root / testing / matgen / HPL_lmul.c
Historique | Voir | Annoter | Télécharger (5,4 ko)
1 | 1 | equemene | /*
|
---|---|---|---|
2 | 1 | equemene | * -- High Performance Computing Linpack Benchmark (HPL)
|
3 | 1 | equemene | * HPL - 2.0 - September 10, 2008
|
4 | 1 | equemene | * Antoine P. Petitet
|
5 | 1 | equemene | * University of Tennessee, Knoxville
|
6 | 1 | equemene | * Innovative Computing Laboratory
|
7 | 1 | equemene | * (C) Copyright 2000-2008 All Rights Reserved
|
8 | 1 | equemene | *
|
9 | 1 | equemene | * -- Copyright notice and Licensing terms:
|
10 | 1 | equemene | *
|
11 | 1 | equemene | * Redistribution and use in source and binary forms, with or without
|
12 | 1 | equemene | * modification, are permitted provided that the following conditions
|
13 | 1 | equemene | * are met:
|
14 | 1 | equemene | *
|
15 | 1 | equemene | * 1. Redistributions of source code must retain the above copyright
|
16 | 1 | equemene | * notice, this list of conditions and the following disclaimer.
|
17 | 1 | equemene | *
|
18 | 1 | equemene | * 2. Redistributions in binary form must reproduce the above copyright
|
19 | 1 | equemene | * notice, this list of conditions, and the following disclaimer in the
|
20 | 1 | equemene | * documentation and/or other materials provided with the distribution.
|
21 | 1 | equemene | *
|
22 | 1 | equemene | * 3. All advertising materials mentioning features or use of this
|
23 | 1 | equemene | * software must display the following acknowledgement:
|
24 | 1 | equemene | * This product includes software developed at the University of
|
25 | 1 | equemene | * Tennessee, Knoxville, Innovative Computing Laboratory.
|
26 | 1 | equemene | *
|
27 | 1 | equemene | * 4. The name of the University, the name of the Laboratory, or the
|
28 | 1 | equemene | * names of its contributors may not be used to endorse or promote
|
29 | 1 | equemene | * products derived from this software without specific written
|
30 | 1 | equemene | * permission.
|
31 | 1 | equemene | *
|
32 | 1 | equemene | * -- Disclaimer:
|
33 | 1 | equemene | *
|
34 | 1 | equemene | * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
35 | 1 | equemene | * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
36 | 1 | equemene | * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
37 | 1 | equemene | * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE UNIVERSITY
|
38 | 1 | equemene | * OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
39 | 1 | equemene | * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
40 | 1 | equemene | * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
41 | 1 | equemene | * DATA OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
42 | 1 | equemene | * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
43 | 1 | equemene | * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
44 | 1 | equemene | * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
45 | 1 | equemene | * ---------------------------------------------------------------------
|
46 | 1 | equemene | */
|
47 | 1 | equemene | /*
|
48 | 1 | equemene | * Include files
|
49 | 1 | equemene | */
|
50 | 1 | equemene | #include "hpl.h" |
51 | 1 | equemene | |
52 | 1 | equemene | #ifdef STDC_HEADERS
|
53 | 1 | equemene | void HPL_lmul
|
54 | 1 | equemene | ( |
55 | 1 | equemene | int * K,
|
56 | 1 | equemene | int * J,
|
57 | 1 | equemene | int * I
|
58 | 1 | equemene | ) |
59 | 1 | equemene | #else
|
60 | 1 | equemene | void HPL_lmul
|
61 | 1 | equemene | ( K, J, I ) |
62 | 1 | equemene | int * K;
|
63 | 1 | equemene | int * J;
|
64 | 1 | equemene | int * I;
|
65 | 1 | equemene | #endif
|
66 | 1 | equemene | { |
67 | 1 | equemene | /*
|
68 | 1 | equemene | * Purpose
|
69 | 1 | equemene | * =======
|
70 | 1 | equemene | *
|
71 | 1 | equemene | * HPL_lmul multiplies without carry two long positive integers K and J
|
72 | 1 | equemene | * and puts the result into I. The long integers I, J, K are encoded on
|
73 | 1 | equemene | * 32 bits using an array of 2 integers. The 32-lower bits are stored in
|
74 | 1 | equemene | * the first entry of each array, the 32-higher bits in the second entry
|
75 | 1 | equemene | * of each array. For efficiency purposes, the intrisic modulo function
|
76 | 1 | equemene | * is inlined.
|
77 | 1 | equemene | *
|
78 | 1 | equemene | * Arguments
|
79 | 1 | equemene | * =========
|
80 | 1 | equemene | *
|
81 | 1 | equemene | * K (local input) int *
|
82 | 1 | equemene | * On entry, K is an integer array of dimension 2 containing the
|
83 | 1 | equemene | * encoded long integer K.
|
84 | 1 | equemene | *
|
85 | 1 | equemene | * J (local input) int *
|
86 | 1 | equemene | * On entry, J is an integer array of dimension 2 containing the
|
87 | 1 | equemene | * encoded long integer J.
|
88 | 1 | equemene | *
|
89 | 1 | equemene | * I (local output) int *
|
90 | 1 | equemene | * On entry, I is an integer array of dimension 2. On exit, this
|
91 | 1 | equemene | * array contains the encoded long integer result.
|
92 | 1 | equemene | *
|
93 | 1 | equemene | * ---------------------------------------------------------------------
|
94 | 1 | equemene | */
|
95 | 1 | equemene | /*
|
96 | 1 | equemene | * .. Local Variables ..
|
97 | 1 | equemene | */
|
98 | 1 | equemene | int r, c;
|
99 | 1 | equemene | unsigned int kk[4], jj[4], res[5]; |
100 | 1 | equemene | /* ..
|
101 | 1 | equemene | * .. Executable Statements ..
|
102 | 1 | equemene | */
|
103 | 1 | equemene | /*
|
104 | 1 | equemene | * Addition is done with 16 bits at a time. Multiplying two 16-bit
|
105 | 1 | equemene | * integers yields a 32-bit result. The lower 16-bits of the result
|
106 | 1 | equemene | * are kept in I, and the higher 16-bits are carried over to the
|
107 | 1 | equemene | * next multiplication.
|
108 | 1 | equemene | */
|
109 | 1 | equemene | for (c = 0; c < 2; ++c) { |
110 | 1 | equemene | kk[2*c] = K[c] & 65535; |
111 | 1 | equemene | kk[2*c+1] = ((unsigned)K[c] >> 16) & 65535; |
112 | 1 | equemene | jj[2*c] = J[c] & 65535; |
113 | 1 | equemene | jj[2*c+1] = ((unsigned)J[c] >> 16) & 65535; |
114 | 1 | equemene | } |
115 | 1 | equemene | |
116 | 1 | equemene | res[0] = 0; |
117 | 1 | equemene | for (c = 0; c < 4; ++c) { |
118 | 1 | equemene | res[c+1] = (res[c] >> 16) & 65535; |
119 | 1 | equemene | res[c] &= 65535;
|
120 | 1 | equemene | for (r = 0; r < c+1; ++r) { |
121 | 1 | equemene | res[c] = kk[r] * jj[c-r] + (res[c] & 65535);
|
122 | 1 | equemene | res[c+1] += (res[c] >> 16) & 65535; |
123 | 1 | equemene | } |
124 | 1 | equemene | } |
125 | 1 | equemene | |
126 | 1 | equemene | for (c = 0; c < 2; ++c) |
127 | 1 | equemene | I[c] = (int)(((res[2*c+1] & 65535) << 16) | (res[2*c] & 65535)); |
128 | 1 | equemene | /*
|
129 | 1 | equemene | * End of HPL_lmul
|
130 | 1 | equemene | */
|
131 | 1 | equemene | } |