fattree.c 59.6 KB
Newer Older
1 2 3 4 5 6 7 8 9
#include <ross.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"
#include "codes/net/fattree.h"
#include "sys/file.h"
10
#include "codes/quickhash.h"
11 12 13 14 15
//#include "codes/map_messages.h"

#define CREDIT_SIZE 8
#define MEAN_PROCESS 1.0

16 17
#define FTREE_HASH_TABLE_SIZE 262144

18
// debugging parameters
19
#define TRACK_PKT -1
20 21 22 23 24 25 26
#define FATTREE_HELLO 0
#define FATTREE_CONNECTIONS 0
#define FATTREE_MSG 0

#define LP_CONFIG_NM (model_net_lp_config_names[FATTREE])
#define LP_METHOD_NM (model_net_method_names[FATTREE])

27 28
long fattree_packet_gen = 0, fattree_packet_fin = 0;

29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46
static double maxd(double a, double b) { return a < b ? b : a; }

// arrival rate
static double MEAN_INTERVAL=200.0;

typedef struct fattree_param fattree_param;
/* annotation-specific parameters (unannotated entry occurs at the 
 * last index) */
static uint64_t                  num_params = 0;
static fattree_param           * all_params = NULL;
static config_anno_map_t * anno_map   = NULL;

/* global variables for codes mapping */
static char lp_group_name[MAX_NAME_LENGTH];
static char def_group_name[MAX_NAME_LENGTH];
static int def_gname_set = 0;
static int mapping_grp_id, mapping_type_id, mapping_rep_id, mapping_offset;

47 48 49
/* switch magic number */
int switch_magic_num = 0;

50
/* terminal magic number */
51
int fattree_terminal_magic_num = 0;
52

53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85
typedef struct fattree_message_list fattree_message_list;
struct fattree_message_list {
    fattree_message msg;
    char* event_data;
    fattree_message_list *next;
    fattree_message_list *prev;
};

void init_fattree_message_list(fattree_message_list *this, 
  fattree_message *inmsg) {
    this->msg = *inmsg;
    this->event_data = NULL;
    this->next = NULL;
    this->prev = NULL;
}

void delete_fattree_message_list(fattree_message_list *this) {
    if(this->event_data != NULL) free(this->event_data);
    free(this);
}

struct fattree_param
{
  int ft_type;
  // configuration parameters
  int num_levels;
  int *num_switches; //switches at various levels
  int *switch_radix; //radix of switches are various levels
  double link_bandwidth;/* bandwidth of a wire connecting switches */
  double cn_bandwidth;/* bandwidth of the compute node channels 
                        connected to switch */
  int vc_size; /* buffer size of the link channels */
  int cn_vc_size; /* buffer size of the compute node channels */
86
  int chunk_size; /* full-sized packets are broken into smaller chunks.*/
87 88 89 90 91 92 93 94 95 96 97
  int packet_size; 
  int num_terminals;
  int l1_set_size;
  int l1_term_size;
  double cn_delay;
  double head_delay;
  double credit_delay;
  double router_delay;
  double soft_delay;
};

98 99 100 101 102 103 104 105 106 107 108 109 110 111 112
struct ftree_hash_key
{
    uint64_t message_id;
    tw_lpid sender_id;
};

struct ftree_qhash_entry
{
   struct ftree_hash_key key;
   char * remote_event_data;
   int num_chunks;
   int remote_event_size;
   struct qhash_head hash_link;
};

113 114 115 116 117 118 119 120 121
/* handles terminal and switch events like packet generate/send/receive/buffer */
typedef enum event_t event_t;
typedef struct ft_terminal_state ft_terminal_state;
typedef struct switch_state switch_state;

/* fattree compute node data structure */
struct ft_terminal_state
{
  unsigned long long packet_counter;
122
  int packet_fin;
123 124 125 126 127 128 129 130 131 132
  // Fattree specific parameters
  unsigned int terminal_id;
  unsigned int switch_id;
  tw_lpid switch_lp;

  // Each terminal will have an input and output channel with the switch
  int vc_occupancy; // NUM_VC
  tw_stime terminal_available_time;
  tw_stime next_credit_available_time;
  
133 134
  struct mn_stats fattree_stats_array[CATEGORY_MAX];

135 136
  fattree_message_list **terminal_msgs;
  fattree_message_list **terminal_msgs_tail;
137
  int terminal_length;
138
  int in_send_loop;
139 140 141
  int issueIdle;

   struct rc_stack * st;
142 143 144

  char * anno;
  fattree_param *params;
145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160

  struct qhash_table *rank_tbl;
  uint64_t rank_tbl_pop;

  tw_stime   total_time;
  uint64_t total_msg_size;
  double total_hops;
  long finished_msgs;
  long finished_chunks;
  long finished_packets;

  /* For sampling */
  long fin_chunks_sample;
  long data_size_sample;
  double fin_hops_sample;
  tw_stime fin_chunks_time;
161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210
};

/* terminal event type (1-4) */
enum event_t
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  S_SEND,
  S_ARRIVE,
  S_BUFFER,
};

enum last_hop
{
   LINK,
   TERMINAL
};

struct switch_state
{
  unsigned int switch_id;
  int switch_level;
  int radix;
  int num_cons;
  int num_lcons;
  int con_per_lneigh;
  int con_per_uneigh;;
  int start_lneigh;
  int end_lneigh;
  int start_uneigh;
  int unused;

  tw_stime* next_output_available_time;
  tw_stime* next_credit_available_time;
  fattree_message_list **pending_msgs;
  fattree_message_list **pending_msgs_tail;
  fattree_message_list **queued_msgs;
  fattree_message_list **queued_msgs_tail;
  int *queued_length;
  int *in_send_loop;
  int* vc_occupancy;
  int64_t* link_traffic;
  tw_lpid *port_connections;

  char * anno;
  fattree_param *params;
};

211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228
static tw_stime         fattree_total_time = 0;
static tw_stime         fattree_max_latency = 0;
static tw_stime         max_collective = 0;


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;

static void free_tmp(void * ptr)
{
    struct ftree_qhash_entry * ftree = ptr; 
    free(ftree->remote_event_data);
    free(ftree);
}

229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312
static void append_to_fattree_message_list(  
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
        int index, 
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
    } 
    thistail[index] = msg;
}

static void prepend_to_fattree_message_list(  
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
        int index, 
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
    } 
    thisq[index] = msg;
}

static void create_prepend_to_fattree_message_list(
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
        int index, 
        fattree_message *msg) {
    fattree_message_list* new_entry = (fattree_message_list*)malloc(
        sizeof(fattree_message_list));
    init_fattree_message_list(new_entry, msg);
    if(msg->remote_event_size_bytes) {
        void *m_data = model_net_method_get_edata(FATTREE, msg);
        new_entry->event_data = (void*)malloc(msg->remote_event_size_bytes);
        memcpy(new_entry->event_data, m_data, msg->remote_event_size_bytes);
    }
    prepend_to_fattree_message_list( thisq, thistail, index, new_entry);
}

static fattree_message_list* return_head(
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
        int index) {
    fattree_message_list *head = thisq[index];
    if(head != NULL) {
        thisq[index] = head->next;
        if(head->next != NULL) {
            head->next->prev = NULL;
            head->next = NULL;
        } else {
            thistail[index] = NULL;
        }
    }
    return head;
}

static fattree_message_list* return_tail(
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
        int index) {
    fattree_message_list *tail = thistail[index];
    if(tail->prev != NULL) {
        tail->prev->next = NULL;
        thistail[index] = tail->prev;
        tail->prev = NULL;
    } else {
        thistail[index] = NULL;
        thisq[index] = NULL;
    }
    return tail;
}

static void copy_fattree_list_entry( fattree_message_list *cur_entry,
    fattree_message *msg) {
    fattree_message *cur_msg = &cur_entry->msg;
    msg->packet_ID = cur_msg->packet_ID;    
    strcpy(msg->category, cur_msg->category);
    msg->final_dest_gid = cur_msg->final_dest_gid;
    msg->sender_lp = cur_msg->sender_lp;
313
    msg->dest_terminal_id = cur_msg->dest_terminal_id;
314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347
    msg->src_terminal_id = cur_msg->src_terminal_id;
    msg->intm_lp_id = cur_msg->intm_lp_id;
    msg->saved_vc = cur_msg->saved_vc;
    msg->saved_off = cur_msg->saved_off;
    msg->last_hop = cur_msg->last_hop;
    msg->intm_id = cur_msg->intm_id;
    msg->vc_index = cur_msg->vc_index;
    msg->vc_off = cur_msg->vc_off;
    msg->packet_size = cur_msg->packet_size;
    msg->msg_size = cur_msg->msg_size;
    msg->src_nic = cur_msg->src_nic;
    msg->uniq_id = cur_msg->uniq_id;
    msg->saved_size = cur_msg->saved_size;
    msg->local_event_size_bytes = cur_msg->local_event_size_bytes;
    msg->remote_event_size_bytes = cur_msg->remote_event_size_bytes;

    if(msg->local_event_size_bytes +  msg->remote_event_size_bytes > 0) {
        void *m_data = model_net_method_get_edata(FATTREE, msg);
        memcpy(m_data, cur_entry->event_data, 
            msg->local_event_size_bytes +  msg->remote_event_size_bytes);
    }
}

//decl
void switch_credit_send(switch_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp, int sq);
int ft_get_output_port( switch_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp, int *out_off );
int get_base_port(switch_state *s, int from_term, int index);


/* returns the fattree switch lp type for lp registration */
static const tw_lptype* fattree_get_switch_lp_type(void);

348
/* returns the fattree message size */
349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518
static int fattree_get_msg_sz(void)
{
  return sizeof(fattree_message);
}

static void fattree_read_config(char * anno, fattree_param *p){
  int i;

  p->ft_type = 0;
  configuration_get_value_int(&config, "PARAMS", "ft_type", anno, 
      &p->ft_type);

  configuration_get_value_int(&config, "PARAMS", "num_levels", anno, 
      &p->num_levels);
  if(p->num_levels <= 0) {
    tw_error(TW_LOC, "Too few num_levels, Aborting\n");
  }
  if(p->num_levels > 3) {
    tw_error(TW_LOC, "Too many num_levels, only upto 3 supported Aborting\n");
  }

  p->num_switches = (int *) malloc (p->num_levels * sizeof(int));
  p->switch_radix = (int*) malloc (p->num_levels * sizeof(int));

  char switch_counts_str[MAX_NAME_LENGTH];
  int rc = configuration_get_value(&config, "PARAMS", "switch_count", anno,
      switch_counts_str, MAX_NAME_LENGTH);
  if (rc == 0){
    tw_error(TW_LOC, "couldn't read PARAMS:switch_count");
  }
  char* token;
  token = strtok(switch_counts_str, ",");
  i = 0;
  while(token != NULL)
  {
    sscanf(token, "%d", &p->num_switches[i]);
    if(p->num_switches[i] <= 0)
    {
      tw_error(TW_LOC, "Invalid switch count  specified "
          "(%d at pos %d), exiting... ", p->num_switches[i], i);
    }
    i++;
    token = strtok(NULL,",");
  }

  //if(i != p->num_levels) {
  //  tw_error(TW_LOC, "Not enough switch counts, Aborting\n");
  //}
  
  char switch_radix_str[MAX_NAME_LENGTH];
  rc = configuration_get_value(&config, "PARAMS", "switch_radix", anno,
      switch_radix_str, MAX_NAME_LENGTH);
  if (rc == 0){
    tw_error(TW_LOC, "couldn't read PARAMS:switch_radix");
  }
  token = strtok(switch_radix_str, ",");
  i = 0;
  while(token != NULL)
  {
    sscanf(token, "%d", &p->switch_radix[i]);
    if(p->switch_radix[i] <= 0)
    {
      tw_error(TW_LOC, "Invalid switch radix  specified "
          "(%d at pos %d), exiting... ", p->switch_radix[i], i);
    }
    i++;
    token = strtok(NULL,",");
  }

  if(p->num_levels == 2) {
    p->num_switches[1] = p->num_switches[0]/2;
    p->switch_radix[1] = p->switch_radix[0];
  } else {
    p->num_switches[1] = p->num_switches[0];
    p->num_switches[2] = p->num_switches[0]/2;
    p->switch_radix[1] = p->switch_radix[2] = p->switch_radix[0];
  }
  //if(i != p->num_levels) {
  //  tw_error(TW_LOC, "Not enough switch radix, Aborting\n");
  //}

  i = 1;
  for(i = 1; i < p->num_levels - 1; i++) {
    if(p->num_switches[i - 1] * p->switch_radix[i - 1] >
       p->num_switches[i] * p->switch_radix[i]) {
      tw_error(TW_LOC, "Not enough switches/radix at level %d for full "
          "bisection bandwidth\n", i);
    }
  }

  if(p->num_switches[i - 1] * p->switch_radix[i - 1] > 2 * p->num_switches[i] *
      p->switch_radix[i]) {
    tw_error(TW_LOC, "Not enough switches/radix at level %d (top) for full "
        "bisection bandwidth\n", i);
  }

  configuration_get_value_int(&config, "PARAMS", "packet_size", anno,
      &p->packet_size);
  if(!p->packet_size) {
    p->packet_size = 512;
    fprintf(stderr, "Packet size is not specified, setting to %d\n", 
        p->packet_size);
  }
    
  p->router_delay = 50;
  configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
      &p->router_delay);
    
  p->soft_delay = 1000;
  configuration_get_value_double(&config, "PARAMS", "soft_delay", anno,
      &p->soft_delay);

  configuration_get_value_int(&config, "PARAMS", "vc_size", anno, &p->vc_size);
  if(!p->vc_size) {
    p->vc_size = 8*p->packet_size;
    fprintf(stderr, "Buffer size of global channels not specified, setting to "
        "%d\n", p->vc_size);
  }

  configuration_get_value_int(&config, "PARAMS", "cn_vc_size", anno, 
      &p->cn_vc_size);
  if(!p->cn_vc_size) {
    p->cn_vc_size = 8*p->packet_size;
    fprintf(stderr, "Buffer size of compute node channels not specified, " 
        "setting to %d\n", p->cn_vc_size);
  }

  configuration_get_value_double(&config, "PARAMS", "link_bandwidth", anno, 
      &p->link_bandwidth);
  if(!p->link_bandwidth) {
    p->link_bandwidth = 5;
    fprintf(stderr, "Bandwidth of links is specified, setting to %lf\n", 
        p->link_bandwidth);
  }

  configuration_get_value_double(&config, "PARAMS", "cn_bandwidth", anno, 
      &p->cn_bandwidth);
  if(!p->cn_bandwidth) {
    p->cn_bandwidth = 5;
    fprintf(stderr, "Bandwidth of compute node channels not specified, " 
        "setting to %lf\n", p->cn_bandwidth);
  }

  p->l1_set_size = p->switch_radix[0]/2;
 
  p->l1_term_size = (p->l1_set_size * (p->switch_radix[0] / 2));

  p->cn_delay = (1.0 / p->cn_bandwidth);
  p->head_delay = (1.0 / p->link_bandwidth);
  p->credit_delay = (1.0 / p->link_bandwidth) * 8; //assume 8 bytes packet
}

static void fattree_configure(){
  anno_map = codes_mapping_get_lp_anno_map(LP_CONFIG_NM);
  assert(anno_map);
  num_params = anno_map->num_annos + (anno_map->has_unanno_lp > 0);
  all_params = malloc(num_params * sizeof(*all_params));

  for (uint64_t i = 0; i < anno_map->num_annos; i++){
    char * anno = anno_map->annotations[i].ptr;
    fattree_read_config(anno, &all_params[i]);
  }
  if (anno_map->has_unanno_lp > 0){
    fattree_read_config(NULL, &all_params[anno_map->num_annos]);
  }
}

/* initialize a fattree compute node terminal */
void ft_terminal_init( ft_terminal_state * s, tw_lp * lp )
{
519 520
    s->packet_fin = 0;

521
    uint32_t h1 = 0, h2 = 0; 
522 523
    bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
    fattree_terminal_magic_num = h1 + h2;
524

525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841
    int i;
    char anno[MAX_NAME_LENGTH];

    if(def_gname_set == 0) {
      def_gname_set = 1;
      codes_mapping_get_lp_info(0, def_group_name, &mapping_grp_id, NULL,
          &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
    }

    // Assign the global switch ID
    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, 1, LP_CONFIG_NM,
           s->anno, 0);

   if(num_lps != (s->params->switch_radix[0]/2)) {
     tw_error(TW_LOC, "Number of NICs per repetition has to be equal to "
         "half the radix of leaf level switches %d vs %d\n", num_lps,
          s->params->switch_radix[0]/2);
   }
   s->terminal_id = (mapping_rep_id * num_lps) + mapping_offset;  
   s->switch_id = s->terminal_id / (s->params->switch_radix[0] / 2);
   codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
           s->switch_id, 0, &s->switch_lp);
   s->terminal_available_time = 0.0;
   s->packet_counter = 0;
   s->terminal_msgs = 
     (fattree_message_list**)malloc(1*sizeof(fattree_message_list*));
   s->terminal_msgs_tail = 
     (fattree_message_list**)malloc(1*sizeof(fattree_message_list*));

#if FATTREE_HELLO
   printf("I am terminal %d (%ld), connected to switch %d\n", s->terminal_id,
       lp->gid, s->switch_id);
#endif
   s->vc_occupancy = 0;
   s->terminal_msgs[0] = NULL;
   s->terminal_msgs_tail[0] = NULL;
   s->terminal_length = 0;
   s->in_send_loop = 0;
   s->issueIdle = 0;

   s->params->num_terminals = codes_mapping_get_lp_count(lp_group_name, 0, 
      LP_CONFIG_NM, s->anno, 0);
   return;
}

/* sets up the switch */
void switch_init(switch_state * r, tw_lp * lp)
{
  char anno[MAX_NAME_LENGTH];
  int num_terminals = -1, num_lps;

  if(def_gname_set == 0) {
    def_gname_set = 1;
    codes_mapping_get_lp_info(0, def_group_name, &mapping_grp_id, NULL,
        &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
    num_terminals = codes_mapping_get_lp_count(def_group_name, 0, 
      LP_CONFIG_NM, anno, 0);
    num_lps = codes_mapping_get_lp_count(def_group_name, 1, LP_CONFIG_NM,
           anno, 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'){
    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];
  }

  // shorthand
  fattree_param *p = r->params;
  if(mapping_offset == p->num_levels - 1) {
    if(mapping_rep_id >= p->num_switches[mapping_offset]) {
      r->unused = 1;
      return;
    }
  }

  r->unused = 0;

  r->switch_id = mapping_rep_id + mapping_offset * p->num_switches[0];

  if(num_terminals != -1) {
    p->num_terminals = num_terminals;
  }

  r->switch_level = mapping_offset;

  r->radix = p->switch_radix[r->switch_level];

  r->next_output_available_time = (tw_stime*) malloc (r->radix * 
      sizeof(tw_stime));
  r->next_credit_available_time = (tw_stime*) malloc (r->radix * 
      sizeof(tw_stime));
  r->vc_occupancy = (int*) malloc (r->radix * sizeof(int));
  r->in_send_loop = (int*) malloc (r->radix * sizeof(int));
  r->link_traffic = (int64_t*) malloc (r->radix * sizeof(int64_t));
  r->port_connections = (tw_lpid*) malloc (r->radix * sizeof(tw_lpid));
  r->pending_msgs = 
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
  r->pending_msgs_tail = 
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
  r->queued_msgs = 
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
  r->queued_msgs_tail = 
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
  r->queued_length = (int*)malloc(r->radix * sizeof(int));

  for(int i = 0; i < r->radix; i++)
  {
    // Set credit & switch occupancy
    r->next_output_available_time[i] = 0;
    r->next_credit_available_time[i] = 0;
    r->vc_occupancy[i] = 0;
    r->in_send_loop[i] = 0;
    r->link_traffic[i] = 0;
    r->pending_msgs[i] = NULL;
    r->pending_msgs_tail[i] = NULL;
    r->queued_msgs[i] = NULL;
    r->queued_msgs_tail[i] = NULL;
    r->queued_length[i] = 0;
  }

  //set lps connected to each port
  r->num_cons = 0;
  r->num_lcons = 0;
#if FATTREE_HELLO
  printf("I am switch %d (%d), level %d, radix %d\n", r->switch_id,
    lp->gid, r->switch_level, r->radix);
#endif
  //if at level 0, first half ports go to terminals
  if(r->switch_level == 0) {
    int start_terminal = r->switch_id * (p->switch_radix[0] / 2);
    int end_terminal = start_terminal + (p->switch_radix[0] / 2);
    for(int term = start_terminal; term < end_terminal; term++) {
      tw_lpid nextTerm;
      int rep = term / (p->switch_radix[0] / 2);
      int off = term % (p->switch_radix[0] / 2);
      codes_mapping_get_lp_id(def_group_name, LP_CONFIG_NM, NULL, 1,
          rep, off, &nextTerm);
      r->port_connections[r->num_cons++] = nextTerm;
      r->num_lcons++;
#if FATTREE_DEBUG
      printf("I am switch %d, connect to terminal %d (%d) at port %d\n",
          r->switch_id, term, nextTerm, r->num_cons - 1);
#endif
    }
    r->start_lneigh = start_terminal;
    r->end_lneigh = end_terminal;
    r->con_per_lneigh = 1;
    assert(r->num_lcons == (r->radix / 2));
    int l1_set;
    if(p->num_levels == 2) {
      l1_set = 0;
    } else {
      l1_set = r->switch_id / p->l1_set_size;
    }
    int l1_base = l1_set * p->l1_set_size;
    r->start_uneigh = p->num_switches[0] + l1_base;
    r->con_per_uneigh = 1;
    for(int l1 = 0; l1 < p->l1_set_size; l1++) {
      tw_lpid nextTerm;
      codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
          l1_base, 1, &nextTerm);
      for(int con = 0; con < r->con_per_uneigh; con++) {
        r->port_connections[r->num_cons++] = nextTerm;
#if FATTREE_DEBUG
      printf("I am switch %d, connect to upper switch %d L1 (%d) at port %d\n",
          r->switch_id, l1_base, nextTerm, r->num_cons - 1);
#endif
      }
      l1_base++;
    }
  } else if (r->switch_level == 1) {
    int l0_set_size, l0_base;
    if(p->num_levels == 2) {
      l0_set_size = p->num_switches[0];
      l0_base = 0;
      r->start_lneigh = 0;
      r->end_lneigh = p->num_switches[0];
    } else {
      l0_set_size = p->l1_set_size;
      l0_base = ((r->switch_id - p->num_switches[0]) / p->l1_set_size) *
        l0_set_size;
      r->start_lneigh = l0_base;
      r->end_lneigh = l0_base + l0_set_size;
    }
    r->con_per_lneigh = 1;
    for(int l0 = 0; l0 < l0_set_size; l0++) {
      tw_lpid nextTerm;
      codes_mapping_get_lp_id(def_group_name, "fattree_switch", NULL, 1,
          l0_base, 0, &nextTerm);
      for(int con = 0; con < r->con_per_lneigh; con++) {
        r->port_connections[r->num_cons++] = nextTerm;
        r->num_lcons++;
#if FATTREE_DEBUG
        printf("I am switch %d, connect to switch %d L0 (%d) at port %d\n",
            r->switch_id, l0_base, nextTerm, r->num_cons - 1);
#endif
      }
      l0_base++;
    }
    if(p->num_levels == 3) {
      if(p->ft_type == 0) {
        int l2_base = 0;
        /* not true anymore */
        r->start_uneigh = p->num_switches[0] + l2_base;
        r->con_per_uneigh = 1;
        if((r->switch_id - p->num_switches[0]) % p->l1_set_size >=
            p->l1_set_size/2) {
          l2_base += (p->num_switches[2]/2);
        }
        for(int l2 = 0; l2 < p->num_switches[2]/2; l2++) {
          tw_lpid nextTerm;
          codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
              l2_base, 2, &nextTerm);
          for(int con = 0; con < r->con_per_uneigh; con++) {
            r->port_connections[r->num_cons++] = nextTerm;
#if FATTREE_DEBUG
            printf("I am switch %d, connect to upper switch %d L2 (%d) at port %d\n",
                r->switch_id, l2_base, nextTerm, r->num_cons - 1);
#endif
          }
          l2_base++;
        }
      } else {
        int l2 = ((r->switch_id - p->num_switches[0]) % p->l1_set_size);
        /* not true anymore */
        r->start_uneigh = p->num_switches[0] + l2;
        r->con_per_uneigh = 2;
        for(; l2 < p->num_switches[2]; l2 += p->l1_set_size) {
          tw_lpid nextTerm;
          codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
              l2, 2, &nextTerm);
          for(int con = 0; con < r->con_per_uneigh; con++) {
            r->port_connections[r->num_cons++] = nextTerm;
#if FATTREE_DEBUG
            printf("I am switch %d, connect to upper switch %d L2 (%d) at port %d\n",
                r->switch_id, l2_base, nextTerm, r->num_cons - 1);
#endif
          }
        }
      }
    }
  } else {
    if(p->ft_type == 0) {
      r->con_per_lneigh = 1;
      /* not true anymore */
      r->start_lneigh = p->num_switches[0];
      r->end_lneigh = r->start_lneigh + p->num_switches[1];
      int l1 = 0;
      if(r->switch_id - p->num_switches[0] - p->num_switches[1] >=
          (p->num_switches[2]/2)) {
        l1 += (p->l1_set_size/2);
      }
      int count = 0;
      for(; l1 < p->num_switches[1]; l1++) {
        tw_lpid nextTerm;
        codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
            l1, 1, &nextTerm);
        for(int con = 0; con < r->con_per_lneigh; con++) {
          r->port_connections[r->num_cons++] = nextTerm;
          r->num_lcons++;
#if FATTREE_DEBUG
          printf("I am switch %d, connect to  switch %d L1 (%d) at port %d\n",
              r->switch_id, l1, nextTerm, r->num_cons - 1);
#endif
        }
        count++;
        if(count == (p->l1_set_size/2)) {
          l1 += (p->l1_set_size/2);
          count = 0;
        }
      }
    } else {
      r->con_per_lneigh = 2;
      /* not true anymore */
      r->start_lneigh = p->num_switches[0];
      r->end_lneigh = r->start_lneigh + p->num_switches[1];
      int l1 = (r->switch_id - p->num_switches[0] - p->num_switches[1]) % p->l1_set_size;
      for(; l1 < p->num_switches[1]; l1 += p->l1_set_size) {
        tw_lpid nextTerm;
        codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
            l1, 1, &nextTerm);
        for(int con = 0; con < r->con_per_lneigh; con++) {
          r->port_connections[r->num_cons++] = nextTerm;
          r->num_lcons++;
#if FATTREE_DEBUG
          printf("I am switch %d, connect to  switch %d L1 (%d) at port %d\n",
              r->switch_id, l1, nextTerm, r->num_cons - 1);
#endif
        }
      }
    }
  }
  return;
}	

/* empty for now.. */
static void fattree_report_stats() { }

/* fattree packet event */
Noah Wolfe's avatar
Noah Wolfe committed
842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867
static tw_stime fattree_packet_event(
        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)
/*	model_net_request* req, 
	char* category, 
    tw_lpid final_dest_lp, 
	uint64_t packet_size, 
	int is_pull, 
    uint64_t pull_size, 
	tw_stime offset, 
	const mn_sched_params *sched_params, 
    int remote_event_size, 
	const void* remote_event, 
	int self_event_size, 
    const void* self_event, 
	tw_lpid src_lp, 
	tw_lp *sender, 
	int is_last_pckt) 
*/{
868 869 870 871 872 873 874 875 876

  tw_event * e_new;
  tw_stime xfer_to_nic_time;
  fattree_message * msg;
  char* tmp_ptr;

  xfer_to_nic_time = codes_local_latency(sender);
  e_new = model_net_method_event_new(sender->gid, xfer_to_nic_time + offset,
      sender, FATTREE, (void**)&msg, (void**)&tmp_ptr);
Noah Wolfe's avatar
Noah Wolfe committed
877 878 879 880 881
  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;
882
  msg->packet_size = packet_size;
Noah Wolfe's avatar
Noah Wolfe committed
883
  msg->travel_start_time = tw_now(sender);
884 885 886
  msg->remote_event_size_bytes = 0;
  msg->local_event_size_bytes = 0;
  msg->type = T_GENERATE;
Noah Wolfe's avatar
Noah Wolfe committed
887 888 889 890
  msg->dest_terminal_id = req->dest_mn_lp;
  msg->message_id = req->msg_id;
  msg->is_pull = req->is_pull;
  msg->pull_size = req->pull_size;
891
  msg->magic = fattree_terminal_magic_num; 
Noah Wolfe's avatar
Noah Wolfe committed
892
  msg->msg_start_time = req->msg_start_time;
893 894 895 896

  /* Its the last packet so pass in remote and local event information*/
  if(is_last_pckt) 
  {
897
    if(req->remote_event_size > 0)
898
    {
899 900 901
      msg->remote_event_size_bytes = req->remote_event_size;
      memcpy(tmp_ptr, remote_event, req->remote_event_size);
      tmp_ptr += req->remote_event_size;
902
    }
903
    if(req->self_event_size > 0)
904
    {
905 906 907
      msg->local_event_size_bytes = req->self_event_size;
      memcpy(tmp_ptr, self_event, req->self_event_size);
      tmp_ptr += req->self_event_size;
908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932
    }
  }
  //printf("[%d] Send to %d\n", sender->gid, sender->gid);
  tw_event_send(e_new);
  return xfer_to_nic_time;
}

/* fattree packet event reverse handler */
static void fattree_packet_event_rc(tw_lp *sender)
{
  codes_local_latency_reverse(sender);
  return;
}

/* generates packet at the current fattree compute node */
void ft_packet_generate(ft_terminal_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp) {

  fattree_param *p = s->params;
  tw_stime ts, nic_ts;
  
  nic_ts = g_tw_lookahead + s->params->cn_delay * msg->packet_size + g_tw_lookahead * tw_rand_unif(lp->rng);

  codes_mapping_get_lp_info(msg->final_dest_gid, lp_group_name, &mapping_grp_id,
      NULL, &mapping_type_id, NULL, &mapping_rep_id, &mapping_offset);
933
  msg->dest_terminal_id = (mapping_rep_id * (p->switch_radix[0] / 2)) +
934 935 936
      (mapping_offset % (p->switch_radix[0]/2));
  
  //message for process on the same terminal
937
  if(msg->dest_terminal_id == s->terminal_id) {
938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007
    bf->c1 = 1;
    model_net_method_idle_event(nic_ts - s->params->cn_delay * msg->packet_size, 0, lp);
    // Trigger an event on receiving server
    if(msg->remote_event_size_bytes)
    {
      void *tmp_ptr = model_net_method_get_edata(FATTREE, msg);
      bf->c2 = 1;
      ts = codes_local_latency(lp);
      tw_event *e = tw_event_new(msg->final_dest_gid, ts, lp);
      fattree_message* m = tw_event_data(e);
      memcpy(m, tmp_ptr, msg->remote_event_size_bytes);
      //printf("[%d] pack gen Send to %d\n", lp->gid, msg->final_dest_gid);
      tw_event_send(e);
    }
    if(msg->local_event_size_bytes > 0)
    {
      tw_event* e_new;
      fattree_message* m_new;
      void* local_event;
      bf->c3 = 1;
      ts = codes_local_latency(lp); 
      e_new = tw_event_new(msg->sender_lp, ts, lp);
      m_new = tw_event_data(e_new);
      local_event = (char*)model_net_method_get_edata(FATTREE, msg) +
        msg->remote_event_size_bytes;
      memcpy(m_new, local_event, msg->local_event_size_bytes);
      tw_event_send(e_new);
    }
    return;
  }

  msg->packet_ID = lp->gid + g_tw_nlp * s->packet_counter;
        
  fattree_message_list * cur_chunk = (fattree_message_list *)malloc( 
      sizeof(fattree_message_list));
  init_fattree_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);
  }

  void *m_data_src = model_net_method_get_edata(FATTREE, msg);
  if (msg->remote_event_size_bytes){
    memcpy(cur_chunk->event_data, m_data_src, msg->remote_event_size_bytes);
  }
  if (msg->local_event_size_bytes){
    m_data_src = (char*)m_data_src + msg->remote_event_size_bytes;
    memcpy((char*)cur_chunk->event_data + msg->remote_event_size_bytes, 
        m_data_src, msg->local_event_size_bytes);
  }

  append_to_fattree_message_list(s->terminal_msgs, s->terminal_msgs_tail, 
    0, cur_chunk);
  s->terminal_length += s->params->packet_size;

  if(s->terminal_length < 2 * s->params->cn_vc_size) {
    model_net_method_idle_event(nic_ts, 0, lp);
  } else {
    bf->c11 = 1;
    s->issueIdle = 1;
  }

  if(s->in_send_loop == 0) {
    fattree_message *m;
    bf->c5 = 1;
    ts = g_tw_lookahead + s->params->cn_delay * msg->packet_size + g_tw_lookahead * tw_rand_unif(lp->rng);
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, FATTREE, 
      (void**)&m, NULL);
    m->type = T_SEND;
1008
	m->magic = fattree_terminal_magic_num;
1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091
    s->in_send_loop = 1;
    tw_event_send(e);
    //printf("[%d] send loop triggered with ts %lf band %lf\n", 
    // lp->gid, ts, s->params->cn_bandwidth);
  }

  return;
}

/* sends the packet from the compute node to the attached switch */
void ft_packet_send(ft_terminal_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp) {

  tw_stime ts;
  tw_event *e;
  fattree_message *m;
 
  fattree_message_list* cur_entry = s->terminal_msgs[0];

  if(s->vc_occupancy + s->params->packet_size > s->params->cn_vc_size ||
    cur_entry == NULL) {
    bf->c1 = 1;
    s->in_send_loop = 0;
    return;
  }

  //  Each packet is broken into chunks and then sent over the channel
  msg->saved_available_time = s->terminal_available_time;
  ts = g_tw_lookahead + s->params->cn_delay * cur_entry->msg.packet_size 
    + g_tw_lookahead * tw_rand_unif(lp->rng);
  s->terminal_available_time = maxd(s->terminal_available_time, tw_now(lp));
  s->terminal_available_time += ts;

  // we are sending an event to the switch, so no method_event here
  ts = s->terminal_available_time - tw_now(lp);
  e = tw_event_new(s->switch_lp, ts, lp);
  m = tw_event_data(e);
  memcpy(m, &cur_entry->msg, sizeof(fattree_message));
  if (m->remote_event_size_bytes){
    memcpy(model_net_method_get_edata(FATTREE, m), cur_entry->event_data,
        m->remote_event_size_bytes);
  }

  m->type = S_ARRIVE;
  m->src_terminal_id = lp->gid;
  m->intm_id = s->terminal_id;
  m->vc_index = 0;
  m->vc_off = 0; //we only have one connection to the terminal NIC
  m->local_event_size_bytes = 0;
  m->last_hop = TERMINAL;
  //printf("[%d] pack send Send to %d\n", lp->gid, s->switch_lp);
  tw_event_send(e);
      
  /* local completion message */
  if(cur_entry->msg.local_event_size_bytes > 0)
  {
    bf->c2 = 1;
    double tsT = codes_local_latency(lp); 
    tw_event *e_new = tw_event_new(cur_entry->msg.sender_lp, tsT, lp);
    fattree_message *m_new = tw_event_data(e_new);
    void *local_event = (char*)cur_entry->event_data + 
                cur_entry->msg.remote_event_size_bytes;
    memcpy(m_new, local_event, cur_entry->msg.local_event_size_bytes);
    tw_event_send(e_new);
  }
   
  s->packet_counter++;
  s->vc_occupancy += s->params->packet_size;
  cur_entry = return_head(s->terminal_msgs, s->terminal_msgs_tail, 0); 
  copy_fattree_list_entry(cur_entry, msg);
  delete_fattree_message_list(cur_entry);
  s->terminal_length -= s->params->packet_size;

  cur_entry = s->terminal_msgs[0];

  if(cur_entry != NULL &&
    s->vc_occupancy + s->params->packet_size <= s->params->cn_vc_size) {
    bf->c3 = 1;
    fattree_message *m;
    ts = ts + g_tw_lookahead * tw_rand_unif(lp->rng);
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, FATTREE, 
      (void**)&m, NULL);
    m->type = T_SEND;
1092
	m->magic = fattree_terminal_magic_num;
1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 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 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310
    tw_event_send(e);
  } else {
    bf->c4 = 1;
    s->in_send_loop = 0;
  }
  
  if(s->issueIdle) {
    bf->c5 = 1;
    s->issueIdle = 0;
    model_net_method_idle_event(codes_local_latency(lp), 0, lp);
  }

  return;
}

/* Packet arrives at the switch and a credit is sent back to the sending 
 * terminal/switch */
void switch_packet_receive( switch_state * s, tw_bf * bf, 
    fattree_message * msg, tw_lp * lp ) {

  tw_event *e;
  fattree_message *m;
  tw_stime ts;

  //printf("[%d] Switch %d recv packet %d\n", lp->gid, msg->vc_index);
  int output_port = -1, out_off = 0;

  output_port = ft_get_output_port(s, bf, msg, lp, &out_off);
  assert(output_port < s->radix);

  int max_vc_size = s->params->vc_size;
  int to_terminal = 0;

  //If going to terminal, use a different max
  if(s->switch_level == 0 && output_port < s->num_lcons) {
    max_vc_size = s->params->cn_vc_size;
    to_terminal = 1;
  }
  
  fattree_message_list * cur_chunk = (fattree_message_list *)malloc( 
      sizeof(fattree_message_list));
  init_fattree_message_list(cur_chunk, msg);
  if(msg->remote_event_size_bytes > 0) {
    void *m_data_src = model_net_method_get_edata(FATTREE, msg);
    cur_chunk->event_data = (char*)malloc(msg->remote_event_size_bytes);
    memcpy(cur_chunk->event_data, m_data_src, 
        msg->remote_event_size_bytes);
  }

  cur_chunk->msg.vc_index = output_port;
  cur_chunk->msg.vc_off = out_off;

  if(s->vc_occupancy[output_port] + s->params->packet_size <= max_vc_size) {
    bf->c1 = 1;
    switch_credit_send(s, bf, msg, lp, -1);
    append_to_fattree_message_list( s->pending_msgs, s->pending_msgs_tail, 
      output_port, cur_chunk);
    s->vc_occupancy[output_port] += s->params->packet_size;
    if(s->in_send_loop[output_port] == 0) {
      bf->c2 = 1;
      fattree_message *m;
      ts = codes_local_latency(lp); 
      tw_event *e = tw_event_new(lp->gid, ts, lp);
      m = tw_event_data(e);
      m->type = S_SEND;
      m->vc_index = output_port;
      //printf("[%d] pack recv Send to %d\n", lp->gid, lp->gid);
      tw_event_send(e);
      s->in_send_loop[output_port] = 1;
    }
  } else {
    bf->c3 = 1;
    cur_chunk->msg.saved_vc = msg->vc_index;
    cur_chunk->msg.saved_off = msg->vc_off;
    append_to_fattree_message_list( s->queued_msgs, s->queued_msgs_tail, 
      output_port, cur_chunk);
    s->queued_length[output_port] += s->params->packet_size;
  }
  //for reverse
  msg->saved_vc = output_port;

  return;
}

/* routes the current packet to the next stop */
void switch_packet_send( switch_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp) {

  tw_stime ts;
  tw_event *e;
  fattree_message *m;

  int output_port = msg->vc_index;
  fattree_message_list *cur_entry = s->pending_msgs[output_port];

  if(cur_entry == NULL) {
    bf->c1 = 1;
    s->in_send_loop[output_port] = 0;
    return;
  }
  
  tw_lpid next_stop = s->port_connections[output_port];
  int to_terminal = 0;
  tw_stime delay = s->params->head_delay;

  // dest can be a switch or a terminal, so we must check
  if(s->switch_level == 0 && output_port < s->num_lcons) {
    to_terminal = 1;
    delay = s->params->cn_delay;
  }

  ts = s->params->router_delay;
  double bytetime = delay * cur_entry->msg.packet_size;
  ts = g_tw_lookahead + g_tw_lookahead * tw_rand_unif( lp->rng) + bytetime + ts;

  msg->saved_available_time = s->next_output_available_time[output_port];
  s->next_output_available_time[output_port] = 
    maxd(s->next_output_available_time[output_port], tw_now(lp));
  s->next_output_available_time[output_port] += ts;

  ts = s->next_output_available_time[output_port] - tw_now(lp);
  void * m_data;
  if (to_terminal) {
    e = model_net_method_event_new(next_stop, ts, lp,
        FATTREE, (void**)&m, &m_data);
  } else {
      e = tw_event_new(next_stop, ts, lp);
      m = tw_event_data(e);
      m_data = model_net_method_get_edata(FATTREE, m);
  }

  memcpy(m, &cur_entry->msg, sizeof(fattree_message));
  if (m->remote_event_size_bytes){
      memcpy(m_data, cur_entry->event_data, m->remote_event_size_bytes);
  }

  m->last_hop = LINK;
  m->intm_lp_id = lp->gid;
  m->intm_id = s->switch_id;
  s->link_traffic[output_port] += cur_entry->msg.packet_size;

  /* Determine the event type. If the packet has arrived at the final destination
     switch then it should arrive at the destination terminal next. */
  if(to_terminal) {
    m->type = T_ARRIVE;
  } else {
    /* The packet has to be sent to another switch */
    m->type = S_ARRIVE;
  }
  tw_event_send(e);
  
  cur_entry = return_head(s->pending_msgs, s->pending_msgs_tail, 
    output_port);
  copy_fattree_list_entry(cur_entry, msg);
  delete_fattree_message_list(cur_entry);
    
  if(bytetime > s->params->router_delay) {
    s->next_output_available_time[output_port] -= s->params->router_delay;
    ts -= s->params->router_delay;
  } else {
    s->next_output_available_time[output_port] -= bytetime;
    ts -= bytetime;
  }

  cur_entry = s->pending_msgs[output_port];
  if(cur_entry != NULL) {
    bf->c3 = 1;
    fattree_message *m;
    ts = ts + g_tw_lookahead * tw_rand_unif(lp->rng);
    tw_event *e = tw_event_new(lp->gid, ts, lp);
    m = tw_event_data(e);
    m->type = S_SEND;
    m->vc_index = output_port;
    //printf("[%d] switch send loop Send to %d\n", lp->gid, lp->gid);
    tw_event_send(e);
  } else {
    bf->c4 = 1;
    s->in_send_loop[output_port] = 0;
  }
  return;
}

/* When a packet is sent from the current switch and a buffer slot 
 * becomes available, a credit is sent back to schedule another packet 
 * event */
void switch_credit_send(switch_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp, int sq) {

  tw_event * buf_e;
  tw_stime ts;
  fattree_message * buf_msg;

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

  fattree_param *p = s->params;
  // Notify sender terminal about available buffer space
  if(msg->last_hop == TERMINAL) {
    dest = msg->src_terminal_id;
    type = T_BUFFER;
    is_terminal = 1;
  } else if(msg->last_hop == LINK) {
    dest = msg->intm_lp_id;
  } 

  // Assume it takes 0.1 ns of serialization latency for processing the 
  // credits in the queue
  //int output_port = msg->vc_off; //src used this offset, so I have to
  //if(sq == 1) {
  //  output_port = msg->saved_off;
  //}
  //output_port += get_base_port(s, is_terminal, msg->intm_id);

  ts = g_tw_lookahead + s->params->credit_delay + g_tw_lookahead * tw_rand_unif(lp->rng);
	
  if (is_terminal) {
    buf_e = model_net_method_event_new(dest, ts, lp, FATTREE, 
      (void**)&buf_msg, NULL);
1311
	buf_msg->magic = fattree_terminal_magic_num;
1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340
  } else {
    buf_e = tw_event_new(dest, ts , lp);
    buf_msg = tw_event_data(buf_e);
  }

  buf_msg->type = type;

  if(sq == 1) {
    buf_msg->vc_index = msg->saved_vc;
  } else {
    buf_msg->vc_index = msg->vc_index; //the port src used to send me this data
  }

  //printf("[%d] credit send Send to %d\n", lp->gid, dest);
  tw_event_send(buf_e);
  return;
}

/* update the compute node-switch channel buffer */
void ft_terminal_buf_update(ft_terminal_state * s, tw_bf * bf,
    fattree_message * msg, tw_lp * lp) {
  s->vc_occupancy -= s->params->packet_size;
  if(s->in_send_loop == 0 && s->terminal_msgs[0] != NULL) {
    fattree_message *m;
    bf->c1 = 1;
    tw_stime ts = codes_local_latency(lp);
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, FATTREE, 
        (void**)&m, NULL);
    m->type = T_SEND;
1341
	m->type = fattree_terminal_magic_num;
1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383
    s->in_send_loop = 1;
    //printf("[%d] term buf Send to %d\n", lp->gid, lp->gid);
    tw_event_send(e);
  }
  return;
}

void switch_buf_update(switch_state * s, tw_bf * bf, fattree_message * msg, 
  tw_lp * lp) {
  int indx = msg->vc_index;
  s->vc_occupancy[indx] -= s->params->packet_size;

  if(s->queued_msgs[indx] != NULL) {
    bf->c1 = 1;
    fattree_message_list *head = return_head( s->queued_msgs,
        s->queued_msgs_tail, indx);
    s->queued_length[indx] -= s->params->packet_size;
    switch_credit_send( s, bf,  &head->msg, lp, 1); 
    append_to_fattree_message_list( s->pending_msgs, s->pending_msgs_tail, 
      indx, head);
    s->vc_occupancy[indx] += s->params->packet_size;
  }

  if(s->in_send_loop[indx] == 0 && s->pending_msgs[indx] != NULL) {
    bf->c2 = 1;
    fattree_message *m;
    tw_stime ts = codes_local_latency(lp);
    tw_event *e = tw_event_new(lp->gid, ts, lp);
    m = tw_event_data(e);
    m->type = S_SEND;
    m->vc_index = indx;
    s->in_send_loop[indx] = 1;
    //printf("[%d] switch buf Send to %d\n", lp->gid, lp->gid);
    tw_event_send(e);
  }
  return;
}

/* packet arrives at the destination terminal */
void ft_packet_arrive(ft_terminal_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp) {

1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401
  //Establish msg hash keys
  struct ftree_hash_key key;
  key.message_id = msg->message_id; 
  key.sender_id = msg->sender_lp;
  //Compute total number of chuncks to expect for the message
  uint64_t total_chunks = msg->total_size / s->params->chunk_size;
  //If total chunks doesn't divid evenly then add one extra for left over
  if(msg->total_size % s->params->chunk_size)
          total_chunks++;
  //If no chunks (sending msgs as whole) then set to 1
  if(!total_chunks)
          total_chunks = 1;
  
  assert(lp->gid == msg->dest_terminal_id);

//  if(msg->packet_ID == LLU(TRACK_PKT))
//        printf("\n Packet %llu arrived at lp %llu hops %d", msg->packet_ID, LLU(lp->gid), msg->my_N_hop);

1402 1403 1404 1405 1406 1407 1408
  // Packet arrives and accumulate # queued
  // Find a queue with an empty buffer slot
  tw_event * e, * buf_e;
  fattree_message * m, * buf_msg;
  tw_stime ts;

  // NIC aggregation - should this be a separate function?
1409
/*  uint64_t recvSize;
1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434
  int eventSize;
  char *data;
  void *tmp_ptr = model_net_method_get_edata(FATTREE, msg);
  int used = addMsgInfo(msg->src_nic, msg->uniq_id, msg->packet_size, 
      msg->remote_event_size_bytes, tmp_ptr);
  getMsgInfo(msg->src_nic, msg->uniq_id, &recvSize, &eventSize, &data);
  // Trigger an event on receiving server
  if(used) bf->c1 = 1;

  if(recvSize >= msg->msg_size && eventSize > 0) {
    bf->c2 = 1;
    void * tmp_ptr = model_net_method_get_edata(FATTREE, msg);
    ts = g_tw_lookahead + g_tw_lookahead * tw_rand_unif(lp->rng) +  
      s->params->cn_delay * eventSize;
    e = tw_event_new(msg->final_dest_gid, ts, lp);
    m = tw_event_data(e);
    memcpy(m, data, eventSize);
    tw_event_send(e); 
    if(!used) {
      msg->remote_event_size_bytes = eventSize;
      memcpy(tmp_ptr, data, eventSize);
    }
    msg->saved_size = recvSize;
    deleteMsgInfo(msg->src_nic, msg->uniq_id);
  }
1435
*/
1436 1437 1438 1439 1440
  ts = g_tw_lookahead + s->params->credit_delay + g_tw_lookahead * tw_rand_unif(lp->rng);
  
  // no method_event here - message going to switch
  buf_e = tw_event_new(s->switch_lp, ts, lp);
  buf_msg = tw_event_data(buf_e);
1441
  buf_msg->magic = switch_magic_num;
1442 1443 1444 1445
  buf_msg->vc_index = msg->vc_index;
  buf_msg->type = S_BUFFER;
  //printf("[%d] pack arrive credit Send to %d\n", lp->gid, s->switch_lp);
  tw_event_send(buf_e);
1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 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 1546 1547 1548 1549 1550 1551 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 1583 1584 1585 1586 1587 1588 1589

  bf->c1 = 0;
  bf->c3 = 0;
  bf->c4 = 0;
  bf->c7 = 0;

  /* 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++;

   s->data_size_sample += msg->total_size;
  /* WE do not allow self messages through fattree */
  assert(lp->gid != msg->src_terminal_id);

  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++;
    fattree_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;
    s->total_time += (tw_now(lp) - msg->travel_start_time); 

    msg->saved_total_time = fattree_total_time;
    fattree_total_time += tw_now( lp ) - msg->travel_start_time;
    total_hops += msg->my_N_hop;
    s->total_hops += msg->my_N_hop;
    s->fin_hops_sample += msg->my_N_hop;

    mn_stats* stat = model_net_find_stats(msg->category, s->fattree_stats_array);
    msg->saved_rcv_time = stat->recv_time;
    stat->recv_time += (tw_now(lp) - msg->travel_start_time);

#if DEBUG == 1
 if( msg->packet_ID == TRACK 
          && msg->chunk_id == num_chunks-1
          && msg->message_id == TRACK_MSG)
  {
    printf( "(%lf) [Terminal %d] packet %lld has arrived  \n",
        tw_now(lp), (int)lp->gid, msg->packet_ID);

    printf("travel start time is %f\n",
        msg->travel_start_time);

    printf("My hop now is %d\n",msg->my_N_hop);
  }
#endif

/* Now retreieve the number of chunks completed from the hash and update
    * them */
   void *m_data_src = model_net_method_get_edata(FATTREE, msg);
    
   struct qhash_head *hash_link = NULL;
   struct ftree_qhash_entry * tmp = NULL;
      
   hash_link = qhash_search(s->rank_tbl, &key);
    
   /* If an entry does not exist then create one */
   if(!hash_link)
   {
       bf->c5 = 1;
       struct ftree_qhash_entry * d_entry = malloc(sizeof (struct ftree_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));
       s->rank_tbl_pop++;
      
       hash_link = &(d_entry->hash_link);
   }
    
    if(hash_link)
        tmp = qhash_entry(hash_link, struct ftree_qhash_entry, hash_link);

    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 */
         tmp->remote_event_data = (void*)malloc(msg->remote_event_size_bytes);
         assert(tmp->remote_event_data);
         tmp->remote_event_size = msg->remote_event_size_bytes; 
         memcpy(tmp->remote_event_data, m_data_src, msg->remote_event_size_bytes);
    }
        if (fattree_max_latency < tw_now( lp ) - msg->travel_start_time) 
        {
          bf->c3 = 1;
          msg->saved_available_time = fattree_max_latency;
          fattree_max_latency = tw_now( lp ) - msg->travel_start_time;
        }
    /* If all chunks of a message have arrived then send a remote event to the
     * callee*/
    /*if(tmp->num_chunks >= total_chunks || tmp->num_chunks < 0)
        {
           //tw_output(lp, "\n invalid number of chunks %d for LP %ld ", tmp->num_chunks, lp->gid);
           tw_lp_suspend(lp, 0, 0);
           return;
        }*/

    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->finished_msgs++;
        
        //assert(tmp->remote_event_data && tmp->remote_event_size > 0);
        send_remote_event(s, msg, lp, bf, tmp->remote_event_data, tmp->remote_event_size);
        /* Remove the hash entry */
        qhash_del(hash_link);
        rc_stack_push(lp, tmp, free_tmp, s->st);
        s->rank_tbl_pop--;
   }

1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602
  return;
}

/* gets the output port corresponding to the next stop of the message */
int ft_get_output_port( switch_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp, int *out_off) {

  int outport = -1;
  int start_port, end_port;
  fattree_param *p = s->params;

  if(s->switch_level == 0) {
    //message for a terminal node
1603 1604
    if(msg->dest_terminal_id >= s->start_lneigh && msg->dest_terminal_id < s->end_lneigh) {
      outport = msg->dest_terminal_id - s->start_lneigh;
1605 1606 1607 1608 1609 1610 1611
      *out_off = 0;
      return outport;
    } else { //go up the least congested path
      start_port = s->num_lcons;
      end_port = s->num_cons;
    }
  } else if(s->switch_level == 1) {
1612
    int dest_switch_id = msg->dest_terminal_id / (p->switch_radix[0] / 2);
1613 1614 1615 1616 1617 1618 1619 1620 1621 1622
    //if only two level or packet going down, send to the right switch
    if(p->num_levels == 2 || (dest_switch_id >= s->start_lneigh && 
      dest_switch_id < s->end_lneigh)) {
      start_port = (dest_switch_id - s->start_lneigh) * s->con_per_lneigh;
      end_port = start_port + s->con_per_lneigh;
    } else {
      start_port = s->num_lcons;
      end_port = s->num_cons;
    }
  } else { //switch level 2
1623
    int dest_l1_group = msg->dest_terminal_id / p->l1_term_size;
1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670
    if(s->params->ft_type == 0) {
      start_port = dest_l1_group * (p->l1_set_size/2) * s->con_per_lneigh;
      end_port = start_port + ((p->l1_set_size/2) * s->con_per_lneigh);
    } else {
      start_port = dest_l1_group * s->con_per_lneigh;
      end_port = start_port + s->con_per_lneigh;
    }
  }

  assert(end_port > start_port);

  outport = start_port;
  int load = s->vc_occupancy[outport] + s->queued_length[outport];
  if(load != 0) {
    for(int port = start_port + 1; port < end_port; port++) {
      if(s->vc_occupancy[port] +  s->queued_length[port] < load) {
        load = s->vc_occupancy[port] +  s->queued_length[port];
        outport = port;
        if(load <= 0) break;
      }
    }
  }
  assert(outport != -1);
  if(outport < s->num_lcons) {
    *out_off = outport % s->con_per_lneigh;
  } else {
    *out_off = (outport - s->num_lcons) % s->con_per_uneigh;
  }
  return outport;
}

/* Currently incomplete. */
int get_base_port(switch_state *s, int from_term, int index) {
  int return_port;
  if(s->switch_level == 2) {
  } else if(from_term || index < s->switch_id) {
    return_port = ((index - s->start_lneigh) * s->con_per_lneigh);
  } else {
    return_port = s->num_lcons;
    return_port += ((index - s->start_uneigh) * s->con_per_uneigh);
  }
  return return_port;
}

void ft_terminal_event( ft_terminal_state * s, tw_bf * bf, fattree_message * msg,
		tw_lp * lp ) {

1671
  assert(msg->magic == fattree_terminal_magic_num);
1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697
  *(int *)bf = (int)0;
  switch(msg->type) {

    case T_GENERATE:
      ft_packet_generate(s, bf, msg, lp);
      break;

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

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

    case T_BUFFER:
      ft_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");
  }
}

1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710
void fattree_terminal_final( ft_terminal_state * s, tw_lp * lp ) 
{ 
    model_net_print_stats(lp->gid, s->fattree_stats_array);
   
    //if(s->packet_gen != s->packet_fin)
    //    printf("\n generated %d finished %d ", s->packet_gen, s->packet_fin);
    
    qhash_finalize(s->rank_tbl);
    rc_stack_destroy(s->st);
    free(s->vc_occupancy);
    free(s->terminal_msgs);
    free(s->terminal_msgs_tail);
//    free(s->children); 
1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746