1 : |
dalvarez |
1 |
/*****************************************************************************/
|
2 : |
|
|
/* */
|
3 : |
|
|
/* Copyright (c) 2008, 2009, 2010 */
|
4 : |
|
|
/* Computer Architecture Group (CAG) */
|
5 : |
|
|
/* University of A Coruña, Spain */
|
6 : |
|
|
/* (http://gac.des.udc.es) */
|
7 : |
|
|
/* Galicia Supercomputing Center (CESGA) */
|
8 : |
|
|
/* (http://www.cesga.es) */
|
9 : |
|
|
/* Hewlett-Packard Spain (HP) */
|
10 : |
|
|
/* (http://www.hp.es) */
|
11 : |
|
|
/* */
|
12 : |
|
|
/* This file is part of UPC Operations Microbenchmarking Suite (UOMS). */
|
13 : |
|
|
/* */
|
14 : |
|
|
/* UOMS is free software: you can redistribute it and/or modify */
|
15 : |
|
|
/* it under the terms of the GNU Lesser General Public License as published */
|
16 : |
|
|
/* by the Free Software Foundation, either version 3 of the License, or */
|
17 : |
|
|
/* (at your option) any later version. */
|
18 : |
|
|
/* */
|
19 : |
|
|
/* UOMS is distributed in the hope that it will be useful, */
|
20 : |
|
|
/* but WITHOUT ANY WARRANTY; without even the implied warranty of */
|
21 : |
|
|
/* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the */
|
22 : |
|
|
/* GNU Lesser General Public License for more details. */
|
23 : |
|
|
/* */
|
24 : |
|
|
/* You should have received a copy of the GNU Lesser General Public License */
|
25 : |
|
|
/* along with UOMS. If not, see <http://www.gnu.org/licenses/>. */
|
26 : |
|
|
/* */
|
27 : |
|
|
/*****************************************************************************/
|
28 : |
|
|
|
29 : |
|
|
/*****************************************************************************/
|
30 : |
|
|
/* */
|
31 : |
|
|
/* FUNDING: This development has been funded by Hewlett-Packard Spain */
|
32 : |
|
|
/* */
|
33 : |
|
|
/* Project Name: */
|
34 : |
|
|
/* UPCHACO (2008-2011) */
|
35 : |
|
|
/* Subproject: */
|
36 : |
|
|
/* Improving UPC Usability and Performance in Constellation Systems: */
|
37 : |
|
|
/* Implementation/Extensions of UPC Libraries. */
|
38 : |
|
|
/* (UPCPUProject -> UPC Performance and Usability Project) */
|
39 : |
|
|
/* */
|
40 : |
|
|
/*****************************************************************************/
|
41 : |
|
|
|
42 : |
|
|
/*****************************************************************************
|
43 : |
|
|
|
44 : |
|
|
For further documentation, see
|
45 : |
|
|
|
46 : |
|
|
[1] Files under doc/
|
47 : |
|
|
|
48 : |
|
|
******************************************************************************/
|
49 : |
|
|
|
50 : |
|
|
#include "../defines.h"
|
51 : |
|
|
|
52 : |
|
|
extern int cache_invalidation;
|
53 : |
|
|
|
54 : |
|
|
/*********************************************
|
55 : |
|
|
Utility functions (niters...)
|
56 : |
|
|
**********************************************/
|
57 : |
|
|
|
58 : |
|
|
int niters(int msize)
|
59 : |
|
|
{
|
60 : |
|
|
/*
|
61 : |
|
|
* With cache invalidation the number of buckets will be equal to the number
|
62 : |
|
|
* of iterations, resulting in excesive memory consumption. To avoid that
|
63 : |
|
|
* the number of iterations will be less than without cache invalidation.
|
64 : |
|
|
*/
|
65 : |
|
|
|
66 : |
|
|
int iters = 0;
|
67 : |
|
|
int factor = 1;
|
68 : |
|
|
|
69 : |
|
|
if(cache_invalidation == 1){
|
70 : |
|
|
if (msize <= 16*1024) iters=100;
|
71 : |
|
|
else if (msize <= 512*1024) iters=50;
|
72 : |
|
|
else if (msize <= 4*1024*1024) iters=25;
|
73 : |
|
|
else if (msize <= 16*1024*1024) iters=10;
|
74 : |
|
|
else iters=5;
|
75 : |
|
|
}
|
76 : |
|
|
else{
|
77 : |
|
|
if (msize <= 16*1024) iters=1000;
|
78 : |
|
|
else if (msize <= 512*1024) iters=500;
|
79 : |
|
|
else if (msize <= 4*1024*1024) iters=250;
|
80 : |
|
|
else if (msize <= 16*1024*1024) iters=100;
|
81 : |
|
|
else iters=50;
|
82 : |
|
|
}
|
83 : |
|
|
|
84 : |
|
|
if(THREADS<16){
|
85 : |
|
|
factor=1;
|
86 : |
|
|
}
|
87 : |
|
|
else if(THREADS<64){
|
88 : |
|
|
factor=2;
|
89 : |
|
|
}
|
90 : |
|
|
else if(THREADS<256){
|
91 : |
|
|
factor=3;
|
92 : |
|
|
}
|
93 : |
|
|
else if(THREADS<1024){
|
94 : |
|
|
factor=4;
|
95 : |
|
|
}
|
96 : |
|
|
else{
|
97 : |
|
|
factor=5;
|
98 : |
|
|
}
|
99 : |
|
|
|
100 : |
|
|
if(iters/factor < 1)
|
101 : |
|
|
return 1;
|
102 : |
|
|
else
|
103 : |
|
|
return iters/factor;
|
104 : |
|
|
|
105 : |
|
|
}
|
106 : |
|
|
|
107 : |
|
|
/*
|
108 : |
|
|
Sets the header type
|
109 : |
|
|
*/
|
110 : |
|
|
int operation_header(int operation_code){
|
111 : |
|
|
|
112 : |
|
|
int headertype=0;
|
113 : |
|
|
|
114 : |
|
|
switch (operation_code) {
|
115 : |
|
|
case BROADCAST:
|
116 : |
|
|
case SCATTER:
|
117 : |
|
|
case GATHER:
|
118 : |
|
|
case GATHERALL:
|
119 : |
|
|
case EXCHANGE:
|
120 : |
|
|
case PERMUTE:
|
121 : |
|
|
case MEMGET:
|
122 : |
|
|
case MEMPUT:
|
123 : |
|
|
case MEMCPY:
|
124 : |
|
|
case LMEMGET:
|
125 : |
|
|
case LMEMPUT:
|
126 : |
|
|
case LMEMCPY:
|
127 : |
|
|
#ifdef ASYNC_MEM_TEST
|
128 : |
|
|
case AMEMGET:
|
129 : |
|
|
case AMEMPUT:
|
130 : |
|
|
case AMEMCPY:
|
131 : |
|
|
case ALMEMGET:
|
132 : |
|
|
case ALMEMPUT:
|
133 : |
|
|
case ALMEMCPY:
|
134 : |
|
|
#endif
|
135 : |
|
|
case SMEMCPY:
|
136 : |
|
|
case MEMMOVE:
|
137 : |
|
|
headertype=3;
|
138 : |
|
|
break;
|
139 : |
|
|
case BARRIER:
|
140 : |
|
|
headertype=1;
|
141 : |
|
|
break;
|
142 : |
|
|
default:
|
143 : |
|
|
headertype=2;
|
144 : |
|
|
break;
|
145 : |
|
|
}
|
146 : |
|
|
return headertype;
|
147 : |
|
|
}
|