.s0 f8 = fNormX, f1, f0 // If |x| >= 2^23, result x | |
151 | br.ret.sptk b0 // Exit main path | |
152 | } | |
153 | ;; | |
154 | ||
155 | ||
156 | TRUNC_UNORM: | |
157 | // Here if x unorm | |
158 | { .mfb | |
159 | getf.exp rSignexp = fNormX // Get signexp, recompute if unorm | |
160 | fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag | |
161 | br.cond.sptk TRUNC_COMMON // Return to main path | |
162 | } | |
163 | ;; | |
164 | ||
165 | GLOBAL_LIBM_END(truncf) | |
aa1142c5 | 166 | libm_alias_float_other (trunc, trunc) |