| 1 |
48 |
alirezamon |
/*
|
| 2 |
|
|
* Copyright (c) 2010-2011 The University of Texas at Austin
|
| 3 |
|
|
* All rights reserved.
|
| 4 |
|
|
*
|
| 5 |
|
|
* Redistribution and use in source and binary forms, with or without
|
| 6 |
|
|
* modification, are permitted provided that the following conditions are
|
| 7 |
|
|
* met: redistributions of source code must retain the above copyright
|
| 8 |
|
|
* notice, this list of conditions and the following disclaimer;
|
| 9 |
|
|
* redistributions in binary form must reproduce the above copyright
|
| 10 |
|
|
* notice, this list of conditions and the following disclaimer in the
|
| 11 |
|
|
* documentation and/or other materials provided with the distribution;
|
| 12 |
|
|
* neither the name of the copyright holders nor the names of its
|
| 13 |
|
|
* contributors may be used to endorse or promote products derived from
|
| 14 |
|
|
* this software without specific prior written permission.
|
| 15 |
|
|
*
|
| 16 |
|
|
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
|
| 17 |
|
|
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
|
| 18 |
|
|
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
|
| 19 |
|
|
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
|
| 20 |
|
|
* OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
|
| 21 |
|
|
* SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
|
| 22 |
|
|
* LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
|
| 23 |
|
|
* DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
|
| 24 |
|
|
* THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
|
| 25 |
|
|
* (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
|
| 26 |
|
|
* OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
|
| 27 |
|
|
*/
|
| 28 |
|
|
|
| 29 |
|
|
#include <stdio.h>
|
| 30 |
|
|
#include <stdlib.h>
|
| 31 |
|
|
#include <math.h>
|
| 32 |
|
|
unsigned long int nt_tr_list_pck=0;
|
| 33 |
|
|
|
| 34 |
|
|
unsigned char done=0;
|
| 35 |
|
|
unsigned long int end_sim_pck_num=0;
|
| 36 |
|
|
|
| 37 |
|
|
#include "queue.h"
|
| 38 |
|
|
#include "netrace.h"
|
| 39 |
|
|
|
| 40 |
|
|
#include "queue.c"
|
| 41 |
|
|
#include "netrace.c"
|
| 42 |
|
|
|
| 43 |
|
|
#include "inttypes.h"
|
| 44 |
|
|
|
| 45 |
|
|
#define L2_LATENCY 8
|
| 46 |
|
|
|
| 47 |
|
|
|
| 48 |
|
|
unsigned long int nt_total_rd_pck=0;
|
| 49 |
|
|
unsigned long int nt_total_ejected_pck=0;
|
| 50 |
|
|
|
| 51 |
|
|
unsigned long long int calc_packet_timing( nt_packet_t* );
|
| 52 |
|
|
|
| 53 |
|
|
int verbosity=1;
|
| 54 |
|
|
|
| 55 |
|
|
nt_header_t* header;
|
| 56 |
|
|
int x_nodes, y_nodes;
|
| 57 |
|
|
unsigned long long int cycle;
|
| 58 |
|
|
|
| 59 |
|
|
|
| 60 |
|
|
unsigned long long int start_cycle=0;
|
| 61 |
|
|
unsigned int router_pipe_latency=3;
|
| 62 |
|
|
|
| 63 |
|
|
typedef struct queue_node queue_node_t;
|
| 64 |
|
|
struct queue_node {
|
| 65 |
|
|
nt_packet_t* packet;
|
| 66 |
|
|
unsigned long long int cycle;
|
| 67 |
|
|
};
|
| 68 |
|
|
|
| 69 |
|
|
int main( int argc, char** argv ) {
|
| 70 |
|
|
|
| 71 |
|
|
int i;
|
| 72 |
|
|
int ignore_dependencies = 0;
|
| 73 |
|
|
int start_region = 0;
|
| 74 |
|
|
int reader_throttling = 0;
|
| 75 |
|
|
|
| 76 |
|
|
|
| 77 |
|
|
char* tracefile;
|
| 78 |
|
|
if( argc > 1 ) {
|
| 79 |
|
|
tracefile = argv[1];
|
| 80 |
|
|
for( i = 2; i < argc; ++i ) {
|
| 81 |
|
|
if( strcmp(argv[i], "-d") == 0 ) {
|
| 82 |
|
|
ignore_dependencies = 1;
|
| 83 |
|
|
} else if( strcmp(argv[i], "-r") == 0 ) {
|
| 84 |
|
|
if( argc > ++i ) {
|
| 85 |
|
|
start_region = atoi(argv[i]);
|
| 86 |
|
|
} else {
|
| 87 |
|
|
fprintf( stderr, "ERROR: Need to specify parameter to -r option\n" );
|
| 88 |
|
|
exit(0);
|
| 89 |
|
|
}
|
| 90 |
|
|
}
|
| 91 |
|
|
else if( strcmp(argv[i], "-n") == 0 ) {
|
| 92 |
|
|
if( argc > ++i ) {
|
| 93 |
|
|
end_sim_pck_num=atoi(argv[i]);
|
| 94 |
|
|
printf ("\tSimulation ends on total packet num of =%ld\n", end_sim_pck_num);
|
| 95 |
|
|
} else {
|
| 96 |
|
|
fprintf( stderr, "ERROR: Need to specify parameter to -n option\n" );
|
| 97 |
|
|
exit(0);
|
| 98 |
|
|
}
|
| 99 |
|
|
}
|
| 100 |
|
|
|
| 101 |
|
|
else if( strcmp(argv[i], "-p") == 0 ) {
|
| 102 |
|
|
if( argc > ++i ) {
|
| 103 |
|
|
router_pipe_latency=atoi(argv[i]);
|
| 104 |
|
|
} else {
|
| 105 |
|
|
fprintf( stderr, "ERROR: Need to specify parameter to -p option\n" );
|
| 106 |
|
|
exit(0);
|
| 107 |
|
|
}
|
| 108 |
|
|
}
|
| 109 |
|
|
|
| 110 |
|
|
|
| 111 |
|
|
|
| 112 |
|
|
|
| 113 |
|
|
else if( strcmp(argv[i], "-v") == 0 ) {
|
| 114 |
|
|
if( argc > ++i ) {
|
| 115 |
|
|
verbosity=atoi(argv[i]);
|
| 116 |
|
|
} else {
|
| 117 |
|
|
fprintf( stderr, "ERROR: Need to specify parameter to -v option\n" );
|
| 118 |
|
|
exit(0);
|
| 119 |
|
|
}
|
| 120 |
|
|
}
|
| 121 |
|
|
|
| 122 |
|
|
|
| 123 |
|
|
|
| 124 |
|
|
else if( strcmp(argv[i], "-t") == 0 ) {
|
| 125 |
|
|
reader_throttling = 1;
|
| 126 |
|
|
} else {
|
| 127 |
|
|
fprintf( stderr, "ERROR: Unknown parameter %s\n", argv[i] );
|
| 128 |
|
|
exit(0);
|
| 129 |
|
|
}
|
| 130 |
|
|
}
|
| 131 |
|
|
} else {
|
| 132 |
|
|
printf( "ERROR: Please specify trace file\n" );
|
| 133 |
|
|
exit(0);
|
| 134 |
|
|
}
|
| 135 |
|
|
|
| 136 |
|
|
|
| 137 |
|
|
|
| 138 |
|
|
int packets_left = 0;
|
| 139 |
|
|
cycle = 0;
|
| 140 |
|
|
nt_packet_t* trace_packet = NULL;
|
| 141 |
|
|
nt_packet_t* packet = NULL;
|
| 142 |
|
|
queue_t** waiting;
|
| 143 |
|
|
queue_t** inject;
|
| 144 |
|
|
queue_t** traverse;
|
| 145 |
|
|
nt_open_trfile( tracefile );
|
| 146 |
|
|
if( ignore_dependencies ) {
|
| 147 |
|
|
nt_disable_dependencies();
|
| 148 |
|
|
}
|
| 149 |
|
|
nt_print_trheader();
|
| 150 |
|
|
header = nt_get_trheader();
|
| 151 |
|
|
nt_seek_region( &header->regions[start_region] );
|
| 152 |
|
|
for( i = 0; i < start_region; i++ ) {
|
| 153 |
|
|
cycle += header->regions[i].num_cycles;
|
| 154 |
|
|
}
|
| 155 |
|
|
start_cycle = cycle;
|
| 156 |
|
|
x_nodes = sqrt( header->num_nodes );
|
| 157 |
|
|
y_nodes = header->num_nodes / x_nodes;
|
| 158 |
|
|
waiting = (queue_t**) malloc( header->num_nodes * sizeof(queue_t*) );
|
| 159 |
|
|
inject = (queue_t**) malloc( header->num_nodes * sizeof(queue_t*) );
|
| 160 |
|
|
traverse = (queue_t**) malloc( header->num_nodes * sizeof(queue_t*) );
|
| 161 |
|
|
if( (waiting == NULL) || (inject == NULL) || (traverse == NULL) ) {
|
| 162 |
|
|
printf( "malloc fail queues\n" );
|
| 163 |
|
|
exit(0);
|
| 164 |
|
|
}
|
| 165 |
|
|
for( i = 0; i < header->num_nodes; ++i ) {
|
| 166 |
|
|
waiting[i] = queue_new();
|
| 167 |
|
|
inject[i] = queue_new();
|
| 168 |
|
|
traverse[i] = queue_new();
|
| 169 |
|
|
}
|
| 170 |
|
|
|
| 171 |
|
|
if( !reader_throttling ) {
|
| 172 |
|
|
trace_packet = nt_read_packet();
|
| 173 |
|
|
|
| 174 |
|
|
} else if( !ignore_dependencies ) {
|
| 175 |
|
|
nt_init_self_throttling();
|
| 176 |
|
|
}
|
| 177 |
|
|
|
| 178 |
|
|
while( (done==0 && ( cycle <= header->num_cycles )) || packets_left!=0 ) {
|
| 179 |
|
|
|
| 180 |
|
|
// Reset packets remaining check
|
| 181 |
|
|
packets_left = 0;
|
| 182 |
|
|
|
| 183 |
|
|
// Get packets for this cycle
|
| 184 |
|
|
if((end_sim_pck_num == 0 ) || (end_sim_pck_num > nt_total_rd_pck )){
|
| 185 |
|
|
if( reader_throttling ) {
|
| 186 |
|
|
nt_packet_list_t* list;
|
| 187 |
|
|
for( list = nt_get_cleared_packets_list(); list != NULL; list = list->next ) {
|
| 188 |
|
|
if( list->node_packet != NULL ) {
|
| 189 |
|
|
trace_packet = list->node_packet;
|
| 190 |
|
|
queue_node_t* new_node = (queue_node_t*) nt_checked_malloc( sizeof(queue_node_t) );
|
| 191 |
|
|
new_node->packet = trace_packet;
|
| 192 |
|
|
new_node->cycle = (trace_packet->cycle > cycle) ? trace_packet->cycle : cycle;
|
| 193 |
|
|
queue_push( inject[trace_packet->src], new_node, new_node->cycle );
|
| 194 |
|
|
nt_total_rd_pck++;
|
| 195 |
|
|
} else {
|
| 196 |
|
|
printf( "Malformed packet list" );
|
| 197 |
|
|
exit(-1);
|
| 198 |
|
|
}
|
| 199 |
|
|
}
|
| 200 |
|
|
nt_empty_cleared_packets_list();
|
| 201 |
|
|
} else {
|
| 202 |
|
|
while( (trace_packet != NULL) && (trace_packet->cycle == cycle) ) {
|
| 203 |
|
|
// Place in appropriate queue
|
| 204 |
|
|
queue_node_t* new_node = (queue_node_t*) nt_checked_malloc( sizeof(queue_node_t) );
|
| 205 |
|
|
new_node->packet = trace_packet;
|
| 206 |
|
|
|
| 207 |
|
|
new_node->cycle = (trace_packet->cycle > cycle) ? trace_packet->cycle : cycle;
|
| 208 |
|
|
|
| 209 |
|
|
if( ignore_dependencies || nt_dependencies_cleared( trace_packet ) ) {
|
| 210 |
|
|
// Add to inject queue
|
| 211 |
|
|
queue_push( inject[trace_packet->src], new_node, new_node->cycle );
|
| 212 |
|
|
} else {
|
| 213 |
|
|
// Add to waiting queue
|
| 214 |
|
|
queue_push( waiting[trace_packet->src], new_node, new_node->cycle );
|
| 215 |
|
|
}
|
| 216 |
|
|
// Get another packet from trace
|
| 217 |
|
|
trace_packet = nt_read_packet();
|
| 218 |
|
|
nt_total_rd_pck++;
|
| 219 |
|
|
|
| 220 |
|
|
}
|
| 221 |
|
|
if( (trace_packet != NULL) && (trace_packet->cycle < cycle) ) {
|
| 222 |
|
|
// Error check: Crash and burn
|
| 223 |
|
|
printf( "Invalid trace_packet cycle time: %llu, current cycle: %llu\n", trace_packet->cycle, cycle );
|
| 224 |
|
|
exit(-1);
|
| 225 |
|
|
}
|
| 226 |
|
|
}
|
| 227 |
|
|
}//endsim packet
|
| 228 |
|
|
else done=1;
|
| 229 |
|
|
|
| 230 |
|
|
// Inject where possible (max one per node)
|
| 231 |
|
|
for( i = 0; i < header->num_nodes; ++i ) {
|
| 232 |
|
|
packets_left |= !queue_empty( inject[i] );
|
| 233 |
|
|
queue_node_t* temp_node = (queue_node_t*) queue_peek_front( inject[i] );
|
| 234 |
|
|
if( temp_node != NULL ) {
|
| 235 |
|
|
packet = temp_node->packet;
|
| 236 |
|
|
if( (packet != NULL) && (temp_node->cycle <= cycle) ) {
|
| 237 |
|
|
|
| 238 |
|
|
if(verbosity>1) {
|
| 239 |
|
|
printf( "Inject: %llu ", cycle );
|
| 240 |
|
|
nt_print_packet( packet );
|
| 241 |
|
|
}
|
| 242 |
|
|
temp_node = (queue_node_t*) queue_pop_front( inject[i] );
|
| 243 |
|
|
temp_node->cycle = cycle + calc_packet_timing( packet );
|
| 244 |
|
|
queue_push( traverse[packet->dst], temp_node, temp_node->cycle );
|
| 245 |
|
|
//printf("\tInject packet to ProNoC:\n");
|
| 246 |
|
|
//printf ("\t\tsize=%u\n", nt_get_packet_size(packet));
|
| 247 |
|
|
//long int ptr_addr = reinterpret_cast<long int> (temp_node);
|
| 248 |
|
|
//printf ("\t\tpacket pointer addr: %p\n" ,temp_node);
|
| 249 |
|
|
//printf ("\t\tpacket pointer addr: %lx\n" ,ptr_addr);
|
| 250 |
|
|
//printf("0x%.16" PRIXPTR " contains 0x%" PRIXPTR "\n", (uintptr_t)&ptr_addr, (uintptr_t)ptr_addr);
|
| 251 |
|
|
}
|
| 252 |
|
|
}
|
| 253 |
|
|
}
|
| 254 |
|
|
|
| 255 |
|
|
|
| 256 |
|
|
|
| 257 |
|
|
// Step all network components, Eject where possible
|
| 258 |
|
|
for( i = 0; i < header->num_nodes; ++i ) {
|
| 259 |
|
|
packets_left |= !queue_empty( traverse[i] );
|
| 260 |
|
|
queue_node_t* temp_node = (queue_node_t*) queue_peek_front( traverse[i] );
|
| 261 |
|
|
if( temp_node != NULL ) {
|
| 262 |
|
|
packet = temp_node->packet;
|
| 263 |
|
|
if( (packet != NULL) && (temp_node->cycle <= cycle) ) {
|
| 264 |
|
|
if(verbosity>1) {
|
| 265 |
|
|
printf( "Eject: %llu ", cycle );
|
| 266 |
|
|
nt_print_packet( packet );
|
| 267 |
|
|
}
|
| 268 |
|
|
nt_total_ejected_pck++;
|
| 269 |
|
|
nt_clear_dependencies_free_packet( packet );
|
| 270 |
|
|
temp_node = (queue_node_t*) queue_pop_front( traverse[i] );
|
| 271 |
|
|
free( temp_node );
|
| 272 |
|
|
}
|
| 273 |
|
|
}
|
| 274 |
|
|
}
|
| 275 |
|
|
|
| 276 |
|
|
// Check for cleared dependences... or not
|
| 277 |
|
|
if( !reader_throttling ) {
|
| 278 |
|
|
for( i = 0; i < header->num_nodes; ++i ) {
|
| 279 |
|
|
packets_left |= !queue_empty( waiting[i] );
|
| 280 |
|
|
node_t* temp = waiting[i]->head;
|
| 281 |
|
|
while( temp != NULL ) {
|
| 282 |
|
|
queue_node_t* temp_node = (queue_node_t*) temp->elem;
|
| 283 |
|
|
packet = temp_node->packet;
|
| 284 |
|
|
temp = temp->next;
|
| 285 |
|
|
if( nt_dependencies_cleared( packet ) ) {
|
| 286 |
|
|
// remove from waiting
|
| 287 |
|
|
queue_remove( waiting[i], temp_node );
|
| 288 |
|
|
// add to inject
|
| 289 |
|
|
queue_node_t* new_node = (queue_node_t*) nt_checked_malloc( sizeof(queue_node_t) );
|
| 290 |
|
|
new_node->packet = packet;
|
| 291 |
|
|
new_node->cycle = cycle + L2_LATENCY;
|
| 292 |
|
|
queue_push( inject[i], new_node, new_node->cycle );
|
| 293 |
|
|
free( temp_node );
|
| 294 |
|
|
}
|
| 295 |
|
|
}
|
| 296 |
|
|
}
|
| 297 |
|
|
}
|
| 298 |
|
|
|
| 299 |
|
|
cycle++;
|
| 300 |
|
|
if(verbosity==1) if(cycle&0xFFF) printf("\rTotal read packet: %9ld", nt_total_rd_pck);
|
| 301 |
|
|
|
| 302 |
|
|
}
|
| 303 |
|
|
for( i = 0; i < header->num_nodes; ++i ) {
|
| 304 |
|
|
queue_delete( waiting[i] );
|
| 305 |
|
|
queue_delete( inject[i] );
|
| 306 |
|
|
queue_delete( traverse[i] );
|
| 307 |
|
|
}
|
| 308 |
|
|
free( waiting );
|
| 309 |
|
|
free( inject );
|
| 310 |
|
|
free( traverse );
|
| 311 |
|
|
nt_close_trfile();
|
| 312 |
|
|
printf ("\nrouter_pipe_latency =%d\n", router_pipe_latency);
|
| 313 |
|
|
printf( "Start cycle: %llu \n",start_cycle);
|
| 314 |
|
|
printf( "End cycle: %llu \n", cycle );
|
| 315 |
|
|
printf( "Sim cycle: %llu \n", cycle - start_cycle);
|
| 316 |
|
|
printf( "total rd pck: %lu \n",nt_total_rd_pck);
|
| 317 |
|
|
printf( "total ejected pck: %lu \n",nt_total_ejected_pck);
|
| 318 |
|
|
return 0;
|
| 319 |
|
|
|
| 320 |
|
|
}
|
| 321 |
|
|
|
| 322 |
|
|
unsigned long long int calc_packet_timing( nt_packet_t* packet ) {
|
| 323 |
|
|
int src_x = packet->src % x_nodes;
|
| 324 |
|
|
int src_y = packet->src / x_nodes;
|
| 325 |
|
|
int dst_x = packet->dst % x_nodes;
|
| 326 |
|
|
int dst_y = packet->dst / x_nodes;
|
| 327 |
|
|
int n_hops = abs( src_x - dst_x ) + abs( src_y - dst_y );
|
| 328 |
|
|
if( n_hops <= 0 ) n_hops = 1;
|
| 329 |
|
|
return router_pipe_latency*n_hops;
|
| 330 |
|
|
}
|
| 331 |
|
|
|
| 332 |
|
|
|
| 333 |
|
|
|
| 334 |
|
|
|
| 335 |
|
|
|
