dragonfly-custom.C 173 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
31
#define DFLY_HASH_TABLE_SIZE 4999
32
// debugging parameters
33
#define BW_MONITOR 1
34
#define DEBUG_LP 892
35
#define T_ID -1
36
#define TRACK -1
37
#define TRACK_PKT -1
38 39 40
#define TRACK_MSG -1
#define DEBUG 0
#define MAX_STATS 65536
41
#define SHOW_ADAP_STATS 1
42 43 44 45 46

#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
47

48
static int debug_cnt = 0;
49 50 51 52 53
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;

54
static int BIAS_MIN = 1;
55
static int DF_DALLY = 0;
56
static int adaptive_threshold = 1024;
57

Neil McGlohon's avatar
Neil McGlohon committed
58
static tw_stime max_qos_monitor = 100000000;
59 60 61
static long num_local_packets_sr = 0;
static long num_local_packets_sg = 0;
static long num_remote_packets = 0;
62 63 64 65

/* time in nanosecs */
static int bw_reset_window = 5000000;

66 67 68 69 70 71
#define indexer3d(_ptr, _x, _y, _z, _maxx, _maxy, _maxz) \
        ((_ptr) + _z * (_maxx * _maxz) + _y * (_maxx) + _x)

#define indexer2d(_ptr, _x, _y, _maxx, _maxy) \
        ((_ptr) + _y * (_maxx) + _x)

Nikhil's avatar
Nikhil committed
72 73 74 75 76 77 78
using namespace std;
struct Link {
  int offset, type;
};
struct bLink {
  int offset, dest;
};
79 80 81 82 83
/* 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
84
vector< map< int, vector<Link> > > intraGroupLinks;
85 86
/* contains mapping between source router and destination group via link (link
 * has dest ID)*/
Nikhil's avatar
Nikhil committed
87
vector< map< int, vector<bLink> > > interGroupLinks;
88
/*MM: Maintains a list of routers connecting the source and destination groups */
Nikhil's avatar
Nikhil committed
89 90 91 92 93 94 95 96 97 98
vector< vector< vector<int> > > connectionList;

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

struct InterGroupLink {
  int src, dest;
};

99 100 101 102 103 104
#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
105

106 107 108
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
109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127

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 */
128
static int router_magic_num = 0;
Nikhil's avatar
Nikhil committed
129 130

/* terminal magic number */
131
static int terminal_magic_num = 0;
Nikhil's avatar
Nikhil committed
132

133 134 135 136
/* Hops within a group */
static int num_intra_nonmin_hops = 4;
static int num_intra_min_hops = 2;

137
static FILE * dragonfly_rtr_bw_log = NULL;
138
//static FILE * dragonfly_term_bw_log = NULL;
Nikhil's avatar
Nikhil committed
139

140 141
static int sample_bytes_written = 0;
static int sample_rtr_bytes_written = 0;
Nikhil's avatar
Nikhil committed
142

143 144
static char cn_sample_file[MAX_NAME_LENGTH];
static char router_sample_file[MAX_NAME_LENGTH];
Nikhil's avatar
Nikhil committed
145

146 147
//don't do overhead here - job of MPI layer
static tw_stime mpi_soft_overhead = 0;
148

149 150 151
typedef struct terminal_custom_message_list terminal_custom_message_list;
struct terminal_custom_message_list {
    terminal_custom_message msg;
Nikhil's avatar
Nikhil committed
152
    char* event_data;
153 154
    terminal_custom_message_list *next;
    terminal_custom_message_list *prev;
Nikhil's avatar
Nikhil committed
155 156
};

157
static void init_terminal_custom_message_list(terminal_custom_message_list *thisO, 
158
    terminal_custom_message *inmsg) {
159 160 161 162
    thisO->msg = *inmsg;
    thisO->event_data = NULL;
    thisO->next = NULL;
    thisO->prev = NULL;
Nikhil's avatar
Nikhil committed
163 164
}

165 166 167 168
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
169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184
}

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;
185 186
    int intra_grp_radix;
    int num_col_chans;
187
    int num_row_chans;
188 189
    int num_router_rows;
    int num_router_cols;
Nikhil's avatar
Nikhil committed
190 191 192 193 194
    int num_groups;
    int radix;
    int total_routers;
    int total_terminals;
    int num_global_channels;
195 196
    int num_qos_levels;
    int * qos_bandwidths;
Nikhil's avatar
Nikhil committed
197 198 199 200 201
    double cn_delay;
    double local_delay;
    double global_delay;
    double credit_delay;
    double router_delay;
202 203

    int max_hops_notify; //maximum number of hops allowed before notifying via printout
Nikhil's avatar
Nikhil committed
204 205
};

206 207 208
static const dragonfly_param* stored_params;


Nikhil's avatar
Nikhil committed
209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265
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
   tw_stime terminal_available_time;
266 267
   terminal_custom_message_list **terminal_msgs;
   terminal_custom_message_list **terminal_msgs_tail;
Nikhil's avatar
Nikhil committed
268 269 270
   int in_send_loop;
   struct mn_stats dragonfly_stats_array[CATEGORY_MAX];

271 272
   int * qos_status;
   int * qos_data;
273 274

   int rc_index;
275 276
   int* last_qos_status;
   int* last_qos_data;
277

278 279 280
   int last_qos_lvl;
   int is_monitoring_bw;

Nikhil's avatar
Nikhil committed
281 282
   struct rc_stack * st;
   int issueIdle;
283
   int* terminal_length;
Nikhil's avatar
Nikhil committed
284 285 286 287 288 289 290 291 292 293 294 295 296 297

   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;

298
   tw_stime last_buf_full;
Nikhil's avatar
Nikhil committed
299
   tw_stime busy_time;
300 301 302 303
   
   tw_stime max_latency;
   tw_stime min_latency;

Nikhil's avatar
Nikhil committed
304
   char output_buf[4096];
305
   char output_buf2[4096];
Nikhil's avatar
Nikhil committed
306 307 308 309 310 311 312 313 314
   /* 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;
315
   int num_term_rc_windows;
Nikhil's avatar
Nikhil committed
316 317 318 319 320

   char sample_buf[4096];
   struct dfly_cn_sample * sample_stat;
   int op_arr_size;
   int max_arr_size;
321
  
Nikhil's avatar
Nikhil committed
322 323 324
   /* for logging forward and reverse events */
   long fwd_events;
   long rev_events;
325 326 327 328 329 330 331 332

   /* 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;
Nikhil's avatar
Nikhil committed
333 334
};

335 336 337 338 339 340 341 342 343 344 345 346 347
typedef enum qos_priority
{
    Q_HIGH =0,
    Q_MEDIUM,
    Q_LOW,
    Q_UNKNOWN,
} qos_priority;

typedef enum qos_status
{
    Q_ACTIVE = 1,
    Q_OVERBW,
} qos_status;
Nikhil's avatar
Nikhil committed
348
/* terminal event type (1-4) */
349
typedef enum event_t
Nikhil's avatar
Nikhil committed
350 351 352 353 354 355 356 357
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  R_SEND,
  R_ARRIVE,
  R_BUFFER,
358 359 360
  R_BANDWIDTH,
  R_BW_HALT,
  T_BANDWIDTH,
361
} event_t;
Nikhil's avatar
Nikhil committed
362 363 364 365

/* whether the last hop of a packet was global, local or a terminal */
enum last_hop
{
366
   GLOBAL=1,
Nikhil's avatar
Nikhil committed
367
   LOCAL,
368 369
   TERMINAL,
   ROOT
Nikhil's avatar
Nikhil committed
370 371 372 373 374 375 376
};

/* 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
{
377
    MINIMAL = 1,
Nikhil's avatar
Nikhil committed
378 379 380 381 382
    NON_MINIMAL,
    ADAPTIVE,
    PROG_ADAPTIVE
};

383 384 385 386 387
enum LINK_TYPE
{
    GREEN,
    BLACK,
};
Nikhil's avatar
Nikhil committed
388 389 390 391 392 393
struct router_state
{
   unsigned int router_id;
   int group_id;
   int op_arr_size;
   int max_arr_size;
394
   int rc_index;
395
   int num_rtr_rc_windows;
Nikhil's avatar
Nikhil committed
396 397 398 399 400

   int* global_channel; 
   
   tw_stime* next_output_available_time;
   tw_stime* cur_hist_start_time;
401
   tw_stime* last_buf_full;
Nikhil's avatar
Nikhil committed
402 403 404 405

   tw_stime* busy_time;
   tw_stime* busy_time_sample;

406 407 408 409
   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
410 411 412
   int *in_send_loop;
   int *queued_count;
   struct rc_stack * st;
413 414

   int* last_sent_chan;
Nikhil's avatar
Nikhil committed
415 416 417 418
   int** vc_occupancy;
   int64_t* link_traffic;
   int64_t * link_traffic_sample;

419 420 421 422
   int is_monitoring_bw;
   int* last_qos_lvl;
   int** qos_status;
   int** qos_data;
423
   /* for reverse handler */
424 425
   int* last_qos_status;
   int* last_qos_data;
426

Nikhil's avatar
Nikhil committed
427 428 429 430 431 432 433 434 435 436 437 438
   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;
439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455

   /* 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[] = {
456
    {(ev_trace_f) custom_dragonfly_event_collect,
457 458 459 460 461 462
     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) } , 
463
    {(ev_trace_f) custom_dragonfly_event_collect,
464 465
     sizeof(int),
     (model_stat_f) custom_dfly_router_model_stat_collect,
466
     0, //updated in router_custom_setup() since it's based on the radix
467 468
     (sample_event_f) ross_custom_dragonfly_rsample_fn,
     (sample_revent_f) ross_custom_dragonfly_rsample_rc_fn,
469
     0 } , //updated in router_custom_setup() since it's based on the radix    
470
    {NULL, 0, NULL, 0, NULL, NULL, 0}
Nikhil's avatar
Nikhil committed
471
};
472
/* End of ROSS model stats collection */
Nikhil's avatar
Nikhil committed
473 474 475 476 477 478 479 480 481 482 483 484 485

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;

486 487 488 489 490 491 492 493 494 495
/* convert ns to seconds */
static tw_stime ns_to_s(tw_stime ns)
{
        return(ns / (1000.0 * 1000.0 * 1000.0));
}

static double bytes_to_gigabytes(double bytes)
{
    return bytes / (double) (1024 * 1024 * 1024);
}
Nikhil's avatar
Nikhil committed
496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
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;
513 514 515 516
    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
517 518 519
    key = key * 21;
    key = ~key ^ (tmp->sender_id >> 4);
    key = key * tmp->sender_id; 
520
    return (int)(key & (table_size - 1));*/
Nikhil's avatar
Nikhil committed
521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538
}

/* 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 */
539
int dragonfly_custom_get_msg_sz(void)
Nikhil's avatar
Nikhil committed
540
{
541
	   return sizeof(terminal_custom_message);
Nikhil's avatar
Nikhil committed
542 543 544 545
}

static void free_tmp(void * ptr)
{
546
    struct dfly_qhash_entry * dfly = (dfly_qhash_entry *)ptr; 
547 548 549 550 551
    if(dfly->remote_event_data)
        free(dfly->remote_event_data);
   
    if(dfly)
        free(dfly);
Nikhil's avatar
Nikhil committed
552
}
553

554 555 556 557 558 559 560 561 562
int get_vcg_from_category(terminal_custom_message * msg)
{
   if(strcmp(msg->category, "high") == 0)
       return Q_HIGH;
   else if(strcmp(msg->category, "medium") == 0)
       return Q_MEDIUM;
   else
       tw_error(TW_LOC, "\n priority needs to be specified with qos_levels>1 %d", msg->category);
}
563 564 565
static void append_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
566
        int index, 
567
        terminal_custom_message_list *msg) {
568
//    printf("\n msg id %d ", msg->msg.packet_ID);
Nikhil's avatar
Nikhil committed
569 570 571
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
572
        assert(thistail[index] != NULL);
Nikhil's avatar
Nikhil committed
573 574 575 576
        thistail[index]->next = msg;
        msg->prev = thistail[index];
    } 
    thistail[index] = msg;
577
//    printf("\n done adding %d ", msg->msg.packet_ID);
Nikhil's avatar
Nikhil committed
578 579
}

580 581 582
static void prepend_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
583
        int index, 
584
        terminal_custom_message_list *msg) {
Nikhil's avatar
Nikhil committed
585 586 587 588 589 590 591 592 593
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
    } 
    thisq[index] = msg;
}

594 595 596
static terminal_custom_message_list* return_head(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
597
        int index) {
598
    terminal_custom_message_list *head = thisq[index];
Nikhil's avatar
Nikhil committed
599 600 601 602 603 604 605 606 607 608 609 610
    if(head != NULL) {
        thisq[index] = head->next;
        if(head->next != NULL) {
            head->next->prev = NULL;
            head->next = NULL;
        } else {
            thistail[index] = NULL;
        }
    }
    return head;
}

611 612 613
static terminal_custom_message_list* return_tail(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
614
        int index) {
615
    terminal_custom_message_list *tail = thistail[index];
Nikhil's avatar
Nikhil committed
616 617 618 619 620 621 622 623 624 625 626
    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;
}
627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652
/* TODO: Differentiate between local and global bandwidths. */
static int get_rtr_bandwidth_consumption(router_state * s, int qos_lvl, int output_port)
{
    assert(qos_lvl >= Q_HIGH && qos_lvl <= Q_LOW);
    assert(output_port < s->params->intra_grp_radix + s->params->num_global_channels + s->params->num_cn);

    int bandwidth = s->params->cn_bandwidth;
    if(output_port < s->params->intra_grp_radix)
        bandwidth = s->params->local_bandwidth;
    else if(output_port < s->params->intra_grp_radix + s->params->num_global_channels)
        bandwidth = s->params->global_bandwidth;

    /* conversion into bytes from GiB */
    double max_bw = bandwidth * 1024.0 * 1024.0 * 1024.0;
    double max_bw_per_ns = max_bw / (1000.0 * 1000.0 * 1000.0);
    double max_bytes_per_win = max_bw_per_ns * bw_reset_window;

    /* bw_consumed would be in Gigabytes per second. */
//    tw_stime reset_window_s = ns_to_s(bw_reset_window);
//    double bw_gib = bytes_to_gigabytes(s->qos_data[output_port][qos_lvl]);
//    double bw_consumed = ((double)bw_gib / (double)reset_window_s);
    int percent_bw = (((double)s->qos_data[output_port][qos_lvl]) / max_bytes_per_win) * 100;
//    printf("\n percent bw consumed by qos_lvl %d is %d bytes transferred %d max_bw %lf ", qos_lvl, percent_bw, s->qos_data[output_port][qos_lvl], max_bw_per_ns);
    return percent_bw;

}
Nikhil's avatar
Nikhil committed
653

654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685
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);
686
        printf("\tmax hops notification =  %d\n",p->max_hops_notify);
687 688 689 690
        printf("------------------------------------------------------\n\n");
    }
}

Nikhil's avatar
Nikhil committed
691
static void dragonfly_read_config(const char * anno, dragonfly_param *params){
692 693 694 695 696 697 698 699
    /*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
700 701
    // shorthand
    dragonfly_param *p = params;
Nikhil's avatar
Nikhil committed
702
    int myRank;
703
    MPI_Comm_rank(MPI_COMM_CODES, &myRank);
Nikhil's avatar
Nikhil committed
704

705
    int rc = configuration_get_value_int(&config, "PARAMS", "local_vc_size", anno, &p->local_vc_size);
Nikhil's avatar
Nikhil committed
706 707 708 709 710
    if(rc) {
        p->local_vc_size = 1024;
        fprintf(stderr, "Buffer size of local channels not specified, setting to %d\n", p->local_vc_size);
    }

711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745
    rc = configuration_get_value_int(&config, "PARAMS", "qos_levels", anno, &p->num_qos_levels);
    if(rc) {
        p->num_qos_levels = 1;
        fprintf(stderr, "Number of QOS levels not specified, setting to %d\n", p->num_qos_levels);
    }

    char qos_levels_str[MAX_NAME_LENGTH];
    rc = configuration_get_value(&config, "PARAMS", "qos_bandwidth", anno, qos_levels_str, MAX_NAME_LENGTH);
    p->qos_bandwidths = (int*)calloc(p->num_qos_levels, sizeof(int));

    if(p->num_qos_levels > 1)
    {
        int total_bw = 0;
        char * token;
        token = strtok(qos_levels_str, ",");
        int i = 0;
        while(token != NULL)
        {
            sscanf(token, "%d", &p->qos_bandwidths[i]);
            total_bw += p->qos_bandwidths[i];
            if(p->qos_bandwidths[i] <= 0)
            {
                tw_error(TW_LOC, "\n Invalid bandwidth levels");
            }
            i++;
            token = strtok(NULL,",");
        }
        assert(total_bw <= 100);
    }
    else
        p->qos_bandwidths[0] = 100;
    rc = configuration_get_value_double(&config, "PARAMS", "max_qos_monitor", anno, &max_qos_monitor);
    if(rc) {
        printf("\n Setting adaptive threshold to %lf ", max_qos_monitor);
	}
746 747 748
    rc = configuration_get_value_int(&config, "PARAMS", "adaptive_threshold", anno, &adaptive_threshold);
    if(rc) {
    	adaptive_threshold = p->local_vc_size / 8;
749
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
750
	}
751 752 753 754
    else
    {
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
    }
755

Nikhil's avatar
Nikhil committed
756 757 758 759 760 761
    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);
    }

762 763 764 765
    rc = configuration_get_value_int(&config, "PARAMS", "df-dally-vc", anno, &DF_DALLY);
    if(rc) {
        DF_DALLY = 0;
    }
766 767 768 769 770
    
    rc = configuration_get_value_int(&config, "PARAMS", "minimal-bias", anno, &BIAS_MIN);
    if(rc) {
        BIAS_MIN = 0;
    }
771 772 773
    else
	printf("\n Setting minimal bias");

Nikhil's avatar
Nikhil committed
774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803
    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);
    }

804
    rc = configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
Nikhil's avatar
Nikhil committed
805
            &p->router_delay);
806
    if(rc) {
807 808
      p->router_delay = 100;
    }
Nikhil's avatar
Nikhil committed
809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825

    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
826
	      routing = PROG_ADAPTIVE;
Nikhil's avatar
Nikhil committed
827 828 829 830 831 832 833
    else
    {
        fprintf(stderr, 
                "No routing protocol specified, setting to minimal routing\n");
        routing = -1;
    }

834 835 836 837 838 839
    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;
    }

840 841 842 843 844 845 846 847 848 849
    // 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);
    // }
850 851 852
   
if(DF_DALLY == 0) 
{
853 854 855
    //if(routing == PROG_ADAPTIVE)
    //    p->num_vcs = 10;
    //else
856
        p->num_vcs = 8;
857 858 859
}
else
{
860
        p->num_vcs = 4;
861
}
862 863 864
    if(p->num_qos_levels > 1)
        p->num_vcs = p->num_qos_levels * p->num_vcs;

Nikhil's avatar
Nikhil committed
865 866 867
    rc = configuration_get_value_int(&config, "PARAMS", "num_groups", anno, &p->num_groups);
    if(rc) {
      printf("Number of groups not specified. Aborting");
868
      MPI_Abort(MPI_COMM_CODES, 1);
Nikhil's avatar
Nikhil committed
869
    }
870 871 872 873 874
    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;
    }
875 876 877 878 879
    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;
    }
880 881 882 883 884 885 886 887 888 889
    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;
    }
890 891 892 893
    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);

894 895
    p->num_routers = p->num_router_rows * p->num_router_cols;
    
896
    rc = configuration_get_value_int(&config, "PARAMS", "num_cns_per_router", anno, &p->num_cn);
Nikhil's avatar
Nikhil committed
897
    if(rc) {
898 899
        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
900
    }
901

902 903
    rc = configuration_get_value_int(&config, "PARAMS", "num_global_channels", anno, &p->num_global_channels);
    if(rc) {
904 905
        printf("\n Number of global channels per router not specified, setting to 10 ");
        p->num_global_channels = 10;
906
    }
907
    p->radix = p->intra_grp_radix + p->num_global_channels + p->num_cn;
Nikhil's avatar
Nikhil committed
908 909
    p->total_routers = p->num_groups * p->num_routers;
    p->total_terminals = p->total_routers * p->num_cn;
Nikhil's avatar
Nikhil committed
910 911 912 913
    
    // read intra group connections, store from a router's perspective
    // all links to the same router form a vector
    char intraFile[MAX_NAME_LENGTH];
914
    configuration_get_value(&config, "PARAMS", "intra-group-connections", 
Nikhil's avatar
Nikhil committed
915
        anno, intraFile, MAX_NAME_LENGTH);
916 917
    if(strlen(intraFile) <= 0) {
      tw_error(TW_LOC, "Intra group connections file not specified. Aborting");
Nikhil's avatar
Nikhil committed
918 919
    }
    FILE *groupFile = fopen(intraFile, "rb");
920 921 922
    if(!groupFile)
        tw_error(TW_LOC, "intra-group file not found ");

Nikhil's avatar
Nikhil committed
923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944
    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];
945
    configuration_get_value(&config, "PARAMS", "inter-group-connections", 
Nikhil's avatar
Nikhil committed
946
        anno, interFile, MAX_NAME_LENGTH);
947 948
    if(strlen(interFile) <= 0) {
      tw_error(TW_LOC, "Inter group connections file not specified. Aborting");
Nikhil's avatar
Nikhil committed
949 950 951
    }
    FILE *systemFile = fopen(interFile, "rb");
    if(!myRank)
952
    {
Nikhil's avatar
Nikhil committed
953
      printf("Reading inter-group connectivity file: %s\n", interFile);
954 955
      printf("\n Total routers %d total groups %d ", p->total_routers, p->num_groups);
    }
Nikhil's avatar
Nikhil committed
956 957 958 959 960 961 962 963 964

    {
      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);
      }
965
      
Nikhil's avatar
Nikhil committed
966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986
      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);

987
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
988 989 990 991 992 993 994 995
    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++) {
996
          // offset is number of local connections
997
          // type is black or green according to Cray architecture 
Nikhil's avatar
Nikhil committed
998 999 1000 1001 1002 1003 1004
          printf("%d,%d ", it->second[l].offset, it->second[l].type);
        }
        printf(")");
      }
      printf("\n");
    }
#endif
1005
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
1006 1007 1008
    printf("Dumping inter-group connections\n");
    for(int a = 0; a < interGroupLinks.size(); a++) {
      printf("Connections for router %d\n", a);
1009 1010
      map< int, vector<bLink> >  &curMap = interGroupLinks[a];
      map< int, vector<bLink> >::iterator it = curMap.begin();
Nikhil's avatar
Nikhil committed
1011
      for(; it != curMap.end(); it++) {
1012
        // dest group ID 
Nikhil's avatar
Nikhil committed
1013 1014
        printf(" ( %d - ", it->first);
        for(int l = 0; l < it->second.size(); l++) {
1015 1016
            // dest is dest router ID
            // offset is number of global connections
Nikhil's avatar
Nikhil committed
1017 1018 1019 1020 1021 1022 1023 1024
          printf("%d,%d ", it->second[l].offset, it->second[l].dest);
        }
        printf(")");
      }
      printf("\n");
    }
#endif

1025
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038
    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) {
1039
        printf("\n Total nodes %d routers %d groups %d routers per group %d radix %d\n",
Nikhil's avatar
Nikhil committed
1040
                p->num_cn * p->total_routers, p->total_routers, p->num_groups,
1041
                p->num_routers, p->radix);
Nikhil's avatar
Nikhil committed
1042
    }
Nikhil's avatar
Nikhil committed
1043

Nikhil's avatar
Nikhil committed
1044 1045 1046
    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);
1047
    p->credit_delay = bytes_to_ns(CREDIT_SIZE, p->local_bandwidth); //assume 8 bytes packet
1048 1049 1050 1051 1052

    if (PRINT_CONFIG) 
        dragonfly_print_params(p);

    stored_params = p;
Nikhil's avatar
Nikhil committed
1053 1054
}

1055
void dragonfly_custom_configure(){
Nikhil's avatar
Nikhil committed
1056 1057 1058
    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);
1059
    all_params = (dragonfly_param *)calloc(num_params, sizeof(*all_params));
Nikhil's avatar
Nikhil committed
1060 1061 1062 1063 1064 1065 1066 1067

    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]);
    }
1068 1069 1070
#ifdef ENABLE_CORTEX
	model_net_topology = dragonfly_custom_cortex_topology;
#endif
Nikhil's avatar
Nikhil committed
1071 1072 1073
}

/* report dragonfly statistics like average and maximum packet latency, average number of hops traversed */
1074
void dragonfly_custom_report_stats()
Nikhil's avatar
Nikhil committed
1075 1076 1077 1078 1079 1080
{
   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;
1081
   long total_local_packets_sr, total_local_packets_sg, total_remote_packets;
Nikhil's avatar
Nikhil committed
1082

1083 1084 1085 1086 1087 1088 1089
   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
1090
   
1091
   MPI_Reduce( &packet_gen, &total_gen, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
1092
   MPI_Reduce(&packet_fin, &total_fin, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
1093 1094
    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);
1095
   MPI_Reduce( &num_remote_packets, &total_remote_packets, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
1096
   if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
Nikhil's avatar
Nikhil committed
1097
    {
1098 1099
	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
1100 1101 1102 1103 1104
    }

   /* print statistics */
   if(!g_tw_mynode)
   {	
1105 1106 1107
    if (PRINT_CONFIG) 
        dragonfly_print_params(stored_params);

Nikhil's avatar
Nikhil committed
1108 1109
      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);
1110
     if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
Nikhil's avatar
Nikhil committed
1111