fattree.c 99.2 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"
Nikhil's avatar
Nikhil committed
13
14
#include <ctype.h>
#include <search.h>
15

Matthieu Dorier's avatar
Matthieu Dorier committed
16
17
18
19
20
#ifdef ENABLE_CORTEX
#include <cortex/cortex.h>
#include <cortex/topology.h>
#endif

21
22
23
#define CREDIT_SIZE 8
#define MEAN_PROCESS 1.0

Nikhil's avatar
Nikhil committed
24
#define TERMINAL_GUID_PREFIX ((uint64_t)(64) << 32)
25
#define FTREE_HASH_TABLE_SIZE 4999
26

27
// debugging parameters
28
29
#define TRACK_PKT -1
//#define TRACK_PKT 2820
30
#define FATTREE_HELLO 0
31
#define FATTREE_DEBUG 0
32
33
#define FATTREE_CONNECTIONS 0
#define FATTREE_MSG 0
Nikhil's avatar
Nikhil committed
34
#define DEBUG_RC 0
35

36
//Data Collection Output Files
Nikhil's avatar
Nikhil committed
37
#define PARAMS_LOG 0
38

39
40
41
#define LP_CONFIG_NM (model_net_lp_config_names[FATTREE])
#define LP_METHOD_NM (model_net_method_names[FATTREE])

42
43
44
45
46
#ifdef ENABLE_CORTEX
/* This structure is defined at the end of the file */
extern cortex_topology fattree_cortex_topology;
#endif

47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
#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

67
long fattree_packet_gen = 0, fattree_packet_fin = 0;
Nikhil's avatar
Nikhil committed
68
int dump_topo = 0;
69

70
71
72
static double maxd(double a, double b) { return a < b ? b : a; }

// arrival rate
Noah Wolfe's avatar
Noah Wolfe committed
73
//static double MEAN_INTERVAL=200.0;
74
75

typedef struct fattree_param fattree_param;
76
/* annotation-specific parameters (unannotated entry occurs at the
77
78
79
 * last index) */
static uint64_t                  num_params = 0;
static fattree_param           * all_params = NULL;
Noah Wolfe's avatar
Noah Wolfe committed
80
static const config_anno_map_t * anno_map   = NULL;
81
82
83
84
85
86
87

/* 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;

Nikhil's avatar
Nikhil committed
88
89
enum ROUTING_ALGO
{
Nikhil's avatar
Nikhil committed
90
    STATIC=1,
Nikhil's avatar
Nikhil committed
91
92
    ADAPTIVE,
};
Nikhil's avatar
Nikhil committed
93
94
95
96
97
98

enum RAIL_SELECTION_ALGO
{
    RAIL_DEDICATED=1,
    RAIL_ADAPTIVE,
};
Nikhil's avatar
Nikhil committed
99
100
101
static char routing_folder[MAX_NAME_LENGTH];
static char dot_file_p[MAX_NAME_LENGTH];

102
103
104
/* switch magic number */
int switch_magic_num = 0;

105
/* terminal magic number */
106
int fattree_terminal_magic_num = 0;
107

Nikhil's avatar
Nikhil committed
108
109
110
/* global variables for DOT file generation */
static FILE *dot_file = NULL;

111
112
113
114
115
116
117
118
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;
};

119
void init_fattree_message_list(fattree_message_list *this,
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
  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;
137
138
139
  int link_repetitions;
  int Ns; // number of switches per pod
  int Np; //Number of pods
140
141
142
  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 */
143
  double cn_bandwidth;/* bandwidth of the compute node channels
144
145
146
                        connected to switch */
  int vc_size; /* buffer size of the link channels */
  int cn_vc_size; /* buffer size of the compute node channels */
147
  int chunk_size; /* full-sized packets are broken into smaller chunks.*/
148
  int packet_size;
149
  int num_terminals;
Nikhil's avatar
Nikhil committed
150
151
152
153
  int l1_set_size, l0_set_size;
  int l1_term_size, l0_term_size;
  int cn_per_switch;
  double tapering_ratio;
154
155
156
157
158
  double cn_delay;
  double head_delay;
  double credit_delay;
  double router_delay;
  double soft_delay;
Nikhil's avatar
Nikhil committed
159
  int routing;
Nikhil's avatar
Nikhil committed
160
161
162
163
  int rail_select;
  int rail_size_limit;
  int num_rails;
  int ports_per_nic;
164
165
};

166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
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;
};

181
182
183
184
185
186
187
188
189
/* 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;
190
  int packet_gen;
191
  int packet_fin;
192
193
194
  // Fattree specific parameters
  unsigned int terminal_id;
  unsigned int switch_id;
Nikhil's avatar
Nikhil committed
195
196
  unsigned int rail_id;
  tw_lpid *switch_lp;
197
198

  // Each terminal will have an input and output channel with the switch
Nikhil's avatar
Nikhil committed
199
200
  int *vc_occupancy; // NUM_VC
  tw_stime* terminal_available_time;
201

202
203
  struct mn_stats fattree_stats_array[CATEGORY_MAX];

204
205
  fattree_message_list **terminal_msgs;
  fattree_message_list **terminal_msgs_tail;
Nikhil's avatar
Nikhil committed
206
207
208
  int *terminal_length;
  int *in_send_loop;
  int *issueIdle;
209
210

   struct rc_stack * st;
211
212
213

  char * anno;
  fattree_param *params;
214
215
216
217
218
219
220
221
222
223
224

  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;

Nikhil's avatar
Nikhil committed
225
226
  tw_stime *last_buf_full;
  tw_stime *busy_time;
227
  char output_buf[4096];
228
  char output_buf3[4096];
229

230
231
232
233
234
  /* For sampling */
  long fin_chunks_sample;
  long data_size_sample;
  double fin_hops_sample;
  tw_stime fin_chunks_time;
Nikhil's avatar
Nikhil committed
235
236
  tw_stime busy_time_sample;

237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
};

/* 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;
Nikhil's avatar
Nikhil committed
261
  unsigned int rail_id;
262
263
264
265
266
267
268
269
270
271
272
  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;
273
274
275
276
277
  tw_stime* last_buf_full;

  tw_stime* busy_time;
  tw_stime* busy_time_sample;

278
279
280
281
282
283
284
285
286
287
  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;

288
289
290
  char output_buf[4096];
  char output_buf2[4096];

Noah Wolfe's avatar
Noah Wolfe committed
291
292
  struct rc_stack * st;

293
294
  char * anno;
  fattree_param *params;
Nikhil's avatar
Nikhil committed
295
296
  /* array to store linear forwaring tables in case we use static routing */
  int *lft;
297
298
};

299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
/* ROSS Instrumentation Support */
struct fattree_sample
{
   tw_lpid lpid;
};

void fattree_event_collect(fattree_message *m, tw_lp *lp, char *buffer, int *collect_flag);
void fattree_model_stat_collect(ft_terminal_state *s, tw_lp *lp, char *buffer);
static void ross_fattree_sample_fn(ft_terminal_state * s, tw_bf * bf, tw_lp * lp, struct fattree_sample *sample);
static void ross_fattree_sample_rc_fn(ft_terminal_state * s, tw_bf * bf, tw_lp * lp, struct fattree_sample *sample);
static void ross_fattree_ssample_fn(switch_state * s, tw_bf * bf, tw_lp * lp, struct fattree_sample *sample);
static void ross_fattree_ssample_rc_fn(switch_state * s, tw_bf * bf, tw_lp * lp, struct fattree_sample *sample);

st_model_types fattree_model_types[] = {
    {(rbev_trace_f) fattree_event_collect,
     sizeof(int),
     (ev_trace_f) fattree_event_collect,
     sizeof(int),
     (model_stat_f) fattree_model_stat_collect,
     0, // update when changing fattree_model_stat_collect
     (sample_event_f) ross_fattree_sample_fn,
     (sample_revent_f) ross_fattree_sample_rc_fn,
     sizeof(struct fattree_sample) } , 
    {(rbev_trace_f) fattree_event_collect,
     sizeof(int),
     (ev_trace_f) fattree_event_collect,
     sizeof(int),
     (model_stat_f) fattree_model_stat_collect,
     0, // update when changing fattree_model_stat_collect
     (sample_event_f) ross_fattree_ssample_fn,
     (sample_revent_f) ross_fattree_ssample_rc_fn,
     sizeof(struct fattree_sample) } , 
    {NULL, 0, NULL, 0, NULL, 0, NULL, NULL, 0}
};
/* End of ROSS model stats collection */

335
336
static tw_stime         fattree_total_time = 0;
static tw_stime         fattree_max_latency = 0;
Noah Wolfe's avatar
Noah Wolfe committed
337
//static tw_stime         max_collective = 0;
338
339
340
341
342
343
344
345


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;

346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
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));
}

370
371
static void free_tmp(void * ptr)
{
372
    struct ftree_qhash_entry * ftree = ptr;
373
374
375
376
    free(ftree->remote_event_data);
    free(ftree);
}

377
static void append_to_fattree_message_list(
378
379
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
380
        int index,
381
382
383
384
385
386
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
387
    }
388
389
390
    thistail[index] = msg;
}

391
static void prepend_to_fattree_message_list(
392
393
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
394
        int index,
395
396
397
398
399
400
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
401
    }
402
403
404
405
406
407
408
409
410
411
412
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
    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
447
//static const tw_lptype* fattree_get_switch_lp_type(void);
448

449
/* returns the fattree message size */
450
451
452
453
454
static int fattree_get_msg_sz(void)
{
  return sizeof(fattree_message);
}

Nikhil's avatar
Nikhil committed
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
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
582
583
584
585
586
587
588
589
590
591
static inline uint64_t get_term_guid(ft_terminal_state *t)
{
  return TERMINAL_GUID_PREFIX + t->terminal_id;
}

static inline uint64_t get_switch_guid(switch_state *s)
{
  return (((uint64_t)(s->switch_level + 1)) << 32) + s->switch_id;
}

typedef struct {
  uint64_t guid;
  uint64_t port;
} guid_port_combi_t;

static int cmp_guids(const void *g1, const void *g2)
{
  uint64_t guid1 = *((uint64_t *) g1);
  uint64_t guid2 = ((guid_port_combi_t *) g2)->guid;
  if (guid1 == guid2)
    return 0;
  else
    return 1;
}

/* parse external file with give forwarding tables
 * file name: 0x<switch guid>.lft
 * content: '0x<terminal guid> <egress port>' per line, e.g.:
 * `head 0x0000000100000000.lf`
 *    0x0000000100000000 0
 *    0x00000040000000ff 22
 *    0x0000000100000001 19
 */
static int read_static_lft(switch_state *s, tw_lp *lp)
{
  if (!s || !lp)
    return -1;

  char dir_name[512];
  char file_name[512];

  char *io_dir = routing_folder;
  sprintf(dir_name, "%s", io_dir);

  sprintf(file_name, "%s/0x%016"PRIx64".lft", dir_name, get_switch_guid(s));
  FILE *file = NULL;
  if (!(file = fopen(file_name, "r")))
    return -1;

  char line[UINT8_MAX];
  char *p = NULL, *e = NULL;
  uint64_t dest_guid = 0, port = 0;

  int num_added = 0;
  guid_port_combi_t *tmpLFT = calloc(s->params->num_terminals,
        sizeof(guid_port_combi_t));
  s->lft = malloc(s->params->num_terminals * sizeof(int));
  /* init all with -1 so that we find missing routing entries */
  for (int i = 0; i < s->params->num_terminals; i++) s->lft[i] = -1;

  while (fgets(line, sizeof(line), file)) {
    p = line;
    while (isspace(*p))
      p++;

    if (*p == '\0' || *p == '\n' || *p == '#')
      continue;

    dest_guid = strtoull(p, &e, 16);
    if (e == p || (!isspace(*e) && *e != '#' && *e != '\0')) {
      errno = EINVAL;
      return -1;
    }

    p = e;
    while (isspace(*p))
      p++;

    port = strtoull(p, &e, 0);
    if (e == p || (!isspace(*e) && *e != '#' && *e != '\0')) {
      errno = EINVAL;
      return -1;
    }

    if (dest_guid < TERMINAL_GUID_PREFIX){
      // we don't need switches
    } else {
      tmpLFT[num_added].guid = dest_guid;
      tmpLFT[num_added].port = port;
      num_added++;
    }
  }

  /* we want a real LFT with terminal_id as lookup index, so have to
   * convert the read lft, which might contain entries for switches
   * or is permutated, into the correct format
   */
  ft_terminal_state tmpTerm;
  for (int dest_num = 0; dest_num < s->params->num_terminals; dest_num++) {
    tmpTerm.terminal_id = dest_num;
    uint64_t key = get_term_guid(&tmpTerm);
    size_t size = s->params->num_terminals;
    guid_port_combi_t *elem = lfind(&key, tmpLFT, &size,
          sizeof(guid_port_combi_t), cmp_guids);
    assert(elem);
    // opensm uses ports=1...n, so convert back here
    //if(NULL != getenv("OSM_ROUTING"))
       s->lft[dest_num] = elem->port - 1;
    //else
    //   s->lft[dest_num] = elem->port;
  }

#if FATTREE_DEBUG
  printf("I am switch %d (guid=%016"PRIx64") and my LFT is:\n",
        s->switch_id, get_switch_guid(s));
  for (int dest_num = 0; dest_num < s->params->num_terminals; dest_num++)
     printf("\tdest %d -> egress port %d\n", dest_num, s->lft[dest_num]);
#endif

  free(tmpLFT);
  fclose(file);
  return 0;
}

static void dot_write_open_file(FILE **fout)
{
  if(!fout || *fout) return;
  if(!dump_topo) return;

  char dir_name[512];
  char file_name[512];

  char *io_dir = routing_folder;
  sprintf(dir_name, "%s", io_dir);

  char *dot_file_prefix = dot_file_p;
  int rank;
592
  MPI_Comm_rank(MPI_COMM_CODES, &rank);
Nikhil's avatar
Nikhil committed
593
594
595
596
597
598
599
  sprintf(file_name, "%s/%s.dot.%d", dir_name, dot_file_prefix, rank);
  *fout = fopen(file_name, "w+");
  if(*fout == NULL) {
    tw_error(TW_LOC, "Error opening dot file for writing");
  }
}

600
/*static void dot_write_close_file(FILE **fout)
Nikhil's avatar
Nikhil committed
601
602
603
604
605
{
  if(!fout || !(*fout)) return;
  if(!dump_topo) return;
  fclose(*fout);
  *fout = NULL;
606
}*/
Nikhil's avatar
Nikhil committed
607
608

/* sw IDs aren't unique, but level+ID info is */
609
static void dot_write_switch_info(switch_state *s, FILE *fout)
Nikhil's avatar
Nikhil committed
610
611
612
{
  if(!s || s->unused || !(s->params) || !fout) return;
  if(!dump_topo) return;
Nikhil's avatar
Nikhil committed
613
  if(s->rail_id) return;
Nikhil's avatar
Nikhil committed
614
615
616
617
618
619

  const char *root_attr = ",root_switch", *empty_attr = "";
  uint64_t switch_guid = get_switch_guid(s);
  fattree_param *p = s->params;

  fprintf(fout, "\t\"S_%d_%d\" [comment=\"0x%016"PRIx64",radix=%d%s\"];\n",
Nikhil's avatar
Nikhil committed
620
      s->switch_level, s->switch_id, switch_guid, s->radix,
Nikhil's avatar
Nikhil committed
621
622
623
624
625
626
627
628
      (p->num_levels == s->switch_level + 1) ? root_attr : empty_attr);
}

/* terminal IDs are unique in this model */
static void dot_write_term_info(ft_terminal_state *t, FILE *fout)
{
  uint64_t term_guid = t->terminal_id;
  if(!dump_topo) return;
Nikhil's avatar
Nikhil committed
629
  if(t->rail_id) return;
Nikhil's avatar
Nikhil committed
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688

  if(!t || !fout) return;
  /* need to shift the term guid, because opensm doesn't like guid=0...0 */
  //if(NULL != getenv("OSM_ROUTING"))
     term_guid = get_term_guid(t);
  fprintf(fout, "\t\"H_%d\" [comment=\"0x%016"PRIx64"\"];\n",
        t->terminal_id, term_guid);
}

static void dot_write_sw2sw_link(int lsw_lvl, int lsw_id, int lsw_port,
                                 int rsw_lvl, int rsw_id, int rsw_port,
                                 FILE *fout)
{
  int tmp_lsw_port = 0, tmp_rsw_port = 0;

  if(!fout) return;
  if(!dump_topo) return;

  /* add +1 to ports, since ibsim/opensm start with physical port number 1
   * (port 0 is internal sw port)
   */
  //if(NULL != getenv("OSM_ROUTING")) {
     tmp_lsw_port = lsw_port + 1;
     tmp_rsw_port = rsw_port + 1;
  //} else {
  //   tmp_lsw_port = lsw_port;
  //   tmp_rsw_port = rsw_port;
  //}

  /* if original right hand side is unknown we write '?' instead
   * and must fix it later ... nice topology definition ... :-(
   */
  if(-1 == rsw_port)
     fprintf(fout, "\t\"S_%d_%d\" -> \"S_%d_%d\" [comment=\"P%d->P?\"];\n",
           lsw_lvl, lsw_id, rsw_lvl, rsw_id, tmp_lsw_port);
  else
     fprintf(fout, "\t\"S_%d_%d\" -> \"S_%d_%d\" [comment=\"P%d->P%d\"];\n",
           lsw_lvl, lsw_id, rsw_lvl, rsw_id, tmp_lsw_port, tmp_rsw_port);
}

static void dot_write_sw2term_link(int lsw_lvl, int lsw_id, int lsw_port,
                                   int term_id, int dump_reverse, FILE *fout)
{
  if(!fout) return;
  if(!dump_topo) return;

  /* add +1 to ports, since ibsim/opensm start with physical port number 1
   * (port 0 is internal sw port)
   */
  //if(NULL != getenv("OSM_ROUTING"))
     lsw_port++;

  fprintf(fout, "\t\"S_%d_%d\" -> \"H_%d\" [comment=\"P%d->P1\"];\n",
        lsw_lvl, lsw_id, term_id, lsw_port);
  if(dump_reverse)
     fprintf(fout, "\t\"H_%d\" -> \"S_%d_%d\" [comment=\"P1->P%d\"];\n",
           term_id, lsw_lvl, lsw_id, lsw_port);
}

689
/*static void dot_write_term2sw_link(int term_id, int rsw_lvl, int rsw_id,
Nikhil's avatar
Nikhil committed
690
691
692
693
694
                                   int rsw_port, FILE *fout)
{
  if(!fout) return;
  if(!dump_topo) return;

695
  *//* add +1 to ports, since ibsim/opensm start with physical port number 1
Nikhil's avatar
Nikhil committed
696
697
698
   * (port 0 is internal sw port)
   */
  //if(NULL != getenv("OSM_ROUTING"))
699
  //   rsw_port++;
Nikhil's avatar
Nikhil committed
700

701
702
703
  //fprintf(fout, "\t\"H_%d\" -> \"S_%d_%d\" [comment=\"P1->P%d\"];\n",
  //      term_id, rsw_lvl, rsw_id, rsw_port);
//}
Nikhil's avatar
Nikhil committed
704
705
706
707
708
709
710
711
712
713
714
715
716
717

void post_switch_init(switch_state *s, tw_lp *lp)
{
  /* read any LFTs which might have been generated by an external routing
   * algorithm, e.g., through the use of opensm\
   */
  if(s->params->routing == STATIC && !dump_topo) {
   if(0 != read_static_lft(s, lp)) {
     tw_error(TW_LOC, "Error while reading the routing table");
   }
  }
}


Noah Wolfe's avatar
Noah Wolfe committed
718
static void fattree_read_config(const char * anno, fattree_param *p){
719
720
721
722
723
724
725
  uint32_t h1 = 0, h2 = 0;
  bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
  switch_magic_num = h1 + h2;
    
  bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
  fattree_terminal_magic_num = h1 + h2;
  
726
727
  int i;

Nikhil's avatar
Nikhil committed
728
  p->ft_type = 1;
729
  configuration_get_value_int(&config, "PARAMS", "ft_type", anno,
730
      &p->ft_type);
Nikhil's avatar
Nikhil committed
731
  if(!g_tw_mynode) printf("FT type is %d\n", p->ft_type);
732

733
  configuration_get_value_int(&config, "PARAMS", "num_levels", anno,
734
735
736
737
738
739
740
      &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");
  }
Nikhil's avatar
Nikhil committed
741
742
743
744
745
746
747
748
  if(!g_tw_mynode) printf("FT num levels is %d\n", p->num_levels);
  
  p->tapering_ratio = 1;
  configuration_get_value_double(&config, "PARAMS", "tapering", anno,
      &p->tapering_ratio);
  if(!g_tw_mynode) printf("FT tapering is %lf\n", p->tapering_ratio);

  p->tapering_ratio += 1;
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793

  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,",");
  }

Nikhil's avatar
Nikhil committed
794
795
796
797
798
  if(fmod(p->switch_radix[0],  p->tapering_ratio)) {
    tw_error(TW_LOC, "Tapering ratio does not fit the given radix; tapering + 1 should"
    " evenly divide the L0 switch radix\n");
  }

799
  p->Ns = p->switch_radix[0]/2;
Nikhil's avatar
Nikhil committed
800
801
802
803
804
805
806
807
808
809
810
  if(p->num_switches[0] % p->Ns) {
    if(!g_tw_mynode) printf("Number of switches has to be a multiple of switch radix/2 (%d);" 
        "next largest value is %d\n", p->Ns, (p->num_switches[0]/p->Ns + 1)*p->Ns);
    tw_error(TW_LOC, "Incorrect number of switches");
  }

  int num_terminals = p->num_switches[0] * (p->tapering_ratio - 1)*(p->switch_radix[0]/p->tapering_ratio);
  p->num_terminals = num_terminals;
    
  p->l0_set_size = p->switch_radix[0]/2;
  p->l0_term_size = (p->tapering_ratio - 1)*(p->switch_radix[0]/p->tapering_ratio);
811

812
  if(p->num_levels == 2) {
Nikhil's avatar
Nikhil committed
813
    p->num_switches[1] = p->num_switches[0]/p->tapering_ratio/2;
814
    p->switch_radix[1] = p->switch_radix[0];
Nikhil's avatar
Nikhil committed
815
816
    p->l1_set_size =  p->num_switches[1];
    p->l1_term_size = num_terminals;
817
  } else {
Nikhil's avatar
Nikhil committed
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
    p->switch_radix[1] = p->switch_radix[2] = p->switch_radix[0];
	  p->l1_set_size = p->switch_radix[0]/p->tapering_ratio;
    p->l1_term_size = p->l0_set_size * p->l0_term_size;
    p->Np = p->num_switches[0]/p->l0_set_size;
    p->num_switches[1] =  p->Np * p->l1_set_size;
    p->link_repetitions = p->switch_radix[0]/p->Np;
    if(p->l1_set_size % p->link_repetitions) {
      while(p->l1_set_size % p->link_repetitions) {
        p->link_repetitions--;
      }
    } 
    int halfPods = p->l1_set_size/p->link_repetitions;
    p->num_switches[2] = halfPods * p->Ns;
    if(!g_tw_mynode) printf("Pods %d; L2 pods %d; Link repetition:%d; l1_set_size %d\n", 
        p->Np, halfPods, p->link_repetitions, p->l1_set_size);
  }

  if(p->ft_type == 1 && p->num_levels == 3) {
    if(p->Np != p->switch_radix[0] && p->Np !=  p->switch_radix[0]/2) {
      if(!g_tw_mynode) {
        printf("ft_type 1 should be used for a system with %d or %d pods, which "
        "needs %d or %d L0 switches, respectively\n",  p->switch_radix[0], 
         p->switch_radix[0]/2,  p->switch_radix[0] *  p->switch_radix[0]/2,
          p->switch_radix[0]/2 *  p->switch_radix[0]/2);
      }
      tw_error(TW_LOC, "Incorrect number of switches for ft_type 1");
    }
845
846
  }

Nikhil's avatar
Nikhil committed
847
  for(int jj=0;jj<3 && !g_tw_mynode;jj++)
848
849
  {
    printf("num_switches[%d]=%d\n",jj,p->num_switches[jj]);
850
851
  }

Nikhil's avatar
Nikhil committed
852
  rc = configuration_get_value_int(&config, "PARAMS", "packet_size", anno,
853
      &p->packet_size);
Nikhil's avatar
Nikhil committed
854
  if(rc) {
855
    p->packet_size = 512;
Nikhil's avatar
Nikhil committed
856
857
858
859
860
861
862
863
864
865
866
  }
  if(!g_tw_mynode) printf("FT packet size is %d\n", p->packet_size);

  p->num_rails = 1;
  configuration_get_value_int(&config, "PARAMS", "num_rails", anno, &p->num_rails);
  if(!g_tw_mynode) printf("FT num rails is %d\n", p->num_rails);
  
  if(p->ft_type == 2) {
    p->ports_per_nic = p->num_rails;
  } else {
    p->ports_per_nic = 1;
867
  }
868

869
870
871
  p->router_delay = 50;
  configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
      &p->router_delay);
Nikhil's avatar
Nikhil committed
872
  if(!g_tw_mynode) printf("FT router delay is %.1lf\n", p->router_delay);
873

874
875
876
877
  p->soft_delay = 1000;
  configuration_get_value_double(&config, "PARAMS", "soft_delay", anno,
      &p->soft_delay);

Nikhil's avatar
Nikhil committed
878
879
880
  rc = configuration_get_value_int(&config, "PARAMS", "vc_size", anno, &p->vc_size);
  if(rc) {
    p->vc_size = 8 * p->packet_size;
881
  }
Nikhil's avatar
Nikhil committed
882
  if(!g_tw_mynode) printf("FT buffer size links is %d\n", p->vc_size);
883

Nikhil's avatar
Nikhil committed
884
  rc = configuration_get_value_int(&config, "PARAMS", "cn_vc_size", anno,
885
      &p->cn_vc_size);
Nikhil's avatar
Nikhil committed
886
887
  if(rc) {
    p->cn_vc_size = 8 * p->packet_size;
888
  }
Nikhil's avatar
Nikhil committed
889
  if(!g_tw_mynode) printf("FT buffer size injection is %d\n", p->cn_vc_size);
890

891
  rc = configuration_get_value_int(&config, "PARAMS", "chunk_size", anno, &p->chunk_size);
Nikhil's avatar
Nikhil committed
892
893
894
  if(rc) {
    p->chunk_size = 512;
  }
895

Nikhil's avatar
Nikhil committed
896
897
898
899
900
901
902
903
904
905
906
  char routing_str[MAX_NAME_LENGTH];
  configuration_get_value(&config, "PARAMS", "routing", anno, routing_str,
      MAX_NAME_LENGTH);
  if(strcmp(routing_str, "static") == 0)
    p->routing = STATIC;
  else if(strcmp(routing_str, "adaptive")==0)
    p->routing = ADAPTIVE;
  else
  {
    p->routing = ADAPTIVE;
  }
Nikhil's avatar
Nikhil committed
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
  if(!g_tw_mynode) printf("FT routing is %d\n", p->routing);
  
  char rail_select_str[MAX_NAME_LENGTH];
  configuration_get_value(&config, "PARAMS", "rail_select", anno, rail_select_str,
      MAX_NAME_LENGTH);
  if(strcmp(rail_select_str, "dedicated") == 0)
    p->rail_select = RAIL_DEDICATED;
  else if(strcmp(rail_select_str, "adaptive")==0)
    p->rail_select = RAIL_ADAPTIVE;
  else {
    p->rail_select = RAIL_DEDICATED;
  }
  if(!g_tw_mynode) printf("FT rail selection is %d\n", p->rail_select);

  p->rail_size_limit = 32768;
  configuration_get_value_int(&config, "PARAMS", "rail_select_limit", anno,
      &p->rail_size_limit);
  if(!g_tw_mynode) printf("FT rail select limit  is %d\n", p->rail_size_limit);
Nikhil's avatar
Nikhil committed
925
926
927
  
  configuration_get_value_int(&config, "PARAMS", "dump_topo", anno,
      &dump_topo);
928

Nikhil's avatar
Nikhil committed
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
  routing_folder[0] = '\0';
  rc = configuration_get_value(&config, "PARAMS", "routing_folder", anno, routing_folder,
      MAX_NAME_LENGTH);

  if(routing_folder[0] == '\0') {
    if(dump_topo || p->routing == STATIC) {
      tw_error(TW_LOC, "routing_folder has to be provided with dump_topo || static routing");
    }
  }
 
  dot_file_p[0] = '\0'; 
  rc = configuration_get_value(&config, "PARAMS", "dot_file", anno, dot_file_p,
      MAX_NAME_LENGTH);
  
  if(dot_file_p[0] == '\0') {
    if(dump_topo) {
      tw_error(TW_LOC, "dot_file has to be provided with dump_topo");
    }
  }
  
Nikhil's avatar
Nikhil committed
949
  rc = configuration_get_value_double(&config, "PARAMS", "link_bandwidth", anno,
950
      &p->link_bandwidth);
Nikhil's avatar
Nikhil committed
951
  if(rc) {
952
953
954
    p->link_bandwidth = 5;
  }

Nikhil's avatar
Nikhil committed
955
  rc = configuration_get_value_double(&config, "PARAMS", "cn_bandwidth", anno,
956
957
958
959
      &p->cn_bandwidth);
  if(!p->cn_bandwidth) {
    p->cn_bandwidth = 5;
  }
Nikhil's avatar
Nikhil committed
960
961
962
  
  if(!g_tw_mynode) printf("FT bandwidths link %.1lf node %.1lf\n", 
      p->link_bandwidth, p->cn_bandwidth);
963

964
965
966
967
#if FATTREE_DEBUG
  printf("l1_set_size:%d l1_term_size:%d\n",p->l1_set_size,p->l1_term_size);
#endif

968
969
970
971
972
973
974
975
976
977
978
  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
979
980
  for (int i = 0; i < anno_map->num_annos; i++){
    const char * anno = anno_map->annotations[i].ptr;
981
982
983
984
985
    fattree_read_config(anno, &all_params[i]);
  }
  if (anno_map->has_unanno_lp > 0){
    fattree_read_config(NULL, &all_params[anno_map->num_annos]);
  }
986
987
988
#ifdef ENABLE_CORTEX
  model_net_topology = fattree_cortex_topology;
#endif
989
990
991
992
993
}

/* initialize a fattree compute node terminal */
void ft_terminal_init( ft_terminal_state * s, tw_lp * lp )
{
994
    s->packet_gen = 0;
995
996
    s->packet_fin = 0;

997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
    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);

Nikhil's avatar
Nikhil committed
1020
1021
   if(s->params->ft_type != 2) {
    num_lps /= s->params->num_rails;
1022
   }
Nikhil's avatar
Nikhil committed
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051

   if(num_lps != s->params->l0_term_size) {
     tw_error(TW_LOC, "Number of NICs per repetition per rail has to be equal to "
         "%d, not the given value of %d\n", s->params->l0_term_size, num_lps);
   }
  
   if(s->params->ft_type != 2) {
    s->terminal_id = (mapping_rep_id * num_lps) + (mapping_offset/s->params->num_rails);
    s->rail_id = (mapping_offset % s->params->num_rails);
   } else {
    s->terminal_id = (mapping_rep_id * num_lps) + mapping_offset;
    s->rail_id = 0;
   }
   s->switch_id = s->terminal_id / s->params->l0_term_size;
   s->switch_lp = (tw_lpid*)malloc(s->params->ports_per_nic * sizeof(tw_lpid));
   if(s->params->ft_type != 2) {
     codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
         s->switch_id, 0 + s->params->num_levels * s->rail_id, &s->switch_lp[0]);
   } else {
     for(int i = 0; i < s->params->ports_per_nic; i++) {
       codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
           s->switch_id, 0 + s->params->num_levels * i, &s->switch_lp[i]);
     }
   }
   s->terminal_available_time = (tw_stime*) malloc(s->params->ports_per_nic * sizeof(tw_stime));
   s->vc_occupancy = (int*) malloc(s->params->ports_per_nic * sizeof(int));
   s->terminal_length = (int*) malloc(s->params->ports_per_nic * sizeof(int));
   s->in_send_loop = (int*) malloc(s->params->ports_per_nic * sizeof(int));
   s->issueIdle = (int*) malloc(s->params->ports_per_nic * sizeof(int));
1052
   s->terminal_msgs =
Nikhil's avatar
Nikhil committed
1053
     (fattree_message_list**)malloc(s->params->ports_per_nic*sizeof(fattree_message_list*));
1054
   s->terminal_msgs_tail =
Nikhil's avatar
Nikhil committed
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
     (fattree_message_list**)malloc(s->params->ports_per_nic*sizeof(fattree_message_list*));
   s->last_buf_full = (tw_stime*) malloc(s->params->ports_per_nic * sizeof(tw_stime));
   s->busy_time = (tw_stime*) malloc(s->params->ports_per_nic * sizeof(tw_stime));
   for(int i = 0; i < s->params->ports_per_nic; i++) {
     s->terminal_available_time[i] = 0.0;
     s->vc_occupancy[i] = 0;
     s->terminal_msgs[i] = NULL;
     s->terminal_msgs_tail[i] = NULL;
     s->terminal_length[i] = 0;
     s->in_send_loop[i] = 0;
     s->issueIdle[i] = 0;
     s->last_buf_full[i] = 0;
     s->busy_time[i] = 0;
   }
   s->packet_counter = 0;
1070
1071
1072
1073
1074
1075
   s->finished_msgs = 0;
   s->finished_chunks = 0;
   s->finished_packets = 0;
   s->total_time = 0.0;
   s->total_msg_size = 0;

Nikhil's avatar
Nikhil committed
1076
1077
1078
#if FATTREE_DEBUG
   printf("I am terminal %d (%llu), connected to switch %d in rail id %d\n", s->terminal_id,
       LLU(lp->gid), s->switch_id, s->rail_id);
1079
#endif
Noah Wolfe's avatar
Noah Wolfe committed
1080
1081
1082

   rc_stack_create(&s->st);

1083
1084
1085
   s->rank_tbl = NULL;
   //if(!s->rank_tbl)
   //  tw_error(TW_LOC, "\n Hash table not initialized! ");
1086

Nikhil's avatar
Nikhil committed
1087
1088
1089
1090
   /* dump partial topology into DOT format
    * skip term2sw link part, because we are missing the remote switch port
    * in this stage of the code
    */
Nikhil's avatar
Nikhil committed
1091
   if(!dot_file && !s->rail_id)
Nikhil's avatar
Nikhil committed
1092
1093
1094
1095
     dot_write_open_file(&dot_file);
   dot_write_term_info(s, dot_file);

   /* ensure the DOT file is out before reading it with external tools */
Nikhil's avatar
Nikhil committed
1096
   if(dot_file && !s->rail_id)
Nikhil's avatar
Nikhil committed
1097
1098
     fflush(dot_file);

1099
1100
1101
1102
1103
1104
   return;
}

/* sets up the switch */
void switch_init(switch_state * r, tw_lp * lp)
{
1105

1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
  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);
  }

  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;
Nikhil's avatar
Nikhil committed
1128
1129
1130
  if(mapping_rep_id >= p->num_switches[mapping_offset % p->num_levels]) {
    r->unused = 1;
    return;
1131
1132
1133
1134
  }

  r->unused = 0;

Nikhil's avatar
Nikhil committed
1135
1136
1137
1138
1139
1140
  if((mapping_offset % p->num_levels) == 0) {
    r->switch_id = mapping_rep_id;
  } else if((mapping_offset % p->num_levels) ==  1) {
    r->switch_id = mapping_rep_id + p->num_switches[0];
  } else {
    r->switch_id = mapping_rep_id + p->num_switches[0] + p->num_switches[1];
1141
  }
Nikhil's avatar
Nikhil committed
1142
  r->rail_id = mapping_offset / p->num_levels;
1143

Nikhil's avatar
Nikhil committed
1144
  r->switch_level = mapping_offset % p->num_levels;
1145
1146
1147

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

1148
  r->next_output_available_time = (tw_stime*) malloc (r->radix *
1149
1150
1151
1152
1153
      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));
1154
  r->pending_msgs =
1155
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
1156
  r->pending_msgs_tail =
1157
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
1158
  r->queued_msgs =
1159
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
1160
  r->queued_msgs_tail =
1161
1162
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
  r->queued_length = (int*)malloc(r->radix * sizeof(int));
Nikhil's avatar
Nikhil committed
1163
  r->lft = NULL;
1164

1165
1166
1167
1168
  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
1169
1170
  rc_stack_create(&r->st);

1171
1172
1173
  for(int i = 0; i < r->radix; i++)
  {
    // Set credit & switch occupancy
1174
1175
1176
    r->last_buf_full[i] = 0.0;
    r->busy_time[i] = 0.0;
    r->busy_time_sample[i] = 0.0;
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
    r->next_output_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;
  }

Nikhil's avatar
Nikhil committed
1188
  /* dump partial topology info into DOT format (switch radix, guid, ...) */
Nikhil's avatar
Nikhil committed
1189
  if(!dot_file && !r->rail_id)
Nikhil's avatar
Nikhil committed
1190
    dot_write_open_file(&dot_file);
1191
  dot_write_switch_info(r, dot_file);
Nikhil's avatar
Nikhil committed
1192

1193
1194
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
    tw_lpid next_switch_lid;
1195
1196
	int written = 0;
    int written_2 = 0;
1197
#endif
1198

1199
1200
1201
  //set lps connected to each port
  r->num_cons = 0;
  r->num_lcons = 0;
Nikhil's avatar
Nikhil committed
1202
1203
1204
#if FATTREE_DEBUG
  printf("I am switch %d (%llu), level %d, radix %d, rail id %d\n", r->switch_id,
    LLU(lp->gid), r->switch_level, r->radix, r->rail_id);
1205
1206
1207
#endif
  //if at level 0, first half ports go to terminals
  if(r->switch_level == 0) {
Nikhil's avatar
Nikhil committed
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
    int term_rails, term_railid;
    if(p->ft_type != 2) {
      term_rails = p->num_rails;
      term_railid = r->rail_id;
    } else {
      term_rails = 1;
      term_railid = 0;
    }
    int start_terminal = r->switch_id * p->l0_term_size;
    int end_terminal = start_terminal + p->l0_term_size;
1218
1219
    for(int term = start_terminal; term < end_terminal; term++) {
      tw_lpid nextTerm;
Nikhil's avatar
Nikhil committed
1220
1221
      int rep = term / p->l0_term_size;
      int off = (term % p->l0_term_size) * term_rails + term_railid;
1222
1223
1224
      codes_mapping_get_lp_id(def_group_name, LP_CONFIG_NM, NULL, 1,
          rep, off, &nextTerm);
      r->port_connections[r->num_cons++] = nextTerm;
1225
#if FATTREE_CONNECTIONS
Nikhil's avatar
Nikhil committed
1226
	    written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(codes_mapping_get_lp_relative_id(nextTerm,0,0)/term_rails));
1227
#endif
1228
1229
      r->num_lcons++;
#if FATTREE_DEBUG
Nikhil's avatar
Nikhil committed
1230
1231
      printf("L0->term I am switch %d, connect to terminal %d (%llu) at port %d, rail id %d yes collecting\n",
          r->switch_id, term, LLU(nextTerm), r->num_cons - 1, r->rail_id);
1232
#endif
Nikhil's avatar
Nikhil committed
1233
      /* write sw2term links and (reverse link, too) into DOT file */
Nikhil's avatar
Nikhil committed
1234
1235
1236
      if(!r->rail_id) {
        dot_write_sw2term_link(r->switch_level, r->switch_id, r->num_cons - 1, term, 1, dot_file);
      }
1237
1238
1239
1240
    }
    r->start_lneigh = start_terminal;
    r->end_lneigh = end_terminal;
    r->con_per_lneigh = 1;
Nikhil's avatar
Nikhil committed
1241
    assert(r->num_lcons == p->l0_term_size);
1242
1243
1244
    int l1_set;
    if(p->num_levels == 2) {
      l1_set = 0;
Nikhil's avatar
Nikhil committed
1245
      r->con_per_uneigh = 2;
1246
    } else {
Nikhil's avatar
Nikhil committed
1247
      l1_set = r->switch_id / p->l0_set_size;
Nikhil's avatar
Nikhil committed
1248
      r->con_per_uneigh = 1;
1249
1250
1251
1252
1253
    }
    int l1_base = l1_set * p->l1_set_size;
    r->start_uneigh = p->num_switches[0] + l1_base;
    for(int l1 = 0; l1 < p->l1_set_size; l1++) {
      tw_lpid nextTerm;
Nikhil's avatar
Nikhil committed
1254
      assert(l1_base < p->num_switches[1]);
1255
      codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,
Nikhil's avatar
Nikhil committed
1256
          l1_base, 1 + r->rail_id * p->num_levels, &nextTerm);
1257
1258
      for(int con = 0; con < r->con_per_uneigh; con++) {
        r->port_connections[r->num_cons++] = nextTerm;
Nikhil's avatar
Nikhil committed
1259
#if FATTREE_CONNECTIONS
1260
1261
        codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
            &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
Nikhil's avatar
Nikhil committed
1262
1263
        next_switch_lid = mapping_rep_id +  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);
1264
#endif
1265
#if FATTREE_DEBUG
Nikhil's avatar
Nikhil committed
1266
1267
    printf("L0->L1 I am switch %d, connect to upper switch %d L1 (%llu) rel_id:%llu at port %d, rail id %d yes collecting\n",
          r->switch_id, l1_base, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1, r->rail_id);
1268
#endif
Nikhil's avatar
Nikhil committed
1269
        /* write all inter switch links to DOT file (initialized before) */
Nikhil's avatar
Nikhil committed
1270
1271
1272
1273
        if(!r->rail_id) {
          dot_write_sw2sw_link(r->switch_level, r->switch_id, r->num_cons - 1,
              r->switch_level + 1, p->num_switches[0] + l1_base, -1, dot_file);
        }
1274
1275
1276
1277
      }
      l1_base++;
    }
  } else if (r->switch_level == 1) {
Nikhil's avatar
Nikhil committed
1278
    int l0_base;
1279
1280
1281
1282
    if(p->num_levels == 2) {
      l0_base = 0;
      r->start_lneigh = 0;
      r->end_lneigh = p->num_switches[0];
Nikhil's avatar
Nikhil committed
1283
      r->con_per_lneigh = 2;
1284
1285
    } else {
      l0_base = ((r->switch_id - p->num_switches[0]) / p->l1_set_size) *
Nikhil's avatar
Nikhil committed
1286
        p->l0_set_size;
1287
      r->start_lneigh = l0_base;
Nikhil's avatar
Nikhil committed
1288
      r->end_lneigh = l0_base + p->l0_set_size;
Nikhil's avatar
Nikhil committed
1289
      r->con_per_lneigh = 1;
1290
    }
Nikhil's avatar
Nikhil committed
1291
    for(int l0 = 0; l0 < p->l0_set_size; l0++) {
1292
1293
      tw_lpid nextTerm;
      codes_mapping_get_lp_id(def_group_name, "fattree_switch", NULL, 1,
Nikhil's avatar
Nikhil committed
1294
          l0_base, 0 + r->rail_id * p->num_levels, &nextTerm);
1295
1296
      for(int con = 0; con < r->con_per_lneigh; con++) {
        r->port_connections[r->num_cons++] = nextTerm;
Nikhil's avatar
Nikhil committed
1297
#if FATTREE_CONNECTIONS
1298
1299
        codes_mapping_get_lp_info(nextTerm, lp_group_name, &mapping_grp_id, NULL,
            &mapping_type_id, anno, &mapping_rep_id, &mapping_offset);
Nikhil's avatar
Nikhil committed
1300
        next_switch_lid = mapping_rep_id;
1301
1302
        written_2 += sprintf(r->output_buf2 + written_2, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
1303
1304
        r->num_lcons++;
#if FATTREE_DEBUG
Nikhil's avatar
Nikhil committed
1305
1306
        printf("L1->L0 I am switch %d, connect to switch %d L0 (%llu) rel_id:%llu at port %d, rail id %d not collecting\n",
            r->switch_id, l0_base, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1, r->rail_id);
1307
#endif
Nikhil's avatar
Nikhil committed
1308
        /* write all inter switch links to DOT file (initialized before) */
Nikhil's avatar
Nikhil committed
1309
1310
1311
1312
        if(!r->rail_id) {
          dot_write_sw2sw_link(r->switch_level, r->switch_id, r->num_cons - 1,
              r->switch_level - 1, l0_base, -1, dot_file);
        }
1313
1314
1315
1316
      }
      l0_base++;
    }
    if(p->num_levels == 3) {
Nikhil's avatar
Nikhil committed
1317
1318
1319
1320
      int rep = p->link_repetitions;
      int l2 = ((r->switch_id - p->num_switches[0]) % p->l1_set_size)/rep * p->Ns;
      r->con_per_uneigh = 1;
      for(int off = 0; off < p->Ns; off++) {
1321
        tw_lpid nextTerm;
Nikhil's avatar
Nikhil committed
1322
        assert(l2 + off < p->num_switches[2]);
1323
        codes_mapping_get_lp_id(lp_group_name, "fattree_switch", NULL, 1,