dragonfly-custom.C 125 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 CREDIT_SIZE 8
Misbah Mubarak's avatar
Misbah Mubarak committed
30
#define DFLY_HASH_TABLE_SIZE 4999
31
// debugging parameters
32
#define DEBUG_LP 892
33
#define T_ID 10
34
#define TRACK -1
35
#define TRACK_PKT 0
36 37 38
#define TRACK_MSG -1
#define DEBUG 0
#define MAX_STATS 65536
39
#define SHOW_ADAP_STATS 1
40 41 42 43 44

#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
45

46
static int BIAS_MIN = 1;
47
static int DF_DALLY = 0;
48
static int adaptive_threshold = 1024;
49

50 51
static long num_nbr_packets = 0;

52 53 54
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
55 56 57 58 59 60 61
using namespace std;
struct Link {
  int offset, type;
};
struct bLink {
  int offset, dest;
};
62 63 64 65 66
/* 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
67
vector< map< int, vector<Link> > > intraGroupLinks;
68 69
/* contains mapping between source router and destination group via link (link
 * has dest ID)*/
Nikhil's avatar
Nikhil committed
70
vector< map< int, vector<bLink> > > interGroupLinks;
71
/*MM: Maintains a list of routers connecting the source and destination groups */
Nikhil's avatar
Nikhil committed
72 73 74 75 76 77 78 79 80 81
vector< vector< vector<int> > > connectionList;

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

struct InterGroupLink {
  int src, dest;
};

82 83 84 85 86 87
#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
88

89 90 91
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
92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110

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 */
111
static int router_magic_num = 0;
Nikhil's avatar
Nikhil committed
112 113

/* terminal magic number */
114
static int terminal_magic_num = 0;
Nikhil's avatar
Nikhil committed
115

116 117 118 119
/* Hops within a group */
static int num_intra_nonmin_hops = 4;
static int num_intra_min_hops = 2;

120
static FILE * dragonfly_log = NULL;
Nikhil's avatar
Nikhil committed
121

122 123
static int sample_bytes_written = 0;
static int sample_rtr_bytes_written = 0;
Nikhil's avatar
Nikhil committed
124

125 126
static char cn_sample_file[MAX_NAME_LENGTH];
static char router_sample_file[MAX_NAME_LENGTH];
Nikhil's avatar
Nikhil committed
127

128 129
//don't do overhead here - job of MPI layer
static tw_stime mpi_soft_overhead = 0;
130

131 132 133
typedef struct terminal_custom_message_list terminal_custom_message_list;
struct terminal_custom_message_list {
    terminal_custom_message msg;
Nikhil's avatar
Nikhil committed
134
    char* event_data;
135 136
    terminal_custom_message_list *next;
    terminal_custom_message_list *prev;
Nikhil's avatar
Nikhil committed
137 138
};

139
static void init_terminal_custom_message_list(terminal_custom_message_list *thisO, 
140
    terminal_custom_message *inmsg) {
141 142 143 144
    thisO->msg = *inmsg;
    thisO->event_data = NULL;
    thisO->next = NULL;
    thisO->prev = NULL;
Nikhil's avatar
Nikhil committed
145 146
}

147 148 149 150
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
151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166
}

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;
167 168
    int intra_grp_radix;
    int num_col_chans;
169
    int num_row_chans;
170 171
    int num_router_rows;
    int num_router_cols;
Nikhil's avatar
Nikhil committed
172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241
    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;
};

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;
242 243
   terminal_custom_message_list **terminal_msgs;
   terminal_custom_message_list **terminal_msgs_tail;
Nikhil's avatar
Nikhil committed
244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263
   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;

264
   tw_stime * last_buf_full;
Nikhil's avatar
Nikhil committed
265
   tw_stime busy_time;
266 267 268 269
   
   tw_stime max_latency;
   tw_stime min_latency;

Nikhil's avatar
Nikhil committed
270 271 272 273 274 275 276 277 278 279 280 281 282 283 284
   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;
285
  
Nikhil's avatar
Nikhil committed
286 287 288 289 290 291
   /* for logging forward and reverse events */
   long fwd_events;
   long rev_events;
};

/* terminal event type (1-4) */
292
typedef enum event_t
Nikhil's avatar
Nikhil committed
293 294 295 296 297 298 299 300
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  R_SEND,
  R_ARRIVE,
  R_BUFFER,
301
} event_t;
Nikhil's avatar
Nikhil committed
302 303 304 305

/* whether the last hop of a packet was global, local or a terminal */
enum last_hop
{
306
   GLOBAL=1,
Nikhil's avatar
Nikhil committed
307
   LOCAL,
308 309
   TERMINAL,
   ROOT
Nikhil's avatar
Nikhil committed
310 311 312 313 314 315 316
};

/* 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
{
317
    MINIMAL = 1,
Nikhil's avatar
Nikhil committed
318 319 320 321 322
    NON_MINIMAL,
    ADAPTIVE,
    PROG_ADAPTIVE
};

323 324 325 326 327
enum LINK_TYPE
{
    GREEN,
    BLACK,
};
Nikhil's avatar
Nikhil committed
328 329 330 331 332 333 334 335 336 337 338
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;
339
   tw_stime** last_buf_full;
Nikhil's avatar
Nikhil committed
340 341 342 343

   tw_stime* busy_time;
   tw_stime* busy_time_sample;

344 345 346 347
   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
348 349 350
   int *in_send_loop;
   int *queued_count;
   struct rc_stack * st;
351 352

   int* last_sent_chan;
Nikhil's avatar
Nikhil committed
353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400
   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;
401 402 403 404
    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
405 406 407
    key = key * 21;
    key = ~key ^ (tmp->sender_id >> 4);
    key = key * tmp->sender_id; 
408
    return (int)(key & (table_size - 1));*/
Nikhil's avatar
Nikhil committed
409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426
}

/* 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 */
427
int dragonfly_custom_get_msg_sz(void)
Nikhil's avatar
Nikhil committed
428
{
429
	   return sizeof(terminal_custom_message);
Nikhil's avatar
Nikhil committed
430 431 432 433
}

static void free_tmp(void * ptr)
{
434
    struct dfly_qhash_entry * dfly = (dfly_qhash_entry *)ptr; 
435 436 437 438 439
    if(dfly->remote_event_data)
        free(dfly->remote_event_data);
   
    if(dfly)
        free(dfly);
Nikhil's avatar
Nikhil committed
440
}
441

442 443 444
static void append_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
445
        int index, 
446
        terminal_custom_message_list *msg) {
Nikhil's avatar
Nikhil committed
447 448 449 450 451 452 453 454 455
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
    } 
    thistail[index] = msg;
}

456 457 458
static void prepend_to_terminal_custom_message_list(  
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
459
        int index, 
460
        terminal_custom_message_list *msg) {
Nikhil's avatar
Nikhil committed
461 462 463 464 465 466 467 468 469
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
    } 
    thisq[index] = msg;
}

470 471 472
static terminal_custom_message_list* return_head(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
473
        int index) {
474
    terminal_custom_message_list *head = thisq[index];
Nikhil's avatar
Nikhil committed
475 476 477 478 479 480 481 482 483 484 485 486
    if(head != NULL) {
        thisq[index] = head->next;
        if(head->next != NULL) {
            head->next->prev = NULL;
            head->next = NULL;
        } else {
            thistail[index] = NULL;
        }
    }
    return head;
}

487 488 489
static terminal_custom_message_list* return_tail(
        terminal_custom_message_list ** thisq,
        terminal_custom_message_list ** thistail,
Nikhil's avatar
Nikhil committed
490
        int index) {
491
    terminal_custom_message_list *tail = thistail[index];
Nikhil's avatar
Nikhil committed
492 493 494 495 496 497 498 499 500 501 502 503 504
    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;
}

static void dragonfly_read_config(const char * anno, dragonfly_param *params){
505 506 507 508 509 510 511 512
    /*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
513 514
    // shorthand
    dragonfly_param *p = params;
Nikhil's avatar
Nikhil committed
515
    int myRank;
516
    MPI_Comm_rank(MPI_COMM_CODES, &myRank);
Nikhil's avatar
Nikhil committed
517

518
    int rc = configuration_get_value_int(&config, "PARAMS", "local_vc_size", anno, &p->local_vc_size);
Nikhil's avatar
Nikhil committed
519 520 521 522 523
    if(rc) {
        p->local_vc_size = 1024;
        fprintf(stderr, "Buffer size of local channels not specified, setting to %d\n", p->local_vc_size);
    }

524 525 526
    rc = configuration_get_value_int(&config, "PARAMS", "adaptive_threshold", anno, &adaptive_threshold);
    if(rc) {
    	adaptive_threshold = p->local_vc_size / 8;
527
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
528
	}
529 530 531 532
    else
    {
        printf("\n Setting adaptive threshold to %d ", adaptive_threshold);
    }
533

Nikhil's avatar
Nikhil committed
534 535 536 537 538 539
    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);
    }

540 541 542 543
    rc = configuration_get_value_int(&config, "PARAMS", "df-dally-vc", anno, &DF_DALLY);
    if(rc) {
        DF_DALLY = 0;
    }
544 545 546 547 548
    
    rc = configuration_get_value_int(&config, "PARAMS", "minimal-bias", anno, &BIAS_MIN);
    if(rc) {
        BIAS_MIN = 0;
    }
549 550 551
    else
	printf("\n Setting minimal bias");

Nikhil's avatar
Nikhil committed
552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581
    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);
    }

582
    rc = configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
Nikhil's avatar
Nikhil committed
583
            &p->router_delay);
584
    if(rc) {
585 586
      p->router_delay = 100;
    }
Nikhil's avatar
Nikhil committed
587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603

    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
604
	      routing = PROG_ADAPTIVE;
Nikhil's avatar
Nikhil committed
605 606 607 608 609 610 611
    else
    {
        fprintf(stderr, 
                "No routing protocol specified, setting to minimal routing\n");
        routing = -1;
    }

612 613 614 615 616 617 618 619 620 621
    // 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);
    // }
622 623 624
   
if(DF_DALLY == 0) 
{
625 626 627 628
    if(routing == PROG_ADAPTIVE)
        p->num_vcs = 10;
    else
        p->num_vcs = 8;
629 630 631
}
else
{
632
    if(routing == PROG_ADAPTIVE)
633
        p->num_vcs = 5;
634
    else
635
        p->num_vcs = 4;
636
}
Nikhil's avatar
Nikhil committed
637 638 639
    rc = configuration_get_value_int(&config, "PARAMS", "num_groups", anno, &p->num_groups);
    if(rc) {
      printf("Number of groups not specified. Aborting");
640
      MPI_Abort(MPI_COMM_CODES, 1);
Nikhil's avatar
Nikhil committed
641
    }
642 643 644 645 646
    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;
    }
647 648 649 650 651
    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;
    }
652 653 654 655 656 657 658 659 660 661
    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;
    }
662 663 664 665
    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);

666 667
    p->num_routers = p->num_router_rows * p->num_router_cols;
    
668
    rc = configuration_get_value_int(&config, "PARAMS", "num_cns_per_router", anno, &p->num_cn);
Nikhil's avatar
Nikhil committed
669
    if(rc) {
670 671
        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
672
    }
673

674 675
    rc = configuration_get_value_int(&config, "PARAMS", "num_global_channels", anno, &p->num_global_channels);
    if(rc) {
676 677
        printf("\n Number of global channels per router not specified, setting to 10 ");
        p->num_global_channels = 10;
678
    }
679
    p->radix = (p->num_router_cols * p->num_row_chans) + (p->num_col_chans * p->num_router_rows) + p->num_global_channels + p->num_cn;
Nikhil's avatar
Nikhil committed
680 681
    p->total_routers = p->num_groups * p->num_routers;
    p->total_terminals = p->total_routers * p->num_cn;
Nikhil's avatar
Nikhil committed
682 683 684 685
    
    // read intra group connections, store from a router's perspective
    // all links to the same router form a vector
    char intraFile[MAX_NAME_LENGTH];
686
    configuration_get_value(&config, "PARAMS", "intra-group-connections", 
Nikhil's avatar
Nikhil committed
687
        anno, intraFile, MAX_NAME_LENGTH);
688 689
    if(strlen(intraFile) <= 0) {
      tw_error(TW_LOC, "Intra group connections file not specified. Aborting");
Nikhil's avatar
Nikhil committed
690 691
    }
    FILE *groupFile = fopen(intraFile, "rb");
692 693 694
    if(!groupFile)
        tw_error(TW_LOC, "intra-group file not found ");

Nikhil's avatar
Nikhil committed
695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716
    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];
717
    configuration_get_value(&config, "PARAMS", "inter-group-connections", 
Nikhil's avatar
Nikhil committed
718
        anno, interFile, MAX_NAME_LENGTH);
719 720
    if(strlen(interFile) <= 0) {
      tw_error(TW_LOC, "Inter group connections file not specified. Aborting");
Nikhil's avatar
Nikhil committed
721 722 723
    }
    FILE *systemFile = fopen(interFile, "rb");
    if(!myRank)
724
    {
Nikhil's avatar
Nikhil committed
725
      printf("Reading inter-group connectivity file: %s\n", interFile);
726 727
      printf("\n Total routers %d total groups %d ", p->total_routers, p->num_groups);
    }
Nikhil's avatar
Nikhil committed
728 729 730 731 732 733 734 735 736

    {
      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);
      }
737
      
Nikhil's avatar
Nikhil committed
738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758
      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);

759
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
760 761 762 763 764 765 766 767
    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++) {
768
          // offset is number of local connections
769
          // type is black or green according to Cray architecture 
Nikhil's avatar
Nikhil committed
770 771 772 773 774 775 776
          printf("%d,%d ", it->second[l].offset, it->second[l].type);
        }
        printf(")");
      }
      printf("\n");
    }
#endif
777
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
778 779 780
    printf("Dumping inter-group connections\n");
    for(int a = 0; a < interGroupLinks.size(); a++) {
      printf("Connections for router %d\n", a);
781 782
      map< int, vector<bLink> >  &curMap = interGroupLinks[a];
      map< int, vector<bLink> >::iterator it = curMap.begin();
Nikhil's avatar
Nikhil committed
783
      for(; it != curMap.end(); it++) {
784
        // dest group ID 
Nikhil's avatar
Nikhil committed
785 786
        printf(" ( %d - ", it->first);
        for(int l = 0; l < it->second.size(); l++) {
787 788
            // dest is dest router ID
            // offset is number of global connections
Nikhil's avatar
Nikhil committed
789 790 791 792 793 794 795 796
          printf("%d,%d ", it->second[l].offset, it->second[l].dest);
        }
        printf(")");
      }
      printf("\n");
    }
#endif

797
#if DUMP_CONNECTIONS == 1
Nikhil's avatar
Nikhil committed
798 799 800 801 802 803 804 805 806 807 808 809 810
    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) {
811
        printf("\n Total nodes %d routers %d groups %d routers per group %d radix %d\n",
Nikhil's avatar
Nikhil committed
812
                p->num_cn * p->total_routers, p->total_routers, p->num_groups,
813
                p->num_routers, p->radix);
Nikhil's avatar
Nikhil committed
814
    }
Nikhil's avatar
Nikhil committed
815

Nikhil's avatar
Nikhil committed
816 817 818
    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);
819
    p->credit_delay = bytes_to_ns(CREDIT_SIZE, p->local_bandwidth); //assume 8 bytes packet
Nikhil's avatar
Nikhil committed
820 821
}

822
void dragonfly_custom_configure(){
Nikhil's avatar
Nikhil committed
823 824 825
    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);
826
    all_params = (dragonfly_param *)malloc(num_params * sizeof(*all_params));
Nikhil's avatar
Nikhil committed
827 828 829 830 831 832 833 834

    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]);
    }
835 836 837
#ifdef ENABLE_CORTEX
	model_net_topology = dragonfly_custom_cortex_topology;
#endif
Nikhil's avatar
Nikhil committed
838 839 840
}

/* report dragonfly statistics like average and maximum packet latency, average number of hops traversed */
841
void dragonfly_custom_report_stats()
Nikhil's avatar
Nikhil committed
842 843 844 845 846 847
{
   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;
848
   long total_local_packets_sr, total_local_packets_sg, total_remote_packets;
Nikhil's avatar
Nikhil committed
849

850 851 852 853 854 855 856
   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
857
   
858
   MPI_Reduce( &packet_gen, &total_gen, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
859
   MPI_Reduce(&packet_fin, &total_fin, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
860 861
    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);
862
   MPI_Reduce( &num_remote_packets, &total_remote_packets, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_CODES);
863
   if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
Nikhil's avatar
Nikhil committed
864
    {
865 866
	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
867 868 869 870 871 872 873
    }

   /* print statistics */
   if(!g_tw_mynode)
   {	
      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);
874
     if(routing == ADAPTIVE || routing == PROG_ADAPTIVE || SHOW_ADAP_STATS)
Nikhil's avatar
Nikhil committed
875 876 877
              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);
 
878
      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
879 880 881 882 883 884 885
   }
   return;
}


/* initialize a dragonfly compute node terminal */
void 
886
terminal_custom_init( terminal_state * s, 
Nikhil's avatar
Nikhil committed
887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911
	       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);

912
   s->terminal_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
913
   s->router_id=(int)s->terminal_id / (s->params->num_cn);
Nikhil's avatar
Nikhil committed
914 915
   s->terminal_available_time = 0.0;
   s->packet_counter = 0;
916 917 918
   s->min_latency = INT_MAX;
   s->max_latency = 0;   

Nikhil's avatar
Nikhil committed
919 920 921 922 923 924 925 926 927 928 929 930 931 932
   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;
   s->vc_occupancy = (int*)malloc(s->num_vcs * sizeof(int));
933
   s->last_buf_full = (tw_stime*)malloc(s->num_vcs * sizeof(tw_stime));
Nikhil's avatar
Nikhil committed
934 935 936

   for( i = 0; i < s->num_vcs; i++ )
    {
937
      s->last_buf_full[i] = 0.0;
Nikhil's avatar
Nikhil committed
938 939 940 941
      s->vc_occupancy[i]=0;
    }


942
   s->rank_tbl = NULL;
Nikhil's avatar
Nikhil committed
943
   s->terminal_msgs = 
944
       (terminal_custom_message_list**)malloc(s->num_vcs*sizeof(terminal_custom_message_list*));
Nikhil's avatar
Nikhil committed
945
   s->terminal_msgs_tail = 
946
       (terminal_custom_message_list**)malloc(s->num_vcs*sizeof(terminal_custom_message_list*));
Nikhil's avatar
Nikhil committed
947 948 949 950 951 952 953 954 955 956 957
   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 */
958
void router_custom_setup(router_state * r, tw_lp * lp)
Nikhil's avatar
Nikhil committed
959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976
{
    
    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;

977
    num_routers_per_mgrp = codes_mapping_get_lp_count (lp_group_name, 1, "modelnet_dragonfly_custom_router",
Nikhil's avatar
Nikhil committed
978 979 980 981 982 983 984
            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);

985
   r->router_id = codes_mapping_get_lp_relative_id(lp->gid, 0, 0);
Nikhil's avatar
Nikhil committed
986
   r->group_id=r->router_id/p->num_routers;
987 988
    
   //printf("\n Local router id %d global id %d ", r->router_id, lp->gid);
Nikhil's avatar
Nikhil committed
989 990 991 992

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

993

Nikhil's avatar
Nikhil committed
994 995 996 997 998 999 1000
   r->global_channel = (int*)malloc(p->num_global_channels * sizeof(int));
   r->next_output_available_time = (tw_stime*)malloc(p->radix * sizeof(tw_stime));
   r->cur_hist_start_time = (tw_stime*)malloc(p->radix * sizeof(tw_stime));
   r->link_traffic = (int64_t*)malloc(p->radix * sizeof(int64_t));
   r->link_traffic_sample = (int64_t*)malloc(p->radix * sizeof(int64_t));
   r->cur_hist_num = (int*)malloc(p->radix * sizeof(int));
   r->prev_hist_num = (int*)malloc(p->radix * sizeof(int));
1001 1002
  
   r->last_sent_chan = (int*) malloc(p->num_router_rows * sizeof(int));
Nikhil's avatar
Nikhil committed
1003 1004 1005
   r->vc_occupancy = (int**)malloc(p->radix * sizeof(int*));
   r->in_send_loop = (int*)malloc(p->radix * sizeof(int));
   r->pending_msgs = 
1006
    (terminal_custom_message_list***)malloc(p->radix * sizeof(terminal_custom_message_list**));
Nikhil's avatar
Nikhil committed
1007
   r->pending_msgs_tail = 
1008
    (terminal_custom_message_list***)malloc(p->radix * sizeof(terminal_custom_message_list**));
Nikhil's avatar
Nikhil committed
1009
   r->queued_msgs = 
1010
    (terminal_custom_message_list***)malloc(p->radix * sizeof(terminal_custom_message_list**));
Nikhil's avatar
Nikhil committed
1011
   r->queued_msgs_tail = 
1012
    (terminal_custom_message_list***)malloc(p->radix * sizeof(terminal_custom_message_list**));
Nikhil's avatar
Nikhil committed
1013
   r->queued_count = (int*)malloc(p->radix * sizeof(int));
1014
   r->last_buf_full = (tw_stime**)malloc(p->radix * sizeof(tw_stime*));
Nikhil's avatar
Nikhil committed
1015 1016 1017 1018
   r->busy_time = (tw_stime*)malloc(p->radix * sizeof(tw_stime));
   r->busy_time_sample = (tw_stime*)malloc(p->radix * sizeof(tw_stime));

   rc_stack_create(&r->st);
1019 1020 1021 1022

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

Nikhil's avatar
Nikhil committed
1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036
   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;
    r->vc_occupancy[i] = (int*)malloc(p->num_vcs * sizeof(int));
1037 1038
    r->pending_msgs[i] = (terminal_custom_message_list**)malloc(p->num_vcs * 
        sizeof(terminal_custom_message_list*));
1039
    r->last_buf_full[i] = (tw_stime*)malloc(p->num_vcs * sizeof(tw_stime));
1040 1041 1042 1043 1044 1045
    r->pending_msgs_tail[i] = (terminal_custom_message_list**)malloc(p->num_vcs * 
        sizeof(terminal_custom_message_list*));
    r->queued_msgs[i] = (terminal_custom_message_list**)malloc(p->num_vcs * 
        sizeof(terminal_custom_message_list*));
    r->queued_msgs_tail[i] = (terminal_custom_message_list**)malloc(p->num_vcs * 
        sizeof(terminal_custom_message_list*));
Nikhil's avatar
Nikhil committed
1046
        for(int j = 0; j < p->num_vcs; j++) {
1047
            r->last_buf_full[i][j] = 0.0;
Nikhil's avatar
Nikhil committed
1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059
            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 */
1060
static tw_stime dragonfly_custom_packet_event(
Nikhil's avatar
Nikhil committed
1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074
        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;
1075
    terminal_custom_message * msg;
Nikhil's avatar
Nikhil committed
1076 1077 1078 1079 1080 1081
    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,
1082
            sender, DRAGONFLY_CUSTOM, (void**)&msg, (void**)&tmp_ptr);
Nikhil's avatar
Nikhil committed
1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
    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*/
      {
1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113
        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
1114 1115 1116 1117 1118 1119 1120
     }
	   //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 */
1121
static void dragonfly_custom_packet_event_rc(tw_lp *sender)
Nikhil's avatar
Nikhil committed
1122 1123 1124 1125 1126 1127
{