dragonfly-custom.C 133 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14
/*
 * Copyright (C) 2013 University of Chicago.
 * See COPYRIGHT notice in top-level directory.
 *
 */

#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"
15
#include "codes/net/dragonfly-custom.h"
16 17 18
#include "sys/file.h"
#include "codes/quickhash.h"
#include "codes/rc-stack.h"
19 20
#include <vector>
#include <map>
21
#include <set>
22

23 24 25 26
#ifdef ENABLE_CORTEX
#include <cortex/cortex.h>
#include <cortex/topology.h>
#endif
27

28
#define DUMP_CONNECTIONS 0
29
#define PRINT_CONFIG 1
30
#define CREDIT_SIZE 8
Misbah Mubarak's avatar
Misbah Mubarak committed
31
#define DFLY_HASH_TABLE_SIZE 4999
32
// debugging parameters
33
#define DEBUG_LP 892
34
#define T_ID -1
35
#define TRACK -1
36
#define TRACK_PKT -1
37 38 39
#define TRACK_MSG -1
#define DEBUG 0
#define MAX_STATS 65536
40
#define SHOW_ADAP_STATS 1
41 42 43 44 45

#define LP_CONFIG_NM_TERM (model_net_lp_config_names[DRAGONFLY_CUSTOM])
#define LP_METHOD_NM_TERM (model_net_method_names[DRAGONFLY_CUSTOM])
#define LP_CONFIG_NM_ROUT (model_net_lp_config_names[DRAGONFLY_CUSTOM_ROUTER])
#define LP_METHOD_NM_ROUT (model_net_method_names[DRAGONFLY_CUSTOM_ROUTER])
46

47
static int BIAS_MIN = 1;
48
static int DF_DALLY = 0;
49
static int adaptive_threshold = 1024;
50

51 52 53
static long num_local_packets_sr = 0;
static long num_local_packets_sg = 0;
static long num_remote_packets = 0;
54 55 56 57 58 59 60
using namespace std;
struct Link {
  int offset, type;
};
struct bLink {
  int offset, dest;
};
61 62 63 64 65
/* Each entry in the vector is for a router id
 * against each router id, there is a map of links (key of the map is the dest
 * router id)
 * link has information on type (green or black) and offset (number of links
 * between that particular source and dest router ID)*/
66
vector< map< int, vector<Link> > > intraGroupLinks;
67 68
/* contains mapping between source router and destination group via link (link
 * has dest ID)*/
69
vector< map< int, vector<bLink> > > interGroupLinks;
70
/*MM: Maintains a list of routers connecting the source and destination groups */
71 72 73 74 75 76 77 78 79 80
vector< vector< vector<int> > > connectionList;

struct IntraGroupLink {
  int src, dest, type;
};

struct InterGroupLink {
  int src, dest;
};

81 82 83 84 85 86
#ifdef ENABLE_CORTEX
/* This structure is defined at the end of the file */
extern "C" {
extern cortex_topology dragonfly_custom_cortex_topology;
}
#endif
87

88 89 90
static int debug_slot_count = 0;
static long term_ecount, router_ecount, term_rev_ecount, router_rev_ecount;
static long packet_gen = 0, packet_fin = 0;
91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109

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

/* minimal and non-minimal packet counts for adaptive routing*/
static int minimal_count=0, nonmin_count=0;
static int num_routers_per_mgrp = 0;

typedef struct dragonfly_param dragonfly_param;
/* annotation-specific parameters (unannotated entry occurs at the 
 * last index) */
static uint64_t                  num_params = 0;
static dragonfly_param         * all_params = NULL;
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 */
110
static int router_magic_num = 0;
111 112

/* terminal magic number */
113
static int terminal_magic_num = 0;
114

115 116 117 118
/* Hops within a group */
static int num_intra_nonmin_hops = 4;
static int num_intra_min_hops = 2;

119
static FILE * dragonfly_log = NULL;
120

121 122
static int sample_bytes_written = 0;
static int sample_rtr_bytes_written = 0;
123

124 125
static char cn_sample_file[MAX_NAME_LENGTH];
static char router_sample_file[MAX_NAME_LENGTH];
126

127 128
//don't do overhead here - job of MPI layer
static tw_stime mpi_soft_overhead = 0;
129

130 131 132
typedef struct terminal_custom_message_list terminal_custom_message_list;
struct terminal_custom_message_list {
    terminal_custom_message msg;
133
    char* event_data;
134 135
    terminal_custom_message_list *next;
    terminal_custom_message_list *prev;
136 137
};

138
static void init_terminal_custom_message_list(terminal_custom_message_list *thisO, 
139
    terminal_custom_message *inmsg) {
140 141 142 143
    thisO->msg = *inmsg;
    thisO->event_data = NULL;
    thisO->next = NULL;
    thisO->prev = NULL;
144 145
}

146 147 148 149
static void delete_terminal_custom_message_list(void *thisO) {
    terminal_custom_message_list* toDel = (terminal_custom_message_list*)thisO;
    if(toDel->event_data != NULL) free(toDel->event_data);
    free(toDel);
150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165
}

struct dragonfly_param
{
    // configuration parameters
    int num_routers; /*Number of routers in a group*/
    double local_bandwidth;/* bandwidth of the router-router channels within a group */
    double global_bandwidth;/* bandwidth of the inter-group router connections */
    double cn_bandwidth;/* bandwidth of the compute node channels connected to routers */
    int num_vcs; /* number of virtual channels */
    int local_vc_size; /* buffer size of the router-router channels */
    int global_vc_size; /* buffer size of the global channels */
    int cn_vc_size; /* buffer size of the compute node channels */
    int chunk_size; /* full-sized packets are broken into smaller chunks.*/
    // derived parameters
    int num_cn;
166 167
    int intra_grp_radix;
    int num_col_chans;
168
    int num_row_chans;
169 170
    int num_router_rows;
    int num_router_cols;
171 172 173 174 175 176 177 178 179 180
    int num_groups;
    int radix;
    int total_routers;
    int total_terminals;
    int num_global_channels;
    double cn_delay;
    double local_delay;
    double global_delay;
    double credit_delay;
    double router_delay;
181 182

    int max_hops_notify; //maximum number of hops allowed before notifying via printout
183 184
};

185 186 187
static const dragonfly_param* stored_params;


188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245
struct dfly_hash_key
{
    uint64_t message_id;
    tw_lpid sender_id;
};

struct dfly_router_sample
{
    tw_lpid router_id;
    tw_stime* busy_time;
    int64_t* link_traffic_sample;
    tw_stime end_time;
    long fwd_events;
    long rev_events;
};

struct dfly_cn_sample
{
   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;
};

struct dfly_qhash_entry
{
   struct dfly_hash_key key;
   char * remote_event_data;
   int num_chunks;
   int remote_event_size;
   struct qhash_head hash_link;
};

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

/* dragonfly compute node data structure */
struct terminal_state
{
   uint64_t packet_counter;

   int packet_gen;
   int packet_fin;

   // Dragonfly specific parameters
   unsigned int router_id;
   unsigned int terminal_id;

   // Each terminal will have an input and output channel with the router
   int* vc_occupancy; // NUM_VC
   int num_vcs;
   tw_stime terminal_available_time;
246 247
   terminal_custom_message_list **terminal_msgs;
   terminal_custom_message_list **terminal_msgs_tail;
248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267
   int in_send_loop;
   struct mn_stats dragonfly_stats_array[CATEGORY_MAX];

   struct rc_stack * st;
   int issueIdle;
   int terminal_length;

   const char * anno;
   const dragonfly_param *params;

   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;

268
   tw_stime last_buf_full;
269
   tw_stime busy_time;
270 271 272 273
   
   tw_stime max_latency;
   tw_stime min_latency;

274 275 276 277 278 279 280 281 282 283 284 285 286 287 288
   char output_buf[4096];
   /* For LP suspend functionality */
   int error_ct;

   /* For sampling */
   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];
   struct dfly_cn_sample * sample_stat;
   int op_arr_size;
   int max_arr_size;
289
  
290 291 292 293 294 295
   /* for logging forward and reverse events */
   long fwd_events;
   long rev_events;
};

/* terminal event type (1-4) */
296
typedef enum event_t
297 298 299 300 301 302 303 304
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  R_SEND,
  R_ARRIVE,
  R_BUFFER,
305
} event_t;
306 307 308 309

/* whether the last hop of a packet was global, local or a terminal */
enum last_hop
{
310
   GLOBAL=1,
311
   LOCAL,
312 313
   TERMINAL,
   ROOT
314 315 316 317 318 319 320
};

/* three forms of routing algorithms available, adaptive routing is not
 * accurate and fully functional in the current version as the formulas
 * for detecting load on global channels are not very accurate */
enum ROUTING_ALGO
{
321
    MINIMAL = 1,
322 323 324 325 326
    NON_MINIMAL,
    ADAPTIVE,
    PROG_ADAPTIVE
};

327 328 329 330 331
enum LINK_TYPE
{
    GREEN,
    BLACK,
};
332 333 334 335 336 337 338 339 340 341 342
struct router_state
{
   unsigned int router_id;
   int group_id;
   int op_arr_size;
   int max_arr_size;

   int* global_channel; 
   
   tw_stime* next_output_available_time;
   tw_stime* cur_hist_start_time;
343
   tw_stime* last_buf_full;
344 345 346 347

   tw_stime* busy_time;
   tw_stime* busy_time_sample;

348 349 350 351
   terminal_custom_message_list ***pending_msgs;
   terminal_custom_message_list ***pending_msgs_tail;
   terminal_custom_message_list ***queued_msgs;
   terminal_custom_message_list ***queued_msgs_tail;
352 353 354
   int *in_send_loop;
   int *queued_count;
   struct rc_stack * st;
355 356

   int* last_sent_chan;
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
   int** vc_occupancy;
   int64_t* link_traffic;
   int64_t * link_traffic_sample;

   const char * anno;
   const dragonfly_param *params;

   int* prev_hist_num;
   int* cur_hist_num;
   
   char output_buf[4096];
   char output_buf2[4096];

   struct dfly_router_sample * rsamples;
   
   long fwd_events;
   long rev_events;
};

static short routing = MINIMAL;

static tw_stime         dragonfly_total_time = 0;
static tw_stime         dragonfly_max_latency = 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 int dragonfly_rank_hash_compare(
        void *key, struct qhash_head *link)
{
    struct dfly_hash_key *message_key = (struct dfly_hash_key *)key;
    struct dfly_qhash_entry *tmp = NULL;

    tmp = qhash_entry(link, struct dfly_qhash_entry, hash_link);
    
    if (tmp->key.message_id == message_key->message_id
            && tmp->key.sender_id == message_key->sender_id)
        return 1;

    return 0;
}
static int dragonfly_hash_func(void *k, int table_size)
{
    struct dfly_hash_key *tmp = (struct dfly_hash_key *)k;
405 406 407 408
    uint32_t pc = 0, pb = 0;	
    bj_hashlittle2(tmp, sizeof(*tmp), &pc, &pb);
    return (int)(pc % (table_size - 1));
    /*uint64_t key = (~tmp->message_id) + (tmp->message_id << 18);
409 410 411
    key = key * 21;
    key = ~key ^ (tmp->sender_id >> 4);
    key = key * tmp->sender_id; 
412
    return (int)(key & (table_size - 1));*/
413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430
}

/* convert GiB/s and bytes to ns */
static tw_stime bytes_to_ns(uint64_t bytes, double GB_p_s)
{
    tw_stime time;

    /* bytes to GB */
    time = ((double)bytes)/(1024.0*1024.0*1024.0);
    /* GiB to s */
    time = time / GB_p_s;
    /* s to ns */
    time = time * 1000.0 * 1000.0 * 1000.0;

    return(time);
}

/* returns the dragonfly message size */
431
int dragonfly_custom_get_msg_sz(void)
432
{
433
	   return sizeof(terminal_custom_message);
434 435 436 437
}

static void free_tmp(void * ptr)
{
438
    struct dfly_qhash_entry * dfly = (dfly_qhash_entry *)ptr; 
439 440 441 442 443
    if(dfly->remote_event_data)
        free(dfly->remote_event_data);
   
    if(dfly)
        free(dfly);
444
}
445

446 447 448
static void append_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
449
        int index, 
450
        terminal_custom_message_list *msg) {
451 452 453 454 455 456 457 458 459
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
    } 
    thistail[index] = msg;
}

460 461 462
static void prepend_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
463
        int index, 
464
        terminal_custom_message_list *msg) {
465 466 467 468 469 470 471 472 473
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
    } 
    thisq[index] = msg;
}

474 475 476
static terminal_custom_message_list* return_head(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
477
        int index) {
478
    terminal_custom_message_list *head = thisq[index];
479 480 481 482 483 484 485 486 487 488 489 490
    if(head != NULL) {
        thisq[index] = head->next;
        if(head->next != NULL) {
            head->next->prev = NULL;
            head->next = NULL;
        } else {
            thistail[index] = NULL;
        }
    }
    return head;
}

491 492 493
static terminal_custom_message_list* return_tail(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
494
        int index) {
495
    terminal_custom_message_list *tail = thistail[index];
496 497 498 499 500 501 502 503 504 505 506 507
    assert(tail);
    if(tail->prev != NULL) {
        tail->prev->next = NULL;
        thistail[index] = tail->prev;
        tail->prev = NULL;
    } else {
        thistail[index] = NULL;
        thisq[index] = NULL;
    }
    return tail;
}

508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539
void dragonfly_print_params(const dragonfly_param *p)
{
    int myRank;
    MPI_Comm_rank(MPI_COMM_CODES, &myRank);
    if (!myRank) { 
        printf("\n------------------ Dragonfly Custom Parameters ---------\n");
        printf("\tnum_routers =            %d\n",p->num_routers);
        printf("\tlocal_bandwidth =        %.2f\n",p->local_bandwidth);
        printf("\tglobal_bandwidth =       %.2f\n",p->global_bandwidth);
        printf("\tcn_bandwidth =           %.2f\n",p->cn_bandwidth);
        printf("\tnum_vcs =                %d\n",p->num_vcs);
        printf("\tlocal_vc_size =          %d\n",p->local_vc_size);
        printf("\tglobal_vc_size =         %d\n",p->global_vc_size);
        printf("\tcn_vc_size =             %d\n",p->cn_vc_size);
        printf("\tchunk_size =             %d\n",p->chunk_size);
        printf("\tnum_cn =                 %d\n",p->num_cn);
        printf("\tintra_grp_radix =        %d\n",p->intra_grp_radix);
        printf("\tnum_col_chans =          %d\n",p->num_col_chans);
        printf("\tnum_row_chans =          %d\n",p->num_row_chans);
        printf("\tnum_router_rows =        %d\n",p->num_router_rows);
        printf("\tnum_router_cols =        %d\n",p->num_router_cols);
        printf("\tnum_groups =             %d\n",p->num_groups);
        printf("\tradix =                  %d\n",p->radix);
        printf("\ttotal_routers =          %d\n",p->total_routers);
        printf("\ttotal_terminals =        %d\n",p->total_terminals);
        printf("\tnum_global_channels =    %d\n",p->num_global_channels);
        printf("\tcn_delay =               %.2f\n",p->cn_delay);
        printf("\tlocal_delay =            %.2f\n",p->local_delay);
        printf("\tglobal_delay =           %.2f\n",p->global_delay);
        printf("\tcredit_delay =           %.2f\n",p->credit_delay);
        printf("\trouter_delay =           %.2f\n",p->router_delay);
        printf("\trouting =                %d\n",routing);
540
        printf("\tmax hops notification =  %d\n",p->max_hops_notify);
541 542 543 544
        printf("------------------------------------------------------\n\n");
    }
}

545
static void dragonfly_read_config(const char * anno, dragonfly_param *params){
546 547 548 549 550 551 552 553
    /*Adding init for router magic number*/
    uint32_t h1 = 0, h2 = 0; 
    bj_hashlittle2(LP_METHOD_NM_ROUT, strlen(LP_METHOD_NM_ROUT), &h1, &h2);
    router_magic_num = h1 + h2;
    
    bj_hashlittle2(LP_METHOD_NM_TERM, strlen(LP_METHOD_NM_TERM), &h1, &h2);
    terminal_magic_num = h1 + h2;
    
554 555
    // shorthand
    dragonfly_param *p = params;
556
    int myRank;
557
    MPI_Comm_rank(MPI_COMM_CODES, &myRank);
558

559
    int rc = configuration_get_value_int(&config, "PARAMS", "local_vc_size", anno, &p->local_vc_size);
560 561 562 563 564
    if(rc) {
        p->local_vc_size = 1024;
        fprintf(stderr, "Buffer size of local channels not specified, setting to %d\n", p->local_vc_size);
    }

565 566 567
    rc = configuration_get_value_int(&config, "PARAMS", "adaptive_threshold", anno, &adaptive_threshold);
    if(rc) {
    	adaptive_threshold = p->local_vc_size / 8;
568
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
569
	}
570 571 572 573
    else
    {
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
    }
574

575 576 577 578 579 580
    rc = configuration_get_value_int(&config, "PARAMS", "global_vc_size", anno, &p->global_vc_size);
    if(rc) {
        p->global_vc_size = 2048;
        fprintf(stderr, "Buffer size of global channels not specified, setting to %d\n", p->global_vc_size);
    }

581 582 583 584
    rc = configuration_get_value_int(&config, "PARAMS", "df-dally-vc", anno, &DF_DALLY);
    if(rc) {
        DF_DALLY = 0;
    }
585 586 587 588 589
    
    rc = configuration_get_value_int(&config, "PARAMS", "minimal-bias", anno, &BIAS_MIN);
    if(rc) {
        BIAS_MIN = 0;
    }
590 591 592
    else
	printf("\n Setting minimal bias");

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
    rc = configuration_get_value_int(&config, "PARAMS", "cn_vc_size", anno, &p->cn_vc_size);
    if(rc) {
        p->cn_vc_size = 1024;
        fprintf(stderr, "Buffer size of compute node channels not specified, setting to %d\n", p->cn_vc_size);
    }

    rc = configuration_get_value_int(&config, "PARAMS", "chunk_size", anno, &p->chunk_size);
    if(rc) {
        p->chunk_size = 512;
        fprintf(stderr, "Chunk size for packets is specified, setting to %d\n", p->chunk_size);
    }

    rc = configuration_get_value_double(&config, "PARAMS", "local_bandwidth", anno, &p->local_bandwidth);
    if(rc) {
        p->local_bandwidth = 5.25;
        fprintf(stderr, "Bandwidth of local channels not specified, setting to %lf\n", p->local_bandwidth);
    }

    rc = configuration_get_value_double(&config, "PARAMS", "global_bandwidth", anno, &p->global_bandwidth);
    if(rc) {
        p->global_bandwidth = 4.7;
        fprintf(stderr, "Bandwidth of global channels not specified, setting to %lf\n", p->global_bandwidth);
    }

    rc = configuration_get_value_double(&config, "PARAMS", "cn_bandwidth", 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);
    }

623
    rc = configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
624
            &p->router_delay);
625
    if(rc) {
626 627
      p->router_delay = 100;
    }
628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644

    configuration_get_value(&config, "PARAMS", "cn_sample_file", anno, cn_sample_file,
            MAX_NAME_LENGTH);
    configuration_get_value(&config, "PARAMS", "rt_sample_file", anno, router_sample_file,
            MAX_NAME_LENGTH);
    
    char routing_str[MAX_NAME_LENGTH];
    configuration_get_value(&config, "PARAMS", "routing", anno, routing_str,
            MAX_NAME_LENGTH);
    if(strcmp(routing_str, "minimal") == 0)
        routing = MINIMAL;
    else if(strcmp(routing_str, "nonminimal")==0 || 
            strcmp(routing_str,"non-minimal")==0)
        routing = NON_MINIMAL;
    else if (strcmp(routing_str, "adaptive") == 0)
        routing = ADAPTIVE;
    else if (strcmp(routing_str, "prog-adaptive") == 0)
645
	      routing = PROG_ADAPTIVE;
646 647 648 649 650 651 652
    else
    {
        fprintf(stderr, 
                "No routing protocol specified, setting to minimal routing\n");
        routing = -1;
    }

653 654 655 656 657 658
    rc = configuration_get_value_int(&config, "PARAMS", "notification_on_hops_greater_than", anno, &p->max_hops_notify);
    if (rc) {
        printf("Maximum hops for notifying not specified, setting to INT MAX\n");
        p->max_hops_notify = INT_MAX;
    }

659 660 661 662 663 664 665 666 667 668
    // rc = configuration_get_value_int(&config, "PARAMS", "num_vcs_override", anno, &p->num_vcs);
    // if(rc) {
    //     if(routing == PROG_ADAPTIVE)
    //         p->num_vcs = 10;
    //     else
    //         p->num_vcs = 8;
    // }
    // else {
    //     printf("Overriding num_vcs: p->num_vcs=%d\n",p->num_vcs);
    // }
669 670 671
   
if(DF_DALLY == 0) 
{
672 673 674
    //if(routing == PROG_ADAPTIVE)
    //    p->num_vcs = 10;
    //else
675
        p->num_vcs = 8;
676 677 678
}
else
{
679
        p->num_vcs = 4;
680
}
681 682 683
    rc = configuration_get_value_int(&config, "PARAMS", "num_groups", anno, &p->num_groups);
    if(rc) {
      printf("Number of groups not specified. Aborting");
684
      MPI_Abort(MPI_COMM_CODES, 1);
685
    }
686 687 688
    if(p->num_groups % 2 == 0 && DF_DALLY == 1)
        tw_error(TW_LOC, "\n 1-D dragonfly expects an odd number of groups");

689 690 691 692 693
    rc = configuration_get_value_int(&config, "PARAMS", "num_col_chans", anno, &p->num_col_chans);
    if(rc) {
//        printf("\n Number of links connecting chassis not specified, setting to default value 3 ");
        p->num_col_chans = 3;
    }
694 695 696 697 698
    rc = configuration_get_value_int(&config, "PARAMS", "num_row_chans", anno, &p->num_row_chans);
    if(rc) {
//        printf("\n Number of links connecting chassis not specified, setting to default value 3 ");
        p->num_row_chans = 1;
    }
699 700 701 702 703 704 705 706 707 708
    rc = configuration_get_value_int(&config, "PARAMS", "num_router_rows", anno, &p->num_router_rows);
    if(rc) {
        printf("\n Number of router rows not specified, setting to 6 ");
        p->num_router_rows = 6;
    }
    rc = configuration_get_value_int(&config, "PARAMS", "num_router_cols", anno, &p->num_router_cols);
    if(rc) {
        printf("\n Number of router columns not specified, setting to 16 ");
        p->num_router_cols = 16;
    }
709 710 711 712
    p->intra_grp_radix = (p->num_router_cols * p->num_row_chans);
    if(p->num_router_rows > 1)
        p->intra_grp_radix += (p->num_router_rows * p->num_col_chans);

713 714
    p->num_routers = p->num_router_rows * p->num_router_cols;
    
715
    rc = configuration_get_value_int(&config, "PARAMS", "num_cns_per_router", anno, &p->num_cn);
716
    if(rc) {
717 718
        printf("\n Number of cns per router not specified, setting to %d ", p->num_routers/2);
        p->num_cn = p->num_routers/2;
719
    }
720

721 722
    rc = configuration_get_value_int(&config, "PARAMS", "num_global_channels", anno, &p->num_global_channels);
    if(rc) {
723 724
        printf("\n Number of global channels per router not specified, setting to 10 ");
        p->num_global_channels = 10;
725
    }
726
    p->radix = p->intra_grp_radix + p->num_global_channels + p->num_cn;
727 728
    p->total_routers = p->num_groups * p->num_routers;
    p->total_terminals = p->total_routers * p->num_cn;
729 730 731 732
    
    // read intra group connections, store from a router's perspective
    // all links to the same router form a vector
    char intraFile[MAX_NAME_LENGTH];
733
    configuration_get_value(&config, "PARAMS", "intra-group-connections", 
734
        anno, intraFile, MAX_NAME_LENGTH);
735 736
    if(strlen(intraFile) <= 0) {
      tw_error(TW_LOC, "Intra group connections file not specified. Aborting");
737 738
    }
    FILE *groupFile = fopen(intraFile, "rb");
739 740 741
    if(!groupFile)
        tw_error(TW_LOC, "intra-group file not found ");

742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763
    if(!myRank)
      printf("Reading intra-group connectivity file: %s\n", intraFile);

    {
      vector< int > offsets;
      offsets.resize(p->num_routers, 0);
      intraGroupLinks.resize(p->num_routers);
      IntraGroupLink newLink;

      while(fread(&newLink, sizeof(IntraGroupLink), 1, groupFile) != 0) {
        Link tmpLink;
        tmpLink.type = newLink.type;
        tmpLink.offset = offsets[newLink.src]++;
        intraGroupLinks[newLink.src][newLink.dest].push_back(tmpLink);
      }
    }

    fclose(groupFile);

    // read inter group connections, store from a router's perspective
    // also create a group level table that tells all the connecting routers
    char interFile[MAX_NAME_LENGTH];
764
    configuration_get_value(&config, "PARAMS", "inter-group-connections", 
765
        anno, interFile, MAX_NAME_LENGTH);
766 767
    if(strlen(interFile) <= 0) {
      tw_error(TW_LOC, "Inter group connections file not specified. Aborting");
768 769 770
    }
    FILE *systemFile = fopen(interFile, "rb");
    if(!myRank)
771
    {
772
      printf("Reading inter-group connectivity file: %s\n", interFile);
773 774
      printf("\n Total routers %d total groups %d ", p->total_routers, p->num_groups);
    }
775 776 777 778 779 780 781 782 783

    {
      vector< int > offsets;
      offsets.resize(p->total_routers, 0);
      interGroupLinks.resize(p->total_routers);
      connectionList.resize(p->num_groups);
      for(int g = 0; g < connectionList.size(); g++) {
        connectionList[g].resize(p->num_groups);
      }
784
      
785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805
      InterGroupLink newLink;

      while(fread(&newLink, sizeof(InterGroupLink), 1, systemFile) != 0) {
        bLink tmpLink;
        tmpLink.dest = newLink.dest;
        int srcG = newLink.src / p->num_routers;
        int destG = newLink.dest / p->num_routers;
        tmpLink.offset = offsets[newLink.src]++;
        interGroupLinks[newLink.src][destG].push_back(tmpLink);
        int r;
        for(r = 0; r < connectionList[srcG][destG].size(); r++) {
          if(connectionList[srcG][destG][r] == newLink.src) break;
        }
        if(r == connectionList[srcG][destG].size()) {
          connectionList[srcG][destG].push_back(newLink.src);
        }
      }
    }

    fclose(systemFile);

806
#if DUMP_CONNECTIONS == 1
807 808 809 810 811 812 813 814
    printf("Dumping intra-group connections\n");
    for(int a = 0; a < intraGroupLinks.size(); a++) {
      printf("Connections for router %d\n", a);
      map< int, vector<Link> >  &curMap = intraGroupLinks[a];
      map< int, vector<Link> >::iterator it = curMap.begin();
      for(; it != curMap.end(); it++) {
        printf(" ( %d - ", it->first);
        for(int l = 0; l < it->second.size(); l++) {
815
          // offset is number of local connections
816
          // type is black or green according to Cray architecture 
817 818 819 820 821 822 823
          printf("%d,%d ", it->second[l].offset, it->second[l].type);
        }
        printf(")");
      }
      printf("\n");
    }
#endif
824
#if DUMP_CONNECTIONS == 1
825 826 827
    printf("Dumping inter-group connections\n");
    for(int a = 0; a < interGroupLinks.size(); a++) {
      printf("Connections for router %d\n", a);
828 829
      map< int, vector<bLink> >  &curMap = interGroupLinks[a];
      map< int, vector<bLink> >::iterator it = curMap.begin();
830
      for(; it != curMap.end(); it++) {
831
        // dest group ID 
832 833
        printf(" ( %d - ", it->first);
        for(int l = 0; l < it->second.size(); l++) {
834 835
            // dest is dest router ID
            // offset is number of global connections
836 837 838 839 840 841 842 843
          printf("%d,%d ", it->second[l].offset, it->second[l].dest);
        }
        printf(")");
      }
      printf("\n");
    }
#endif

844
#if DUMP_CONNECTIONS == 1
845 846 847 848 849 850 851 852 853 854 855 856 857
    printf("Dumping source aries for global connections\n");
    for(int g = 0; g < p->num_groups; g++) {
      for(int g1 = 0; g1 < p->num_groups; g1++) {
        printf(" ( ");
        for(int l = 0; l < connectionList[g][g1].size(); l++) {
          printf("%d ", connectionList[g][g1][l]);
        }
        printf(")");
      }
      printf("\n");
    }
#endif
    if(!myRank) {
858
        printf("\n Total nodes %d routers %d groups %d routers per group %d radix %d\n",
859
                p->num_cn * p->total_routers, p->total_routers, p->num_groups,
860
                p->num_routers, p->radix);
861
    }
862

863 864 865
    p->cn_delay = bytes_to_ns(p->chunk_size, p->cn_bandwidth);
    p->local_delay = bytes_to_ns(p->chunk_size, p->local_bandwidth);
    p->global_delay = bytes_to_ns(p->chunk_size, p->global_bandwidth);
866
    p->credit_delay = bytes_to_ns(CREDIT_SIZE, p->local_bandwidth); //assume 8 bytes packet
867 868 869 870 871

    if (PRINT_CONFIG) 
        dragonfly_print_params(p);

    stored_params = p;
872 873
}

874
void dragonfly_custom_configure(){
875 876 877
    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);
878
    all_params = (dragonfly_param *)calloc(num_params, sizeof(*all_params));
879 880 881 882 883 884 885 886

    for (int i = 0; i < anno_map->num_annos; i++){
        const char * anno = anno_map->annotations[i].ptr;
        dragonfly_read_config(anno, &all_params[i]);
    }
    if (anno_map->has_unanno_lp > 0){
        dragonfly_read_config(NULL, &all_params[anno_map->num_annos]);
    }
887 888 889
#ifdef ENABLE_CORTEX
	model_net_topology = dragonfly_custom_cortex_topology;
#endif
890 891 892
}

/* report dragonfly statistics like average and maximum packet latency, average number of hops traversed */
893
void dragonfly_custom_report_stats()
894 895 896 897 898 899
{
   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;
900
   long total_local_packets_sr, total_local_packets_sg, total_remote_packets;
901

902 903 904 905 906 907 908
   MPI_Reduce( &total_hops, &avg_hops, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
   MPI_Reduce( &N_finished_packets, &total_finished_packets, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
   MPI_Reduce( &N_finished_msgs, &total_finished_msgs, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
   MPI_Reduce( &N_finished_chunks, &total_finished_chunks, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
   MPI_Reduce( &total_msg_sz, &final_msg_sz, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
   MPI_Reduce( &dragonfly_total_time, &avg_time, 1,MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_CODES);
   MPI_Reduce( &dragonfly_max_latency, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_CODES);
909
   
910
   MPI_Reduce( &packet_gen, &total_gen, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
911
   MPI_Reduce(&packet_fin, &total_fin, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
912 913
    MPI_Reduce( &num_local_packets_sr, &total_local_packets_sr, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
    MPI_Reduce( &num_local_packets_sg, &total_local_packets_sg, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
914
   MPI_Reduce( &num_remote_packets, &total_remote_packets, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
915
   if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
916
    {
917 918
	MPI_Reduce(&minimal_count, &total_minimal_packets, 1, MPI_INT, MPI_SUM, 0, MPI_COMM_CODES);
 	MPI_Reduce(&nonmin_count, &total_nonmin_packets, 1, MPI_INT, MPI_SUM, 0, MPI_COMM_CODES);
919 920 921 922 923
    }

   /* print statistics */
   if(!g_tw_mynode)
   {	
924 925 926
    if (PRINT_CONFIG) 
        dragonfly_print_params(stored_params);

927 928
      printf(" Average number of hops traversed %f average chunk latency %lf us maximum chunk latency %lf us avg message size %lf bytes finished 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, total_finished_chunks);
929
     if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
930 931 932
              printf("\n ADAPTIVE ROUTING STATS: %d chunks routed minimally %d chunks routed non-minimally completed packets %lld \n", 
                      total_minimal_packets, total_nonmin_packets, total_finished_chunks);
 
933
      printf("\n Total packets generated %ld finished %ld Locally routed- same router %ld different-router %ld Remote (inter-group) %ld \n", total_gen, total_fin, total_local_packets_sr, total_local_packets_sg, total_remote_packets);
934 935 936 937 938 939 940
   }
   return;
}


/* initialize a dragonfly compute node terminal */
void 
941
terminal_custom_init( terminal_state * s, 
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
	       tw_lp * lp )
{
    s->packet_gen = 0;
    s->packet_fin = 0;

    int i;
    char anno[MAX_NAME_LENGTH];

    // Assign the global router ID
    // TODO: be annotation-aware
    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_TERM,
           s->anno, 0);

967
   s->terminal_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
968
   s->router_id=(int)s->terminal_id / (s->params->num_cn);
969 970
   s->terminal_available_time = 0.0;
   s->packet_counter = 0;
971 972 973
   s->min_latency = INT_MAX;
   s->max_latency = 0;   

974 975 976 977 978 979 980 981 982 983 984 985 986
   s->finished_msgs = 0;
   s->finished_chunks = 0;
   s->finished_packets = 0;
   s->total_time = 0.0;
   s->total_msg_size = 0;

   s->busy_time = 0.0;

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

   rc_stack_create(&s->st);
   s->num_vcs = 1;
987
   s->vc_occupancy = (int*)calloc(s->num_vcs, sizeof(int));
988
   s->last_buf_full = 0.0;
989 990 991 992 993 994 995

   for( i = 0; i < s->num_vcs; i++ )
    {
      s->vc_occupancy[i]=0;
    }


996
   s->rank_tbl = NULL;
997
   s->terminal_msgs = 
998
       (terminal_custom_message_list**)calloc(s->num_vcs, sizeof(terminal_custom_message_list*));
999
   s->terminal_msgs_tail = 
1000
       (terminal_custom_message_list**)calloc(s->num_vcs, sizeof(terminal_custom_message_list*));
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011
   s->terminal_msgs[0] = NULL;
   s->terminal_msgs_tail[0] = NULL;
   s->terminal_length = 0;
   s->in_send_loop = 0;
   s->issueIdle = 0;

   return;
}

/* sets up the router virtual channels, global channels, 
 * local channels, compute node channels */
1012
void router_custom_setup(router_state * r, tw_lp * lp)
1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030
{
    
    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];
    }

    // shorthand
    const dragonfly_param *p = r->params;

1031
    num_routers_per_mgrp = codes_mapping_get_lp_count (lp_group_name, 1, "modelnet_dragonfly_custom_router",
1032 1033 1034 1035 1036 1037 1038
            NULL, 0);
    int num_grp_reps = codes_mapping_get_group_reps(lp_group_name);
    if(p->total_routers != num_grp_reps * num_routers_per_mgrp)
        tw_error(TW_LOC, "\n Config error: num_routers specified %d total routers computed in the network %d "
                "does not match with repetitions * dragonfly_router %d  ",
                p->num_routers, p->total_routers, num_grp_reps * num_routers_per_mgrp);

1039
   r->router_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
1040
   r->group_id=r->router_id/p->num_routers;
1041 1042
    
   //printf("\n Local router id %d global id %d ", r->router_id, lp->gid);
1043 1044 1045 1046

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

1047

1048 1049 1050 1051 1052 1053 1054
   r->global_channel = (int*)calloc(p->num_global_channels, sizeof(int));
   r->next_output_available_time = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
   r->cur_hist_start_time = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
   r->link_traffic = (int64_t*)calloc(p->radix, sizeof(int64_t));
   r->link_traffic_sample = (int64_t*)calloc(p->radix, sizeof(int64_t));
   r->cur_hist_num = (int*)calloc(p->radix, sizeof(int));
   r->prev_hist_num = (int*)calloc(p->radix, sizeof(int));
1055
  
1056 1057 1058
   r->last_sent_chan = (int*) calloc(p->num_router_rows, sizeof(int));
   r->vc_occupancy = (int**)calloc(p->radix, sizeof(int*));
   r->in_send_loop = (int*)calloc(p->radix, sizeof(int));
1059
   r->pending_msgs = 
1060
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
1061
   r->pending_msgs_tail = 
1062
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
1063
   r->queued_msgs = 
1064
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
1065
   r->queued_msgs_tail = 
1066 1067
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
   r->queued_count = (int*)calloc(p->radix, sizeof(int));
1068
   r->last_buf_full = (tw_stime*)calloc(p->radix, sizeof(tw_stime*));
1069 1070
   r->busy_time = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
   r->busy_time_sample = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
1071 1072

   rc_stack_create(&r->st);
1073 1074 1075 1076

   for(int i = 0; i < p->num_router_rows; i++)
       r->last_sent_chan[i] = 0;

1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089
   for(int i=0; i < p->radix; i++)
    {
       // Set credit & router occupancy
    r->busy_time[i] = 0.0;
    r->busy_time_sample[i] = 0.0;
	r->next_output_available_time[i]=0;
	r->cur_hist_start_time[i] = 0;
    r->link_traffic[i]=0;
    r->link_traffic_sample[i] = 0;
	r->cur_hist_num[i] = 0;
	r->prev_hist_num[i] = 0;
    r->queued_count[i] = 0;    
    r->in_send_loop[i] = 0;
1090 1091
    r->vc_occupancy[i] = (int*)calloc(p->num_vcs, sizeof(int));
    r->pending_msgs[i] = (terminal_custom_message_list**)calloc(p->num_vcs, 
1092
        sizeof(terminal_custom_message_list*));
1093
    r->last_buf_full[i] = 0.0;
1094
    r->pending_msgs_tail[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1095
        sizeof(terminal_custom_message_list*));
1096
    r->queued_msgs[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1097
        sizeof(terminal_custom_message_list*));
1098
    r->queued_msgs_tail[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1099
        sizeof(terminal_custom_message_list*));
1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112
        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;
        }
    }
   return;
}	


/* dragonfly packet event , generates a dragonfly packet on the compute node */
1113
static tw_stime dragonfly_custom_packet_event(
1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127
        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;
1128
    terminal_custom_message * msg;
1129 1130 1131 1132 1133 1134
    char* tmp_ptr;

    xfer_to_nic_time = codes_local_latency(sender); 
    //e_new = tw_event_new(sender->gid, xfer_to_nic_time+offset, sender);
    //msg = tw_event_data(e_new);
    e_new = model_net_method_event_new(sender->gid, xfer_to_nic_time+offset,
1135
            sender, DRAGONFLY_CUSTOM, (void**)&msg, (void**)&tmp_ptr);
1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
    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->message_id = req->msg_id;
    msg->is_pull = req->is_pull;
    msg->pull_size = req->pull_size;
    msg->magic = terminal_magic_num; 
    msg->msg_start_time = req->msg_start_time;

    if(is_last_pckt) /* Its the last packet so pass in remote and local event information*/
      {
1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166
        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;
        }
1167 1168 1169 1170 1171 1172 1173
     }
	   //printf("\n dragonfly remote event %d local event %d last packet %d %lf ", msg->remote_event_size_bytes, msg->local_event_size_bytes, is_last_pckt, xfer_to_nic_time);
    tw_event_send(e_new);
    return xfer_to_nic_time;
}

/* dragonfly packet event reverse handler */
1174
static void dragonfly_custom_packet_event_rc(tw_lp *sender)
1175 1176 1177 1178 1179 1180
{
	  codes_local_latency_reverse(sender);
	    return;
}

/*When a packet is sent from the current router and a buffer slot becomes available, a credit is sent back to schedule another packet event*/
1181
static void router_credit_send(router_state * s, terminal_custom_message * msg, 
1182 1183 1184
  tw_lp * lp, int sq) {
  tw_event * buf_e;
  tw_stime ts;
1185
  terminal_custom_message * buf_msg;
1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196

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

  const dragonfly_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;
1197 1198 1199 1200
  } else if(msg->last_hop == GLOBAL 
          || msg->last_hop == LOCAL
          || msg->last_hop == ROOT)
  {
1201 1202 1203 1204 1205 1206 1207
    dest = msg->intm_lp_id;
  } else
    printf("\n Invalid message type");

  ts = g_tw_lookahead + p->credit_delay +  tw_rand_unif(lp->rng);
	
  if (is_terminal) {
1208
    buf_e = model_net_method_event_new(dest, ts, lp, DRAGONFLY_CUSTOM, 
1209 1210 1211
      (void**)&buf_msg, NULL);
    buf_msg->magic = terminal_magic_num;
  } else {
1212
    buf_e = model_net_method_event_new(dest, ts, lp, DRAGONFLY_CUSTOM_ROUTER,
1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230
            (void**)&buf_msg, NULL);
    buf_msg->magic = router_magic_num;
  }
 
  if(sq == -1) {
    buf_msg->vc_index = msg->vc_index;
    buf_msg->output_chan = msg->output_chan;
  } else {
    buf_msg->vc_index = msg->saved_vc;
    buf_msg->output_chan = msg->saved_channel;
  }
  
  buf_msg->type = type;

  tw_event_send(buf_e);
  return;
}

1231
static void packet_generate_rc(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, tw_lp * lp)
1232 1233 1234
{
   s->packet_gen--;
   packet_gen--;
1235 1236
   s->packet_counter--;

1237 1238 1239
   tw_rand_reverse_unif(lp->rng);

   int num_chunks = msg->packet_size/s->params->chunk_size;
1240
   if(msg->packet_size < s->params->chunk_size)
1241 1242 1243 1244
       num_chunks++;

   int i;
   for(i = 0; i < num_chunks; i++) {
1245
        delete_terminal_custom_message_list(return_tail(s->terminal_msgs, 
1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263
          s->terminal_msgs_tail, 0));
        s->terminal_length -= s->params->chunk_size;
   }
    if(bf->c5) {
        codes_local_latency_reverse(lp);
        s->in_send_loop = 0;
    }
      if(bf->c11) {
        s->issueIdle = 0;
      }
     struct mn_stats* stat;
     stat = model_net_find_stats(msg->category, s->dragonfly_stats_array);
     stat->send_count--;
     stat->send_bytes -= msg->packet_size;
     stat->send_time -= (1/s->params->cn_bandwidth) * msg->packet_size;
}

/* generates packet at the current dragonfly compute node */
1264
static void packet_generate(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, 
1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275
  tw_lp * lp) {
  packet_gen++;
  s->packet_gen++;

  tw_stime ts, nic_ts;

  assert(lp->gid != msg->dest_terminal_id);
  const dragonfly_param *p = s->params;

  int total_event_size;
  uint64_t num_chunks = msg->packet_size / p->chunk_size;
1276 1277
  double cn_delay = s->params->cn_delay;

1278
  if (msg->packet_size < s->params->chunk_size) 
1279 1280
      num_chunks++;

1281 1282 1283
  if(msg->packet_size < s->params->chunk_size)
      cn_delay = bytes_to_ns(msg->packet_size % s->params->chunk_size, s->params->cn_bandwidth);

1284 1285 1286 1287 1288
  int dest_router_id = codes_mapping_get_lp_relative_id(msg->dest_terminal_id, 0, 0) / s->params->num_cn;
  int dest_grp_id = dest_router_id / s->params->num_routers; 
  int src_grp_id = s->router_id / s->params->num_routers; 

  if(src_grp_id == dest_grp_id)
1289 1290 1291 1292 1293 1294
  {
      if(dest_router_id == s->router_id)
          num_local_packets_sr++;
      else
          num_local_packets_sg++;
  }
1295
  else
1296
  {
1297
      num_remote_packets++;
1298
  }
1299
  nic_ts = g_tw_lookahead + (num_chunks * cn_delay) + tw_rand_unif(lp->rng);
1300
  
1301
  msg->packet_ID = s->packet_counter;
1302
  s->packet_counter++;
1303 1304 1305 1306 1307
  msg->my_N_hop = 0;
  msg->my_l_hop = 0;
  msg->my_g_hop = 0;


1308
  for(int i = 0; i < num_chunks; i++)
1309
  {
1310
    terminal_custom_message_list *cur_chunk = (terminal_custom_message_list*)calloc(1,
1311
      sizeof(terminal_custom_message_list));
1312
    msg->origin_router_id = s->router_id;
1313
    init_terminal_custom_message_list(cur_chunk, msg);
1314 1315
  
    if(msg->remote_event_size_bytes + msg->local_event_size_bytes > 0) {
1316
      cur_chunk->event_data = (char*)calloc(1,
1317 1318 1319
          msg->remote_event_size_bytes + msg->local_event_size_bytes);
    }
    
1320
    void * m_data_src = model_net_method_get_edata(DRAGONFLY_CUSTOM, msg);
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331
    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);
    }

    cur_chunk->msg.chunk_id = i;
    cur_chunk->msg.origin_router_id = s->router_id;
1332
    append_to_terminal_custom_message_list(s->terminal_msgs, s->terminal_msgs_tail,
1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346
      0, cur_chunk);
    s->terminal_length += s->params->chunk_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) {
    bf->c5 = 1;
    ts = codes_local_latency(lp);
1347
    terminal_custom_message *m;
1348
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, DRAGONFLY_CUSTOM, 
1349 1350 1351 1352 1353 1354 1355
      (void**)&m, NULL);
    m->type = T_SEND;
    m->magic = terminal_magic_num;
    s->in_send_loop = 1;
    tw_event_send(e);
  }

1356
  total_event_size = model_net_get_msg_sz(DRAGONFLY_CUSTOM) + 
1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368
      msg->remote_event_size_bytes + msg->local_event_size_bytes;
  mn_stats* stat;
  stat = model_net_find_stats(msg->category, s->dragonfly_stats_array);
  stat->send_count++;
  stat->send_bytes += msg->packet_size;
  stat->send_time += (1/p->cn_bandwidth) * msg->packet_size;
  if(stat->max_event_size < total_event_size)
	  stat->max_event_size = total_event_size;

  return;
}

1369
static void packet_send_rc(terminal_state * s, tw_bf * bf, terminal_custom_message * msg,
1370 1371
        tw_lp * lp)
{
1372
      
1373 1374
      if(bf->c1) {
        s->in_send_loop = 1;
1375 1376 1377 1378
        if(bf->c3)
        {
         s->last_buf_full = msg->saved_busy_time;
        }
1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390
        return;
      }
      
      tw_rand_reverse_unif(lp->rng);
      s->terminal_available_time = msg->saved_available_time;
      if(bf->c2) {
        codes_local_latency_reverse(lp);
      }
     
      s->terminal_length += s->params->chunk_size;
      s->vc_occupancy[0] -= s->params->chunk_size;

1391
      terminal_custom_message_list* cur_entry = (terminal_custom_message_list *)rc_stack_pop(s->st);
1392

1393
      prepend_to_terminal_custom_message_list(s->terminal_msgs, 
1394 1395 1396 1397 1398 1399 1400
              s->terminal_msgs_tail, 0, cur_entry);
      if(bf->c3) {
        tw_rand_reverse_unif(lp->rng);
      }
      if(bf->c4) {
        s->in_send_loop = 1;
      }
1401
      if(bf->c5)
1402
      {
1403
          tw_rand_reverse_unif(lp->rng);
1404
          s->issueIdle = 1;
1405 1406 1407
          if(bf->c6)
          {
            s->busy_time = msg->saved_total_time;
1408
            s->last_buf_full = msg->saved_busy_time;
1409 1410
            s->busy_time_sample = msg->saved_sample_time;
          }
1411
      }
1412 1413 1414
      return;
}
/* sends the packet from the current dragonfly compute node to the attached router */
1415
static void packet_send(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, 
1416 1417 1418 1419
  tw_lp * lp) {
  
  tw_stime ts;
  tw_event *e;
1420
  terminal_custom_message *m;
1421 1422
  tw_lpid router_id;

1423
  terminal_custom_message_list* cur_entry = s->terminal_msgs[0];
1424
  bool noEmptyVC = false;
1425

1426
  if(s->vc_occupancy[0] + s->params->chunk_size > s->params->cn_vc_size)
1427
      noEmptyVC = true;
1428

1429 1430 1431 1432
  if(s->vc_occupancy[0] + s->params->chunk_size > s->params->cn_vc_size 
      || cur_entry == NULL) {
    bf->c1 = 1;
    s->in_send_loop = 0;
1433 1434 1435 1436 1437 1438
    if(noEmptyVC && !s->last_buf_full)
    {
        bf->c3 = 1;
        msg->saved_busy_time = s->last_buf_full;
        s->last_buf_full = tw_now(lp); 
    }
1439 1440 1441 1442
    return;
  }

  uint64_t num_chunks = cur_entry->msg.packet_size/s->params->chunk_size;
1443
  if(cur_entry->msg.packet_size < s->params->chunk_size)
1444 1445 1446
    num_chunks++;

  tw_stime delay = s->params->cn_delay;
1447
  if((cur_entry->msg.packet_size < s->params->chunk_size) && (cur_entry->msg.chunk_id == num_chunks - 1))
1448 1449 1450 1451 1452 1453 1454
       delay = bytes_to_ns(cur_entry->msg.packet_size % s->params->chunk_size, s->params->cn_bandwidth); 

  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;

1455
  ts = s->terminal_available_time - tw_now(lp);
1456 1457
  codes_mapping_get_lp_info(lp->gid, lp_group_name, &mapping_grp_id, NULL,
      &mapping_type_id, NULL, &mapping_rep_id, &mapping_offset);
1458
  codes_mapping_get_lp_id(lp_group_name, LP_CONFIG_NM_ROUT, NULL, 0,
1459
      s->router_id / num_routers_per_mgrp, s->router_id % num_routers_per_mgrp, &router_id);
1460 1461 1462

//  if(s->router_id == 1)
//   printf("\n Local router id %d global router id %d ", s->router_id, router_id);
1463 1464
  // we are sending an event to the router, so no method_event here
  void * remote_event;
1465
  e = model_net_method_event_new(router_id, ts, lp,
1466
          DRAGONFLY_CUSTOM_ROUTER, (void**)&m, &remote_event);
1467
  memcpy(m, &cur_entry->msg, sizeof(terminal_custom_message));
1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478
  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->last_hop = TERMINAL;
  m->magic = router_magic_num;
  m->path_type = -1;
  m->local_event_size_bytes = 0;
1479
  m->intm_rtr_id = -1;
1480 1481 1482
  tw_event_send(e);


1483 1484 1485 1486 1487
  if(cur_entry->msg.packet_ID == LLU(TRACK_PKT) && lp->gid == T_ID)
    printf("\n Packet %llu generated at terminal %d dest %llu size %llu num chunks %llu router-id %d %d", 
            cur_entry->msg.packet_ID, s->terminal_id, LLU(cur_entry->msg.dest_terminal_id),
            LLU(cur_entry->msg.packet_size), LLU(num_chunks), s->router_id, router_id);

1488 1489 1490
  if(cur_entry->msg.chunk_id == num_chunks - 1 && 
      (cur_entry->msg.local_event_size_bytes > 0)) {
    bf->c2 = 1;
1491 1492
    tw_stime local_ts = codes_local_latency(lp); 
    tw_event *e_new = tw_event_new(cur_entry->msg.sender_lp, local_ts, lp);
1493 1494 1495 1496 1497 1498 1499 1500
    void * 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->vc_occupancy[0] += s->params->chunk_size;
  cur_entry = return_head(s->terminal_msgs, s->terminal_msgs_tail, 0); 
1501
  rc_stack_push(lp, cur_entry, delete_terminal_custom_message_list, s->st);
1502 1503 1504 1505 1506 1507 1508 1509
  s->terminal_length -= s->params->chunk_size;

  cur_entry = s->terminal_msgs[0];

  /* if there is another packet inline then schedule another send event */
  if(cur_entry != NULL &&
    s->vc_occupancy[0] + s->params->chunk_size <= s->params->cn_vc_size) {
    bf->c3 = 1;
1510
    terminal_custom_message *m_new;
1511
    ts += tw_rand_unif(lp->rng);
1512
    e = model_net_method_event_new(lp->gid, ts, lp, DRAGONFLY_CUSTOM, 
1513 1514 1515 1516 1517 1518 1519 1520 1521
      (void**)&m_new, NULL);
    m_new->type = T_SEND;
    m_new->magic = terminal_magic_num;
    tw_event_send(e);
  } else {
      /* If not then the LP will wait for another credit or packet generation */
    bf->c4 = 1;
    s->in_send_loop = 0;
  }
1522
  if(s->issueIdle) {
1523 1524
    bf->c5 = 1;
    s->issueIdle = 0;