fattree.c 83.4 KB
Newer Older
1
2
#include <ross.h>

Noah Wolfe's avatar
Noah Wolfe committed
3
#include "codes/jenkins-hash.h"
4
5
6
7
8
9
10
#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"
#include "codes/net/fattree.h"
#include "sys/file.h"
11
#include "codes/quickhash.h"
Noah Wolfe's avatar
Noah Wolfe committed
12
#include "codes/rc-stack.h"
13
14
15
16

#define CREDIT_SIZE 8
#define MEAN_PROCESS 1.0

17
18
#define FTREE_HASH_TABLE_SIZE 262144

19
// debugging parameters
Noah Wolfe's avatar
Noah Wolfe committed
20
21
//#define TRACK_PKT -1
#define TRACK_PKT 2820
22
#define FATTREE_HELLO 0
23
#define FATTREE_DEBUG 0
24
#define FATTREE_CONNECTIONS 1
25
#define FATTREE_MSG 0
Nikhil's avatar
Nikhil committed
26
#define DEBUG_RC 0
27

28
//Data Collection Output Files
29
#define PARAMS_LOG 1
30

31
32
33
#define LP_CONFIG_NM (model_net_lp_config_names[FATTREE])
#define LP_METHOD_NM (model_net_method_names[FATTREE])

34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
#if DEBUG_RC
  //Reverse Compute Debug Variables
  long long packet_event_f = 0;
  long long packet_event_r = 0;
  long long t_generate_f = 0;
  long long t_generate_r = 0;
  long long t_send_f = 0;
  long long t_send_r = 0;
  long long t_arrive_f = 0;
  long long t_arrive_r = 0;
  long long t_buffer_f = 0;
  long long t_buffer_r = 0;
  long long s_send_f = 0;
  long long s_send_r = 0;
  long long s_arrive_f = 0;
  long long s_arrive_r = 0;
  long long s_buffer_f = 0;
  long long s_buffer_r = 0;
#endif

54
55
long fattree_packet_gen = 0, fattree_packet_fin = 0;

56
57
58
static double maxd(double a, double b) { return a < b ? b : a; }

// arrival rate
Noah Wolfe's avatar
Noah Wolfe committed
59
//static double MEAN_INTERVAL=200.0;
60
61

typedef struct fattree_param fattree_param;
62
/* annotation-specific parameters (unannotated entry occurs at the
63
64
65
 * last index) */
static uint64_t                  num_params = 0;
static fattree_param           * all_params = NULL;
Noah Wolfe's avatar
Noah Wolfe committed
66
static const config_anno_map_t * anno_map   = NULL;
67
68
69
70
71
72
73

/* global variables for codes mapping */
static char lp_group_name[MAX_NAME_LENGTH];
static char def_group_name[MAX_NAME_LENGTH];
static int def_gname_set = 0;
static int mapping_grp_id, mapping_type_id, mapping_rep_id, mapping_offset;

74
75
76
/* switch magic number */
int switch_magic_num = 0;

77
/* terminal magic number */
78
int fattree_terminal_magic_num = 0;
79

80
81
82
83
84
85
86
87
typedef struct fattree_message_list fattree_message_list;
struct fattree_message_list {
    fattree_message msg;
    char* event_data;
    fattree_message_list *next;
    fattree_message_list *prev;
};

88
void init_fattree_message_list(fattree_message_list *this,
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
  fattree_message *inmsg) {
    this->msg = *inmsg;
    this->event_data = NULL;
    this->next = NULL;
    this->prev = NULL;
}

void delete_fattree_message_list(fattree_message_list *this) {
    if(this->event_data != NULL) free(this->event_data);
    free(this);
}

struct fattree_param
{
  int ft_type;
  // configuration parameters
  int num_levels;
106
107
108
109
  int link_repetitions;
  int requested_terminals; //Number of terminals to shoot for if possible
  int Ns; // number of switches per pod
  int Np; //Number of pods
110
111
112
  int *num_switches; //switches at various levels
  int *switch_radix; //radix of switches are various levels
  double link_bandwidth;/* bandwidth of a wire connecting switches */
113
  double cn_bandwidth;/* bandwidth of the compute node channels
114
115
116
                        connected to switch */
  int vc_size; /* buffer size of the link channels */
  int cn_vc_size; /* buffer size of the compute node channels */
117
  int chunk_size; /* full-sized packets are broken into smaller chunks.*/
118
  int packet_size;
119
120
121
122
123
124
125
126
127
128
  int num_terminals;
  int l1_set_size;
  int l1_term_size;
  double cn_delay;
  double head_delay;
  double credit_delay;
  double router_delay;
  double soft_delay;
};

129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
struct ftree_hash_key
{
    uint64_t message_id;
    tw_lpid sender_id;
};

struct ftree_qhash_entry
{
   struct ftree_hash_key key;
   char * remote_event_data;
   int num_chunks;
   int remote_event_size;
   struct qhash_head hash_link;
};

144
145
146
147
/* convert GiB/s and bytes to ns */
 static tw_stime bytes_to_ns(uint64_t bytes, double GB_p_s)
 {
     tw_stime time;
148

149
150
151
152
153
154
     /* 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;
155

156
157
158
     return(time);
 }

159
160
161
162
163
164
165
166
167
/* handles terminal and switch events like packet generate/send/receive/buffer */
typedef enum event_t event_t;
typedef struct ft_terminal_state ft_terminal_state;
typedef struct switch_state switch_state;

/* fattree compute node data structure */
struct ft_terminal_state
{
  unsigned long long packet_counter;
168
  int packet_gen;
169
  int packet_fin;
170
171
172
173
174
175
176
177
178
  // Fattree specific parameters
  unsigned int terminal_id;
  unsigned int switch_id;
  tw_lpid switch_lp;

  // Each terminal will have an input and output channel with the switch
  int vc_occupancy; // NUM_VC
  tw_stime terminal_available_time;
  tw_stime next_credit_available_time;
179

180
181
  struct mn_stats fattree_stats_array[CATEGORY_MAX];

182
183
  fattree_message_list **terminal_msgs;
  fattree_message_list **terminal_msgs_tail;
184
  int terminal_length;
185
  int in_send_loop;
186
187
188
  int issueIdle;

   struct rc_stack * st;
189
190
191

  char * anno;
  fattree_param *params;
192
193
194
195
196
197
198
199
200
201
202

  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;

203
  tw_stime last_buf_full;
204
  tw_stime busy_time;
205
  char output_buf[4096];
206
  char output_buf3[4096];
207

208
209
210
211
212
  /* For sampling */
  long fin_chunks_sample;
  long data_size_sample;
  double fin_hops_sample;
  tw_stime fin_chunks_time;
213
 tw_stime busy_time_sample;
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
};

/* terminal event type (1-4) */
enum event_t
{
  T_GENERATE=1,
  T_ARRIVE,
  T_SEND,
  T_BUFFER,
  S_SEND,
  S_ARRIVE,
  S_BUFFER,
};

enum last_hop
{
   LINK,
   TERMINAL
};

struct switch_state
{
  unsigned int switch_id;
  int switch_level;
  int radix;
  int num_cons;
  int num_lcons;
  int con_per_lneigh;
  int con_per_uneigh;;
  int start_lneigh;
  int end_lneigh;
  int start_uneigh;
  int unused;

  tw_stime* next_output_available_time;
  tw_stime* next_credit_available_time;
250
251
252
253
254
  tw_stime* last_buf_full;

  tw_stime* busy_time;
  tw_stime* busy_time_sample;

255
256
257
258
259
260
261
262
263
264
  fattree_message_list **pending_msgs;
  fattree_message_list **pending_msgs_tail;
  fattree_message_list **queued_msgs;
  fattree_message_list **queued_msgs_tail;
  int *queued_length;
  int *in_send_loop;
  int* vc_occupancy;
  int64_t* link_traffic;
  tw_lpid *port_connections;

265
266
267
  char output_buf[4096];
  char output_buf2[4096];

Noah Wolfe's avatar
Noah Wolfe committed
268
269
  struct rc_stack * st;

270
271
272
273
  char * anno;
  fattree_param *params;
};

274
275
static tw_stime         fattree_total_time = 0;
static tw_stime         fattree_max_latency = 0;
Noah Wolfe's avatar
Noah Wolfe committed
276
//static tw_stime         max_collective = 0;
277
278
279
280
281
282
283
284


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;

285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
static int fattree_rank_hash_compare(
void *key, struct qhash_head *link)
{
    struct ftree_hash_key *message_key = (struct ftree_hash_key *)key;
    struct ftree_qhash_entry *tmp = NULL;

    tmp = qhash_entry(link, struct ftree_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 fattree_hash_func(void *k, int table_size)
{
    struct ftree_hash_key *tmp = (struct ftree_hash_key *)k;
    uint64_t key = (~tmp->message_id) + (tmp->message_id << 18);
    key = key * 21;
    key = ~key ^ (tmp->sender_id >> 4);
    key = key * tmp->sender_id;
    return (int)(key & (table_size - 1));
}

309
310
static void free_tmp(void * ptr)
{
311
    struct ftree_qhash_entry * ftree = ptr;
312
313
314
315
    free(ftree->remote_event_data);
    free(ftree);
}

316
static void append_to_fattree_message_list(
317
318
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
319
        int index,
320
321
322
323
324
325
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
326
    }
327
328
329
    thistail[index] = msg;
}

330
static void prepend_to_fattree_message_list(
331
332
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
333
        int index,
334
335
336
337
338
339
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
340
    }
341
342
343
344
345
346
347
348
349
350
351
352
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
    thisq[index] = msg;
}

static fattree_message_list* return_head(
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
        int index) {
    fattree_message_list *head = thisq[index];
    if(head != NULL) {
        thisq[index] = head->next;
        if(head->next != NULL) {
            head->next->prev = NULL;
            head->next = NULL;
        } else {
            thistail[index] = NULL;
        }
    }
    return head;
}

static fattree_message_list* return_tail(
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
        int index) {
    fattree_message_list *tail = thistail[index];
    if(tail->prev != NULL) {
        tail->prev->next = NULL;
        thistail[index] = tail->prev;
        tail->prev = NULL;
    } else {
        thistail[index] = NULL;
        thisq[index] = NULL;
    }
    return tail;
}

//decl
void switch_credit_send(switch_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp, int sq);
int ft_get_output_port( switch_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp, int *out_off );
int get_base_port(switch_state *s, int from_term, int index);


/* returns the fattree switch lp type for lp registration */
Noah Wolfe's avatar
Noah Wolfe committed
386
//static const tw_lptype* fattree_get_switch_lp_type(void);
387

388
/* returns the fattree message size */
389
390
391
392
393
static int fattree_get_msg_sz(void)
{
  return sizeof(fattree_message);
}

Noah Wolfe's avatar
Noah Wolfe committed
394
static void fattree_read_config(const char * anno, fattree_param *p){
395
396
  int i;

Nikhil's avatar
Nikhil committed
397
  p->ft_type = 1;
398
  configuration_get_value_int(&config, "PARAMS", "ft_type", anno,
399
400
      &p->ft_type);

401
  configuration_get_value_int(&config, "PARAMS", "num_levels", anno,
402
403
404
405
406
407
408
409
      &p->num_levels);
  if(p->num_levels <= 0) {
    tw_error(TW_LOC, "Too few num_levels, Aborting\n");
  }
  if(p->num_levels > 3) {
    tw_error(TW_LOC, "Too many num_levels, only upto 3 supported Aborting\n");
  }

410
411
412
  configuration_get_value_int(&config, "PARAMS", "requested_terminals", anno,
          &p->requested_terminals);

413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
  p->num_switches = (int *) malloc (p->num_levels * sizeof(int));
  p->switch_radix = (int*) malloc (p->num_levels * sizeof(int));

  char switch_counts_str[MAX_NAME_LENGTH];
  int rc = configuration_get_value(&config, "PARAMS", "switch_count", anno,
      switch_counts_str, MAX_NAME_LENGTH);
  if (rc == 0){
    tw_error(TW_LOC, "couldn't read PARAMS:switch_count");
  }
  char* token;
  token = strtok(switch_counts_str, ",");
  i = 0;
  while(token != NULL)
  {
    sscanf(token, "%d", &p->num_switches[i]);
    if(p->num_switches[i] <= 0)
    {
      tw_error(TW_LOC, "Invalid switch count  specified "
          "(%d at pos %d), exiting... ", p->num_switches[i], i);
    }
    i++;
    token = strtok(NULL,",");
  }

  char switch_radix_str[MAX_NAME_LENGTH];
  rc = configuration_get_value(&config, "PARAMS", "switch_radix", anno,
      switch_radix_str, MAX_NAME_LENGTH);
  if (rc == 0){
    tw_error(TW_LOC, "couldn't read PARAMS:switch_radix");
  }
  token = strtok(switch_radix_str, ",");
  i = 0;
  while(token != NULL)
  {
    sscanf(token, "%d", &p->switch_radix[i]);
    if(p->switch_radix[i] <= 0)
    {
      tw_error(TW_LOC, "Invalid switch radix  specified "
          "(%d at pos %d), exiting... ", p->switch_radix[i], i);
    }
    i++;
    token = strtok(NULL,",");
  }

457
458
  p->Ns = p->switch_radix[0]/2;

459
460
461
462
  if(p->num_levels == 2) {
    p->num_switches[1] = p->num_switches[0]/2;
    p->switch_radix[1] = p->switch_radix[0];
  } else {
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
    if(p->ft_type == 0){
      p->Np = ceil((double)p->requested_terminals/(double)(p->Ns*p->Ns));
      p->num_switches[1] = p->num_switches[0];
      p->num_switches[2] = ceil((p->Ns*p->Ns) / floor(p->switch_radix[0]/p->Np));
      p->switch_radix[1] = p->switch_radix[2] = p->switch_radix[0];
      p->link_repetitions = floor(p->switch_radix[2] / p->Np);
      printf("Np:%d Ns:%d requested_terminals:%d\n",p->Np,p->Ns,p->requested_terminals);
    }else{
      p->num_switches[1] = p->num_switches[0];
      p->num_switches[2] = p->num_switches[0]/2;
      p->switch_radix[1] = p->switch_radix[2] = p->switch_radix[0];
    }
  }

  if(p->num_levels == 2){
    if(p->requested_terminals > 2*(p->switch_radix[0]/2)*(p->switch_radix[0]/2)){
      tw_error(TW_LOC, "too many terminals requested. Can't maintain full bisection bandwidth for given radix.");
    }
  }else{
    if(p->requested_terminals > 2*(p->switch_radix[0]/2)*(p->switch_radix[0]/2)*(p->switch_radix[0]/2)){
      tw_error(TW_LOC, "too many terminals requested. Can't maintain full bisection bandwidth for given radix.");
    }
485
486
  }

487
488
489
490
491
492
493
494
#if FATTREE_DEBUG
  for(int jj=0;jj<3;jj++)
  {
    printf("num_switches[%d]=%d\n",jj,p->num_switches[jj]);
    printf("switch_radix[%d]=%d\n",jj,p->switch_radix[jj]);
  }
#endif

495
496
497
498
  i = 1;
  for(i = 1; i < p->num_levels - 1; i++) {
    if(p->num_switches[i - 1] * p->switch_radix[i - 1] >
       p->num_switches[i] * p->switch_radix[i]) {
499
500
//      tw_error(TW_LOC, "Not enough switches/radix at level %d for full "
//          "bisection bandwidth\n", i);
501
502
503
504
505
    }
  }

  if(p->num_switches[i - 1] * p->switch_radix[i - 1] > 2 * p->num_switches[i] *
      p->switch_radix[i]) {
506
507
//    tw_error(TW_LOC, "Not enough switches/radix at level %d (top) for full "
//        "bisection bandwidth\n", i);
508
509
510
511
512
513
  }

  configuration_get_value_int(&config, "PARAMS", "packet_size", anno,
      &p->packet_size);
  if(!p->packet_size) {
    p->packet_size = 512;
514
    fprintf(stderr, "Packet size is not specified, setting to %d\n",
515
516
        p->packet_size);
  }
517

518
519
520
  p->router_delay = 50;
  configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
      &p->router_delay);
521

522
523
524
525
526
527
528
529
530
531
532
  p->soft_delay = 1000;
  configuration_get_value_double(&config, "PARAMS", "soft_delay", anno,
      &p->soft_delay);

  configuration_get_value_int(&config, "PARAMS", "vc_size", anno, &p->vc_size);
  if(!p->vc_size) {
    p->vc_size = 8*p->packet_size;
    fprintf(stderr, "Buffer size of global channels not specified, setting to "
        "%d\n", p->vc_size);
  }

533
  configuration_get_value_int(&config, "PARAMS", "cn_vc_size", anno,
534
535
536
      &p->cn_vc_size);
  if(!p->cn_vc_size) {
    p->cn_vc_size = 8*p->packet_size;
537
    fprintf(stderr, "Buffer size of compute node channels not specified, "
538
539
540
        "setting to %d\n", p->cn_vc_size);
  }

541
542
543
544
545
546
  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);
    }

547
  configuration_get_value_double(&config, "PARAMS", "link_bandwidth", anno,
548
549
550
      &p->link_bandwidth);
  if(!p->link_bandwidth) {
    p->link_bandwidth = 5;
551
    fprintf(stderr, "Bandwidth of links is specified, setting to %lf\n",
552
553
554
        p->link_bandwidth);
  }

555
  configuration_get_value_double(&config, "PARAMS", "cn_bandwidth", anno,
556
557
558
      &p->cn_bandwidth);
  if(!p->cn_bandwidth) {
    p->cn_bandwidth = 5;
559
    fprintf(stderr, "Bandwidth of compute node channels not specified, "
560
561
562
563
        "setting to %lf\n", p->cn_bandwidth);
  }

  p->l1_set_size = p->switch_radix[0]/2;
564

565
566
  p->l1_term_size = (p->l1_set_size * (p->switch_radix[0] / 2));

567
568
569
570
#if FATTREE_DEBUG
  printf("l1_set_size:%d l1_term_size:%d\n",p->l1_set_size,p->l1_term_size);
#endif

571
572
573
574
575
576
577
578
579
580
581
  p->cn_delay = (1.0 / p->cn_bandwidth);
  p->head_delay = (1.0 / p->link_bandwidth);
  p->credit_delay = (1.0 / p->link_bandwidth) * 8; //assume 8 bytes packet
}

static void fattree_configure(){
  anno_map = codes_mapping_get_lp_anno_map(LP_CONFIG_NM);
  assert(anno_map);
  num_params = anno_map->num_annos + (anno_map->has_unanno_lp > 0);
  all_params = malloc(num_params * sizeof(*all_params));

Noah Wolfe's avatar
Noah Wolfe committed
582
583
  for (int i = 0; i < anno_map->num_annos; i++){
    const char * anno = anno_map->annotations[i].ptr;
584
585
586
587
588
589
590
591
592
593
    fattree_read_config(anno, &all_params[i]);
  }
  if (anno_map->has_unanno_lp > 0){
    fattree_read_config(NULL, &all_params[anno_map->num_annos]);
  }
}

/* initialize a fattree compute node terminal */
void ft_terminal_init( ft_terminal_state * s, tw_lp * lp )
{
594
    s->packet_gen = 0;
595
596
    s->packet_fin = 0;

597
    uint32_t h1 = 0, h2 = 0;
598
599
    bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
    fattree_terminal_magic_num = h1 + h2;
600

601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
    char anno[MAX_NAME_LENGTH];

    if(def_gname_set == 0) {
      def_gname_set = 1;
      codes_mapping_get_lp_info(0, def_group_name, &mapping_grp_id, NULL,
          &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
    }

    // Assign the global switch ID
    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,
           s->anno, 0);

   if(num_lps != (s->params->switch_radix[0]/2)) {
     tw_error(TW_LOC, "Number of NICs per repetition has to be equal to "
         "half the radix of leaf level switches %d vs %d\n", num_lps,
          s->params->switch_radix[0]/2);
   }
629
   s->terminal_id = (mapping_rep_id * num_lps) + mapping_offset;
630
631
632
633
634
   s->switch_id = s->terminal_id / (s->params->switch_radix[0] / 2);
   codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
           s->switch_id, 0, &s->switch_lp);
   s->terminal_available_time = 0.0;
   s->packet_counter = 0;
635
   s->terminal_msgs =
636
     (fattree_message_list**)malloc(1*sizeof(fattree_message_list*));
637
   s->terminal_msgs_tail =
638
639
     (fattree_message_list**)malloc(1*sizeof(fattree_message_list*));

640
641
642
643
644
645
   s->finished_msgs = 0;
   s->finished_chunks = 0;
   s->finished_packets = 0;
   s->total_time = 0.0;
   s->total_msg_size = 0;

646
647
648
   s->last_buf_full = 0;
   s->busy_time = 0;

649
#if FATTREE_HELLO
650
651
   printf("I am terminal %d (%llu), connected to switch %d\n", s->terminal_id,
       LLU(lp->gid), s->switch_id);
652
#endif
Noah Wolfe's avatar
Noah Wolfe committed
653
654
655

   rc_stack_create(&s->st);

656
657
658
659
660
661
662
   s->vc_occupancy = 0;
   s->terminal_msgs[0] = NULL;
   s->terminal_msgs_tail[0] = NULL;
   s->terminal_length = 0;
   s->in_send_loop = 0;
   s->issueIdle = 0;

663
664
665
666
   s->rank_tbl = qhash_init(fattree_rank_hash_compare, fattree_hash_func, FTREE_HASH_TABLE_SIZE);
   if(!s->rank_tbl)
     tw_error(TW_LOC, "\n Hash table not initialized! ");

667
   s->params->num_terminals = codes_mapping_get_lp_count(lp_group_name, 0,
668
669
670
671
672
673
674
      LP_CONFIG_NM, s->anno, 0);
   return;
}

/* sets up the switch */
void switch_init(switch_state * r, tw_lp * lp)
{
675
676
677
678
  uint32_t h1 = 0, h2 = 0;
  bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
  switch_magic_num = h1 + h2;

679
  char anno[MAX_NAME_LENGTH];
Noah Wolfe's avatar
Noah Wolfe committed
680
  int num_terminals = -1;
681
682
683
684
685

  if(def_gname_set == 0) {
    def_gname_set = 1;
    codes_mapping_get_lp_info(0, def_group_name, &mapping_grp_id, NULL,
        &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
686
    num_terminals = codes_mapping_get_lp_count(def_group_name, 0,
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
      LP_CONFIG_NM, anno, 0);
  }

  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
  fattree_param *p = r->params;
  if(mapping_offset == p->num_levels - 1) {
    if(mapping_rep_id >= p->num_switches[mapping_offset]) {
      r->unused = 1;
      return;
    }
  }

  r->unused = 0;

  r->switch_id = mapping_rep_id + mapping_offset * p->num_switches[0];

  if(num_terminals != -1) {
    p->num_terminals = num_terminals;
  }

  r->switch_level = mapping_offset;

  r->radix = p->switch_radix[r->switch_level];

723
  r->next_output_available_time = (tw_stime*) malloc (r->radix *
724
      sizeof(tw_stime));
725
  r->next_credit_available_time = (tw_stime*) malloc (r->radix *
726
727
728
729
730
      sizeof(tw_stime));
  r->vc_occupancy = (int*) malloc (r->radix * sizeof(int));
  r->in_send_loop = (int*) malloc (r->radix * sizeof(int));
  r->link_traffic = (int64_t*) malloc (r->radix * sizeof(int64_t));
  r->port_connections = (tw_lpid*) malloc (r->radix * sizeof(tw_lpid));
731
  r->pending_msgs =
732
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
733
  r->pending_msgs_tail =
734
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
735
  r->queued_msgs =
736
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
737
  r->queued_msgs_tail =
738
739
740
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
  r->queued_length = (int*)malloc(r->radix * sizeof(int));

741

742
743
744
745
  r->last_buf_full = (tw_stime*)malloc(r->radix * sizeof(tw_stime));
  r->busy_time = (tw_stime*)malloc(r->radix * sizeof(tw_stime));
  r->busy_time_sample = (tw_stime*)malloc(r->radix * sizeof(tw_stime));

Noah Wolfe's avatar
Noah Wolfe committed
746
747
  rc_stack_create(&r->st);

748
749
750
  for(int i = 0; i < r->radix; i++)
  {
    // Set credit & switch occupancy
751
752
753
    r->last_buf_full[i] = 0.0;
    r->busy_time[i] = 0.0;
    r->busy_time_sample[i] = 0.0;
754
755
756
757
758
759
760
761
762
763
764
765
    r->next_output_available_time[i] = 0;
    r->next_credit_available_time[i] = 0;
    r->vc_occupancy[i] = 0;
    r->in_send_loop[i] = 0;
    r->link_traffic[i] = 0;
    r->pending_msgs[i] = NULL;
    r->pending_msgs_tail[i] = NULL;
    r->queued_msgs[i] = NULL;
    r->queued_msgs_tail[i] = NULL;
    r->queued_length[i] = 0;
  }

766
#if FATTREE_CONNECTIONS
767
768
	int written = 0;
    int written_2 = 0;
769
770
    tw_lpid next_switch_lid;
#endif
771

772
773
774
775
  //set lps connected to each port
  r->num_cons = 0;
  r->num_lcons = 0;
#if FATTREE_HELLO
776
777
  printf("I am switch %d (%llu), level %d, radix %d\n", r->switch_id,
    LLU(lp->gid), r->switch_level, r->radix);
778
779
780
781
782
783
784
785
786
787
788
789
#endif
  //if at level 0, first half ports go to terminals
  if(r->switch_level == 0) {
    int start_terminal = r->switch_id * (p->switch_radix[0] / 2);
    int end_terminal = start_terminal + (p->switch_radix[0] / 2);
    for(int term = start_terminal; term < end_terminal; term++) {
      tw_lpid nextTerm;
      int rep = term / (p->switch_radix[0] / 2);
      int off = term % (p->switch_radix[0] / 2);
      codes_mapping_get_lp_id(def_group_name, LP_CONFIG_NM, NULL, 1,
          rep, off, &nextTerm);
      r->port_connections[r->num_cons++] = nextTerm;
790
791
792
#if FATTREE_CONNECTIONS
	  written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(codes_mapping_get_lp_relative_id(nextTerm,0,0)));
#endif
793
794
      r->num_lcons++;
#if FATTREE_DEBUG
795
796
      printf("I am switch %d, connect to terminal %d (%llu) at port %d yes collecting\n",
          r->switch_id, term, LLU(nextTerm), r->num_cons - 1);
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
#endif
    }
    r->start_lneigh = start_terminal;
    r->end_lneigh = end_terminal;
    r->con_per_lneigh = 1;
    assert(r->num_lcons == (r->radix / 2));
    int l1_set;
    if(p->num_levels == 2) {
      l1_set = 0;
    } else {
      l1_set = r->switch_id / p->l1_set_size;
    }
    int l1_base = l1_set * p->l1_set_size;
    r->start_uneigh = p->num_switches[0] + l1_base;
    r->con_per_uneigh = 1;
    for(int l1 = 0; l1 < p->l1_set_size; l1++) {
      tw_lpid nextTerm;
      codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
          l1_base, 1, &nextTerm);
      for(int con = 0; con < r->con_per_uneigh; con++) {
        r->port_connections[r->num_cons++] = nextTerm;
818
#if FATTREE_CONNECTIONS
819
820
821
        codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
            &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
        next_switch_lid = mapping_rep_id + mapping_offset * p->num_switches[0];
822
823
	    written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
824
#if FATTREE_DEBUG
825
826
      printf("I am switch %d, connect to upper switch %d L1 (%llu) rel_id:%llu at port %d yes collecting\n",
          r->switch_id, l1_base, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1);
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
#endif
      }
      l1_base++;
    }
  } else if (r->switch_level == 1) {
    int l0_set_size, l0_base;
    if(p->num_levels == 2) {
      l0_set_size = p->num_switches[0];
      l0_base = 0;
      r->start_lneigh = 0;
      r->end_lneigh = p->num_switches[0];
    } else {
      l0_set_size = p->l1_set_size;
      l0_base = ((r->switch_id - p->num_switches[0]) / p->l1_set_size) *
        l0_set_size;
      r->start_lneigh = l0_base;
      r->end_lneigh = l0_base + l0_set_size;
    }
    r->con_per_lneigh = 1;
    for(int l0 = 0; l0 < l0_set_size; l0++) {
      tw_lpid nextTerm;
      codes_mapping_get_lp_id(def_group_name, "fattree_switch", NULL, 1,
          l0_base, 0, &nextTerm);
      for(int con = 0; con < r->con_per_lneigh; con++) {
        r->port_connections[r->num_cons++] = nextTerm;
852
#if FATTREE_CONNECTIONS
853
854
855
        codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
            &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
        next_switch_lid = mapping_rep_id + mapping_offset * p->num_switches[0];
856
857
        written_2 += sprintf(r->output_buf2 + written_2, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
858
859
        r->num_lcons++;
#if FATTREE_DEBUG
860
861
        printf("I am switch %d, connect to switch %d L0 (%llu) at port %d not collecting\n",
            r->switch_id, l0_base, LLU(nextTerm), r->num_cons - 1);
862
863
864
865
866
#endif
      }
      l0_base++;
    }
    if(p->num_levels == 3) {
867
      int l2_base = 0;
868
      if(p->ft_type == 0) {
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
        for(int rep=0;rep<p->link_repetitions;rep++){
            int l2 = ((r->switch_id - p->num_switches[0]) % p->l1_set_size + rep*p->l1_set_size);
            printf("link_repetitions:%d\n",p->link_repetitions);
            /* not true anymore */
            r->start_uneigh = p->num_switches[0] + l2;
            r->con_per_uneigh = p->link_repetitions;
            for(; l2 < p->num_switches[2]; l2 += p->l1_set_size) {
              tw_lpid nextTerm;
              codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
                      l2, 2, &nextTerm);
              for(int con = 0; con < r->con_per_uneigh; con++) {
                r->port_connections[r->num_cons++] = nextTerm;
#if FATTREE_CONNECTIONS
                codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
                    &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
                next_switch_lid = mapping_rep_id + mapping_offset * p->num_switches[0];
                written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
#if FATTREE_DEBUG
                printf("I am switch %d, connect to upper switch %d L2 (%llu) at port %d yes collecting\n",
                    r->switch_id, l2+rep*p->l1_set_size, LLU(nextTerm), r->num_cons - 1);
#endif
              }
            }
        }
894
895
//        int l2_base = 0;
        /* not true anymore */
896
/*        r->start_uneigh = p->num_switches[0] + l2_base;
897
898
899
900
901
902
903
904
905
906
907
        r->con_per_uneigh = 1;
        if(((int)r->switch_id - p->num_switches[0]) % p->l1_set_size >=
            p->l1_set_size/2) {
          l2_base += (p->num_switches[2]/2);
        }
        for(int l2 = 0; l2 < p->num_switches[2]/2; l2++) {
          tw_lpid nextTerm;
          codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
              l2_base, 2, &nextTerm);
          for(int con = 0; con < r->con_per_uneigh; con++) {
            r->port_connections[r->num_cons++] = nextTerm;
908
#if FATTREE_CONNECTIONS
909
910
911
            codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
                &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
            next_switch_lid = mapping_rep_id + mapping_offset * p->num_switches[0];
912
913
	        written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
914
915
916
917
918
919
#if FATTREE_DEBUG
            printf("I am switch %d, connect to upper switch %d L2 (%llu) at port %d yes collecting\n",
                r->switch_id, l2_base, LLU(nextTerm), r->num_cons - 1);
#endif
          }
          l2_base++;
920
 */ 
921
      } else {
922
923
924
        int l2 = ((r->switch_id - p->num_switches[0]) % p->l1_set_size);
        /* not true anymore */
        r->start_uneigh = p->num_switches[0] + l2;
925
926
927
928
929
930
        if(p->ft_type == 1){
          r->con_per_uneigh = 1;
        }
        else{
          r->con_per_uneigh = 2;
        }
931
932
933
934
935
936
        for(; l2 < p->num_switches[2]; l2 += p->l1_set_size) {
          tw_lpid nextTerm;
          codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
              l2, 2, &nextTerm);
          for(int con = 0; con < r->con_per_uneigh; con++) {
            r->port_connections[r->num_cons++] = nextTerm;
937
#if FATTREE_CONNECTIONS
938
939
940
            codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
                &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
            next_switch_lid = mapping_rep_id + mapping_offset * p->num_switches[0];
941
942
            written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
943
#if FATTREE_DEBUG
944
945
            printf("I am switch %d, connect to upper switch %d L2 (%llu) at port %d yes collecting\n",
                r->switch_id, l2, LLU(nextTerm), r->num_cons - 1);
946
947
948
949
950
951
#endif
          }
        }
      }
    }
  } else {
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
    if(p->ft_type == 0) {
      r->con_per_lneigh = 1;
      /* not true anymore */
      r->start_lneigh = p->num_switches[0];
      r->end_lneigh = r->start_lneigh + p->num_switches[1];
      int l1 = 0;
      if((int)r->switch_id - p->num_switches[0] - p->num_switches[1] >=
          (p->num_switches[2]/2)) {
        l1 += (p->l1_set_size/2);
      }
      int count = 0;
      for(; l1 < p->num_switches[1]; l1++) {
        tw_lpid nextTerm;
        codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
            l1, 1, &nextTerm);
        for(int con = 0; con < r->con_per_lneigh; con++) {
          r->port_connections[r->num_cons++] = nextTerm;
969
#if FATTREE_CONNECTIONS
970
971
972
          codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
              &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
          next_switch_lid = mapping_rep_id + mapping_offset * p->num_switches[0];
973
974
	      written_2 += sprintf(r->output_buf2 + written_2, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
975
976
977
978
979
980
981
982
983
984
985
986
987
          r->num_lcons++;
#if FATTREE_DEBUG
          printf("I am switch %d, connect to  switch %d L1 (%llu) at port %d not collecting\n",
              r->switch_id, l1, LLU(nextTerm), r->num_cons - 1);
#endif
        }
        count++;
        if(count == (p->l1_set_size/2)) {
          l1 += (p->l1_set_size/2);
          count = 0;
        }
      }
    } else {
988
989
990
991
992
993
994
995
996
997
998
      r->con_per_lneigh = 2;
      /* not true anymore */
      r->start_lneigh = p->num_switches[0];
      r->end_lneigh = r->start_lneigh + p->num_switches[1];
      int l1 = (r->switch_id - p->num_switches[0] - p->num_switches[1]) % p->l1_set_size;
      for(; l1 < p->num_switches[1]; l1 += p->l1_set_size) {
        tw_lpid nextTerm;
        codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
            l1, 1, &nextTerm);
        for(int con = 0; con < r->con_per_lneigh; con++) {
          r->port_connections[r->num_cons++] = nextTerm;
999
#if FATTREE_CONNECTIONS
1000
1001
1002
          codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
              &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
          next_switch_lid = mapping_rep_id + mapping_offset * p->num_switches[0];
1003
1004
	      written_2 += sprintf(r->output_buf2 + written_2, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
1005
1006
          r->num_lcons++;
#if FATTREE_DEBUG
1007
1008
          printf("I am switch %d, connect to  switch %d L1 (%llu) at port %d not collecting\n",
              r->switch_id, l1, LLU(nextTerm), r->num_cons - 1);
1009
1010
1011
1012
1013
#endif
        }
      }
    }
  }
1014
#if FATTREE_CONNECTIONS
1015
1016
  lp_io_write(lp->gid, "fattree-config-up-connections", written, r->output_buf);
  lp_io_write(lp->gid, "fattree-config-down-connections", written_2, r->output_buf2);
1017
#endif
1018
  return;
1019
}
1020
1021

/* empty for now.. */
1022
static void fattree_report_stats()
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
{
#if DEBUG_RC
	long long t_packet_event_f = 0;
	long long t_packet_event_r = 0;
	long long tt_generate_f = 0;
	long long tt_generate_r = 0;
	long long tt_send_f = 0;
	long long tt_send_r = 0;
	long long tt_arrive_f = 0;
	long long tt_arrive_r = 0;
	long long tt_buffer_f = 0;
	long long tt_buffer_r = 0;
	long long ts_send_f = 0;
	long long ts_send_r = 0;
	long long ts_arrive_f = 0;
	long long ts_arrive_r = 0;
	long long ts_buffer_f = 0;
	long long ts_buffer_r = 0;
1041

1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
	MPI_Reduce( &packet_event_f, &t_packet_event_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &packet_event_r, &t_packet_event_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_generate_f, &tt_generate_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_generate_r, &tt_generate_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_send_f, &tt_send_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_send_r, &tt_send_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_arrive_f, &tt_arrive_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_arrive_r, &tt_arrive_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_buffer_f, &tt_buffer_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &t_buffer_r, &tt_buffer_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &s_send_f, &ts_send_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &s_send_r, &ts_send_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &s_arrive_f, &ts_arrive_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &s_arrive_r, &ts_arrive_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &s_buffer_f, &ts_buffer_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
	MPI_Reduce( &s_buffer_r, &ts_buffer_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);

   if(!g_tw_mynode)
   {
	printf("Reverse Computation Counters:\n");
	printf("packet_event_f:%llu\n",t_packet_event_f);
	printf("packet_event_r:%llu\n",t_packet_event_r);
	printf("t_gen_f:%llu\n",tt_generate_f);
	printf("t_gen_r:%llu\n",tt_generate_r);
	printf("t_gen_net:%llu\n",tt_generate_f-tt_generate_r);
	printf("t_send_f:%llu\n",tt_send_f);
	printf("t_send_r:%llu\n",tt_send_r);
	printf("t_send_net:%llu\n",tt_send_f-tt_send_r);
	printf("t_arrive_f:%llu\n",tt_arrive_f);
	printf("t_arrive_r:%llu\n",tt_arrive_r);
	printf("t_arrive_net:%llu\n",tt_arrive_f-tt_arrive_r);
	printf("t_buf_f:%llu\n",tt_buffer_f);
	printf("t_buf_r:%llu\n",tt_buffer_r);
	printf("t_buf_net:%llu\n",tt_buffer_f-tt_buffer_r);
	printf("s_send_f:%llu\n",ts_send_f);
	printf("s_send_r:%llu\n",ts_send_r);
	printf("s_send_net:%llu\n",ts_send_f-ts_send_r);
	printf("s_arrive_f:%llu\n",ts_arrive_f);
	printf("s_arrive_r:%llu\n",ts_arrive_r);
	printf("s_arrive_net:%llu\n",ts_arrive_f-ts_arrive_r);
	printf("s_buffer_f:%llu\n",ts_buffer_f);
	printf("s_buffer_r:%llu\n",ts_buffer_r);
	printf("s_buffer_net:%llu\n",ts_buffer_f-ts_buffer_r);
   }
#endif
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099

   long long avg_hops, total_finished_packets, total_finished_chunks;
   long long total_finished_msgs, final_msg_sz;
   tw_stime avg_time, max_time;
   long total_gen, total_fin;

   MPI_Reduce( &total_hops, &avg_hops, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
   MPI_Reduce( &N_finished_packets, &total_finished_packets, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
   MPI_Reduce( &N_finished_msgs, &total_finished_msgs, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
   MPI_Reduce( &N_finished_chunks, &total_finished_chunks, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
   MPI_Reduce( &total_msg_sz, &final_msg_sz, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
   MPI_Reduce( &fattree_total_time, &avg_time, 1,MPI_DOUBLE, MPI_SUM, 0, MPI_COMM_WORLD);
   MPI_Reduce( &fattree_max_latency, &max_time, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
1100

1101
1102
1103
1104
1105
   MPI_Reduce( &fattree_packet_gen, &total_gen, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_WORLD);
   MPI_Reduce( &fattree_packet_fin, &total_fin, 1, MPI_LONG, MPI_SUM, 0, MPI_COMM_WORLD);

   /* print statistics */
   if(!g_tw_mynode)
1106
1107
   {
      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",
1108
1109
1110
              (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);
      printf(" Total packets generated %ld finished %ld \n", total_gen, total_fin);
   }
1111
1112

   if(!g_tw_mynode)
1113
   {
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
      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 \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);

#if PARAMS_LOG
//    throughput_avg = throughput_avg / (float)slim_total_terminals_noah;
//    throughput_avg2 = throughput_avg2 / (float)slim_total_terminals_noah;

	//Open file to append simulation results
	char temp_filename[1024];
	char temp_filename_header[1024];
	sprintf(temp_filename,"%s/sim_log.txt",modelnet_stats_dir);
	sprintf(temp_filename_header,"%s/sim_log_header.txt",modelnet_stats_dir);
	FILE *fattree_results_log=fopen(temp_filename, "a");
	FILE *fattree_results_log_header=fopen(temp_filename_header, "a");
	if(fattree_results_log == NULL)
		printf("\n Failed to open results log file %s \n",temp_filename);
	if(fattree_results_log_header == NULL)
		printf("\n Failed to open results log header file %s \n",temp_filename_header);
	printf("Printing Simulation Parameters/Results Log File\n");
	fprintf(fattree_results_log_header,"<Avg Hops/Total Packets>, <Avg Time/Total Packets>, <Max Latency>, <Total Finished Chunks>");
Noah Wolfe's avatar
Noah Wolfe committed
1133
	fprintf(fattree_results_log,"%24.3lf, %24.3lf, %13.3lf, %23.3lld, ", (float)avg_hops/total_finished_packets, avg_time/(total_finished_packets),max_time,total_finished_chunks);
1134
1135
	fclose(fattree_results_log_header);
	fclose(fattree_results_log);
1136
#endif
1137
  }
1138
}
1139
1140

/* fattree packet event */
Noah Wolfe's avatar
Noah Wolfe committed
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
static tw_stime fattree_packet_event(
        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)
Nikhil's avatar
Nikhil committed
1151
{
1152
1153
1154
#if DEBUG_RC
  packet_event_f++;
#endif
1155

Noah Wolfe's avatar
Noah Wolfe committed
1156
1157
  (void)message_offset;
  (void)sched_params;
1158
1159
1160
1161
1162
1163
1164
1165
  tw_event * e_new;
  tw_stime xfer_to_nic_time;
  fattree_message * msg;
  char* tmp_ptr;

  xfer_to_nic_time = codes_local_latency(sender);
  e_new = model_net_method_event_new(sender->gid, xfer_to_nic_time + offset,
      sender, FATTREE, (void**)&msg, (void**)&tmp_ptr);
Noah Wolfe's avatar
Noah Wolfe committed
1166
1167
1168
1169
1170
  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;
1171
  msg->packet_size = packet_size;
Noah Wolfe's avatar
Noah Wolfe committed
1172
  msg->travel_start_time = tw_now(sender);
1173
1174
1175
  msg->remote_event_size_bytes = 0;
  msg->local_event_size_bytes = 0;
  msg->type = T_GENERATE;
Noah Wolfe's avatar
Noah Wolfe committed
1176
1177
1178
1179
  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;
1180
  msg->magic = fattree_terminal_magic_num;
Noah Wolfe's avatar
Noah Wolfe committed
1181
  msg->msg_start_time = req->msg_start_time;
1182
1183

  /* Its the last packet so pass in remote and local event information*/
1184
  if(is_last_pckt)
1185
  {
1186
    if(req->remote_event_size > 0)
1187
    {
1188
1189
1190
      msg->remote_event_size_bytes = req->remote_event_size;
      memcpy(tmp_ptr, remote_event, req->remote_event_size);
      tmp_ptr += req->remote_event_size;
1191
    }
1192
    if(req->self_event_size > 0)
1193
    {
1194
1195
1196
      msg->local_event_size_bytes = req->self_event_size;
      memcpy(tmp_ptr, self_event, req->self_event_size);
      tmp_ptr += req->self_event_size;
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
    }
  }
  //printf("[%d] Send to %d\n", sender->gid, sender->gid);
  tw_event_send(e_new);
  return xfer_to_nic_time;
}

/* fattree packet event reverse handler */
static void fattree_packet_event_rc(tw_lp *sender)
{
1207
1208
1209
#if DEBUG_RC
  packet_event_r++;
#endif
1210
1211
1212
1213
  codes_local_latency_reverse(sender);
  return;
}

1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
void ft_packet_generate_rc(ft_terminal_state * s, tw_bf * bf, fattree_message * msg, tw_lp * lp)
{
#if DEBUG_RC
    t_generate_r++;
#endif
    fattree_packet_gen--;
    s->packet_gen--;

    tw_rand_reverse_unif(lp->rng);

    int num_chunks = msg->packet_size/s->params->chunk_size;
    if(msg->packet_size % s->params->chunk_size)
	num_chunks++;

    if(!num_chunks)
	num_chunks = 1;

    int i;
    for(i = 0; i < num_chunks; i++) {
1233
	delete_fattree_message_list(return_tail(s->terminal_msgs,
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
	    s->terminal_msgs_tail, 0));
	s->terminal_length -= s->params->chunk_size;
    }
    if(bf->c11) {
	s->issueIdle = 0;
	s->last_buf_full = msg->saved_busy_time;
    }
    if(bf->c5) {
	codes_local_latency_reverse(lp);
	s->in_send_loop = 0;
    }
1245

1246
1247
1248
1249
1250
1251
1252
    struct mn_stats* stat;
    stat = model_net_find_stats(msg->category, s->fattree_stats_array);
    stat->send_count--;
    stat->send_bytes -= msg->packet_size;
    stat->send_time -= (1/s->params->cn_bandwidth) * msg->packet_size;
}

1253
1254
1255
/* generates packet at the current fattree compute node */
void ft_packet_generate(ft_terminal_state * s, tw_bf * bf, fattree_message * msg,
    tw_lp * lp) {
1256
1257
1258
  bf->c11 = 0;
  bf->c5 = 0;

1259
1260
  fattree_packet_gen++;
  s->packet_gen++;
1261

1262
1263
1264
  fattree_param *p = s->params;
  tw_stime ts, nic_ts;

1265
1266
1267
1268
1269
1270
  int total_event_size;
  uint64_t num_chunks = msg->packet_size / p->chunk_size;
  if(msg->packet_size % s->params->chunk_size)
    num_chunks++;
  if(!num_chunks)
    num_chunks = 1;
1271

Nikhil's avatar
Nikhil committed
1272
  nic_ts = g_tw_lookahead + (msg->packet_size * s->params->cn_delay) + tw_rand_unif(lp->rng);
1273

1274
1275
  msg->my_N_hop = 0;

1276
  msg->packet_ID = lp->gid + g_tw_nlp * s->packet_counter;
1277
1278
  if(msg->packet_ID == LLU(TRACK_PKT))
    printf("\n Packet %llu generated at terminal %d terminal_gid %llu dest_terminal_id %llu final_dest_gid %llu size %llu num chunks %llu \n",
Noah Wolfe's avatar
Noah Wolfe committed
1279
       msg->packet_ID, s->terminal_id, LLU(lp->gid), LLU(msg->dest_terminal_id), LLU(msg->final_dest_gid),
1280
1281
1282
1283
       LLU(msg->packet_size), LLU(num_chunks));

  for(uint64_t i = 0; i < num_chunks; i++)
  {
1284
    fattree_message_list * cur_chunk = (fattree_message_list *)malloc(
1285
      sizeof(fattree_message_list));
1286
1287
    msg->origin_switch_id = s->switch_id;
    init_fattree_message_list(cur_chunk, msg);
1288

1289
1290
    if(msg->remote_event_size_bytes + msg->local_event_size_bytes > 0) {
      cur_chunk->event_data = (char*)malloc(
1291
        msg->remote_event_size_bytes + msg->local_event_size_bytes);
1292
    }
1293

1294
1295
1296
1297
1298
1299
    void *m_data_src = model_net_method_get_edata(FATTREE, msg);
    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;
1300
      memcpy((char*)cur_chunk->event_data + msg->remote_event_size_bytes,
1301
        m_data_src, msg->local_event_size_bytes);
1302
    }
1303

1304
1305
    cur_chunk->msg.chunk_id = i;
    cur_chunk->msg.origin_switch_id = s->switch_id;
1306
    append_to_fattree_message_list(s->terminal_msgs, s->terminal_msgs_tail,
1307
1308
1309
      0, cur_chunk);
    s->terminal_length += s->params->chunk_size;
  }
1310
1311
1312
1313
1314
1315

  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;
1316
1317
    msg->saved_busy_time = s->last_buf_full;
    s->last_buf_full = tw_now(lp);
1318
1319
1320
1321
1322
  }

  if(s->in_send_loop == 0) {
    fattree_message *m;
    bf->c5 = 1;
1323
    ts = codes_local_latency(lp);
1324
    tw_event* e = model_net_method_event_new(lp->gid, ts, lp, FATTREE,
1325
1326
      (void**)&m, NULL);
    m->type = T_SEND;
1327
    m->magic = fattree_terminal_magic_num;