dragonfly-custom.C 129 KB
Newer Older
Nikhil's avatar
Nikhil committed
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"
Nikhil's avatar
Nikhil committed
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>
Nikhil's avatar
Nikhil committed
22

23 24 25 26
#ifdef ENABLE_CORTEX
#include <cortex/cortex.h>
#include <cortex/topology.h>
#endif
Nikhil's avatar
Nikhil committed
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])
Nikhil's avatar
Nikhil committed
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;
Nikhil's avatar
Nikhil committed
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)*/
Nikhil's avatar
Nikhil committed
66
vector< map< int, vector<Link> > > intraGroupLinks;
67 68
/* contains mapping between source router and destination group via link (link
 * has dest ID)*/
Nikhil's avatar
Nikhil committed
69
vector< map< int, vector<bLink> > > interGroupLinks;
70
/*MM: Maintains a list of routers connecting the source and destination groups */
Nikhil's avatar
Nikhil committed
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
Nikhil's avatar
Nikhil committed
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;
Nikhil's avatar
Nikhil committed
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;
Nikhil's avatar
Nikhil committed
111 112

/* terminal magic number */
113
static int terminal_magic_num = 0;
Nikhil's avatar
Nikhil committed
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;
Nikhil's avatar
Nikhil committed
120

121 122
static int sample_bytes_written = 0;
static int sample_rtr_bytes_written = 0;
Nikhil's avatar
Nikhil committed
123

124 125
static char cn_sample_file[MAX_NAME_LENGTH];
static char router_sample_file[MAX_NAME_LENGTH];
Nikhil's avatar
Nikhil committed
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;
Nikhil's avatar
Nikhil committed
133
    char* event_data;
134 135
    terminal_custom_message_list *next;
    terminal_custom_message_list *prev;
Nikhil's avatar
Nikhil committed
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;
Nikhil's avatar
Nikhil committed
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);
Nikhil's avatar
Nikhil committed
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;
Nikhil's avatar
Nikhil committed
171 172 173 174 175 176 177 178 179 180 181 182
    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;
};

183 184 185
static const dragonfly_param* stored_params;


Nikhil's avatar
Nikhil committed
186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243
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;
244 245
   terminal_custom_message_list **terminal_msgs;
   terminal_custom_message_list **terminal_msgs_tail;
Nikhil's avatar
Nikhil committed
246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265
   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;

266
   tw_stime last_buf_full;
Nikhil's avatar
Nikhil committed
267
   tw_stime busy_time;
268 269 270 271
   
   tw_stime max_latency;
   tw_stime min_latency;

Nikhil's avatar
Nikhil committed
272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
   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;
287
  
Nikhil's avatar
Nikhil committed
288 289 290 291 292 293
   /* for logging forward and reverse events */
   long fwd_events;
   long rev_events;
};

/* terminal event type (1-4) */
294
typedef enum event_t
Nikhil's avatar
Nikhil committed
295 296 297 298 299 300 301 302
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  R_SEND,
  R_ARRIVE,
  R_BUFFER,
303
} event_t;
Nikhil's avatar
Nikhil committed
304 305 306 307

/* whether the last hop of a packet was global, local or a terminal */
enum last_hop
{
308
   GLOBAL=1,
Nikhil's avatar
Nikhil committed
309
   LOCAL,
310 311
   TERMINAL,
   ROOT
Nikhil's avatar
Nikhil committed
312 313 314 315 316 317 318
};

/* 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
{
319
    MINIMAL = 1,
Nikhil's avatar
Nikhil committed
320 321 322 323 324
    NON_MINIMAL,
    ADAPTIVE,
    PROG_ADAPTIVE
};

325 326 327 328 329
enum LINK_TYPE
{
    GREEN,
    BLACK,
};
Nikhil's avatar
Nikhil committed
330 331 332 333 334 335 336 337 338 339 340
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;
341
   tw_stime* last_buf_full;
Nikhil's avatar
Nikhil committed
342 343 344 345

   tw_stime* busy_time;
   tw_stime* busy_time_sample;

346 347 348 349
   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;
Nikhil's avatar
Nikhil committed
350 351 352
   int *in_send_loop;
   int *queued_count;
   struct rc_stack * st;
353 354

   int* last_sent_chan;
Nikhil's avatar
Nikhil committed
355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402
   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;
403 404 405 406
    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);
Nikhil's avatar
Nikhil committed
407 408 409
    key = key * 21;
    key = ~key ^ (tmp->sender_id >> 4);
    key = key * tmp->sender_id; 
410
    return (int)(key & (table_size - 1));*/
Nikhil's avatar
Nikhil committed
411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428
}

/* 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 */
429
int dragonfly_custom_get_msg_sz(void)
Nikhil's avatar
Nikhil committed
430
{
431
	   return sizeof(terminal_custom_message);
Nikhil's avatar
Nikhil committed
432 433 434 435
}

static void free_tmp(void * ptr)
{
436
    struct dfly_qhash_entry * dfly = (dfly_qhash_entry *)ptr; 
437 438 439 440 441
    if(dfly->remote_event_data)
        free(dfly->remote_event_data);
   
    if(dfly)
        free(dfly);
Nikhil's avatar
Nikhil committed
442
}
443

444 445 446
static void append_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
447
        int index, 
448
        terminal_custom_message_list *msg) {
Nikhil's avatar
Nikhil committed
449 450 451 452 453 454 455 456 457
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
    } 
    thistail[index] = msg;
}

458 459 460
static void prepend_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
461
        int index, 
462
        terminal_custom_message_list *msg) {
Nikhil's avatar
Nikhil committed
463 464 465 466 467 468 469 470 471
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
    } 
    thisq[index] = msg;
}

472 473 474
static terminal_custom_message_list* return_head(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
475
        int index) {
476
    terminal_custom_message_list *head = thisq[index];
Nikhil's avatar
Nikhil committed
477 478 479 480 481 482 483 484 485 486 487 488
    if(head != NULL) {
        thisq[index] = head->next;
        if(head->next != NULL) {
            head->next->prev = NULL;
            head->next = NULL;
        } else {
            thistail[index] = NULL;
        }
    }
    return head;
}

489 490 491
static terminal_custom_message_list* return_tail(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
492
        int index) {
493
    terminal_custom_message_list *tail = thistail[index];
Nikhil's avatar
Nikhil committed
494 495 496 497 498 499 500 501 502 503 504 505
    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;
}

506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541
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);
        printf("------------------------------------------------------\n\n");
    }
}

Nikhil's avatar
Nikhil committed
542
static void dragonfly_read_config(const char * anno, dragonfly_param *params){
543 544 545 546 547 548 549 550
    /*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;
    
Nikhil's avatar
Nikhil committed
551 552
    // shorthand
    dragonfly_param *p = params;
Nikhil's avatar
Nikhil committed
553
    int myRank;
554
    MPI_Comm_rank(MPI_COMM_CODES, &myRank);
Nikhil's avatar
Nikhil committed
555

556
    int rc = configuration_get_value_int(&config, "PARAMS", "local_vc_size", anno, &p->local_vc_size);
Nikhil's avatar
Nikhil committed
557 558 559 560 561
    if(rc) {
        p->local_vc_size = 1024;
        fprintf(stderr, "Buffer size of local channels not specified, setting to %d\n", p->local_vc_size);
    }

562 563 564
    rc = configuration_get_value_int(&config, "PARAMS", "adaptive_threshold", anno, &adaptive_threshold);
    if(rc) {
    	adaptive_threshold = p->local_vc_size / 8;
565
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
566
	}
567 568 569 570
    else
    {
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
    }
571

Nikhil's avatar
Nikhil committed
572 573 574 575 576 577
    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);
    }

578 579 580 581
    rc = configuration_get_value_int(&config, "PARAMS", "df-dally-vc", anno, &DF_DALLY);
    if(rc) {
        DF_DALLY = 0;
    }
582 583 584 585 586
    
    rc = configuration_get_value_int(&config, "PARAMS", "minimal-bias", anno, &BIAS_MIN);
    if(rc) {
        BIAS_MIN = 0;
    }
587 588 589
    else
	printf("\n Setting minimal bias");

Nikhil's avatar
Nikhil committed
590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619
    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);
    }

620
    rc = configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
Nikhil's avatar
Nikhil committed
621
            &p->router_delay);
622
    if(rc) {
623 624
      p->router_delay = 100;
    }
Nikhil's avatar
Nikhil committed
625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641

    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)
Nikhil's avatar
Nikhil committed
642
	      routing = PROG_ADAPTIVE;
Nikhil's avatar
Nikhil committed
643 644 645 646 647 648 649
    else
    {
        fprintf(stderr, 
                "No routing protocol specified, setting to minimal routing\n");
        routing = -1;
    }

650 651 652 653 654 655 656 657 658 659
    // 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);
    // }
660 661 662
   
if(DF_DALLY == 0) 
{
663 664 665
    //if(routing == PROG_ADAPTIVE)
    //    p->num_vcs = 10;
    //else
666
        p->num_vcs = 8;
667 668 669
}
else
{
670
        p->num_vcs = 4;
671
}
Nikhil's avatar
Nikhil committed
672 673 674
    rc = configuration_get_value_int(&config, "PARAMS", "num_groups", anno, &p->num_groups);
    if(rc) {
      printf("Number of groups not specified. Aborting");
675
      MPI_Abort(MPI_COMM_CODES, 1);
Nikhil's avatar
Nikhil committed
676
    }
677 678 679 680 681
    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;
    }
682 683 684 685 686
    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;
    }
687 688 689 690 691 692 693 694 695 696
    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;
    }
697 698 699 700
    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);

701 702
    p->num_routers = p->num_router_rows * p->num_router_cols;
    
703
    rc = configuration_get_value_int(&config, "PARAMS", "num_cns_per_router", anno, &p->num_cn);
Nikhil's avatar
Nikhil committed
704
    if(rc) {
705 706
        printf("\n Number of cns per router not specified, setting to %d ", p->num_routers/2);
        p->num_cn = p->num_routers/2;
Nikhil's avatar
Nikhil committed
707
    }
708

709 710
    rc = configuration_get_value_int(&config, "PARAMS", "num_global_channels", anno, &p->num_global_channels);
    if(rc) {
711 712
        printf("\n Number of global channels per router not specified, setting to 10 ");
        p->num_global_channels = 10;
713
    }
714
    p->radix = p->intra_grp_radix + p->num_global_channels + p->num_cn;
Nikhil's avatar
Nikhil committed
715 716
    p->total_routers = p->num_groups * p->num_routers;
    p->total_terminals = p->total_routers * p->num_cn;
Nikhil's avatar
Nikhil committed
717 718 719 720
    
    // read intra group connections, store from a router's perspective
    // all links to the same router form a vector
    char intraFile[MAX_NAME_LENGTH];
721
    configuration_get_value(&config, "PARAMS", "intra-group-connections", 
Nikhil's avatar
Nikhil committed
722
        anno, intraFile, MAX_NAME_LENGTH);
723 724
    if(strlen(intraFile) <= 0) {
      tw_error(TW_LOC, "Intra group connections file not specified. Aborting");
Nikhil's avatar
Nikhil committed
725 726
    }
    FILE *groupFile = fopen(intraFile, "rb");
727 728 729
    if(!groupFile)
        tw_error(TW_LOC, "intra-group file not found ");

Nikhil's avatar
Nikhil committed
730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751
    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];
752
    configuration_get_value(&config, "PARAMS", "inter-group-connections", 
Nikhil's avatar
Nikhil committed
753
        anno, interFile, MAX_NAME_LENGTH);
754 755
    if(strlen(interFile) <= 0) {
      tw_error(TW_LOC, "Inter group connections file not specified. Aborting");
Nikhil's avatar
Nikhil committed
756 757 758
    }
    FILE *systemFile = fopen(interFile, "rb");
    if(!myRank)
759
    {
Nikhil's avatar
Nikhil committed
760
      printf("Reading inter-group connectivity file: %s\n", interFile);
761 762
      printf("\n Total routers %d total groups %d ", p->total_routers, p->num_groups);
    }
Nikhil's avatar
Nikhil committed
763 764 765 766 767 768 769 770 771

    {
      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);
      }
772
      
Nikhil's avatar
Nikhil committed
773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793
      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);

794
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
795 796 797 798 799 800 801 802
    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++) {
803
          // offset is number of local connections
804
          // type is black or green according to Cray architecture 
Nikhil's avatar
Nikhil committed
805 806 807 808 809 810 811
          printf("%d,%d ", it->second[l].offset, it->second[l].type);
        }
        printf(")");
      }
      printf("\n");
    }
#endif
812
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
813 814 815
    printf("Dumping inter-group connections\n");
    for(int a = 0; a < interGroupLinks.size(); a++) {
      printf("Connections for router %d\n", a);
816 817
      map< int, vector<bLink> >  &curMap = interGroupLinks[a];
      map< int, vector<bLink> >::iterator it = curMap.begin();
Nikhil's avatar
Nikhil committed
818
      for(; it != curMap.end(); it++) {
819
        // dest group ID 
Nikhil's avatar
Nikhil committed
820 821
        printf(" ( %d - ", it->first);
        for(int l = 0; l < it->second.size(); l++) {
822 823
            // dest is dest router ID
            // offset is number of global connections
Nikhil's avatar
Nikhil committed
824 825 826 827 828 829 830 831
          printf("%d,%d ", it->second[l].offset, it->second[l].dest);
        }
        printf(")");
      }
      printf("\n");
    }
#endif

832
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
833 834 835 836 837 838 839 840 841 842 843 844 845
    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) {
846
        printf("\n Total nodes %d routers %d groups %d routers per group %d radix %d\n",
Nikhil's avatar
Nikhil committed
847
                p->num_cn * p->total_routers, p->total_routers, p->num_groups,
848
                p->num_routers, p->radix);
Nikhil's avatar
Nikhil committed
849
    }
Nikhil's avatar
Nikhil committed
850

Nikhil's avatar
Nikhil committed
851 852 853
    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);
854
    p->credit_delay = bytes_to_ns(CREDIT_SIZE, p->local_bandwidth); //assume 8 bytes packet
855 856 857 858 859

    if (PRINT_CONFIG) 
        dragonfly_print_params(p);

    stored_params = p;
Nikhil's avatar
Nikhil committed
860 861
}

862
void dragonfly_custom_configure(){
Nikhil's avatar
Nikhil committed
863 864 865
    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);
866
    all_params = (dragonfly_param *)calloc(num_params, sizeof(*all_params));
Nikhil's avatar
Nikhil committed
867 868 869 870 871 872 873 874

    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]);
    }
875 876 877
#ifdef ENABLE_CORTEX
	model_net_topology = dragonfly_custom_cortex_topology;
#endif
Nikhil's avatar
Nikhil committed
878 879 880
}

/* report dragonfly statistics like average and maximum packet latency, average number of hops traversed */
881
void dragonfly_custom_report_stats()
Nikhil's avatar
Nikhil committed
882 883 884 885 886 887
{
   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;
888
   long total_local_packets_sr, total_local_packets_sg, total_remote_packets;
Nikhil's avatar
Nikhil committed
889

890 891 892 893 894 895 896
   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);
Nikhil's avatar
Nikhil committed
897
   
898
   MPI_Reduce( &packet_gen, &total_gen, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
899
   MPI_Reduce(&packet_fin, &total_fin, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
900 901
    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);
902
   MPI_Reduce( &num_remote_packets, &total_remote_packets, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
903
   if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
Nikhil's avatar
Nikhil committed
904
    {
905 906
	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);
Nikhil's avatar
Nikhil committed
907 908 909 910 911
    }

   /* print statistics */
   if(!g_tw_mynode)
   {	
912 913 914
    if (PRINT_CONFIG) 
        dragonfly_print_params(stored_params);

Nikhil's avatar
Nikhil committed
915 916
      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);
917
     if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
Nikhil's avatar
Nikhil committed
918 919 920
              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);
 
921
      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);
Nikhil's avatar
Nikhil committed
922 923 924 925 926 927 928
   }
   return;
}


/* initialize a dragonfly compute node terminal */
void 
929
terminal_custom_init( terminal_state * s, 
Nikhil's avatar
Nikhil committed
930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954
	       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);

955
   s->terminal_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
956
   s->router_id=(int)s->terminal_id / (s->params->num_cn);
Nikhil's avatar
Nikhil committed
957 958
   s->terminal_available_time = 0.0;
   s->packet_counter = 0;
959 960 961
   s->min_latency = INT_MAX;
   s->max_latency = 0;   

Nikhil's avatar
Nikhil committed
962 963 964 965 966 967 968 969 970 971 972 973 974
   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;
975
   s->vc_occupancy = (int*)calloc(s->num_vcs, sizeof(int));
976
   s->last_buf_full = 0.0;
Nikhil's avatar
Nikhil committed
977 978 979 980 981 982 983

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


984
   s->rank_tbl = NULL;
Nikhil's avatar
Nikhil committed
985
   s->terminal_msgs = 
986
       (terminal_custom_message_list**)calloc(s->num_vcs, sizeof(terminal_custom_message_list*));
Nikhil's avatar
Nikhil committed
987
   s->terminal_msgs_tail = 
988
       (terminal_custom_message_list**)calloc(s->num_vcs, sizeof(terminal_custom_message_list*));
Nikhil's avatar
Nikhil committed
989 990 991 992 993 994 995 996 997 998 999
   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 */
1000
void router_custom_setup(router_state * r, tw_lp * lp)
Nikhil's avatar
Nikhil committed
1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018
{
    
    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;

1019
    num_routers_per_mgrp = codes_mapping_get_lp_count (lp_group_name, 1, "modelnet_dragonfly_custom_router",
Nikhil's avatar
Nikhil committed
1020 1021 1022 1023 1024 1025 1026
            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);

1027
   r->router_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
Nikhil's avatar
Nikhil committed
1028
   r->group_id=r->router_id/p->num_routers;
1029 1030
    
   //printf("\n Local router id %d global id %d ", r->router_id, lp->gid);
Nikhil's avatar
Nikhil committed
1031 1032 1033 1034

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

1035

1036 1037 1038 1039 1040 1041 1042
   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));
1043
  
1044 1045 1046
   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));
Nikhil's avatar
Nikhil committed
1047
   r->pending_msgs = 
1048
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
Nikhil's avatar
Nikhil committed
1049
   r->pending_msgs_tail = 
1050
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
Nikhil's avatar
Nikhil committed
1051
   r->queued_msgs = 
1052
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
Nikhil's avatar
Nikhil committed
1053
   r->queued_msgs_tail = 
1054 1055
    (terminal_custom_message_list***)calloc(p->radix, sizeof(terminal_custom_message_list**));
   r->queued_count = (int*)calloc(p->radix, sizeof(int));
1056
   r->last_buf_full = (tw_stime*)calloc(p->radix, sizeof(tw_stime*));
1057 1058
   r->busy_time = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
   r->busy_time_sample = (tw_stime*)calloc(p->radix, sizeof(tw_stime));
Nikhil's avatar
Nikhil committed
1059 1060

   rc_stack_create(&r->st);
1061 1062 1063 1064

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

Nikhil's avatar
Nikhil committed
1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077
   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;
1078 1079
    r->vc_occupancy[i] = (int*)calloc(p->num_vcs, sizeof(int));
    r->pending_msgs[i] = (terminal_custom_message_list**)calloc(p->num_vcs, 
1080
        sizeof(terminal_custom_message_list*));
1081
    r->last_buf_full[i] = 0.0;
1082
    r->pending_msgs_tail[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1083
        sizeof(terminal_custom_message_list*));
1084
    r->queued_msgs[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1085
        sizeof(terminal_custom_message_list*));
1086
    r->queued_msgs_tail[i] = (terminal_custom_message_list**)calloc(p->num_vcs,
1087
        sizeof(terminal_custom_message_list*));
Nikhil's avatar
Nikhil committed
1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100
        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 */
1101
static tw_stime dragonfly_custom_packet_event(
Nikhil's avatar
Nikhil committed
1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115
        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;
1116
    terminal_custom_message * msg;
Nikhil's avatar
Nikhil committed
1117 1118 1119 1120 1121 1122
    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,
1123
            sender, DRAGONFLY_CUSTOM, (void**)&msg, (void**)&tmp_ptr);
Nikhil's avatar
Nikhil committed
1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142
    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*/
      {
1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154
        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;
        }
Nikhil's avatar
Nikhil committed
1155 1156 1157 1158 1159 1160 1161
     }
	   //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 */
1162
static void dragonfly_custom_packet_event_rc(tw_lp *sender)
Nikhil's avatar
Nikhil committed
1163 1164 1165 1166 1167 1168
{
	  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*/
1169
static void router_credit_send(router_state * s, terminal_custom_message * msg, 
Nikhil's avatar
Nikhil committed
1170 1171 1172
  tw_lp * lp, int sq) {
  tw_event * buf_e;
  tw_stime ts;
1173
  terminal_custom_message * buf_msg;
Nikhil's avatar
Nikhil committed
1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184

  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;
1185 1186 1187 1188
  } else if(msg->last_hop == GLOBAL 
          || msg->last_hop == LOCAL
          || msg->last_hop == ROOT)
  {
Nikhil's avatar
Nikhil committed
1189 1190 1191 1192 1193 1194 1195
    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) {
1196
    buf_e = model_net_method_event_new(dest, ts, lp, DRAGONFLY_CUSTOM, 
Nikhil's avatar
Nikhil committed
1197 1198 1199
      (void**)&buf_msg, NULL);
    buf_msg->magic = terminal_magic_num;
  } else {
1200
    buf_e = model_net_method_event_new(dest, ts, lp, DRAGONFLY_CUSTOM_ROUTER,
Nikhil's avatar
Nikhil committed
1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218
            (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;
}

1219
static void packet_generate_rc(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, tw_lp * lp)
Nikhil's avatar
Nikhil committed
1220 1221 1222
{
   s->packet_gen--;
   packet_gen--;
1223 1224
   s->packet_counter--;

Nikhil's avatar
Nikhil committed
1225 1226 1227
   tw_rand_reverse_unif(lp->rng);

   int num_chunks = msg->packet_size/s->params->chunk_size;
1228
   if(msg->packet_size < s->params->chunk_size)
Nikhil's avatar
Nikhil committed
1229 1230 1231 1232
       num_chunks++;

   int i;
   for(i = 0; i < num_chunks; i++) {
1233
        delete_terminal_custom_message_list(return_tail(s->terminal_msgs, 
Nikhil's avatar
Nikhil committed
1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251
          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 */
1252
static void packet_generate(terminal_state * s, tw_bf * bf, terminal_custom_message * msg, 
Nikhil's avatar
Nikhil committed
1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263
  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;
1264 1265
  double cn_delay = s->params->cn_delay;

1266
  if (msg->packet_size < s->params->chunk_size) 
Nikhil's avatar
Nikhil committed
1267 1268
      num_chunks++;

1269 1270 1271
  if(msg->packet_size < s->params->chunk_size)
      cn_delay = bytes_to_ns(msg->packet_size % s->params->chunk_size, s->params->cn_bandwidth);

1272 1273 1274 1275 1276
  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)
1277 1278 1279 1280 1281 1282
  {
      if(dest_router_id == s->router_id)
          num_local_packets_sr++;
      else
          num_local_packets_sg++;
  }
1283
  else
1284
  {
1285
      num_remote_packets++;
1286
  }
1287
  nic_ts = g_tw_lookahead + (num_chunks * cn_delay) + tw_rand_unif(lp->rng);
Nikhil's avatar
Nikhil committed
1288
  
1289
  msg->packet_ID = s->packet_counter;
1290
  s->packet_counter++;
Nikhil's avatar
Nikhil committed
1291 1292 1293 1294 1295
  msg->my_N_hop = 0;
  msg->my_l_hop = 0;
  msg->my_g_hop = 0;


1296
  for(int i = 0; i < num_chunks; i++)
Nikhil's avatar
Nikhil committed
1297
  {
1298
    terminal_custom_message_list *cur_chunk = (terminal_custom_message_list*)calloc(1,
1299
      sizeof(terminal_custom_message_list));
Nikhil's avatar
Nikhil committed
1300
    msg->origin_router_id = s->router_id;
1301
    init_terminal_custom_message_list(cur_chunk, msg);
Nikhil's avatar
Nikhil committed
1302 1303
  
    if(msg->remote_event_size_bytes + msg->local_event_size_bytes > 0) {
1304
      cur_chunk->event_data = (char*)calloc(1,
Nikhil's avatar
Nikhil committed
1305 1306 1307
          msg->remote_event_size_bytes + msg->local_event_size_bytes);
    }
    
1308
    void * m_data_src = model_net_method_get_edata(DRAGONFLY_CUSTOM, msg);
Nikhil's avatar
Nikhil committed
1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319
    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;
1320
    append_to_terminal_custom_message_list(s->terminal_msgs, s->terminal_msgs_tail,
Nikhil's avatar
Nikhil committed
1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334
      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);
1335
    terminal_custom_message *m;
1336
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, DRAGONFLY_CUSTOM, 
Nikhil's avatar
Nikhil committed
1337 1338 1339 1340 1341 1342 1343
      (void**)&m, NULL);
    m->type = T_SEND;
    m->magic = terminal_magic_num;
    s->in_send_loop = 1;
    tw_event_send(e);
  }

1344
  total_event_size = model_net_get_msg_sz(DRAGONFLY_CUSTOM) + 
Nikhil's avatar
Nikhil committed
1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356
      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;
}

1357
static void packet_send_rc(terminal_state * s, tw_bf * bf, terminal_custom_message * msg,
Nikhil's avatar
Nikhil committed
1358 1359
        tw_lp * lp)
{
1360
      
Nikhil's avatar
Nikhil committed
1361 1362
      if(bf->c1) {
        s->in_send_loop = 1;
1363 1364 1365 1366
        if(bf->c3)
        {
         s->last_buf_full = msg->saved_busy_time;
        }
Nikhil's avatar
Nikhil committed
1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378
        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;

1379
      terminal_custom_message_list* cur_entry = (terminal_custom_message_list *)rc_stack_pop(s->st);
Nikhil's avatar
Nikhil committed
1380

1381
      prepend_to_terminal_custom_message_list(s->terminal_msgs, 
Nikhil's avatar
Nikhil committed
1382 1383 1384 1385 1386 1387 1388
              s->terminal_msgs_tail, 0, cur_entry);
      if(bf->c3) {
        tw_rand_reverse_unif(lp->rng);
      }
      if(bf->c4) {
        s->in_send_loop = 1;
      }
Misbah Mubarak's avatar