express-mesh.C 76.6 KB
Newer Older
1 2 3 4 5 6 7 8
#include <ross.h>

#include "codes/jenkins-hash.h"
#include "codes/codes_mapping.h"
#include "codes/codes.h"
#include "codes/model-net.h"
#include "codes/model-net-method.h"
#include "codes/model-net-lp.h"
9 10

//CHANGE: use the network file created
11
#include "codes/net/express-mesh.h"
12

13 14 15 16 17 18 19
#include "codes/net/common-net.h"
#include "sys/file.h"
#include "codes/quickhash.h"
#include "codes/rc-stack.h"
#include <vector>

#define CREDIT_SZ 8
20 21
#define HASH_TABLE_SIZE 262144
#define MULT_FACTOR 2
22 23 24 25

#define DEBUG 0
#define MAX_STATS 65536

26 27 28 29 30 31 32 33 34 35
//CHANGE: define them for the local network
#define LOCAL_NETWORK_NAME EXPRESS_MESH
#define LOCAL_NETWORK_ROUTER_NAME EXPRESS_MESH_ROUTER
#define LOCAL_MSG_STRUCT em_message
#define LOCAL_MSG_NAME_FROM_UNION em_msg

#define LP_CONFIG_NM_TERM (model_net_lp_config_names[LOCAL_NETWORK_NAME])
#define LP_METHOD_NM_TERM (model_net_method_names[LOCAL_NETWORK_NAME])
#define LP_CONFIG_NM_ROUT (model_net_lp_config_names[LOCAL_NETWORK_ROUTER_NAME])
#define LP_METHOD_NM_ROUT (model_net_method_names[LOCAL_NETWORK_ROUTER_NAME])
36 37 38 39 40

static long packet_gen = 0, packet_fin = 0;

static double maxd(double a, double b) { return a < b ? b : a; }

41
typedef struct local_param local_param;
42
static uint64_t                  num_params = 0;
43
static local_param         * all_params = NULL;
44 45 46 47 48 49 50 51 52 53 54 55 56 57 58
static const config_anno_map_t * anno_map   = NULL;

/* global variables for codes mapping */
static char lp_group_name[MAX_NAME_LENGTH];
static int mapping_grp_id, mapping_type_id, mapping_rep_id, mapping_offset;

/* router magic number */
static int router_magic_num = 0;

/* terminal magic number */
static int terminal_magic_num = 0;

static int sample_bytes_written = 0;
static int sample_rtr_bytes_written = 0;

59 60
static char local_cn_sample_file[MAX_NAME_LENGTH];
static char local_rtr_sample_file[MAX_NAME_LENGTH];
61

62 63 64
static void init_message_list(message_list *thism,
    LOCAL_MSG_STRUCT *inmsg) {
  thism->LOCAL_MSG_NAME_FROM_UNION = *inmsg;
65 66 67 68 69 70 71 72
  thism->event_data = NULL;
  thism->next = NULL;
  thism->prev = NULL;
  thism->in_alt_q = 0;
  thism->altq_next = NULL;
  thism->altq_prev = NULL;
}

73
struct local_param
74 75 76 77 78 79 80 81
{
  double link_bandwidth;/* bandwidth of each link */
  double cn_bandwidth;/* injection bandwidth */
  int num_cn; // number of nodes per router
  int num_vcs; /* number of virtual channels */
  int vc_size; /* buffer size of the router-router channels */
  int cn_vc_size; /* buffer size of the compute node channels */
  int chunk_size; /* full-sized packets are broken into smaller chunks.*/
82 83 84
  int router_delay; /* delay at each router */
  int routing; /* type of routing */

85
  //derived param
86 87 88 89 90 91 92 93 94 95 96 97 98
  int radix; /* radix of the routers */
  int total_routers; /* how many routers in the system */
  int total_terminals; /* how many terminals in the system */
  double cn_delay; /* bandwidth based time for 1 byte */
  double link_delay; /* bandwidth based time for 1 byte */
  double credit_delay; /* how long for credit to arrive - all bytes */

  //CHANGE: add network specific data here
  int n_dims; // Dimensions in the base torus layout
  int *dim_length;
  int gap; // Gap at which nodes are connected (0 for log)
  int * factor; /* used in torus coordinate calculation */
  int *cons_per_dim, *offset_per_dim;
99 100
};

101
struct local_router_sample
102 103 104 105 106 107 108 109 110
{
  tw_lpid router_id;
  tw_stime* busy_time;
  int64_t* link_traffic_sample;
  tw_stime end_time;
  long fwd_events;
  long rev_events;
};

111
struct local_cn_sample
112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127
{
  tw_lpid terminal_id;
  long fin_chunks_sample;
  long data_size_sample;
  double fin_hops_sample;
  tw_stime fin_chunks_time;
  tw_stime busy_time_sample;
  tw_stime end_time;
  long fwd_events;
  long rev_events;
};

/* handles terminal and router events like packet generate/send/receive/buffer */
typedef struct terminal_state terminal_state;
typedef struct router_state router_state;

128
/* compute node data (think NIC) structure */
129 130 131
struct terminal_state
{
  unsigned int terminal_id; //what is my local id
132 133
  const char * anno;
  const local_param *params;
134

135 136
  //which router I am connected to
  unsigned int router_id;
137 138
  tw_lpid router_gid;

139
  // Each terminal will have input/output channel(s) with the router
140 141 142
  int** vc_occupancy; // NUM_VC
  tw_stime terminal_available_time;

143
  //available messages to be sent
144 145 146 147 148 149 150 151
  message_list ***terminal_msgs;
  message_list ***terminal_msgs_tail;
  int terminal_length;
  int in_send_loop;
  int issueIdle;

  //packet aggregation
  struct qhash_table *rank_tbl;
152
  //transient storage for reverse computation
153 154
  struct rc_stack * st;

155 156 157 158 159
  //stats collection
  uint64_t packet_counter;
  int packet_gen;
  int packet_fin;
  struct mn_stats local_stats_array[CATEGORY_MAX];
160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176
  tw_stime   total_time;
  uint64_t total_msg_size;
  double total_hops;
  long finished_msgs;
  long finished_chunks;
  long finished_packets;

  //sampling
  tw_stime last_buf_full;
  tw_stime busy_time;
  char output_buf[4096];
  long fin_chunks_sample;
  long data_size_sample;
  double fin_hops_sample;
  tw_stime fin_chunks_time;
  tw_stime busy_time_sample;
  char sample_buf[4096];
177
  struct local_cn_sample * sample_stat;
178 179 180 181 182 183 184
  int op_arr_size;
  int max_arr_size;
  /* for logging forward and reverse events */
  long fwd_events;
  long rev_events;
};

185 186
//CHANGE: may need to change if more functionality is desired
/* event types */
187 188 189 190 191 192 193 194 195 196
enum event_t
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  R_SEND,
  R_ARRIVE,
  R_BUFFER,
};
Nikhil's avatar
Nikhil committed
197
typedef enum event_t event_t;
198

199
//CHANGE: may need to change if more functionality is desired
200 201 202 203 204 205 206
/* whether the last hop of a packet was global, local or a terminal */
enum last_hop
{
  ROUTER=1,
  TERMINAL
};

207
//CHANGE: may need to change if more functionality is desired
208 209 210 211 212 213 214 215 216 217
enum ROUTING_ALGO
{
  STATIC = 0,
  ADAPTIVE,
};

struct router_state
{
  //who am I
  unsigned int router_id;
218 219
  const char * anno;
  const local_param *params;
220

221 222
  //CHANGE: may need to be changed if linear storage is not desired
  //array/linked list based storage of info about ports/vcs
223 224 225 226 227 228
  tw_lpid* link_connections;
  tw_stime* next_output_available_time;
  message_list ***pending_msgs;
  message_list ***pending_msgs_tail;
  message_list ***queued_msgs;
  message_list ***queued_msgs_tail;
229
  int** vc_occupancy;
230 231 232
  int *in_send_loop;
  int *queued_count;

233 234
  //for reverse computation
  struct rc_stack * st;
235 236

  //sampling and stats
237
  int64_t* link_traffic;
238 239 240
  tw_stime* last_buf_full;
  tw_stime* busy_time;
  tw_stime* busy_time_sample;
241
  struct local_router_sample * rsamples;
242 243
  int op_arr_size;
  int max_arr_size;
244
  long fwd_events, rev_events;
245
  int64_t * link_traffic_sample;
246 247 248 249 250
  char output_buf[4096];
  char output_buf2[4096];

  //CHANGE: add network specific data here
  int* dim_position;
251 252 253 254 255 256 257 258
};

struct VC_Entry {
  int vc;
  message_list* entry;
};

//global stats
259 260
static tw_stime         local_total_time = 0;
static tw_stime         local_max_latency = 0;
261 262 263 264 265 266 267

static long long       total_hops = 0;
static long long       N_finished_packets = 0;
static long long       total_msg_sz = 0;
static long long       N_finished_msgs = 0;
static long long       N_finished_chunks = 0;

268 269
/* returns the message size */
static int local_get_msg_sz(void)
270
{
271
  return sizeof(LOCAL_MSG_STRUCT);
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
}

/* helper functions - convert between flat ids and torus n-dimensional ids */
static void to_dim_id(
    int flat_id,
    int ndims,
    const int *dim_lens,
    int *out_dim_ids)
{
  for (int i = 0; i < ndims; i++) {
    out_dim_ids[i] = flat_id % dim_lens[i];
    flat_id /= dim_lens[i];
  }
}

static int to_flat_id(
    int ndims,
    const int *dim_lens,
    const int *dim_ids)
{
  int flat_id = dim_ids[0];
  int mult = dim_lens[0];
  for (int i = 1; i < ndims; i++) {
    flat_id += dim_ids[i] * mult;
    mult *= dim_lens[i];
  }
  return flat_id;
}

301 302 303 304 305
//CHANGE: network specific params have to be read here
static void local_read_config(const char * anno, local_param *params){
  local_param *p = params;

  // general params - do not change unless you intent to modify them
Nikhil's avatar
Nikhil committed
306
  int rc = configuration_get_value_double(&config, "PARAMS", "link_bandwidth",
307 308 309
      anno, &p->link_bandwidth);
  if(rc) {
    p->link_bandwidth = 5.25;
310
    fprintf(stderr, "Bandwidth of links  not specified, setting to %lf\n",
311 312 313
        p->link_bandwidth);
  }

314
  rc = configuration_get_value_double(&config, "PARAMS", "cn_bandwidth",
315 316 317 318 319 320 321
      anno, &p->cn_bandwidth);
  if(rc) {
    p->cn_bandwidth = 5.25;
    fprintf(stderr, "Bandwidth of compute node channels not specified, setting "
        "to %lf\n", p->cn_bandwidth);
  }

322
  rc = configuration_get_value_int(&config, "PARAMS", "num_cn", anno,
323 324 325 326 327
      &p->num_cn);
  if(rc) {
    tw_error(TW_LOC, "Nodes per router (num_cn) not specified\n");
  }

328
  rc = configuration_get_value_int(&config, "PARAMS", "num_vcs", anno,
329 330 331 332
      &p->num_vcs);
  if(rc) {
    p->num_vcs = 1;
  }
333 334

  rc = configuration_get_value_int(&config, "PARAMS", "chunk_size", anno,
335 336 337
      &p->chunk_size);
  if(rc) {
    p->chunk_size = 512;
338
    fprintf(stderr, "Chunk size for packets is not specified, setting to %d\n",
339 340 341
      p->chunk_size);
  }

342
  rc = configuration_get_value_int(&config, "PARAMS", "vc_size", anno,
343 344 345
      &p->vc_size);
  if(rc) {
    p->vc_size = 32768;
346
    fprintf(stderr, "Buffer size of link channels not specified, setting to %d\n",
347 348 349
      p->vc_size);
  }

350
  rc = configuration_get_value_int(&config, "PARAMS", "cn_vc_size", anno,
351 352 353 354 355 356 357 358 359 360 361
      &p->cn_vc_size);
  if(rc) {
    p->cn_vc_size = 65536;
    fprintf(stderr, "Buffer size of compute node channels not specified, "
        "setting to %d\n", p->cn_vc_size);
  }

  p->router_delay = 50;
  configuration_get_value_int(&config, "PARAMS", "router_delay", anno,
      &p->router_delay);

362 363 364 365 366 367
  configuration_get_value(&config, "PARAMS", "cn_sample_file", anno,
      local_cn_sample_file, MAX_NAME_LENGTH);
  configuration_get_value(&config, "PARAMS", "rt_sample_file", anno,
      local_rtr_sample_file, MAX_NAME_LENGTH);

  //CHANGE: add network specific parameters here
Nikhil's avatar
Nikhil committed
368
  rc = configuration_get_value_int(&config, "PARAMS", "n_dims", anno,
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
      &p->n_dims);
  if(rc) {
    tw_error(TW_LOC, "Number of dimensions not specified\n");
  }

  rc = configuration_get_value_int(&config, "PARAMS", "gap", anno, &p->gap);
  if(rc) {
    tw_error(TW_LOC, "Gap not specified\n");
  }

  char dim_length_str[MAX_NAME_LENGTH];
  rc = configuration_get_value(&config, "PARAMS", "dim_length", anno,
      dim_length_str, MAX_NAME_LENGTH);
  if (rc == 0){
    tw_error(TW_LOC, "couldn't read PARAMS:dim_length");
  }
  char* token;
  p->dim_length= (int*)malloc(p->n_dims * sizeof(*p->dim_length));
  token = strtok(dim_length_str, ",");
  int i = 0;
  while(token != NULL)
  {
    sscanf(token, "%d", &p->dim_length[i]);
    if(p->dim_length[i] <= 0)
    {
      tw_error(TW_LOC, "Invalid torus dimension specified "
          "(%d at pos %d), exiting... ", p->dim_length[i], i);
    }
    i++;
    token = strtok(NULL,",");
  }
400 401 402 403 404 405 406 407 408 409 410 411 412 413 414

  char routing_str[MAX_NAME_LENGTH];
  configuration_get_value(&config, "PARAMS", "routing", anno, routing_str,
      MAX_NAME_LENGTH);
  if(strcmp(routing_str, "static") == 0)
    p->routing = STATIC;
  else if (strcmp(routing_str, "adaptive") == 0) {
    p->routing = ADAPTIVE;
    if(p->num_vcs < 2) {
      p->num_vcs = 2;
    }
  }
  else
  {
    p->routing = STATIC;
415
    fprintf(stderr,
416 417 418
        "No routing protocol specified, setting to static routing\n");
  }

419
  //CHANGE: derived parameters often are computed based on network specifics
420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437
  p->radix = 0;
  p->total_routers = 1;
  p->cons_per_dim = (int *)malloc(p->n_dims * sizeof(int));
  p->offset_per_dim = (int *)malloc(p->n_dims * sizeof(int));
  for(int i = 0; i < p->n_dims; i++) {
    p->cons_per_dim[i] = 2 + (p->dim_length[i] - 3)/p->gap;
    p->radix += p->cons_per_dim[i];
    p->total_routers *= p->dim_length[i];
    if(i == 0) {
      p->offset_per_dim[i] = p->num_cn;
    } else {
      p->offset_per_dim[i] = p->offset_per_dim[i - 1] + p->cons_per_dim[i - 1];
    }
  }
  p->radix += p->num_cn;
  p->total_terminals = p->total_routers * p->num_cn;

  int rank;
438
  MPI_Comm_rank(MPI_COMM_CODES, &rank);
439 440 441 442 443
  if(!rank) {
    printf("\n Total nodes %d routers %d radix %d \n",
        p->total_terminals, p->total_routers, p->radix);
  }

444
  //general derived parameters
445 446 447
  p->cn_delay = bytes_to_ns(1, p->cn_bandwidth);
  p->link_delay = bytes_to_ns(1, p->link_bandwidth);
  p->credit_delay = bytes_to_ns(CREDIT_SZ, p->link_bandwidth);
448 449 450 451 452 453 454 455

  uint32_t h1 = 0, h2 = 0;
  bj_hashlittle2(LP_METHOD_NM_TERM, strlen(LP_METHOD_NM_TERM), &h1, &h2);
  terminal_magic_num = h1 + h2;

  bj_hashlittle2(LP_METHOD_NM_ROUT, strlen(LP_METHOD_NM_ROUT), &h1, &h2);
  router_magic_num = h1 + h2;

456 457
}

458
static void local_configure(){
459 460 461
  anno_map = codes_mapping_get_lp_anno_map(LP_CONFIG_NM_TERM);
  assert(anno_map);
  num_params = anno_map->num_annos + (anno_map->has_unanno_lp > 0);
462
  all_params = (local_param *)malloc(num_params * sizeof(*all_params));
463 464 465

  for (int i = 0; i < anno_map->num_annos; i++){
    const char * anno = anno_map->annotations[i].ptr;
466
    local_read_config(anno, &all_params[i]);
467 468
  }
  if (anno_map->has_unanno_lp > 0){
469
    local_read_config(NULL, &all_params[anno_map->num_annos]);
470 471 472
  }
}

473 474
/* report statistics like average and maximum packet latency, average number of hops traversed */
static void local_report_stats()
475 476 477 478 479 480 481
{
  long long avg_hops, total_finished_packets, total_finished_chunks;
  long long total_finished_msgs, final_msg_sz;
  tw_stime avg_time, max_time;
  int total_minimal_packets, total_nonmin_packets;
  long total_gen, total_fin;

482
  MPI_Reduce( &total_hops, &avg_hops, 1, MPI_LONG_LONG, MPI_SUM, 0,
483
      MPI_COMM_CODES);
484
  MPI_Reduce( &N_finished_packets, &total_finished_packets, 1, MPI_LONG_LONG,
485
      MPI_SUM, 0, MPI_COMM_CODES);
486
  MPI_Reduce( &N_finished_msgs, &total_finished_msgs, 1, MPI_LONG_LONG, MPI_SUM,
487
      0, MPI_COMM_CODES);
488
  MPI_Reduce( &N_finished_chunks, &total_finished_chunks, 1, MPI_LONG_LONG,
489
      MPI_SUM, 0, MPI_COMM_CODES);
490
  MPI_Reduce( &total_msg_sz, &final_msg_sz, 1, MPI_LONG_LONG, MPI_SUM, 0,
491
      MPI_COMM_CODES);
492
  MPI_Reduce( &local_total_time, &avg_time, 1,MPI_DOUBLE, MPI_SUM, 0,
493
      MPI_COMM_CODES);
494
  MPI_Reduce( &local_max_latency, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0,
495
      MPI_COMM_CODES);
496

497 498
  MPI_Reduce( &packet_gen, &total_gen, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
  MPI_Reduce( &packet_fin, &total_fin, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
499 500 501

  /* print statistics */
  if(!g_tw_mynode)
502
  {
503 504
    printf(" Average number of hops traversed %f average chunk latency %lf us "
      "maximum chunk latency %lf us avg message size %lf bytes finished "
505 506 507 508
      "messages %lld finished chunks %lld \n",
      (float)avg_hops/total_finished_chunks,
      avg_time/(total_finished_chunks*1000), max_time/1000,
      (float)final_msg_sz/total_finished_msgs, total_finished_msgs,
509 510 511 512 513 514
      total_finished_chunks);
    printf("\n Total packets generated %ld finished %ld \n", total_gen, total_fin);
  }
  return;
}

515 516
/* initialize a compute node terminal */
static void terminal_init( terminal_state * s, tw_lp * lp )
517
{
518 519 520
  int i;
  char anno[MAX_NAME_LENGTH];

521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542
  s->packet_gen = 0;
  s->packet_fin = 0;

  codes_mapping_get_lp_info(lp->gid, lp_group_name, &mapping_grp_id, NULL,
      &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);

  if (anno[0] == '\0') {
    s->anno = NULL;
    s->params = &all_params[num_params-1];
  } else {
    s->anno = strdup(anno);
    int id = configuration_get_annotation_index(anno, anno_map);
    s->params = &all_params[id];
  }

  int num_lps = codes_mapping_get_lp_count(lp_group_name, 0, LP_CONFIG_NM_TERM,
      s->anno, 0);

  if(num_lps != s->params->total_terminals) {
    tw_error(TW_LOC, "Number of terminals LP does not match number of nodes\n");
  }

543
  s->terminal_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
544 545

  s->router_id = (int)s->terminal_id / s->params->num_cn;
546
  s->router_gid = codes_mapping_get_lpid_from_relative(s->router_id, NULL,
547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570
      LP_CONFIG_NM_ROUT, s->anno, 1);

  s->terminal_available_time = 0.0;
  s->packet_counter = 0;
  s->finished_msgs = 0;
  s->finished_chunks = 0;
  s->finished_packets = 0;
  s->total_time = 0.0;
  s->total_msg_size = 0;

  s->last_buf_full = 0.0;
  s->busy_time = 0.0;

  s->fwd_events = 0;
  s->rev_events = 0;

  rc_stack_create(&s->st);
  s->vc_occupancy = (int **)malloc(sizeof(int*));

  s->vc_occupancy[0] = (int*)malloc(s->params->num_vcs * sizeof(int));
  for(i = 0; i < s->params->num_vcs; i++ ) {
    s->vc_occupancy[0][i] = 0;
  }

571
  s->rank_tbl = qhash_init(mn_rank_hash_compare, mn_hash_func, HASH_TABLE_SIZE);
572 573 574 575 576 577

  if(!s->rank_tbl)
    tw_error(TW_LOC, "\n Hash table not initialized! ");

  s->terminal_msgs = (message_list ***)malloc(sizeof(message_list**));
  s->terminal_msgs_tail = (message_list ***)malloc(sizeof(message_list**));
578
  s->terminal_msgs[0] =
579
    (message_list **)malloc(s->params->num_vcs * sizeof(message_list*));
580
  s->terminal_msgs_tail[0] =
581 582 583 584 585 586 587 588 589 590 591 592
    (message_list **)malloc(s->params->num_vcs * sizeof(message_list*));
  for(i = 0; i < s->params->num_vcs; i++ ) {
    s->terminal_msgs[0][i] = NULL;
    s->terminal_msgs_tail[0][i] = NULL;
  }
  s->terminal_length = 0;
  s->in_send_loop = 0;
  s->issueIdle = 0;

  return;
}

593 594 595 596
//CHANGE: fill this function to set up router's connectivity and network
//specific data structures
static void create_router_connections(router_state * r, tw_lp * lp) {
  local_param *p = (local_param *)r->params;
597 598 599

  r->dim_position = (int *)malloc(p->n_dims * sizeof(int));

600
  to_dim_id(r->router_id, r->params->n_dims, r->params->dim_length,
601 602 603 604 605 606 607 608 609 610
      r->dim_position);

  //set up connections
  int curr_con = 0;
  int first = r->router_id * p->num_cn;
  for(; curr_con < p->num_cn; curr_con++) {
    r->link_connections[curr_con] = codes_mapping_get_lpid_from_relative(
      first, NULL, LP_CONFIG_NM_TERM, r->anno, 1);
    first++;
  }
611

612 613 614 615 616 617 618 619
  int temp_dim_pos[p->n_dims];
  for(int i = 0; i < p->n_dims; i++)
    temp_dim_pos[i] = r->dim_position[i];

  for(int curr_dim = 0; curr_dim < p->n_dims; curr_dim++) {
    curr_con = p->offset_per_dim[curr_dim];
    for(int loc = 0; loc < p->dim_length[curr_dim]; loc++) {
      if(loc != r->dim_position[curr_dim]) {
620
        if(loc == r->dim_position[curr_dim] - 1 ||
621
           loc == r->dim_position[curr_dim] + 1 ||
622 623 624
           (loc < r->dim_position[curr_dim] &&
           ((r->dim_position[curr_dim] - 1 - loc) % p->gap == 0)) ||
           (loc > r->dim_position[curr_dim] &&
625 626 627 628 629 630 631 632 633 634 635
           ((loc - r->dim_position[curr_dim] - 1) % p->gap == 0))) {
          temp_dim_pos[curr_dim] = loc;
          int neigh_id = to_flat_id(p->n_dims, p->dim_length, temp_dim_pos);
          temp_dim_pos[curr_dim] = r->dim_position[curr_dim];
          r->link_connections[curr_con] = codes_mapping_get_lpid_from_relative(
              neigh_id, NULL, LP_CONFIG_NM_ROUT, r->anno, 1);
          curr_con++;
        }
      }
    }
  }
636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657
}

static void router_setup(router_state * r, tw_lp * lp)
{
  char anno[MAX_NAME_LENGTH];
  codes_mapping_get_lp_info(lp->gid, lp_group_name, &mapping_grp_id, NULL,
      &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);

  if (anno[0] == '\0'){
    r->anno = NULL;
    r->params = &all_params[num_params-1];
  } else{
    r->anno = strdup(anno);
    int id = configuration_get_annotation_index(anno, anno_map);
    r->params = &all_params[id];
  }

  local_param *p = (local_param *)r->params;

  r->link_connections = (tw_lpid *)malloc(p->radix * sizeof(tw_lpid));

  r->router_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
658 659 660 661 662 663 664 665 666 667

  r->fwd_events = 0;
  r->rev_events = 0;

  r->next_output_available_time = (tw_stime*)malloc(p->radix * sizeof(tw_stime));
  r->link_traffic = (int64_t*)malloc(p->radix * sizeof(int64_t));
  r->link_traffic_sample = (int64_t*)malloc(p->radix * sizeof(int64_t));

  r->vc_occupancy = (int**)malloc(p->radix * sizeof(int*));
  r->in_send_loop = (int*)malloc(p->radix * sizeof(int));
668
  r->pending_msgs =
669
    (message_list ***)malloc(p->radix * sizeof(message_list**));
670
  r->pending_msgs_tail =
671
    (message_list ***)malloc(p->radix * sizeof(message_list**));
672
  r->queued_msgs =
673
    (message_list ***)malloc(p->radix * sizeof(message_list**));
674
  r->queued_msgs_tail =
675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690
    (message_list ***)malloc(p->radix * sizeof(message_list**));
  r->queued_count = (int*)malloc(p->radix * sizeof(int));
  r->last_buf_full = (tw_stime*)malloc(p->radix * sizeof(tw_stime));
  r->busy_time = (tw_stime*)malloc(p->radix * sizeof(tw_stime));
  r->busy_time_sample = (tw_stime*)malloc(p->radix * sizeof(tw_stime));

  rc_stack_create(&r->st);
  for(int i = 0; i < p->radix; i++)
  {
    // Set credit & router occupancy
    r->last_buf_full[i] = 0.0;
    r->busy_time[i] = 0.0;
    r->busy_time_sample[i] = 0.0;
    r->next_output_available_time[i] = 0;
    r->link_traffic[i] = 0;
    r->link_traffic_sample[i] = 0;
691
    r->queued_count[i] = 0;
692 693
    r->in_send_loop[i] = 0;
    r->vc_occupancy[i] = (int *)malloc(p->num_vcs * sizeof(int));
694
    r->pending_msgs[i] = (message_list **)malloc(p->num_vcs *
695
        sizeof(message_list*));
696
    r->pending_msgs_tail[i] = (message_list **)malloc(p->num_vcs *
697
        sizeof(message_list*));
698
    r->queued_msgs[i] = (message_list **)malloc(p->num_vcs *
699
        sizeof(message_list*));
700
    r->queued_msgs_tail[i] = (message_list **)malloc(p->num_vcs *
701 702 703 704 705 706 707 708 709 710
        sizeof(message_list*));
    for(int j = 0; j < p->num_vcs; j++) {
      r->vc_occupancy[i][j] = 0;
      r->pending_msgs[i][j] = NULL;
      r->pending_msgs_tail[i][j] = NULL;
      r->queued_msgs[i][j] = NULL;
      r->queued_msgs_tail[i][j] = NULL;
    }
  }

711 712 713
  create_router_connections(r, lp);
  return;
}
714

715 716
/* packet event , generates a packet on the compute node */
static tw_stime local_packet_event(
717 718 719 720 721 722 723 724 725 726 727 728 729 730
    model_net_request const * req,
    uint64_t message_offset,
    uint64_t packet_size,
    tw_stime offset,
    mn_sched_params const * sched_params,
    void const * remote_event,
    void const * self_event,
    tw_lp *sender,
    int is_last_pckt)
{
  (void)message_offset;
  (void)sched_params;
  tw_event * e_new;
  tw_stime xfer_to_nic_time;
731
  LOCAL_MSG_STRUCT * msg;
732 733
  char* tmp_ptr;

734
  xfer_to_nic_time = codes_local_latency(sender);
735
  e_new = model_net_method_event_new(sender->gid, xfer_to_nic_time+offset,
736
      sender, LOCAL_NETWORK_NAME, (void**)&msg, (void**)&tmp_ptr);
737 738 739 740 741 742 743 744 745 746 747 748 749 750 751
  strcpy(msg->category, req->category);
  msg->final_dest_gid = req->final_dest_lp;
  msg->total_size = req->msg_size;
  msg->sender_lp = req->src_lp;
  msg->sender_mn_lp = sender->gid;
  msg->packet_size = packet_size;
  msg->travel_start_time = tw_now(sender);
  msg->remote_event_size_bytes = 0;
  msg->local_event_size_bytes = 0;
  msg->type = T_GENERATE;
  msg->dest_terminal_id = req->dest_mn_lp;
  msg->dest_terminal = codes_mapping_get_lp_relative_id(msg->dest_terminal_id, 0, 0);
  msg->message_id = req->msg_id;
  msg->is_pull = req->is_pull;
  msg->pull_size = req->pull_size;
752
  msg->magic = terminal_magic_num;
753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773
  msg->msg_start_time = req->msg_start_time;

  if(is_last_pckt) /* Its the last packet so pass in remote and local event information*/
  {
    if(req->remote_event_size > 0)
    {
      msg->remote_event_size_bytes = req->remote_event_size;
      memcpy(tmp_ptr, remote_event, req->remote_event_size);
      tmp_ptr += req->remote_event_size;
    }
    if(req->self_event_size > 0)
    {
      msg->local_event_size_bytes = req->self_event_size;
      memcpy(tmp_ptr, self_event, req->self_event_size);
      tmp_ptr += req->self_event_size;
    }
  }
  tw_event_send(e_new);
  return xfer_to_nic_time;
}

774 775
/* packet event reverse handler */
static void local_packet_event_rc(tw_lp *sender)
776 777 778 779 780
{
  codes_local_latency_reverse(sender);
  return;
}

781 782
/* generates packet at the current compute node */
static void packet_generate(terminal_state * s, tw_bf * bf, LOCAL_MSG_STRUCT * msg,
783 784 785 786 787 788 789
    tw_lp * lp) {
  packet_gen++;
  s->packet_gen++;

  tw_stime ts, nic_ts;

  assert(lp->gid != msg->dest_terminal_id);
790
  const local_param *p = s->params;
791 792 793

  int total_event_size;
  uint64_t num_chunks = msg->packet_size / p->chunk_size;
794
  if (msg->packet_size % s->params->chunk_size)
795 796 797 798 799
    num_chunks++;

  if(!num_chunks)
    num_chunks = 1;

800
  nic_ts = g_tw_lookahead + (msg->packet_size * s->params->cn_delay) +
801 802 803 804 805 806 807 808 809
    tw_rand_unif(lp->rng);

  msg->packet_ID = lp->gid + g_tw_nlp * s->packet_counter;
  for(int i = 0; i < p->n_dims; i++) {
    msg->hops[i] = 0;
  }
  msg->my_N_hop = 0;

  /* TODO: how do we get a better vc selection mechanism */
810
  //CHANGE: use only 1 VC for NIC to router transfer; VC selection mechanism on the NIC can be added here
811 812 813 814 815 816 817 818 819 820 821 822 823 824
  int use_vc = 0;
  msg->saved_channel = use_vc;

  for(uint64_t i = 0; i < num_chunks; i++)
  {
    message_list *cur_chunk = (message_list*)malloc(
        sizeof(message_list));
    init_message_list(cur_chunk, msg);

    if(msg->remote_event_size_bytes + msg->local_event_size_bytes > 0) {
      cur_chunk->event_data = (char*)malloc(
          msg->remote_event_size_bytes + msg->local_event_size_bytes);
    }

825
    void * m_data_src = model_net_method_get_edata(LOCAL_NETWORK_NAME, msg);
826 827 828
    if (msg->remote_event_size_bytes){
      memcpy(cur_chunk->event_data, m_data_src, msg->remote_event_size_bytes);
    }
829
    if (msg->local_event_size_bytes){
830
      m_data_src = (char*)m_data_src + msg->remote_event_size_bytes;
831
      memcpy((char*)cur_chunk->event_data + msg->remote_event_size_bytes,
832 833 834
          m_data_src, msg->local_event_size_bytes);
    }

835 836
    cur_chunk->LOCAL_MSG_NAME_FROM_UNION.chunk_id = i;
    cur_chunk->port = 0; cur_chunk->index = use_vc;
837 838 839 840 841
    append_to_message_list(s->terminal_msgs[0], s->terminal_msgs_tail[0],
        use_vc, cur_chunk);
    s->terminal_length += s->params->chunk_size;
  }

842
  if(s->terminal_length < s->params->num_vcs * s->params->cn_vc_size) {
843 844 845 846 847 848 849 850 851 852 853
    model_net_method_idle_event(nic_ts, 0, lp);
  } else {
    bf->c11 = 1;
    s->issueIdle = 1;
    msg->saved_busy_time = s->last_buf_full;
    s->last_buf_full = tw_now(lp);
  }

  if(s->in_send_loop == 0) {
    bf->c5 = 1;
    ts = codes_local_latency(lp);
854 855
    LOCAL_MSG_STRUCT *m;
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, LOCAL_NETWORK_NAME,
856 857 858 859 860 861 862
        (void**)&m, NULL);
    m->type = T_SEND;
    m->magic = terminal_magic_num;
    s->in_send_loop = 1;
    tw_event_send(e);
  }

863
  total_event_size = model_net_get_msg_sz(LOCAL_NETWORK_NAME) +
864 865
    msg->remote_event_size_bytes + msg->local_event_size_bytes;
  mn_stats* stat;
866
  stat = model_net_find_stats(msg->category, s->local_stats_array);
867 868 869 870 871 872 873 874 875
  stat->send_count++;
  stat->send_bytes += msg->packet_size;
  stat->send_time += p->cn_delay * msg->packet_size;
  if(stat->max_event_size < total_event_size)
    stat->max_event_size = total_event_size;

  return;
}

876
static void packet_generate_rc(terminal_state * s, tw_bf * bf, LOCAL_MSG_STRUCT * msg, tw_lp * lp)
877 878 879 880 881 882 883 884 885 886 887 888 889 890 891
{
  s->packet_gen--;
  packet_gen--;

  tw_rand_reverse_unif(lp->rng);

  int num_chunks = msg->packet_size/s->params->chunk_size;
  if(msg->packet_size % s->params->chunk_size)
    num_chunks++;

  if(!num_chunks)
    num_chunks = 1;

  int i;
  for(i = 0; i < num_chunks; i++) {
892
    delete_message_list(return_tail(s->terminal_msgs[0],
893 894 895 896 897 898 899 900 901 902 903 904
          s->terminal_msgs_tail[0], msg->saved_channel));
    s->terminal_length -= s->params->chunk_size;
  }
  if(bf->c11) {
    s->issueIdle = 0;
    s->last_buf_full = msg->saved_busy_time;
  }
  if(bf->c5) {
    codes_local_latency_reverse(lp);
    s->in_send_loop = 0;
  }
  struct mn_stats* stat;
905
  stat = model_net_find_stats(msg->category, s->local_stats_array);
906 907 908 909 910 911
  stat->send_count--;
  stat->send_bytes -= msg->packet_size;
  stat->send_time -= s->params->cn_delay * msg->packet_size;
}


912 913
/* sends the packet from the current compute node to the attached router */
static void packet_send(terminal_state * s, tw_bf * bf, LOCAL_MSG_STRUCT * msg,
914 915 916 917
    tw_lp * lp) {

  tw_stime ts;
  tw_event *e;
918
  LOCAL_MSG_STRUCT *m;
919 920 921
  tw_lpid router_id;

  std::vector<VC_Entry> entries;
922

923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940
  for(int i = 0; i < s->params->num_vcs; i++) {
    if(s->terminal_msgs[0][i] != NULL &&
      s->vc_occupancy[0][i] + s->params->chunk_size <= s->params->cn_vc_size) {
      VC_Entry tmp;
      tmp.vc = i; tmp.entry = s->terminal_msgs[0][i];
      entries.push_back(tmp);
    }
  }

  if(entries.size() == 0) {
    bf->c1 = 1;
    s->in_send_loop = 0;

    msg->saved_busy_time = s->last_buf_full;
    s->last_buf_full = tw_now(lp);
    return;
  }

941
  //CHANGE: randomly pick the VC to send on (by default only VC 0 is filed); can be changed.
942 943 944 945 946
  int pick = tw_rand_integer(lp->rng, 0, entries.size() - 1);
  message_list* cur_entry = entries[pick].entry;
  int use_vc = entries[pick].vc;
  msg->saved_channel = use_vc;

947 948
  uint64_t num_chunks = cur_entry->LOCAL_MSG_NAME_FROM_UNION.packet_size/s->params->chunk_size;
  if(cur_entry->LOCAL_MSG_NAME_FROM_UNION.packet_size % s->params->chunk_size)
949 950 951 952 953 954
    num_chunks++;

  if(!num_chunks)
    num_chunks = 1;

  tw_stime delay;
955 956 957
  if((cur_entry->LOCAL_MSG_NAME_FROM_UNION.packet_size % s->params->chunk_size)
      && (cur_entry->LOCAL_MSG_NAME_FROM_UNION.chunk_id == num_chunks - 1))
    delay = (cur_entry->LOCAL_MSG_NAME_FROM_UNION.packet_size % s->params->chunk_size) *
958
      s->params->cn_delay;
959
  else
960 961 962 963 964 965
    delay = s->params->chunk_size * s->params->cn_delay;

  msg->saved_available_time = s->terminal_available_time;
  ts = g_tw_lookahead + delay + tw_rand_unif(lp->rng);
  s->terminal_available_time = maxd(s->terminal_available_time, tw_now(lp));
  s->terminal_available_time += ts;
966

967 968
  ts = s->terminal_available_time - tw_now(lp);
  void * remote_event;
969
  e = model_net_method_event_new(s->router_gid, ts, lp, LOCAL_NETWORK_ROUTER_NAME,
970
      (void**)&m, &remote_event);
971
  memcpy(m, &cur_entry->LOCAL_MSG_NAME_FROM_UNION, sizeof(LOCAL_MSG_STRUCT));
972 973 974 975 976 977 978 979 980 981 982 983 984
  if (m->remote_event_size_bytes){
    memcpy(remote_event, cur_entry->event_data, m->remote_event_size_bytes);
  }

  m->type = R_ARRIVE;
  m->src_terminal_id = lp->gid;
  m->vc_index = 0;
  m->output_chan = use_vc;
  m->last_hop = TERMINAL;
  m->magic = router_magic_num;
  m->local_event_size_bytes = 0;
  tw_event_send(e);

985 986
  if(cur_entry->LOCAL_MSG_NAME_FROM_UNION.chunk_id == num_chunks - 1 &&
      (cur_entry->LOCAL_MSG_NAME_FROM_UNION.local_event_size_bytes > 0)) {
987
    bf->c2 = 1;
988 989
    tw_stime local_ts = codes_local_latency(lp);
    tw_event *e_new = tw_event_new(cur_entry->LOCAL_MSG_NAME_FROM_UNION.sender_lp, local_ts, lp);
990
    void * m_new = tw_event_data(e_new);
991 992 993
    void *local_event = (char*)cur_entry->event_data +
      cur_entry->LOCAL_MSG_NAME_FROM_UNION.remote_event_size_bytes;
    memcpy(m_new, local_event, cur_entry->LOCAL_MSG_NAME_FROM_UNION.local_event_size_bytes);
994 995 996 997
    tw_event_send(e_new);
  }
  s->packet_counter++;
  s->vc_occupancy[0][use_vc] += s->params->chunk_size;
998
  cur_entry = return_head(s->terminal_msgs[0], s->terminal_msgs_tail[0], use_vc);
999 1000 1001
  rc_stack_push(lp, cur_entry, delete_message_list, s->st);
  s->terminal_length -= s->params->chunk_size;

1002
  LOCAL_MSG_STRUCT *m_new;
1003
  ts += tw_rand_unif(lp->rng);
1004
  e = model_net_method_event_new(lp->gid, ts, lp, LOCAL_NETWORK_NAME,
1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030
      (void**)&m_new, NULL);
  m_new->type = T_SEND;
  m_new->magic = terminal_magic_num;
  tw_event_send(e);

  if(s->issueIdle) {
    bf->c5 = 1;
    s->issueIdle = 0;
    ts += tw_rand_unif(lp->rng);
    model_net_method_idle_event(ts, 0, lp);

    if(s->last_buf_full > 0.0)
    {
      bf->c6 = 1;
      msg->saved_total_time = s->busy_time;
      msg->saved_busy_time = s->last_buf_full;
      msg->saved_sample_time = s->busy_time_sample;

      s->busy_time += (tw_now(lp) - s->last_buf_full);
      s->busy_time_sample += (tw_now(lp) - s->last_buf_full);
      s->last_buf_full = 0.0;
    }
  }
  return;
}

1031
static void packet_send_rc(terminal_state * s, tw_bf * bf, LOCAL_MSG_STRUCT * msg,
1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053
    tw_lp * lp)
{
  if(bf->c1) {
    s->in_send_loop = 1;
    s->last_buf_full = msg->saved_busy_time;
    return;
  }

  tw_rand_reverse_unif(lp->rng);
  tw_rand_reverse_unif(lp->rng);
  s->terminal_available_time = msg->saved_available_time;
  if(bf->c2) {
    codes_local_latency_reverse(lp);
  }

  int use_vc = msg->saved_channel;

  s->packet_counter--;
  s->vc_occupancy[0][use_vc] -= s->params->chunk_size;

  message_list* cur_entry = (message_list *)rc_stack_pop(s->st);
  cur_entry->port = 0; cur_entry->index = use_vc;
1054
  prepend_to_message_list(s->terminal_msgs[0], s->terminal_msgs_tail[0],
1055 1056
      use_vc, cur_entry);
  s->terminal_length += s->params->chunk_size;
1057

1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072
  tw_rand_reverse_unif(lp->rng);
  if(bf->c5)
  {
    tw_rand_reverse_unif(lp->rng);
    s->issueIdle = 1;
    if(bf->c6)
    {
      s->busy_time = msg->saved_total_time;
      s->last_buf_full = msg->saved_busy_time;
      s->busy_time_sample = msg->saved_sample_time;
    }
  }
  return;
}

1073
static void send_remote_event(terminal_state * s, LOCAL_MSG_STRUCT * msg,
1074 1075
  tw_lp * lp, tw_bf * bf, char * event_data, int remote_event_size)
{
1076 1077
  void * tmp_ptr = model_net_method_get_edata(LOCAL_NETWORK_NAME, msg);
  //tw_stime ts = g_tw_lookahead + s->params->cn_delay *
1078 1079 1080 1081 1082 1083 1084 1085 1086 1087
  //    msg->remote_event_size_bytes + tw_rand_unif(lp->rng);;
  tw_stime ts = g_tw_lookahead + tw_rand_unif(lp->rng);
  if (msg->is_pull){
    bf->c4 = 1;
    struct codes_mctx mc_dst =
      codes_mctx_set_global_direct(msg->sender_mn_lp);
    struct codes_mctx mc_src =
      codes_mctx_set_global_direct(lp->gid);
    int net_id = model_net_get_id(LP_METHOD_NM_TERM);

1088
    model_net_set_msg_param(MN_MSG_PARAM_START_TIME,
1089 1090 1091 1092 1093 1094 1095 1096 1097
        MN_MSG_PARAM_START_TIME_VAL, &(msg->msg_start_time));

    msg->event_rc = model_net_event_mctx(net_id, &mc_src, &mc_dst, msg->category,
        msg->sender_lp, msg->pull_size, ts,
        remote_event_size, tmp_ptr, 0, NULL, lp);
  } else {
    tw_event * e = tw_event_new(msg->final_dest_gid, ts, lp);
    void * m_remote = tw_event_data(e);
    memcpy(m_remote, event_data, remote_event_size);
1098
    tw_event_send(e);
1099 1100 1101 1102 1103
  }
  return;
}

/* packet arrives at the destination terminal */
1104
static void packet_arrive(terminal_state * s, tw_bf * bf, LOCAL_MSG_STRUCT * msg,
1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118
    tw_lp * lp) {

  assert(lp->gid == msg->dest_terminal_id);

  //total chunks expected in this message
  uint64_t total_chunks = msg->total_size / s->params->chunk_size;
  if(msg->total_size % s->params->chunk_size)
    total_chunks++;
  if(!total_chunks)
    total_chunks = 1;

  /* send credit back to router */
  tw_stime ts = g_tw_lookahead + s->params->credit_delay + tw_rand_unif(lp->rng);
  tw_event * buf_e;
1119
  LOCAL_MSG_STRUCT * buf_msg;
1120
  buf_e = model_net_method_event_new(msg->intm_lp_id, ts, lp,
1121
      LOCAL_NETWORK_ROUTER_NAME, (void**)&buf_msg, NULL);
1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137
  buf_msg->magic = router_magic_num;
  buf_msg->vc_index = msg->vc_index;
  buf_msg->output_chan = msg->output_chan;
  buf_msg->type = R_BUFFER;
  tw_event_send(buf_e);

  //save stats
  /* Total overall finished chunks in simulation */
  N_finished_chunks++;
  /* Finished chunks on a LP basis */
  s->finished_chunks++;
  /* Finished chunks per sample */
  s->fin_chunks_sample++;

  assert(lp->gid != msg->src_terminal_id);

1138
  // chunks part of this packet
1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156
  uint64_t num_chunks = msg->packet_size / s->params->chunk_size;
  if (msg->packet_size % s->params->chunk_size)
    num_chunks++;
  if(!num_chunks)
    num_chunks = 1;

  if(msg->chunk_id == num_chunks - 1)
  {
    bf->c31 = 1;
    s->packet_fin++;
    packet_fin++;
  }

  /* save the sample time */
  msg->saved_sample_time = s->fin_chunks_time;
  s->fin_chunks_time += (tw_now(lp) - msg->travel_start_time);
  /* save the total time per LP */
  msg->saved_avg_time = s->total_time;
1157 1158 1159
  s->total_time += (tw_now(lp) - msg->travel_start_time);
  msg->saved_total_time = local_total_time;
  local_total_time += tw_now( lp ) - msg->travel_start_time;
1160 1161 1162 1163
  total_hops += msg->my_N_hop;
  s->total_hops += msg->my_N_hop;
  s->fin_hops_sample += msg->my_N_hop;

1164
  mn_stats* stat = model_net_find_stats(msg->category, s->local_stats_array);
1165 1166 1167 1168 1169
  msg->saved_rcv_time = stat->recv_time;
  stat->recv_time += (tw_now(lp) - msg->travel_start_time);

  /* Now retreieve the number of chunks completed from the hash and update
   * them */
1170

1171
  struct mn_hash_key key;
1172
  key.message_id = msg->message_id;
1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211
  key.sender_id = msg->sender_lp;
  struct qhash_head *hash_link = NULL;
  struct mn_qhash_entry * tmp = NULL;
  hash_link = qhash_search(s->rank_tbl, &key);
  if(hash_link)
    tmp = qhash_entry(hash_link, struct mn_qhash_entry, hash_link);

  /* If an entry does not exist then create one */
  if(!tmp)
  {
    bf->c5 = 1;
    struct mn_qhash_entry * d_entry = (struct mn_qhash_entry *)
        malloc(sizeof(struct mn_qhash_entry));
    d_entry->num_chunks = 0;
    d_entry->key = key;
    d_entry->remote_event_data = NULL;
    d_entry->remote_event_size = 0;
    qhash_add(s->rank_tbl, &key, &(d_entry->hash_link));

    hash_link = &(d_entry->hash_link);
    tmp = d_entry;
  }

  assert(tmp);
  tmp->num_chunks++;

  if(msg->chunk_id == num_chunks - 1)
  {
    bf->c1 = 1;
    stat->recv_count++;
    stat->recv_bytes += msg->packet_size;
    N_finished_packets++;
    s->finished_packets++;
  }

  /* if its the last chunk of the packet then handle the remote event data */
  if(msg->remote_event_size_bytes > 0 && !tmp->remote_event_data)
  {
    /* Retreive the remote event entry */
1212
    void *m_data_src = model_net_method_get_edata(LOCAL_NETWORK_NAME, msg);
1213 1214
    tmp->remote_event_data = (char*)malloc(msg->remote_event_size_bytes);
    assert(tmp->remote_event_data);
1215
    tmp->remote_event_size = msg->remote_event_size_bytes;
1216 1217 1218
    memcpy(tmp->remote_event_data, m_data_src, msg->remote_event_size_bytes);
  }

1219
  if (local_max_latency < tw_now( lp ) - msg->travel_start_time) {
1220
    bf->c3 = 1;
1221 1222
    msg->saved_available_time = local_max_latency;
    local_max_latency = tw_now( lp ) - msg->travel_start_time;
1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236
  }

  if(tmp->num_chunks >= total_chunks)
  {
    bf->c7 = 1;

    N_finished_msgs++;
    total_msg_sz += msg->total_size;
    s->total_msg_size += msg->total_size;
    s->data_size_sample += msg->total_size;
    s->finished_msgs++;

    if(tmp->remote_event_data && tmp->remote_event_size > 0) {
      bf->c8 = 1;
1237
      send_remote_event(s, msg, lp, bf, tmp->remote_event_data,
1238 1239 1240 1241 1242 1243 1244 1245 1246 1247
          tmp->remote_event_size);
    }

    /* Remove the hash entry */
    qhash_del(hash_link);
    rc_stack_push(lp, tmp, free_tmp, s->st);
  }
  return;
}

1248
static void packet_arrive_rc(terminal_state * s, tw_bf * bf, LOCAL_MSG_STRUCT * msg, tw_lp * lp)
1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261
{
  tw_rand_reverse_unif(lp->rng);
  N_finished_chunks--;
  s->finished_chunks--;
  s->fin_chunks_sample--;
  if(bf->c31)
  {
    s->packet_fin--;
    packet_fin--;
  }

  s->fin_chunks_time = msg->saved_sample_time;
  s->total_time = msg->saved_avg_time;
1262
  local_total_time  = msg->saved_total_time;
1263 1264 1265 1266 1267
  total_hops -= msg->my_N_hop;
  s->total_hops -= msg->my_N_hop;
  s->fin_hops_sample -= msg->my_N_hop;

  mn_stats* stat;
1268
  stat = model_net_find_stats(msg->category, s->local_stats_array);
1269 1270 1271
  stat->recv_time = msg->saved_rcv_time;

  struct qhash_head * hash_link = NULL;
1272
  struct mn_qhash_entry * tmp = NULL;
1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290

  struct mn_hash_key key;
  key.message_id = msg->message_id;
  key.sender_id = msg->sender_lp;

  hash_link = qhash_search(s->rank_tbl, &key);
  if(hash_link)
    tmp = qhash_entry(hash_link, struct mn_qhash_entry, hash_link);

  if(bf->c1)
  {
    stat->recv_count--;
    stat->recv_bytes -= msg->packet_size;
    N_finished_packets--;
    s->finished_packets--;
  }

  if(bf->c3)
1291
    local_max_latency = msg->saved_available_time;
1292 1293 1294 1295 1296 1297 1298 1299

  if(bf->c7)
  {
    N_finished_msgs--;
    total_msg_sz -= msg->total_size;
    s->total_msg_size -= msg->total_size;
    s->data_size_sample -= msg->total_size;
    s->finished_msgs--;
1300 1301

    if(bf->c8)
1302 1303 1304 1305 1306 1307 1308
      tw_rand_reverse_unif(lp->rng);

    struct mn_qhash_entry * d_entry_pop = (struct mn_qhash_entry * )
        rc_stack_pop(s->st);
    qhash_add(s->rank_tbl, &key, &(d_entry_pop->hash_link));

    hash_link = &(d_entry_pop->hash_link);
1309
    tmp = d_entry_pop;
1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320

    if(bf->c4)
      model_net_event_rc2(lp, &msg->event_rc);
  }

  assert(tmp);
  tmp->num_chunks--;

  if(bf->c5)
  {
    qhash_del(hash_link);
1321
    free_tmp(tmp);
1322 1323 1324 1325 1326
  }
  return;
}

/* update the compute node-router channel buffer */
1327 1328 1329 1330
static void
terminal_buf_update(terminal_state * s,
    tw_bf * bf,
    LOCAL_MSG_STRUCT * msg,
1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343
    tw_lp * lp)
{
  s->vc_occupancy[0][msg->output_chan] -= s->params->chunk_size;

  if(s->in_send_loop == 0) {
    int do_send = 0;
    for(int i = 0; i < s->params->num_vcs; i++) {
      if(s->terminal_msgs[0][i] != NULL) {
        do_send = 1;
        break;
      }
    }
    if(do_send) {
1344
      LOCAL_MSG_STRUCT *m;
1345 1346
      bf->c1 = 1;
      tw_stime ts = codes_local_latency(lp);
1347
      tw_event* e = model_net_method_event_new(lp->gid, ts, lp, LOCAL_NETWORK_NAME,
1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358
          (void**)&m, NULL);
      m->type = T_SEND;
      m->magic = terminal_magic_num;
      s->in_send_loop = 1;
      tw_event_send(e);
    }
  }
  return;
}

static void terminal_buf_update_rc(terminal_state * s,
1359 1360
    tw_bf * bf,
    LOCAL_MSG_STRUCT * msg,
1361 1362 1363 1364 1365 1366 1367 1368 1369 1370
    tw_lp * lp)
{
  s->vc_occupancy[0][msg->output_chan] += s->params->chunk_size;
  if(bf->c1) {
    codes_local_latency_reverse(lp);
    s->in_send_loop = 0;
  }
  return;
}

1371 1372 1373
static void terminal_event( terminal_state * s,
    tw_bf * bf,
    LOCAL_MSG_STRUCT * msg,
1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403
    tw_lp * lp )
{
  s->fwd_events++;
  assert(msg->magic == terminal_magic_num);
  rc_stack_gc(lp, s->st);
  switch(msg->type)
  {
    case T_GENERATE:
      packet_generate(s,bf,msg,lp);
      break;

    case T_ARRIVE:
      packet_arrive(s,bf,msg,lp);
      break;

    case T_SEND:
      packet_send(s,bf,msg,lp);
      break;

    case T_BUFFER:
      terminal_buf_update(s, bf, msg, lp);
      break;

    default:
      printf("\n LP %d Terminal message type not supported %d ", (int)lp->gid, msg->type);
      tw_error(TW_LOC, "Msg type not supported");
  }
}

/* Reverse computation handler for a terminal event */
1404 1405 1406 1407
static void terminal_event_rc (terminal_state * s,
    tw_bf * bf,
    LOCAL_MSG_STRUCT * msg,
    tw_lp * lp)
1408 1409 1410 1411 1412
{
  s->rev_events++;
  switch(msg->type)
  {
    case T_GENERATE:
1413
      packet_generate_rc(s, bf, msg, lp);
1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424
      break;

    case T_SEND:
      packet_send_rc(s, bf, msg, lp);
      break;

    case T_ARRIVE:
      packet_arrive_rc(s, bf, msg, lp);
      break;

    case T_BUFFER:
1425
      terminal_buf_update_rc(s, bf, msg, lp);
1426 1427 1428 1429 1430
      break;

  }
}

1431
static void
1432 1433
terminal_final( terminal_state * s, tw_lp * lp )
{
1434
  model_net_print_stats(lp->gid, s->local_stats_array);
1435 1436 1437 1438 1439

  int written = 0;
  if(!s->terminal_id)
    written = sprintf(s->output_buf, "# Format <LP id> <Terminal ID> <Total Data Size> <Avg packet latency> <# Flits/Packets finished> <Avg hops> <Busy Time>");

1440
  written += sprintf(s->output_buf + written, "\n %llu %u %llu %lf %ld %lf %lf",
1441
      LLU(lp->gid), s->terminal_id, s->total_msg_size, s->total_time,
1442 1443 1444
      s->finished_packets, (double)s->total_hops/s->finished_chunks,
      s->busy_time);

1445
  lp_io_write(lp->gid, (char*)"msg-stats", written, s->output_buf);
1446 1447

  for(int i = 0; i < s->params->num_vcs; i++) {
1448
    if(s->terminal_msgs[0][i] != NULL)
1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469
      printf("[%llu] leftover terminal messages \n", LLU(lp->gid));
  }

  qhash_finalize(s->rank_tbl);
  rc_stack_destroy(s->st);
  free(s->vc_occupancy[0]);
  free(s->vc_occupancy);
  free(s->terminal_msgs[0]);
  free(s->terminal_msgs_tail[0]);
  free(s->terminal_msgs);
  free(s->terminal_msgs_tail);
}

struct NextHop {
  int dim, port;
  NextHop(int _dim, int _port) {
    dim = _dim;
    port = _port;
  }
};

1470 1471
//CHANGE: implement the get next stop function for the network - some
//implementations may choose to change the function prototype
1472 1473
/* get the next stop for the current packet */
static void
1474 1475
get_next_stop(router_state * s,
    LOCAL_MSG_STRUCT * msg,
1476 1477 1478 1479 1480
    tw_bf *bf,
    int *port,
    int *vc,
    int *src_dim,
    int *dst_dim,
1481
    int *static_port)
1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497
{
  *static_port = -1;
  *src_dim = -2;
  if(msg->last_hop == TERMINAL) {
    *src_dim = -1;
  } else {
    for(int i = s->params->n_dims - 1; i >= 0; i--) {
      if(msg->vc_index >= s->params->offset_per_dim[i]) {
        *src_dim = i;
        break;
      }
    }
  }
  assert(*src_dim > -2);

  int dest[s->params->n_dims];
1498
  to_dim_id(msg->dest_terminal/s->params->num_cn, s->params->n_dims,
1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545
    s->params->dim_length, dest);

  std::vector<NextHop> port_options;
  bool at_dest = true;
  int first_dim = -1;
  for(int i = 0; i < s->params->n_dims; i++)
  {
    if(s->dim_position[i] != dest[i]) {
      at_dest = false;
      *port = s->params->offset_per_dim[i];
      if(first_dim  == -1) {
        first_dim = i;
      }

      int first_dim_con;
      if(s->dim_position[i] == 0) {
        first_dim_con = -1;
      } else {
        first_dim_con = (s->dim_position[i] - 1) % s->params->gap;
      }

      if(dest[i] == s->dim_position[i] - 1) {
        *port += (s->dim_position[i] - 1 - first_dim_con) / s->params->gap;
        port_options.push_back(NextHop(i, *port));
      } else if(dest[i] == s->dim_position[i] + 1) {
        *port += 1 + (s->dim_position[i] - 1 - first_dim_con) / s->params->gap;
        *port -= (s->dim_position[i] == 0 ? 1 : 0);
        port_options.push_back(NextHop(i, *port));
      } else {
        if(dest[i] < s->dim_position[i]) {
          if((s->dim_position[i] - 1 - dest[i]) % s->params->gap == 0) {
            *port += (dest[i] - first_dim_con) / s->params->gap;
            port_options.push_back(NextHop(i, *port));
          } else {
            int long_hop;
            if(dest[i] < first_dim_con) {
              long_hop = *port;
            } else {
              long_hop = *port + (dest[i] - first_dim_con) / s->params->gap + 1;
            }
            //int long_hop = *port + (dest[i] - first_dim_con) / s->params->gap + 1;
            int short_hop = *port + (s->dim_position[i] - 1 - first_dim_con) / s->params->gap;
            port_options.push_back(NextHop(i, long_hop));
            port_options.push_back(NextHop(i, short_hop));
          }
        } else if(dest[i] > s->dim_position[i]) {
          if((dest[i] - s->dim_position[i] - 1) % s->params->gap == 0) {
1546
            *port += (s->dim_position[i] - 1 - first_dim_con) / s->params->gap
1547 1548 1549 1550
              + 1 + (dest[i] - s->dim_position[i] - 1) / s->params->gap;
            *port -= (s->dim_position[i] == 0 ? 1 : 0);
            port_options.push_back(NextHop(i, *port));
          } else {
1551
            int long_hop = *port + (s->dim_position[i] - 1 - first_dim_con) / s->params->gap
1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582
              + 1 + (dest[i] - s->dim_position[i] - 1) / s->params->gap;
            int short_hop = *port + 1 + (s->dim_position[i] - 1 - first_dim_con) / s->params->gap;
            short_hop -= (s->dim_position[i] == 0 ? 1 : 0);
            long_hop -= (s->dim_position[i] == 0 ? 1 : 0);
            port_options.push_back(NextHop(i, long_hop));
            port_options.push_back(NextHop(i, short_hop));
          }
        } else {
          tw_error(TW_LOC, "Impossible condition in get_next_stop");
        }
      }
      if(s->params->routing == STATIC) {
        break;
      }
    }
  }

  int try_vcs = s->params->num_vcs;
  if(at_dest) {
    *port = (msg->dest_terminal % s->params->num_cn);
    port_options.push_back(NextHop(-1, *port));
    try_vcs = 1;
    assert(port_options.size() == 1);
  }
  if(s->params->routing == STATIC || at_dest) {
    *vc = 0;
    *port = port_options[0].port;
    *dst_dim = port_options[0].dim;
    int vc_s = s->vc_occupancy[*port][*vc] + s->queued_count[*port];
    for(int i = 0; i < port_options.size(); i++) {
      for(int j = 0; j < try_vcs; j++) {
1583
        if(s->vc_occupancy[port_options[i].port][j] +
1584
           s->queued_count[port_options[i].port] < vc_s) {
1585
          vc_s = s->vc_occupancy[port_options[i].port][j] +
1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599
                 s->queued_count[port_options[i].port];
          *port = port_options[i].port;
          *dst_dim = port_options[i].dim;
          *vc = j;
        }
      }
    }
  } else {
    *vc = 1; //dynamic VC
    *port = port_options[0].port;
    *dst_dim = port_options[0].dim;
    int vc_s = s->vc_occupancy[*port][*vc] + s->queued_count[*port];
    for(int i = 0; i < port_options.size(); i++) {
      for(int j = 1; j < try_vcs; j++) {
1600
        if(s->vc_occupancy[port_options[i].port][j] +
1601
           s->queued_count[port_options[i].port] < vc_s) {
1602
          vc_s = s->vc_occupancy[port_options[i].port][j] +
1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615
                 s->queued_count[port_options[i].port];
          *port = port_options[i].port;
          *dst_dim = port_options[i].dim;
          *vc = j;
        }
      }
    }
    //if going to queued list, find the first dim port also
    if((s->vc_occupancy[*port][*vc] + s->params->chunk_size > s->params->vc_size)) {
      int start_vc = 1;
      int end_vc = s->params->num_vcs;
      *static_port = port_options[0].port;
      int vc_s = s->vc_occupancy[*static_port][start_vc] + s->queued_count[*static_port];
1616
      for(int i = 0; (i < port_options.size()) &&
1617 1618
                     (port_options[i].dim == first_dim); i++) {
        for(int j = start_vc; j < end_vc; j++) {
1619
          if(s->vc_occupancy[port_options[i].port][j] +
1620
              s->queued_count[port_options[i].port] < vc_s) {
1621
            vc_s = s->vc_occupancy[port_options[i].port][j] +
1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635
              s->queued_count[port_options[i].port];
            *static_port = port_options[i].port;
          }
        }
      }
    }
  }
#if DEBUG
  printf("[%lld-%d] Me: %d %d %d: Dest %d %d %d Found %d %d\n", msg->packet_ID, s->params->n_dims,
      s->dim_position[0], s->dim_position[1], s->dim_position[2],
      dest[0], dest[1], dest[2], *dst_dim, *port);
#endif
}

1636
//CHANGE: reverse computation for get_next_stop
1637 1638
/* get the next stop for the current packet */
static void
1639 1640
get_next_stop_rc(router_state * s,
    LOCAL_MSG_STRUCT * msg,
1641 1642 1643 1644
    tw_bf *bf)
{
}

1645
/*When a packet is sent from the current router and a buffer slot becomes
1646
 * available, a credit is sent back to schedule another packet event*/
1647
static void router_credit_send(router_state * s, LOCAL_MSG_STRUCT * msg,
1648 1649 1650
    tw_lp * lp) {
  tw_event * buf_e;
  tw_stime ts;
1651
  LOCAL_MSG_STRUCT * buf_msg;
1652 1653 1654 1655

  int dest = 0,  type = R_BUFFER;
  int is_terminal = 0;

1656
  const local_param *p = s->params;
1657 1658 1659 1660 1661 1662 1663 1664

  // Notify sender terminal