| [658] | 1 | /***************************************************************************
|
|---|
| 2 | * blitz/tuning.h Platform-specific code tuning
|
|---|
| 3 | *
|
|---|
| 4 | * $Id: tuning.h,v 1.1.1.1 1999-11-26 16:37:05 ansari Exp $
|
|---|
| 5 | *
|
|---|
| 6 | * Copyright (C) 1997,1998 Todd Veldhuizen <tveldhui@seurat.uwaterloo.ca>
|
|---|
| 7 | *
|
|---|
| 8 | * This program is free software; you can redistribute it and/or
|
|---|
| 9 | * modify it under the terms of the GNU General Public License
|
|---|
| 10 | * as published by the Free Software Foundation; either version 2
|
|---|
| 11 | * of the License, or (at your option) any later version.
|
|---|
| 12 | *
|
|---|
| 13 | * This program is distributed in the hope that it will be useful,
|
|---|
| 14 | * but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|---|
| 15 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|---|
| 16 | * GNU General Public License for more details.
|
|---|
| 17 | *
|
|---|
| 18 | * Suggestions: blitz-suggest@cybervision.com
|
|---|
| 19 | * Bugs: blitz-bugs@cybervision.com
|
|---|
| 20 | *
|
|---|
| 21 | * For more information, please see the Blitz++ Home Page:
|
|---|
| 22 | * http://seurat.uwaterloo.ca/blitz/
|
|---|
| 23 | *
|
|---|
| 24 | ***************************************************************************
|
|---|
| 25 | * $Log: not supported by cvs2svn $
|
|---|
| 26 | * Revision 1.1.1.1 1999/04/09 17:58:59 ansari
|
|---|
| 27 | * Creation module DPC/Blitz (blitz 0.4) Reza 09/04/99
|
|---|
| 28 | *
|
|---|
| 29 | * Revision 1.6 1998/03/14 00:04:47 tveldhui
|
|---|
| 30 | * 0.2-alpha-05
|
|---|
| 31 | *
|
|---|
| 32 | * Revision 1.5 1997/07/16 14:51:20 tveldhui
|
|---|
| 33 | * Update: Alpha release 0.2 (Arrays)
|
|---|
| 34 | *
|
|---|
| 35 | * Revision 1.4 1997/01/24 14:42:00 tveldhui
|
|---|
| 36 | * Periodic RCS update
|
|---|
| 37 | *
|
|---|
| 38 | * Revision 1.3 1997/01/23 03:28:28 tveldhui
|
|---|
| 39 | * Periodic RCS update
|
|---|
| 40 | *
|
|---|
| 41 | * Revision 1.2 1997/01/13 22:19:58 tveldhui
|
|---|
| 42 | * Periodic RCS update
|
|---|
| 43 | *
|
|---|
| 44 | * Revision 1.1 1996/11/11 17:29:13 tveldhui
|
|---|
| 45 | * Initial revision
|
|---|
| 46 | *
|
|---|
| 47 | *
|
|---|
| 48 | */
|
|---|
| 49 |
|
|---|
| 50 | #ifndef BZ_TUNING_H
|
|---|
| 51 | #define BZ_TUNING_H
|
|---|
| 52 |
|
|---|
| 53 | // These estimates should be conservative (i.e. underestimate the
|
|---|
| 54 | // cache sizes).
|
|---|
| 55 | #define BZ_L1_CACHE_ESTIMATED_SIZE 8192
|
|---|
| 56 | #define BZ_L2_CACHE_ESTIMATED_SIZE 65536
|
|---|
| 57 |
|
|---|
| 58 |
|
|---|
| 59 | #undef BZ_PARTIAL_LOOP_UNROLL
|
|---|
| 60 | #define BZ_PASS_EXPR_BY_VALUE
|
|---|
| 61 | #undef BZ_PTR_INC_FASTER_THAN_INDIRECTION
|
|---|
| 62 | #define BZ_MANUAL_VECEXPR_COPY_CONSTRUCTOR
|
|---|
| 63 | #undef BZ_KCC_COPY_PROPAGATION_KLUDGE
|
|---|
| 64 | #undef BZ_ALTERNATE_FORWARD_BACKWARD_TRAVERSALS
|
|---|
| 65 | #undef BZ_ARRAY_EXPR_PASS_INDEX_BY_VALUE
|
|---|
| 66 | #define BZ_INLINE_GROUP1
|
|---|
| 67 | #define BZ_INLINE_GROUP2
|
|---|
| 68 | #define BZ_COLLAPSE_LOOPS
|
|---|
| 69 | #define BZ_USE_FAST_READ_ARRAY_EXPR
|
|---|
| 70 | #define BZ_ARRAY_EXPR_USE_COMMON_STRIDE
|
|---|
| 71 | #undef BZ_ARRAY_FAST_TRAVERSAL_UNROLL
|
|---|
| 72 | #undef BZ_ARRAY_STACK_TRAVERSAL_CSE_AND_ANTIALIAS
|
|---|
| 73 | #undef BZ_ARRAY_STACK_TRAVERSAL_UNROLL
|
|---|
| 74 | #define BZ_ARRAY_2D_STENCIL_TILING
|
|---|
| 75 | #define BZ_ARRAY_2D_STENCIL_TILE_SIZE 128
|
|---|
| 76 | #undef BZ_INTERLACE_ARRAYS
|
|---|
| 77 |
|
|---|
| 78 |
|
|---|
| 79 | /*
|
|---|
| 80 | * Platform-specific tuning
|
|---|
| 81 | */
|
|---|
| 82 |
|
|---|
| 83 | #ifdef _CRAYT3E
|
|---|
| 84 | // The backend compiler on the T3E does a better job of
|
|---|
| 85 | // loop unrolling.
|
|---|
| 86 | #undef BZ_PARTIAL_LOOP_UNROLL
|
|---|
| 87 | #undef BZ_ARRAY_FAST_TRAVERSAL_UNROLL
|
|---|
| 88 | #undef BZ_ARRAY_STACK_TRAVERSAL_UNROLL
|
|---|
| 89 | #endif
|
|---|
| 90 |
|
|---|
| 91 | #ifdef __GNUC__
|
|---|
| 92 | // The egcs compiler does a good job of loop unrolling, if
|
|---|
| 93 | // -funroll-loops is used.
|
|---|
| 94 | #undef BZ_PARTIAL_LOOP_UNROLL
|
|---|
| 95 | #undef BZ_ARRAY_FAST_TRAVERSAL_UNROLL
|
|---|
| 96 | #undef BZ_ARRAY_STACK_TRAVERSAL_UNROLL
|
|---|
| 97 | #endif
|
|---|
| 98 |
|
|---|
| 99 | #ifdef BZ_DISABLE_KCC_COPY_PROPAGATION_KLUDGE
|
|---|
| 100 | #undef BZ_KCC_COPY_PROPAGATION_KLUDGE
|
|---|
| 101 | #endif
|
|---|
| 102 |
|
|---|
| 103 | #ifdef BZ_INLINE_GROUP1
|
|---|
| 104 | #define _bz_inline1 inline
|
|---|
| 105 | #else
|
|---|
| 106 | #define _bz_inline1
|
|---|
| 107 | #endif
|
|---|
| 108 |
|
|---|
| 109 | #ifdef BZ_INLINE_GROUP2
|
|---|
| 110 | #define _bz_inline2 inline
|
|---|
| 111 | #else
|
|---|
| 112 | #define _bz_inline2
|
|---|
| 113 | #endif
|
|---|
| 114 |
|
|---|
| 115 | #ifdef __DECCXX
|
|---|
| 116 | // The DEC cxx compiler has problems with loop unrolling
|
|---|
| 117 | // because of aliasing. Loop unrolling and anti-aliasing
|
|---|
| 118 | // is done by Blitz++.
|
|---|
| 119 |
|
|---|
| 120 | #define BZ_PARTIAL_LOOP_UNROLL
|
|---|
| 121 | #define BZ_ARRAY_STACK_TRAVERSAL_CSE_AND_ANTIALIAS
|
|---|
| 122 | #define BZ_ARRAY_STACK_TRAVERSAL_UNROLL
|
|---|
| 123 | #endif
|
|---|
| 124 |
|
|---|
| 125 | /*
|
|---|
| 126 | * BZ_NO_PROPAGATE(X) prevents the compiler from performing
|
|---|
| 127 | * copy propagation on a variable. This is used for loop
|
|---|
| 128 | * unrolling to prevent KAI C++ from rearranging the
|
|---|
| 129 | * ordering of memory accesses.
|
|---|
| 130 | */
|
|---|
| 131 |
|
|---|
| 132 | #define BZ_NO_PROPAGATE(X) X
|
|---|
| 133 |
|
|---|
| 134 | #ifdef __KCC
|
|---|
| 135 | #ifdef BZ_USE_NO_PROPAGATE
|
|---|
| 136 | extern "C" int __kai_apply(const char*, ...);
|
|---|
| 137 |
|
|---|
| 138 | #undef BZ_NO_PROPAGATE(X)
|
|---|
| 139 | #define BZ_NO_PROPAGATE(X) __kai_apply("(%a)",&X)
|
|---|
| 140 | #endif
|
|---|
| 141 | #endif
|
|---|
| 142 |
|
|---|
| 143 | #endif // BZ_TUNING_H
|
|---|
![(please configure the [header_logo] section in trac.ini)](/PI/chrome/site/your_project_logo.png){kind=logo}
