root / src / blas / HPL_dcopy.c @ 1
Historique | Voir | Annoter | Télécharger (6,51 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 | #ifndef HPL_dcopy
|
| 53 | 1 | equemene | |
| 54 | 1 | equemene | #ifdef STDC_HEADERS
|
| 55 | 1 | equemene | void HPL_dcopy
|
| 56 | 1 | equemene | ( |
| 57 | 1 | equemene | const int N, |
| 58 | 1 | equemene | const double * X, |
| 59 | 1 | equemene | const int INCX, |
| 60 | 1 | equemene | double * Y,
|
| 61 | 1 | equemene | const int INCY |
| 62 | 1 | equemene | ) |
| 63 | 1 | equemene | #else
|
| 64 | 1 | equemene | void HPL_dcopy
|
| 65 | 1 | equemene | ( N, X, INCX, Y, INCY ) |
| 66 | 1 | equemene | const int N; |
| 67 | 1 | equemene | const double * X; |
| 68 | 1 | equemene | const int INCX; |
| 69 | 1 | equemene | double * Y;
|
| 70 | 1 | equemene | const int INCY; |
| 71 | 1 | equemene | #endif
|
| 72 | 1 | equemene | {
|
| 73 | 1 | equemene | /*
|
| 74 | 1 | equemene | * Purpose
|
| 75 | 1 | equemene | * =======
|
| 76 | 1 | equemene | *
|
| 77 | 1 | equemene | * HPL_dcopy copies the vector x into the vector y.
|
| 78 | 1 | equemene | *
|
| 79 | 1 | equemene | *
|
| 80 | 1 | equemene | * Arguments
|
| 81 | 1 | equemene | * =========
|
| 82 | 1 | equemene | *
|
| 83 | 1 | equemene | * N (local input) const int
|
| 84 | 1 | equemene | * On entry, N specifies the length of the vectors x and y. N
|
| 85 | 1 | equemene | * must be at least zero.
|
| 86 | 1 | equemene | *
|
| 87 | 1 | equemene | * X (local input) const double *
|
| 88 | 1 | equemene | * On entry, X is an incremented array of dimension at least
|
| 89 | 1 | equemene | * ( 1 + ( n - 1 ) * abs( INCX ) ) that contains the vector x.
|
| 90 | 1 | equemene | *
|
| 91 | 1 | equemene | * INCX (local input) const int
|
| 92 | 1 | equemene | * On entry, INCX specifies the increment for the elements of X.
|
| 93 | 1 | equemene | * INCX must not be zero.
|
| 94 | 1 | equemene | *
|
| 95 | 1 | equemene | * Y (local input/output) double *
|
| 96 | 1 | equemene | * On entry, Y is an incremented array of dimension at least
|
| 97 | 1 | equemene | * ( 1 + ( n - 1 ) * abs( INCY ) ) that contains the vector y.
|
| 98 | 1 | equemene | * On exit, the entries of the incremented array Y are updated
|
| 99 | 1 | equemene | * with the entries of the incremented array X.
|
| 100 | 1 | equemene | *
|
| 101 | 1 | equemene | * INCY (local input) const int
|
| 102 | 1 | equemene | * On entry, INCY specifies the increment for the elements of Y.
|
| 103 | 1 | equemene | * INCY must not be zero.
|
| 104 | 1 | equemene | *
|
| 105 | 1 | equemene | * ---------------------------------------------------------------------
|
| 106 | 1 | equemene | */
|
| 107 | 1 | equemene | #ifdef HPL_CALL_CBLAS
|
| 108 | 1 | equemene | cblas_dcopy( N, X, INCX, Y, INCY ); |
| 109 | 1 | equemene | #endif
|
| 110 | 1 | equemene | #ifdef HPL_CALL_VSIPL
|
| 111 | 1 | equemene | register double x0, x1, x2, x3, x4, x5, x6, x7; |
| 112 | 1 | equemene | const double * StX; |
| 113 | 1 | equemene | register int i; |
| 114 | 1 | equemene | int nu;
|
| 115 | 1 | equemene | const int incX2 = 2 * INCX, incY2 = 2 * INCY, |
| 116 | 1 | equemene | incX3 = 3 * INCX, incY3 = 3 * INCY, |
| 117 | 1 | equemene | incX4 = 4 * INCX, incY4 = 4 * INCY, |
| 118 | 1 | equemene | incX5 = 5 * INCX, incY5 = 5 * INCY, |
| 119 | 1 | equemene | incX6 = 6 * INCX, incY6 = 6 * INCY, |
| 120 | 1 | equemene | incX7 = 7 * INCX, incY7 = 7 * INCY, |
| 121 | 1 | equemene | incX8 = 8 * INCX, incY8 = 8 * INCY; |
| 122 | 1 | equemene | |
| 123 | 1 | equemene | if( N > 0 ) |
| 124 | 1 | equemene | {
|
| 125 | 1 | equemene | if( ( nu = ( N >> 3 ) << 3 ) != 0 ) |
| 126 | 1 | equemene | {
|
| 127 | 1 | equemene | StX = X + nu * INCX; |
| 128 | 1 | equemene | |
| 129 | 1 | equemene | do
|
| 130 | 1 | equemene | {
|
| 131 | 1 | equemene | x0 = (*X); x4 = X[incX4]; x1 = X[INCX ]; x5 = X[incX5]; |
| 132 | 1 | equemene | x2 = X[incX2]; x6 = X[incX6]; x3 = X[incX3]; x7 = X[incX7]; |
| 133 | 1 | equemene | |
| 134 | 1 | equemene | *Y = x0; Y[incY4] = x4; Y[INCY ] = x1; Y[incY5] = x5; |
| 135 | 1 | equemene | Y[incY2] = x2; Y[incY6] = x6; Y[incY3] = x3; Y[incY7] = x7; |
| 136 | 1 | equemene | |
| 137 | 1 | equemene | X += incX8; |
| 138 | 1 | equemene | Y += incY8; |
| 139 | 1 | equemene | |
| 140 | 1 | equemene | } while( X != StX );
|
| 141 | 1 | equemene | } |
| 142 | 1 | equemene | |
| 143 | 1 | equemene | for( i = N - nu; i != 0; i-- ) |
| 144 | 1 | equemene | {
|
| 145 | 1 | equemene | x0 = (*X); |
| 146 | 1 | equemene | *Y = x0; |
| 147 | 1 | equemene | |
| 148 | 1 | equemene | X += INCX; |
| 149 | 1 | equemene | Y += INCY; |
| 150 | 1 | equemene | } |
| 151 | 1 | equemene | } |
| 152 | 1 | equemene | #endif
|
| 153 | 1 | equemene | #ifdef HPL_CALL_FBLAS
|
| 154 | 1 | equemene | #ifdef HPL_USE_F77_INTEGER_DEF
|
| 155 | 1 | equemene | const F77_INTEGER F77N = N, F77incx = INCX, F77incy = INCY;
|
| 156 | 1 | equemene | #else
|
| 157 | 1 | equemene | #define F77N N
|
| 158 | 1 | equemene | #define F77incx INCX
|
| 159 | 1 | equemene | #define F77incy INCY
|
| 160 | 1 | equemene | #endif
|
| 161 | 1 | equemene | F77dcopy( &F77N, X, &F77incx, Y, &F77incy ); |
| 162 | 1 | equemene | #endif
|
| 163 | 1 | equemene | /*
|
| 164 | 1 | equemene | * End of HPL_dcopy
|
| 165 | 1 | equemene | */
|
| 166 | 1 | equemene | } |
| 167 | 1 | equemene | |
| 168 | 1 | equemene | #endif |