fattree.c 95.4 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"
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
90
91
92
93
94
95
enum ROUTING_ALGO
{
    STATIC=0,
    ADAPTIVE,
};
static char routing_folder[MAX_NAME_LENGTH];
static char dot_file_p[MAX_NAME_LENGTH];

96
97
98
/* switch magic number */
int switch_magic_num = 0;

99
/* terminal magic number */
100
int fattree_terminal_magic_num = 0;
101

Nikhil's avatar
Nikhil committed
102
103
104
/* global variables for DOT file generation */
static FILE *dot_file = NULL;

105
106
107
108
109
110
111
112
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;
};

113
void init_fattree_message_list(fattree_message_list *this,
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
  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;
131
132
133
  int link_repetitions;
  int Ns; // number of switches per pod
  int Np; //Number of pods
134
135
136
  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 */
137
  double cn_bandwidth;/* bandwidth of the compute node channels
138
139
140
                        connected to switch */
  int vc_size; /* buffer size of the link channels */
  int cn_vc_size; /* buffer size of the compute node channels */
141
  int chunk_size; /* full-sized packets are broken into smaller chunks.*/
142
  int packet_size;
143
144
145
146
147
148
149
150
  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;
Nikhil's avatar
Nikhil committed
151
  int routing;
152
153
};

154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
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;
};

169
170
171
172
/* convert GiB/s and bytes to ns */
 static tw_stime bytes_to_ns(uint64_t bytes, double GB_p_s)
 {
     tw_stime time;
173

174
175
176
177
178
179
     /* 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;
180

181
182
183
     return(time);
 }

184
185
186
187
188
189
190
191
192
/* 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;
193
  int packet_gen;
194
  int packet_fin;
195
196
197
198
199
200
201
202
203
  // 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;
204

205
206
  struct mn_stats fattree_stats_array[CATEGORY_MAX];

207
208
  fattree_message_list **terminal_msgs;
  fattree_message_list **terminal_msgs_tail;
209
  int terminal_length;
210
  int in_send_loop;
211
212
213
  int issueIdle;

   struct rc_stack * st;
214
215
216

  char * anno;
  fattree_param *params;
217
218
219
220
221
222
223
224
225
226
227

  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;

228
  tw_stime last_buf_full;
229
  tw_stime busy_time;
230
  char output_buf[4096];
231
  char output_buf3[4096];
232

233
234
235
236
237
  /* For sampling */
  long fin_chunks_sample;
  long data_size_sample;
  double fin_hops_sample;
  tw_stime fin_chunks_time;
Nikhil's avatar
Nikhil committed
238
239
  tw_stime busy_time_sample;

240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
};

/* 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;
276
277
278
279
280
  tw_stime* last_buf_full;

  tw_stime* busy_time;
  tw_stime* busy_time_sample;

281
282
283
284
285
286
287
288
289
290
  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;

291
292
293
  char output_buf[4096];
  char output_buf2[4096];

Noah Wolfe's avatar
Noah Wolfe committed
294
295
  struct rc_stack * st;

296
297
  char * anno;
  fattree_param *params;
Nikhil's avatar
Nikhil committed
298
299
  /* array to store linear forwaring tables in case we use static routing */
  int *lft;
300
301
};

302
303
static tw_stime         fattree_total_time = 0;
static tw_stime         fattree_max_latency = 0;
Noah Wolfe's avatar
Noah Wolfe committed
304
//static tw_stime         max_collective = 0;
305
306
307
308
309
310
311
312


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;

313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
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));
}

337
338
static void free_tmp(void * ptr)
{
339
    struct ftree_qhash_entry * ftree = ptr;
340
341
342
343
    free(ftree->remote_event_data);
    free(ftree);
}

344
static void append_to_fattree_message_list(
345
346
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
347
        int index,
348
349
350
351
352
353
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thisq[index] = msg;
    } else {
        thistail[index]->next = msg;
        msg->prev = thistail[index];
354
    }
355
356
357
    thistail[index] = msg;
}

358
static void prepend_to_fattree_message_list(
359
360
        fattree_message_list ** thisq,
        fattree_message_list ** thistail,
361
        int index,
362
363
364
365
366
367
        fattree_message_list *msg) {
    if(thisq[index] == NULL) {
        thistail[index] = msg;
    } else {
        thisq[index]->prev = msg;
        msg->next = thisq[index];
368
    }
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
    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
414
//static const tw_lptype* fattree_get_switch_lp_type(void);
415

416
/* returns the fattree message size */
417
418
419
420
421
static int fattree_get_msg_sz(void)
{
  return sizeof(fattree_message);
}

Nikhil's avatar
Nikhil committed
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
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
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;
559
  MPI_Comm_rank(MPI_COMM_CODES, &rank);
Nikhil's avatar
Nikhil committed
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
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
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
  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");
  }
}

static void dot_write_close_file(FILE **fout)
{
  if(!fout || !(*fout)) return;
  if(!dump_topo) return;
  fclose(*fout);
  *fout = NULL;
}

/* sw IDs aren't unique, but level+ID info is */
static void dot_write_switch_info(switch_state *s, int sw_gid, FILE *fout)
{
  if(!s || s->unused || !(s->params) || !fout) return;
  if(!dump_topo) return;

  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",
      s->switch_level, sw_gid, switch_guid, s->radix,
      (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;

  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);
}

static void dot_write_term2sw_link(int term_id, int rsw_lvl, int rsw_id,
                                   int rsw_port, 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"))
     rsw_port++;

  fprintf(fout, "\t\"H_%d\" -> \"S_%d_%d\" [comment=\"P1->P%d\"];\n",
        term_id, rsw_lvl, rsw_id, rsw_port);
}

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
683
static void fattree_read_config(const char * anno, fattree_param *p){
684
685
  int i;

Nikhil's avatar
Nikhil committed
686
  p->ft_type = 1;
687
  configuration_get_value_int(&config, "PARAMS", "ft_type", anno,
688
689
      &p->ft_type);

690
  configuration_get_value_int(&config, "PARAMS", "num_levels", anno,
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
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
      &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");
  }

  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,",");
  }

743
  p->Ns = p->switch_radix[0]/2;
744
  int num_terminals = p->num_switches[0]*p->switch_radix[0]/2;
745

746
747
748
  if(p->num_levels == 2) {
    p->num_switches[1] = p->num_switches[0]/2;
    p->switch_radix[1] = p->switch_radix[0];
Nikhil's avatar
Nikhil committed
749
750
751
    p->l1_set_size = p->switch_radix[0]/4;
    p->l1_term_size = (p->l1_set_size * p->switch_radix[0]);

752
  } else {
753
	if(p->ft_type == 0){
754
      p->Np = ceil((num_terminals)/(double)(p->Ns*p->Ns));
755
756
757
758
      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);
759
      printf("Np:%d Ns:%d\n",p->Np,p->Ns);
760
761
	}else{
	  p->num_switches[1] = p->num_switches[0];
762
763
      p->num_switches[2] = p->num_switches[0]/2;
      p->switch_radix[1] = p->switch_radix[2] = p->switch_radix[0];
764
765
	}
	p->l1_set_size = p->switch_radix[0]/2;
Nikhil's avatar
Nikhil committed
766
    p->l1_term_size = (p->l1_set_size * (p->switch_radix[0] / 2));
767
768
  }

769
#if FATTREE_CONNECTIONS
770
771
772
773
774
775
776
  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

777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
  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]) {
      tw_error(TW_LOC, "Not enough switches/radix at level %d for full "
          "bisection bandwidth\n", i);
    }
  }

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

  configuration_get_value_int(&config, "PARAMS", "packet_size", anno,
      &p->packet_size);
  if(!p->packet_size) {
    p->packet_size = 512;
796
    fprintf(stderr, "Packet size is not specified, setting to %d\n",
797
798
        p->packet_size);
  }
799

800
801
802
  p->router_delay = 50;
  configuration_get_value_double(&config, "PARAMS", "router_delay", anno,
      &p->router_delay);
803

804
805
806
807
808
809
810
811
812
813
814
  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);
  }

815
  configuration_get_value_int(&config, "PARAMS", "cn_vc_size", anno,
816
817
818
      &p->cn_vc_size);
  if(!p->cn_vc_size) {
    p->cn_vc_size = 8*p->packet_size;
819
    fprintf(stderr, "Buffer size of compute node channels not specified, "
820
821
822
        "setting to %d\n", p->cn_vc_size);
  }

823
824
825
826
827
828
  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);
    }

Nikhil's avatar
Nikhil committed
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
  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
  {
    fprintf(stderr, 
        "No routing protocol specified, setting to adaptive routing\n");
    p->routing = ADAPTIVE;
  }
  
  configuration_get_value_int(&config, "PARAMS", "dump_topo", anno,
      &dump_topo);
845

Nikhil's avatar
Nikhil committed
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
  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");
    }
  }
  
866
  configuration_get_value_double(&config, "PARAMS", "link_bandwidth", anno,
867
868
869
      &p->link_bandwidth);
  if(!p->link_bandwidth) {
    p->link_bandwidth = 5;
870
    fprintf(stderr, "Bandwidth of links is specified, setting to %lf\n",
871
872
873
        p->link_bandwidth);
  }

874
  configuration_get_value_double(&config, "PARAMS", "cn_bandwidth", anno,
875
876
877
      &p->cn_bandwidth);
  if(!p->cn_bandwidth) {
    p->cn_bandwidth = 5;
878
    fprintf(stderr, "Bandwidth of compute node channels not specified, "
879
880
881
        "setting to %lf\n", p->cn_bandwidth);
  }

882
883
884
885
#if FATTREE_DEBUG
  printf("l1_set_size:%d l1_term_size:%d\n",p->l1_set_size,p->l1_term_size);
#endif

886
887
888
889
890
891
892
893
894
895
896
  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
897
898
  for (int i = 0; i < anno_map->num_annos; i++){
    const char * anno = anno_map->annotations[i].ptr;
899
900
901
902
903
    fattree_read_config(anno, &all_params[i]);
  }
  if (anno_map->has_unanno_lp > 0){
    fattree_read_config(NULL, &all_params[anno_map->num_annos]);
  }
904
905
906
#ifdef ENABLE_CORTEX
  model_net_topology = fattree_cortex_topology;
#endif
907
908
909
910
911
}

/* initialize a fattree compute node terminal */
void ft_terminal_init( ft_terminal_state * s, tw_lp * lp )
{
912
    s->packet_gen = 0;
913
914
    s->packet_fin = 0;

915
    uint32_t h1 = 0, h2 = 0;
916
917
    bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
    fattree_terminal_magic_num = h1 + h2;
918

919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
    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);
   }
947
   s->terminal_id = (mapping_rep_id * num_lps) + mapping_offset;
948
949
950
951
952
   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;
953
   s->terminal_msgs =
954
     (fattree_message_list**)malloc(1*sizeof(fattree_message_list*));
955
   s->terminal_msgs_tail =
956
957
     (fattree_message_list**)malloc(1*sizeof(fattree_message_list*));

958
959
960
961
962
963
   s->finished_msgs = 0;
   s->finished_chunks = 0;
   s->finished_packets = 0;
   s->total_time = 0.0;
   s->total_msg_size = 0;

964
965
966
   s->last_buf_full = 0;
   s->busy_time = 0;

967
#if FATTREE_HELLO
968
969
   printf("I am terminal %d (%llu), connected to switch %d\n", s->terminal_id,
       LLU(lp->gid), s->switch_id);
970
#endif
Noah Wolfe's avatar
Noah Wolfe committed
971
972
973

   rc_stack_create(&s->st);

974
975
976
977
978
979
980
   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;

981
982
983
   s->rank_tbl = NULL;
   //if(!s->rank_tbl)
   //  tw_error(TW_LOC, "\n Hash table not initialized! ");
984

985
   s->params->num_terminals = codes_mapping_get_lp_count(lp_group_name, 0,
986
      LP_CONFIG_NM, s->anno, 0);
Nikhil's avatar
Nikhil committed
987
988
989
990
991
992
993
994
995
996
997
998
   /* dump partial topology into DOT format
    * skip term2sw link part, because we are missing the remote switch port
    * in this stage of the code
    */
   if(!dot_file)
     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 */
   if(dot_file)
     fflush(dot_file);

999
1000
1001
1002
1003
1004
   return;
}

/* sets up the switch */
void switch_init(switch_state * r, tw_lp * lp)
{
1005
1006
1007
1008
  uint32_t h1 = 0, h2 = 0;
  bj_hashlittle2(LP_METHOD_NM, strlen(LP_METHOD_NM), &h1, &h2);
  switch_magic_num = h1 + h2;

1009
  char anno[MAX_NAME_LENGTH];
Noah Wolfe's avatar
Noah Wolfe committed
1010
  int num_terminals = -1;
1011
1012
1013
1014
1015

  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);
1016
    num_terminals = codes_mapping_get_lp_count(def_group_name, 0,
1017
1018
1019
1020
1021
1022
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
1052
      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];

1053
  r->next_output_available_time = (tw_stime*) malloc (r->radix *
1054
      sizeof(tw_stime));
1055
  r->next_credit_available_time = (tw_stime*) malloc (r->radix *
1056
1057
1058
1059
1060
      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));
1061
  r->pending_msgs =
1062
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
1063
  r->pending_msgs_tail =
1064
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
1065
  r->queued_msgs =
1066
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
1067
  r->queued_msgs_tail =
1068
1069
    (fattree_message_list**)malloc(r->radix * sizeof(fattree_message_list*));
  r->queued_length = (int*)malloc(r->radix * sizeof(int));
Nikhil's avatar
Nikhil committed
1070
  r->lft = NULL;
1071

1072
1073
1074
1075
  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
1076
1077
  rc_stack_create(&r->st);

1078
1079
1080
  for(int i = 0; i < r->radix; i++)
  {
    // Set credit & switch occupancy
1081
1082
1083
    r->last_buf_full[i] = 0.0;
    r->busy_time[i] = 0.0;
    r->busy_time_sample[i] = 0.0;
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
    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;
  }

Nikhil's avatar
Nikhil committed
1096
1097
1098
1099
1100
  /* dump partial topology info into DOT format (switch radix, guid, ...) */
  if(!dot_file)
    dot_write_open_file(&dot_file);
  dot_write_switch_info(r, lp->gid, dot_file);

1101
1102
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
    tw_lpid next_switch_lid;
1103
1104
	int written = 0;
    int written_2 = 0;
1105
#endif
1106

1107
1108
1109
1110
  //set lps connected to each port
  r->num_cons = 0;
  r->num_lcons = 0;
#if FATTREE_HELLO
1111
1112
  printf("I am switch %d (%llu), level %d, radix %d\n", r->switch_id,
    LLU(lp->gid), r->switch_level, r->radix);
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
#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;
1125
1126
1127
#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
1128
1129
      r->num_lcons++;
#if FATTREE_DEBUG
1130
      printf("L0->term I am switch %d, connect to terminal %d (%llu) at port %d yes collecting\n",
1131
          r->switch_id, term, LLU(nextTerm), r->num_cons - 1);
1132
#endif
Nikhil's avatar
Nikhil committed
1133
1134
      /* write sw2term links and (reverse link, too) into DOT file */
      dot_write_sw2term_link(r->switch_level, lp->gid, r->num_cons - 1, term, 1, dot_file);
1135
1136
1137
1138
1139
1140
1141
1142
    }
    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;
Nikhil's avatar
Nikhil committed
1143
      r->con_per_uneigh = 2;
1144
1145
    } else {
      l1_set = r->switch_id / p->l1_set_size;
Nikhil's avatar
Nikhil committed
1146
      r->con_per_uneigh = 1;
1147
1148
1149
1150
1151
1152
1153
1154
1155
    }
    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;
      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;
1156
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
1157
1158
1159
        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];
1160
1161
	    written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
1162
#if FATTREE_DEBUG
1163
    printf("L0->L1 I am switch %d, connect to upper switch %d L1 (%llu) rel_id:%llu at port %d yes collecting\n",
1164
          r->switch_id, l1_base, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1);
1165
#endif
Nikhil's avatar
Nikhil committed
1166
1167
1168
        /* write all inter switch links to DOT file (initialized before) */
        dot_write_sw2sw_link(r->switch_level, lp->gid, r->num_cons - 1,
              r->switch_level + 1, nextTerm, -1, dot_file);
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
      }
      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];
Nikhil's avatar
Nikhil committed
1179
      r->con_per_lneigh = 2;
1180
1181
1182
1183
1184
1185
    } 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;
Nikhil's avatar
Nikhil committed
1186
      r->con_per_lneigh = 1;
1187
1188
1189
1190
1191
1192
1193
    }
    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;
1194
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
1195
1196
1197
        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];
1198
1199
        written_2 += sprintf(r->output_buf2 + written_2, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
1200
1201
        r->num_lcons++;
#if FATTREE_DEBUG
1202
1203
        printf("L1->L0 I am switch %d, connect to switch %d L0 (%llu) rel_id:%llu at port %d not collecting\n",
            r->switch_id, l0_base, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1);
1204
#endif
Nikhil's avatar
Nikhil committed
1205
1206
1207
        /* write all inter switch links to DOT file (initialized before) */
        dot_write_sw2sw_link(r->switch_level, lp->gid, r->num_cons - 1,
              r->switch_level - 1, nextTerm, -1, dot_file);
1208
1209
1210
1211
      }
      l0_base++;
    }
    if(p->num_levels == 3) {
1212
      int l2_base = 0;
1213
      if(p->ft_type == 0) {
1214
1215
        /*for(int rep=0;rep<p->link_repetitions;rep++)*/{
            int rep = 0;
1216
1217
1218
1219
            int l2 = ((r->switch_id - p->num_switches[0]) % p->l1_set_size + rep*p->l1_set_size);
            /* not true anymore */
            r->start_uneigh = p->num_switches[0] + l2;
            r->con_per_uneigh = p->link_repetitions;
1220
            printf("link_repetitions:%d rep:%d l2:%d\n",p->link_repetitions,rep,l2);
1221
1222
1223
1224
1225
1226
            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;
1227
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
1228
1229
1230
1231
1232
1233
                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
1234
1235
                printf("L1->L2:t=0 I am switch %d, connect to upper switch %d L2 (%llu) rel_id:%llu at port %d yes collecting\n",
                    r->switch_id, l2+rep*p->l1_set_size, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1);
1236
#endif
1237
1238
1239
				/* write all inter switch links to DOT file (initialized before) */
            	dot_write_sw2sw_link(r->switch_level, lp->gid, r->num_cons - 1,
                  r->switch_level + 1, nextTerm, -1, dot_file);
1240
1241
1242
              }
            }
        }
1243
      } else {
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
        int l2 = ((r->switch_id - p->num_switches[0]) % p->l1_set_size);
        /* not true anymore */
        r->start_uneigh = p->num_switches[0] + l2;
        r->con_per_uneigh = 2;
        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;
1254
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
1255
1256
1257
            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];
1258
1259
            written += sprintf(r->output_buf + written, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
1260
#if FATTREE_DEBUG
1261
1262
            printf("L1->L2:t!=0 I am switch %d, connect to upper switch %d L2 (%llu) rel_id:%llu at port %d yes collecting\n",
                r->switch_id, l2, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1);
1263
#endif
Nikhil's avatar
Nikhil committed
1264
1265
1266
            /* write all inter switch links to DOT file (initialized before) */
            dot_write_sw2sw_link(r->switch_level, lp->gid, r->num_cons - 1,
                  r->switch_level + 1, nextTerm, -1, dot_file);
1267
1268
1269
1270
1271
          }
        }
      }
    }
  } else {
1272
    if(p->ft_type == 0) {
1273
      r->con_per_lneigh = p->link_repetitions;
1274
1275
1276
      /* not true anymore */
      r->start_lneigh = p->num_switches[0];
      r->end_lneigh = r->start_lneigh + p->num_switches[1];
1277
1278
      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) {
1279
1280
1281
1282
1283
        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;
1284
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
1285
1286
1287
          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];
1288
1289
	      written_2 += sprintf(r->output_buf2 + written_2, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
1290
1291
          r->num_lcons++;
#if FATTREE_DEBUG
1292
1293
          printf("L2->L1:t=0 I am switch %d, connect to  switch %d L1 (%llu) rel_id:%llu at port %d not collecting\n",
                  r->switch_id, l1, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1);
1294
#endif
1295
1296
1297
          /* write all inter switch links to DOT file (initialized before) */
          dot_write_sw2sw_link(r->switch_level, lp->gid, r->num_cons - 1,
                r->switch_level - 1, nextTerm, -1, dot_file);
1298
1299
1300
        }
      }
    } else {
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
      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;
1312
#if FATTREE_CONNECTIONS || FATTREE_DEBUG
1313
1314
1315
          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];
1316
1317
	      written_2 += sprintf(r->output_buf2 + written_2, "%u, %llu, ", r->switch_id+p->num_terminals,LLU(next_switch_lid)+p->num_terminals);
#endif
1318
1319
          r->num_lcons++;
#if FATTREE_DEBUG
1320
1321
          printf("L2->L1:t!=0 I am switch %d, connect to  switch %d L1 (%llu) rel_id:%llu at port %d not collecting\n",
              r->switch_id, l1, LLU(nextTerm), LLU(next_switch_lid), r->num_cons - 1);
1322
#endif
Nikhil's avatar
Nikhil committed
1323
1324
1325
          /* write all inter switch links to DOT file (initialized before) */
          dot_write_sw2sw_link(r->switch_level, lp->gid, r->num_cons - 1,
                r->switch_level - 1, nextTerm, -1, dot_file);
1326
1327
1328
1329
        }
      }
    }
  }
Nikhil's avatar
Nikhil committed
1330
1331
1332
1333
1334

  /* ensure the DOT file is out before reading it with external tools */
  if(dot_file)
    fflush(dot_file);

1335
#if FATTREE_CONNECTIONS
1336
1337
  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);
1338
#endif
1339
  return;
1340
}
1341
1342

/* empty for now.. */
1343
static void fattree_report_stats()
1344
{
Nikhil's avatar
Nikhil committed
1345
  if(dump_topo) return;
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
#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;
1363

1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
	MPI_Reduce( &packet_event_f, &t_packet_event_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
	MPI_Reduce( &packet_event_r, &t_packet_event_r, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
	MPI_Reduce( &t_generate_f, &tt_generate_f, 1, MPI_LONG_LONG, MPI_SUM, 0, MPI_COMM_CODES);
	<