]>
Commit | Line | Data |
---|---|---|
8da2915d UD |
1 | .file "trunc.s" |
2 | ||
0ecb606c JJ |
3 | |
4 | // Copyright (c) 2000 - 2003, Intel Corporation | |
8da2915d | 5 | // All rights reserved. |
0ecb606c JJ |
6 | // |
7 | // Contributed 2000 by the Intel Numerics Group, 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. | |
0ecb606c JJ |
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 |
0ecb606c | 27 | // A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL INTEL OR ITS |
8da2915d | 28 | // CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
0ecb606c JJ |
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 |
0ecb606c JJ |
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 |
0ecb606c JJ |
37 | // problem reports or change requests be submitted to it directly at |
38 | // http://www.intel.com/software/products/opensource/libraries/num.htm. | |
8da2915d | 39 | // |
8da2915d UD |
40 | // History |
41 | //============================================================== | |
0ecb606c JJ |
42 | // 07/07/00 Created |
43 | // 05/20/02 Cleaned up namespace and sf0 syntax | |
44 | // 01/20/03 Improved performance and reduced code size | |
8da2915d UD |
45 | //============================================================== |
46 | ||
47 | // API | |
48 | //============================================================== | |
49 | // double trunc(double x) | |
50 | //============================================================== | |
51 | ||
0ecb606c JJ |
52 | // general input registers: |
53 | // r14 - r18 | |
8da2915d | 54 | |
0ecb606c JJ |
55 | rExpBias = r14 |
56 | rSignexp = r15 | |
57 | rExp = r16 | |
58 | rExpMask = r17 | |
59 | rBigexp = r18 | |
8da2915d UD |
60 | |
61 | // floating-point registers: | |
0ecb606c JJ |
62 | // f8 - f10 |
63 | ||
64 | fXtruncInt = f9 | |
65 | fNormX = f10 | |
8da2915d | 66 | |
0ecb606c JJ |
67 | // predicate registers used: |
68 | // p6, p7 | |
8da2915d UD |
69 | |
70 | // Overview of operation | |
71 | //============================================================== | |
72 | // double trunc(double x) | |
0ecb606c | 73 | // Return an integer value (represented as a double) less than or |
8da2915d UD |
74 | // equal to x in magnitude. |
75 | // This is x rounded toward zero to an integral value. | |
76 | //============================================================== | |
77 | ||
78 | // double_extended | |
79 | // if the exponent is > 1003e => 3F(true) = 63(decimal) | |
80 | // we have a significand of 64 bits 1.63-bits. | |
81 | // If we multiply by 2^63, we no longer have a fractional part | |
82 | // So input is an integer value already. | |
83 | ||
84 | // double | |
85 | // if the exponent is >= 10033 => 34(true) = 52(decimal) | |
86 | // 34 + 3ff = 433 | |
87 | // we have a significand of 53 bits 1.52-bits. (implicit 1) | |
88 | // If we multiply by 2^52, we no longer have a fractional part | |
89 | // So input is an integer value already. | |
90 | ||
91 | // single | |
92 | // if the exponent is > 10016 => 17(true) = 23(decimal) | |
93 | // we have a significand of 24 bits 1.23-bits. (implicit 1) | |
94 | // If we multiply by 2^23, we no longer have a fractional part | |
95 | // So input is an integer value already. | |
96 | ||
0ecb606c JJ |
97 | .section .text |
98 | GLOBAL_LIBM_ENTRY(trunc) | |
8da2915d UD |
99 | |
100 | { .mfi | |
0ecb606c JJ |
101 | getf.exp rSignexp = f8 // Get signexp, recompute if unorm |
102 | fcvt.fx.trunc.s1 fXtruncInt = f8 // Convert to int in significand | |
103 | addl rBigexp = 0x10033, r0 // Set exponent at which is integer | |
8da2915d UD |
104 | } |
105 | { .mfi | |
0ecb606c JJ |
106 | mov rExpBias = 0x0FFFF // Form exponent bias |
107 | fnorm.s1 fNormX = f8 // Normalize input | |
108 | mov rExpMask = 0x1FFFF // Form exponent mask | |
109 | } | |
110 | ;; | |
8da2915d UD |
111 | |
112 | { .mfi | |
113 | nop.m 0 | |
0ecb606c | 114 | fclass.m p7,p0 = f8, 0x0b // Test x unorm |
8da2915d UD |
115 | nop.i 0 |
116 | } | |
0ecb606c JJ |
117 | ;; |
118 | ||
8da2915d UD |
119 | { .mfb |
120 | nop.m 0 | |
0ecb606c JJ |
121 | fclass.m p6,p0 = f8, 0x1e3 // Test x natval, nan, inf |
122 | (p7) br.cond.spnt TRUNC_UNORM // Branch if x unorm | |
8da2915d | 123 | } |
0ecb606c JJ |
124 | ;; |
125 | ||
126 | TRUNC_COMMON: | |
127 | // Return here from TRUNC_UNORM | |
8da2915d | 128 | { .mfb |
0ecb606c JJ |
129 | and rExp = rSignexp, rExpMask // Get biased exponent |
130 | (p6) fma.d.s0 f8 = f8, f1, f0 // Result if x natval, nan, inf | |
131 | (p6) br.ret.spnt b0 // Exit if x natval, nan, inf | |
132 | } | |
133 | ;; | |
134 | ||
135 | { .mfi | |
136 | cmp.lt p6,p0 = rExp, rExpBias // Is |x| < 1? | |
137 | fcvt.xf f8 = fXtruncInt // Result, assume 1 <= |x| < 2^52 | |
138 | cmp.ge p7,p0 = rExp, rBigexp // Is |x| >= 2^52? | |
139 | } | |
140 | ;; | |
141 | ||
142 | // We must correct result if |x| < 1, or |x| >= 2^52 | |
143 | .pred.rel "mutex",p6,p7 | |
144 | { .mfi | |
8da2915d | 145 | nop.m 0 |
0ecb606c JJ |
146 | (p6) fmerge.s f8 = fNormX, f0 // If |x| < 1, result sgn(x)*0 |
147 | nop.i 0 | |
148 | } | |
8da2915d | 149 | { .mfb |
0ecb606c JJ |
150 | nop.m 0 |
151 | (p7) fma.d.s0 f8 = fNormX, f1, f0 // If |x| >= 2^52, result x | |
152 | br.ret.sptk b0 // Exit main path | |
8da2915d | 153 | } |
0ecb606c JJ |
154 | ;; |
155 | ||
156 | ||
157 | TRUNC_UNORM: | |
158 | // Here if x unorm | |
8da2915d | 159 | { .mfb |
0ecb606c JJ |
160 | getf.exp rSignexp = fNormX // Get signexp, recompute if unorm |
161 | fcmp.eq.s0 p7,p0 = f8, f0 // Dummy op to set denormal flag | |
162 | br.cond.sptk TRUNC_COMMON // Return to main path | |
163 | } | |
164 | ;; | |
165 | ||
166 | GLOBAL_LIBM_END(trunc) |