dragonfly-custom.C 146 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 48 49 50 51
static int max_lvc_src_g = 1;
static int max_lvc_intm_g = 3;
static int min_gvc_src_g = 0;
static int min_gvc_intm_g = 1;

52
static int BIAS_MIN = 1;
53
static int DF_DALLY = 0;
54
static int adaptive_threshold = 1024;
55

56 57 58
static long num_local_packets_sr = 0;
static long num_local_packets_sg = 0;
static long num_remote_packets = 0;
59 60 61 62 63 64 65
using namespace std;
struct Link {
  int offset, type;
};
struct bLink {
  int offset, dest;
};
66 67 68 69 70
/* 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)*/
71
vector< map< int, vector<Link> > > intraGroupLinks;
72 73
/* contains mapping between source router and destination group via link (link
 * has dest ID)*/
74
vector< map< int, vector<bLink> > > interGroupLinks;
75
/*MM: Maintains a list of routers connecting the source and destination groups */
76 77 78 79 80 81 82 83 84 85
vector< vector< vector<int> > > connectionList;

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

struct InterGroupLink {
  int src, dest;
};

86 87 88 89 90 91
#ifdef ENABLE_CORTEX
/* This structure is defined at the end of the file */
extern "C" {
extern cortex_topology dragonfly_custom_cortex_topology;
}
#endif
92

93 94 95
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;
96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114

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 */
115
static int router_magic_num = 0;
116 117

/* terminal magic number */
118
static int terminal_magic_num = 0;
119

120 121 122 123
/* Hops within a group */
static int num_intra_nonmin_hops = 4;
static int num_intra_min_hops = 2;

124
static FILE * dragonfly_log = NULL;
125

126 127
static int sample_bytes_written = 0;
static int sample_rtr_bytes_written = 0;
128

129 130
static char cn_sample_file[MAX_NAME_LENGTH];
static char router_sample_file[MAX_NAME_LENGTH];
131

132 133
//don't do overhead here - job of MPI layer
static tw_stime mpi_soft_overhead = 0;
134

135 136 137
typedef struct terminal_custom_message_list terminal_custom_message_list;
struct terminal_custom_message_list {
    terminal_custom_message msg;
138
    char* event_data;
139 140
    terminal_custom_message_list *next;
    terminal_custom_message_list *prev;
141 142
};

143
static void init_terminal_custom_message_list(terminal_custom_message_list *thisO, 
144
    terminal_custom_message *inmsg) {
145 146 147 148
    thisO->msg = *inmsg;
    thisO->event_data = NULL;
    thisO->next = NULL;
    thisO->prev = NULL;
149 150
}

151 152 153 154
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);
155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170
}

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;
171 172
    int intra_grp_radix;
    int num_col_chans;
173
    int num_row_chans;
174 175
    int num_router_rows;
    int num_router_cols;
176 177 178 179 180 181 182 183 184 185
    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;
186 187

    int max_hops_notify; //maximum number of hops allowed before notifying via printout
188 189
};

190 191 192
static const dragonfly_param* stored_params;


193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250
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;
251 252
   terminal_custom_message_list **terminal_msgs;
   terminal_custom_message_list **terminal_msgs_tail;
253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272
   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;

273
   tw_stime last_buf_full;
274
   tw_stime busy_time;
275 276 277 278
   
   tw_stime max_latency;
   tw_stime min_latency;

279
   char output_buf[4096];
280
   char output_buf2[4096];
281 282 283 284 285 286 287 288 289 290 291 292 293 294
   /* 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;
295
  
296 297 298
   /* for logging forward and reverse events */
   long fwd_events;
   long rev_events;
299 300 301 302 303 304 305 306

   /* following used for ROSS model-level stats collection */
   long fin_chunks_ross_sample;
   long data_size_ross_sample;
   long fin_hops_ross_sample;
   tw_stime fin_chunks_time_ross_sample;
   tw_stime busy_time_ross_sample;
   struct dfly_cn_sample ross_sample;
307 308 309
};

/* terminal event type (1-4) */
310
typedef enum event_t
311 312 313 314 315 316 317 318
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  R_SEND,
  R_ARRIVE,
  R_BUFFER,
319
} event_t;
320 321 322 323

/* whether the last hop of a packet was global, local or a terminal */
enum last_hop
{
324
   GLOBAL=1,
325
   LOCAL,
326 327
   TERMINAL,
   ROOT
328 329 330 331 332 333 334
};

/* 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
{
335
    MINIMAL = 1,
336 337 338 339 340
    NON_MINIMAL,
    ADAPTIVE,
    PROG_ADAPTIVE
};

341 342 343 344 345
enum LINK_TYPE
{
    GREEN,
    BLACK,
};
346 347 348 349 350 351 352 353 354 355 356
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;
357
   tw_stime* last_buf_full;
358 359 360 361

   tw_stime* busy_time;
   tw_stime* busy_time_sample;

362 363 364 365
   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;
366 367 368
   int *in_send_loop;
   int *queued_count;
   struct rc_stack * st;
369 370

   int* last_sent_chan;
371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386
   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];

   struct dfly_router_sample * rsamples;
   
   long fwd_events;
   long rev_events;
387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403

   /* following used for ROSS model-level stats collection */
   tw_stime* busy_time_ross_sample;
   int64_t * link_traffic_ross_sample;
   struct dfly_router_sample ross_rsample;
};

/* had to pull some of the ROSS model stats collection stuff up here */
void custom_dragonfly_event_collect(terminal_custom_message *m, tw_lp *lp, char *buffer, int *collect_flag);
void custom_dragonfly_model_stat_collect(terminal_state *s, tw_lp *lp, char *buffer);
void custom_dfly_router_model_stat_collect(router_state *s, tw_lp *lp, char *buffer);
static void ross_custom_dragonfly_rsample_fn(router_state * s, tw_bf * bf, tw_lp * lp, struct dfly_router_sample *sample);
static void ross_custom_dragonfly_rsample_rc_fn(router_state * s, tw_bf * bf, tw_lp * lp, struct dfly_router_sample *sample);
static void ross_custom_dragonfly_sample_fn(terminal_state * s, tw_bf * bf, tw_lp * lp, struct dfly_cn_sample *sample);
static void ross_custom_dragonfly_sample_rc_fn(terminal_state * s, tw_bf * bf, tw_lp * lp, struct dfly_cn_sample *sample);

st_model_types custom_dragonfly_model_types[] = {
404
    {(ev_trace_f) custom_dragonfly_event_collect,
405 406 407 408 409 410
     sizeof(int),
     (model_stat_f) custom_dragonfly_model_stat_collect,
     sizeof(tw_lpid) + sizeof(long) * 2 + sizeof(double) + sizeof(tw_stime) *2,
     (sample_event_f) ross_custom_dragonfly_sample_fn,
     (sample_revent_f) ross_custom_dragonfly_sample_rc_fn,
     sizeof(struct dfly_cn_sample) } , 
411
    {(ev_trace_f) custom_dragonfly_event_collect,
412 413
     sizeof(int),
     (model_stat_f) custom_dfly_router_model_stat_collect,
414
     0, //updated in router_custom_setup() since it's based on the radix
415 416
     (sample_event_f) ross_custom_dragonfly_rsample_fn,
     (sample_revent_f) ross_custom_dragonfly_rsample_rc_fn,
417
     0 } , //updated in router_custom_setup() since it's based on the radix    
418
    {NULL, 0, NULL, 0, NULL, NULL, 0}
419
};
420
/* End of ROSS model stats collection */
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

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;
451 452 453 454
    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);
455 456 457
    key = key * 21;
    key = ~key ^ (tmp->sender_id >> 4);
    key = key * tmp->sender_id; 
458
    return (int)(key & (table_size - 1));*/
459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476
}

/* 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 */
477
int dragonfly_custom_get_msg_sz(void)
478
{
479
	   return sizeof(terminal_custom_message);
480 481 482 483
}

static void free_tmp(void * ptr)
{
484
    struct dfly_qhash_entry * dfly = (dfly_qhash_entry *)ptr; 
485 486 487 488 489
    if(dfly->remote_event_data)
        free(dfly->remote_event_data);
   
    if(dfly)
        free(dfly);
490
}
491

492 493 494
static void append_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
495
        int index, 
496
        terminal_custom_message_list *msg) {
497 498 499 500 501 502 503 504 505
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
    } 
    thistail[index] = msg;
}

506 507 508
static void prepend_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
509
        int index, 
510
        terminal_custom_message_list *msg) {
511 512 513 514 515 516 517 518 519
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
    } 
    thisq[index] = msg;
}

520 521 522
static terminal_custom_message_list* return_head(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
523
        int index) {
524
    terminal_custom_message_list *head = thisq[index];
525 526 527 528 529 530 531 532 533 534 535 536
    if(head != NULL) {
        thisq[index] = head->next;
        if(head->next != NULL) {
            head->next->prev = NULL;
            head->next = NULL;
        } else {
            thistail[index] = NULL;
        }
    }
    return head;
}

537 538 539
static terminal_custom_message_list* return_tail(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
540
        int index) {
541
    terminal_custom_message_list *tail = thistail[index];
542 543 544 545 546 547 548 549 550 551 552 553
    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;
}

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
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);
586
        printf("\tmax hops notification =  %d\n",p->max_hops_notify);
587 588 589 590
        printf("------------------------------------------------------\n\n");
    }
}

591
static void dragonfly_read_config(const char * anno, dragonfly_param *params){
592 593 594 595 596 597 598 599
    /*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;
    
600 601
    // shorthand
    dragonfly_param *p = params;
602
    int myRank;
603
    MPI_Comm_rank(MPI_COMM_CODES, &myRank);
604

605
    int rc = configuration_get_value_int(&config, "PARAMS", "local_vc_size", anno, &p->local_vc_size);
606 607 608 609 610
    if(rc) {
        p->local_vc_size = 1024;
        fprintf(stderr, "Buffer size of local channels not specified, setting to %d\n", p->local_vc_size);
    }

611 612 613
    rc = configuration_get_value_int(&config, "PARAMS", "adaptive_threshold", anno, &adaptive_threshold);
    if(rc) {
    	adaptive_threshold = p->local_vc_size / 8;
614
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
615
	}
616 617 618 619
    else
    {
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
    }
620

621 622 623 624 625 626
    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);
    }

627 628 629 630
    rc = configuration_get_value_int(&config, "PARAMS", "df-dally-vc", anno, &DF_DALLY);
    if(rc) {
        DF_DALLY = 0;
    }
631 632 633 634 635
    
    rc = configuration_get_value_int(&config, "PARAMS", "minimal-bias", anno, &BIAS_MIN);
    if(rc) {
        BIAS_MIN = 0;
    }
636 637 638
    else
	printf("\n Setting minimal bias");

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
    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);
    }

669
    rc = configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
670
            &p->router_delay);
671
    if(rc) {
672 673
      p->router_delay = 100;
    }
674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690

    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)
691
	      routing = PROG_ADAPTIVE;
692 693 694 695 696 697 698
    else
    {
        fprintf(stderr, 
                "No routing protocol specified, setting to minimal routing\n");
        routing = -1;
    }

699 700 701 702 703 704
    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;
    }

705 706 707 708 709 710 711 712 713 714
    // 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);
    // }
715 716 717
   
if(DF_DALLY == 0) 
{
718 719 720
    //if(routing == PROG_ADAPTIVE)
    //    p->num_vcs = 10;
    //else
721
        p->num_vcs = 8;
722 723 724
}
else
{
725
        p->num_vcs = 4;
726
}
727 728 729
    rc = configuration_get_value_int(&config, "PARAMS", "num_groups", anno, &p->num_groups);
    if(rc) {
      printf("Number of groups not specified. Aborting");
730
      MPI_Abort(MPI_COMM_CODES, 1);
731
    }
732 733 734 735 736
    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;
    }
737 738 739 740 741
    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;
    }
742 743 744 745 746 747 748 749 750 751
    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;
    }
752 753 754 755
    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);

756 757
    p->num_routers = p->num_router_rows * p->num_router_cols;
    
758
    rc = configuration_get_value_int(&config, "PARAMS", "num_cns_per_router", anno, &p->num_cn);
759
    if(rc) {
760 761
        printf("\n Number of cns per router not specified, setting to %d ", p->num_routers/2);
        p->num_cn = p->num_routers/2;
762
    }
763

764 765
    rc = configuration_get_value_int(&config, "PARAMS", "num_global_channels", anno, &p->num_global_channels);
    if(rc) {
766 767
        printf("\n Number of global channels per router not specified, setting to 10 ");
        p->num_global_channels = 10;
768
    }
769
    p->radix = p->intra_grp_radix + p->num_global_channels + p->num_cn;
770 771
    p->total_routers = p->num_groups * p->num_routers;
    p->total_terminals = p->total_routers * p->num_cn;
772 773 774 775
    
    // read intra group connections, store from a router's perspective
    // all links to the same router form a vector
    char intraFile[MAX_NAME_LENGTH];
776
    configuration_get_value(&config, "PARAMS", "intra-group-connections", 
777
        anno, intraFile, MAX_NAME_LENGTH);
778 779
    if(strlen(intraFile) <= 0) {
      tw_error(TW_LOC, "Intra group connections file not specified. Aborting");
780 781
    }
    FILE *groupFile = fopen(intraFile, "rb");
782 783 784
    if(!groupFile)
        tw_error(TW_LOC, "intra-group file not found ");

785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806
    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];
807
    configuration_get_value(&config, "PARAMS", "inter-group-connections", 
808
        anno, interFile, MAX_NAME_LENGTH);
809 810
    if(strlen(interFile) <= 0) {
      tw_error(TW_LOC, "Inter group connections file not specified. Aborting");
811 812 813
    }
    FILE *systemFile = fopen(interFile, "rb");
    if(!myRank)
814
    {
815
      printf("Reading inter-group connectivity file: %s\n", interFile);
816 817
      printf("\n Total routers %d total groups %d ", p->total_routers, p->num_groups);
    }
818 819 820 821 822 823 824 825 826

    {
      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);
      }
827
      
828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848
      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);

849
#if DUMP_CONNECTIONS == 1
850 851 852 853 854 855 856 857
    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++) {
858
          // offset is number of local connections
859
          // type is black or green according to Cray architecture 
860 861 862 863 864 865 866
          printf("%d,%d ", it->second[l].offset, it->second[l].type);
        }
        printf(")");
      }
      printf("\n");
    }
#endif
867
#if DUMP_CONNECTIONS == 1
868 869 870
    printf("Dumping inter-group connections\n");
    for(int a = 0; a < interGroupLinks.size(); a++) {
      printf("Connections for router %d\n", a);
871 872
      map< int, vector<bLink> >  &curMap = interGroupLinks[a];
      map< int, vector<bLink> >::iterator it = curMap.begin();
873
      for(; it != curMap.end(); it++) {
874
        // dest group ID 
875 876
        printf(" ( %d - ", it->first);
        for(int l = 0; l < it->second.size(); l++) {
877 878
            // dest is dest router ID
            // offset is number of global connections
879 880 881 882 883 884 885 886
          printf("%d,%d ", it->second[l].offset, it->second[l].dest);
        }
        printf(")");
      }
      printf("\n");
    }
#endif

887
#if DUMP_CONNECTIONS == 1
888 889 890 891 892 893 894 895 896 897 898 899 900
    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) {
901
        printf("\n Total nodes %d routers %d groups %d routers per group %d radix %d\n",
902
                p->num_cn * p->total_routers, p->total_routers, p->num_groups,
903
                p->num_routers, p->radix);
904
    }
905

906 907 908
    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);
909
    p->credit_delay = bytes_to_ns(CREDIT_SIZE, p->local_bandwidth); //assume 8 bytes packet
910 911 912 913 914

    if (PRINT_CONFIG) 
        dragonfly_print_params(p);

    stored_params = p;
915 916
}

917
void dragonfly_custom_configure(){
918 919 920
    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);
921
    all_params = (dragonfly_param *)calloc(num_params, sizeof(*all_params));
922 923 924 925 926 927 928 929

    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]);
    }
930 931 932
#ifdef ENABLE_CORTEX
	model_net_topology = dragonfly_custom_cortex_topology;
#endif
933 934 935
}

/* report dragonfly statistics like average and maximum packet latency, average number of hops traversed */
936
void dragonfly_custom_report_stats()
937 938 939 940 941 942
{
   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;
943
   long total_local_packets_sr, total_local_packets_sg, total_remote_packets;
944

945 946 947 948 949 950 951
   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);
952
   
953
   MPI_Reduce( &packet_gen, &total_gen, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
954
   MPI_Reduce(&packet_fin, &total_fin, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
955 956
    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);
957
   MPI_Reduce( &num_remote_packets, &total_remote_packets, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
958
   if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
959
    {
960 961
	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);
962 963 964 965 966
    }

   /* print statistics */
   if(!g_tw_mynode)
   {	
967 968 969
    if (PRINT_CONFIG) 
        dragonfly_print_params(stored_params);

970 971
      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);
972
     if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
973 974 975
              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);
 
976
      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);
977 978 979 980 981 982 983
   }
   return;
}


/* initialize a dragonfly compute node terminal */
void 
984
terminal_custom_init( terminal_state * s, 
985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009
	       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);

1010
   s->terminal_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
1011
   s->router_id=(int)s->terminal_id / (s->params->num_cn);
1012 1013
   s->terminal_available_time = 0.0;
   s->packet_counter = 0;
1014 1015 1016
   s->min_latency = INT_MAX;
   s->max_latency = 0;   

1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029
   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;
1030
   s->vc_occupancy = (int*)calloc(s->num_vcs, sizeof(int));
1031
   s->last_buf_full = 0.0;
1032 1033 1034 1035 1036 1037 1038

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


1039
   s->rank_tbl = NULL;
1040
   s->terminal_msgs = 
1041
       (terminal_custom_message_list**)calloc(s->num_vcs, sizeof(terminal_custom_message_list*));
1042
   s->terminal_msgs_tail = 
1043
       (terminal_custom_message_list**)calloc(s->num_vcs, sizeof(terminal_custom_message_list*));
1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054
   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 */
1055
void router_custom_setup(router_state * r, tw_lp * lp)
1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073
{
    
    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;

1074
    num_routers_per_mgrp = codes_mapping_get_lp_count (lp_group_name, 1, "modelnet_dragonfly_custom_router",
1075 1076 1077 1078 1079 1080 1081
            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);

1082
   r->router_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
1083
   r->group_id=r->router_id/p->num_routers;
1084 1085
    
   //printf("\n Local router id %d global id %d ", r->router_id, lp->gid);
1086 1087 1088

   r->fwd_events = 0;
   r->rev_events = 0;
1089 1090
   r->ross_rsample.fwd_events = 0;
   r->ross_rsample.rev_events = 0;
1091

1092

1093 1094 1095 1096 1097 1098 1099
   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));
1100
  
1101 1102 1103
   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));
1104
   r->pending_msgs = 
1105
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
1106
   r->pending_msgs_tail = 
1107
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
1108
   r->queued_msgs = 
1109
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
1110
   r->queued_msgs_tail = 
1111 1112
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
   r->queued_count = (int*)calloc(p->radix, sizeof(int));
1113
   r->last_buf_full = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
1114 1115
   r->busy_time = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
   r->busy_time_sample = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
1116

1117 1118 1119 1120 1121
   /* set up for ROSS stats sampling */
   r->link_traffic_ross_sample = (int64_t*)calloc(p->radix, sizeof(int64_t));
   r->busy_time_ross_sample = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
   if (g_st_model_stats)
       lp->model_types->mstat_sz = sizeof(tw_lpid) + (sizeof(int64_t) + sizeof(tw_stime)) * p->radix;
1122
   if (g_st_use_analysis_lps && g_st_model_stats)
1123 1124 1125 1126
       lp->model_types->sample_struct_sz = sizeof(struct dfly_router_sample) + (sizeof(tw_stime) + sizeof(int64_t)) * p->radix;
   r->ross_rsample.busy_time = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
   r->ross_rsample.link_traffic_sample = (int64_t*)calloc(p->radix, sizeof(int64_t));

1127
   rc_stack_create(&r->st);
1128 1129 1130 1131

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

1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144
   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;
1145 1146
    r->vc_occupancy[i] = (int*)calloc(p->num_vcs, sizeof(int));
    r->pending_msgs[i] = (terminal_custom_message_list**)calloc(p->num_vcs, 
1147
        sizeof(terminal_custom_message_list*));
1148
    r->last_buf_full[i] = 0.0;
1149
    r->pending_msgs_tail[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1150
        sizeof(terminal_custom_message_list*));
1151
    r->queued_msgs[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1152
        sizeof(terminal_custom_message_list*));
1153
    r->queued_msgs_tail[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1154
        sizeof(terminal_custom_message_list*));
1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167
        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 */
1168
static tw_stime dragonfly_custom_packet_event(
1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182
        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;
1183
    terminal_custom_message * msg;
1184 1185 1186 1187 1188 1189
    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,
1190
            sender, DRAGONFLY_CUSTOM, (void**)&msg, (void**)&tmp_ptr);
1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209
    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*/
      {
1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221
        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;
        }
1222 1223 1224 1225 1226 1227 1228
     }
	   //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 */
1229
static void dragonfly_custom_packet_event_rc(tw_lp *sender)
1230 1231 1232 1233 1234 1235
{
	  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*/
1236
static void router_credit_send(router_state * s, terminal_custom_message * msg, 
1237 1238 1239
  tw_lp * lp, int sq) {
  tw_event * buf_e;
  tw_stime ts;
1240
  terminal_custom_message * buf_msg;
1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251

  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;
1252 1253 1254 1255
  } else if(msg->last_hop == GLOBAL 
          || msg->last_hop == LOCAL
          || msg->last_hop == ROOT)
  {
1256 1257 1258 1259 1260 1261 1262
    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) {
1263
    buf_e = model_net_method_event_new(dest, ts, lp, DRAGONFLY_CUSTOM, 
1264 1265 1266
      (void**)&buf_msg, NULL);
    buf_msg->magic = terminal_magic_num;
  } else {
1267
    buf_e = model_net_method_event_new(dest, ts, lp, DRAGONFLY_CUSTOM_ROUTER,
1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285
            (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;
}

1286
static void packet_generate_rc(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, tw_lp * lp)
1287 1288 1289
{
   s->packet_gen--;
   packet_gen--;
1290 1291
   s->packet_counter--;

1292 1293 1294
   tw_rand_reverse_unif(lp->rng);

   int num_chunks = msg->packet_size/s->params->chunk_size;
1295
   if(msg->packet_size < s->params->chunk_size)
1296 1297 1298 1299
       num_chunks++;

   int i;
   for(i = 0; i < num_chunks; i++) {
1300
        delete_terminal_custom_message_list(return_tail(s->terminal_msgs, 
1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318
          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 */
1319
static void packet_generate(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, 
1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330
  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;
1331 1332
  double cn_delay = s->params->cn_delay;

1333
  if (msg->packet_size < s->params->chunk_size) 
1334 1335
      num_chunks++;

1336 1337 1338
  if(msg->packet_size < s->params->chunk_size)
      cn_delay = bytes_to_ns(msg->packet_size % s->params->chunk_size, s->params->cn_bandwidth);

1339 1340 1341 1342 1343
  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)
1344 1345 1346 1347 1348 1349
  {
      if(dest_router_id == s->router_id)
          num_local_packets_sr++;
      else
          num_local_packets_sg++;
  }
1350
  else
1351
  {
1352
      num_remote_packets++;
1353
  }
1354
  nic_ts = g_tw_lookahead + (num_chunks * cn_delay) + tw_rand_unif(lp->rng);
1355
  
1356
  msg->packet_ID = s->packet_counter;
1357
  s->packet_counter++;
1358 1359 1360 1361 1362
  msg->my_N_hop = 0;
  msg->my_l_hop = 0;
  msg->my_g_hop = 0;


1363
  for(int i = 0; i < num_chunks; i++)
1364
  {
1365
    terminal_custom_message_list *cur_chunk = (terminal_custom_message_list*)calloc(1,
1366
      sizeof(terminal_custom_message_list));
1367
    msg->origin_router_id = s->router_id;
1368
    init_terminal_custom_message_list(cur_chunk, msg);
1369 1370
  
    if(msg->remote_event_size_bytes + msg->local_event_size_bytes > 0) {
1371
      cur_chunk->event_data = (char*)calloc(1,
1372 1373 1374
          msg->remote_event_size_bytes + msg->local_event_size_bytes);
    }
    
1375
    void * m_data_src = model_net_method_get_edata(DRAGONFLY_CUSTOM, msg);
1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386
    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;
1387
    append_to_terminal_custom_message_list(s->terminal_msgs, s->terminal_msgs_tail,
1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401
      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);
1402
    terminal_custom_message *m;
1403
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, DRAGONFLY_CUSTOM, 
1404 1405 1406 1407 1408 1409 1410
      (void**)&m, NULL);
    m->type = T_SEND;
    m->magic = terminal_magic_num;
    s->in_send_loop = 1;
    tw_event_send(e);
  }

1411
  total_event_size = model_net_get_msg_sz(DRAGONFLY_CUSTOM) + 
1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423
      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;
}

1424
static void packet_send_rc(terminal_state * s, tw_bf * bf, terminal_custom_message * msg,
1425 1426
        tw_lp * lp)
{
1427
      
1428 1429
      if(bf->c1) {
        s->in_send_loop = 1;
1430 1431 1432 1433
        if(bf->c3)
        {
         s->last_buf_full = msg->saved_busy_time;
        }
1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445
        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;

1446
      terminal_custom_message_list* cur_entry = (terminal_custom_message_list *)rc_stack_pop(s->st);
1447

1448
      prepend_to_terminal_custom_message_list(s->terminal_msgs, 
1449 1450 1451 1452 1453 1454 1455
              s->terminal_msgs_tail, 0, cur_entry);
      if(bf->c3) {
        tw_rand_reverse_unif(lp->rng);
      }
      if(bf->c4) {
        s->in_send_loop = 1;
      }
1456
      if(bf->c5)
1457
      {
1458
          tw_rand_reverse_unif(lp->rng);
1459
          s->issueIdle = 1;
1460 1461 1462
          if(bf->c6)
          {
            s->busy_time = msg->saved_total_time;
1463
            s->last_buf_full = msg->saved_busy_time;
1464
            s->busy_time_sample = msg->saved_sample_time;
1465 1466
            s->ross_sample.busy_time_sample = msg->saved_sample_time;
            s->busy_time_ross_sample = msg->saved_busy_time_ross;
1467
          }
1468
      }
1469 1470 1471
      return;
}
/* sends the packet from the current dragonfly compute node to the attached router */
1472
static void packet_send(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, 
1473 1474 1475 1476
  tw_lp * lp) {
  
  tw_stime ts;
  tw_event *e;
1477
  terminal_custom_message *m;
1478 1479
  tw_lpid router_id;

1480
  terminal_custom_message_list* cur_entry = s->terminal_msgs[0];
1481
  bool noEmptyVC = false;
1482

1483
  if(s->vc_occupancy[0] + s->params->chunk_size > s->params->cn_vc_size)
1484
      noEmptyVC = true;
1485

1486 1487 1488 1489
  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;
1490 1491 1492 1493 1494 1495
    if(noEmptyVC && !s->last_buf_full)
    {
        bf->c3 = 1;
        msg->saved_busy_time = s->last_buf_full;
        s->last_buf_full = tw_now(lp); 
    }
1496 1497 1498 1499
    return;
  }

  uint64_t num_chunks = cur_entry->msg.packet_size/s->params->chunk_size;
1500
  if(cur_entry->msg.packet_size < s->params->chunk_size)
1501 1502 1503
    num_chunks++;

  tw_stime delay = s->params->cn_delay;
1504
  if((cur_entry->msg.packet_size < s->params->chunk_size) && (cur_entry->msg.chunk_id == num_chunks - 1))
1505 1506 1507 1508 1509 1510 1511
       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;

1512
  ts = s->terminal_available_time - tw_now(lp);
1513 1514
  codes_mapping_get_lp_info(lp->gid, lp_group_name, &mapping_grp_id, NULL,
      &mapping_type_id, NULL, &mapping_rep_id, &mapping_offset);
1515
  codes_mapping_get_lp_id(lp_group_name, LP_CONFIG_NM_ROUT, NULL, 0,