model-net.c 11.8 KB
Newer Older
1
/*
Philip Carns's avatar
Philip Carns committed
2
 * Copyright (C) 2013 University of Chicago.
3
 * See COPYRIGHT notice in top-level directory.
Philip Carns's avatar
Philip Carns committed
4
 *
5
 */
Philip Carns's avatar
Philip Carns committed
6

7 8 9
#include <string.h>
#include <assert.h>

10
#include "codes/model-net.h"
11
#include "codes/model-net-method.h"
12
#include "codes/model-net-lp.h"
13
#include "codes/model-net-sched.h"
14
#include "codes/codes.h"
15
#include <codes/codes_mapping.h>
16

17 18 19 20
#define STR_SIZE 16
#define PROC_TIME 10.0

extern struct model_net_method simplenet_method;
21
extern struct model_net_method simplewan_method;
22
extern struct model_net_method torus_method;
23
extern struct model_net_method dragonfly_method;
Philip Carns's avatar
Philip Carns committed
24
extern struct model_net_method loggp_method;
25

26 27 28 29 30 31 32 33 34 35 36 37
#define X(a,b,c,d) b,
char * model_net_lp_config_names[] = {
    NETWORK_DEF
};
#undef X

#define X(a,b,c,d) c,
char * model_net_method_names[] = {
    NETWORK_DEF
};
#undef X

38
/* Global array initialization, terminated with a NULL entry */
39 40 41 42 43
#define X(a,b,c,d) d,
struct model_net_method* method_array[] = { 
    NETWORK_DEF
};
#undef X
44

45
// counter and offset for the MN_START_SEQ / MN_END_SEQ macros
46 47 48
int in_sequence = 0;
tw_stime mn_msg_offset = 0.0;

49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121
// global listing of lp types found by model_net_register
// - needs to be held between the register and configure calls
static int do_config_nets[MAX_NETS];

void model_net_register(){
    // first set up which networks need to be registered, then pass off to base
    // LP to do its thing
    memset(do_config_nets, 0, MAX_NETS * sizeof(*do_config_nets));
    for (int grp = 0; grp < lpconf.lpgroups_count; grp++){
        config_lpgroup_t *lpgroup = &lpconf.lpgroups[grp];
        for (int lpt = 0; lpt < lpgroup->lptypes_count; lpt++){
            char *nm = lpgroup->lptypes[lpt].name;
            for (int n = 0; n < MAX_NETS; n++){
                if (!do_config_nets[n] && 
                        strcmp(model_net_lp_config_names[n], nm) == 0){
                    do_config_nets[n] = 1;
                    break;
                }
            }
        }
    }
    model_net_base_register(do_config_nets);
}

int* model_net_configure(int *id_count){
    // first call the base LP configure, which sets up the general parameters
    model_net_base_configure();

    // do network-specific configures
    *id_count = 0;
    for (int i = 0; i < MAX_NETS; i++) {
        if (do_config_nets[i]){
            method_array[i]->mn_configure();
            (*id_count)++;
        }
    }

    // allocate the output
    int *ids = malloc(*id_count * sizeof(int));
    // read the ordering provided by modelnet_order
    char **values;
    size_t length;
    int ret = configuration_get_multivalue(&config, "PARAMS", "modelnet_order",
            NULL, &values, &length);
    if (ret != 1){
        tw_error(TW_LOC, "unable to read PARAMS:modelnet_order variable\n");
    }
    if (length != (size_t) *id_count){
        tw_error(TW_LOC, "number of networks in PARAMS:modelnet_order "
                "do not match number in LPGROUPS\n");
    }
    // set the index
    for (int i = 0; i < *id_count; i++){
        ids[i] = -1;
        for (int n = 0; n < MAX_NETS; n++){
            if (strcmp(values[i], model_net_method_names[n]) == 0){
                if (!do_config_nets[n]){
                    tw_error(TW_LOC, "network in PARAMS:modelnet_order not "
                            "present in LPGROUPS: %s\n", values[i]);
                }
                ids[i] = n;
                break;
            }
        }
        if (ids[i] == -1){
            tw_error(TW_LOC, "unknown network in PARAMS:modelnet_order: %s\n",
                    values[i]);
        }
        free(values[i]);
    }
    free(values);

    return ids;
122 123
}

124 125 126
int model_net_get_id(char *name){
    int i;
    for(i=0; method_array[i] != NULL; i++) {
127
        if(strcmp(model_net_method_names[i], name) == 0) {
128 129 130 131 132 133
            return i;
        }
    }
    return -1;
}

134 135 136
void model_net_write_stats(tw_lpid lpid, struct mn_stats* stat)
{
    int ret;
137
    char id[19+CATEGORY_NAME_MAX+1];
138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215
    char data[1024];

    sprintf(id, "model-net-category-%s", stat->category);
    sprintf(data, "lp:%ld\tsend_count:%ld\tsend_bytes:%ld\tsend_time:%f\t" 
        "recv_count:%ld\trecv_bytes:%ld\trecv_time:%f\tmax_event_size:%ld\n",
        (long)lpid,
        stat->send_count,
        stat->send_bytes,
        stat->send_time,
        stat->recv_count,
        stat->recv_bytes,
        stat->recv_time,
        stat->max_event_size);

    ret = lp_io_write(lpid, id, strlen(data), data);
    assert(ret == 0);

    return;
}

void model_net_print_stats(tw_lpid lpid, mn_stats mn_stats_array[])
{

    int i;
    struct mn_stats all;

    memset(&all, 0, sizeof(all));
    sprintf(all.category, "all");

    for(i=0; i<CATEGORY_MAX; i++)
    {
        if(strlen(mn_stats_array[i].category) > 0)
        {
            all.send_count += mn_stats_array[i].send_count;
            all.send_bytes += mn_stats_array[i].send_bytes;
            all.send_time += mn_stats_array[i].send_time;
            all.recv_count += mn_stats_array[i].recv_count;
            all.recv_bytes += mn_stats_array[i].recv_bytes;
            all.recv_time += mn_stats_array[i].recv_time;
            if(mn_stats_array[i].max_event_size > all.max_event_size)
                all.max_event_size = mn_stats_array[i].max_event_size;

            model_net_write_stats(lpid, &mn_stats_array[i]);
        }
    }
    model_net_write_stats(lpid, &all);
}

struct mn_stats* model_net_find_stats(const char* category, mn_stats mn_stats_array[])
{
    int i;
    int new_flag = 0;
    int found_flag = 0;

    for(i=0; i<CATEGORY_MAX; i++)
    {
        if(strlen(mn_stats_array[i].category) == 0)
        {
            found_flag = 1;
            new_flag = 1;
            break;
        }
        if(strcmp(category, mn_stats_array[i].category) == 0)
        {
            found_flag = 1;
            new_flag = 0;
            break;
        }
    }
    assert(found_flag);

    if(new_flag)
    {
        strcpy(mn_stats_array[i].category, category);
    }
    return(&mn_stats_array[i]);
}

216
static void model_net_event_impl_base(
217 218 219 220 221 222 223 224 225 226 227
        int net_id,
        char* category, 
        tw_lpid final_dest_lp, 
        uint64_t message_size, 
        int is_pull,
        tw_stime offset,
        int remote_event_size,
        const void* remote_event,
        int self_event_size,
        const void* self_event,
        tw_lp *sender) {
228

229 230 231 232 233 234 235 236
    if (remote_event_size + self_event_size + sizeof(model_net_wrap_msg) 
            > g_tw_msg_sz){
        tw_error(TW_LOC, "Error: model_net trying to transmit an event of size "
                         "%d but ROSS is configured for events of size %zd\n",
                         remote_event_size+self_event_size+sizeof(model_net_wrap_msg),
                         g_tw_msg_sz);
        return;
    }
237

238
    tw_lpid mn_lp = model_net_find_local_device(net_id, sender);
239
    tw_stime poffset = codes_local_latency(sender);
240 241 242 243 244 245 246 247 248 249 250 251 252 253 254
    if (in_sequence){
        tw_stime tmp = mn_msg_offset;
        mn_msg_offset += poffset;
        poffset += tmp;
    }
    tw_event *e = codes_event_new(mn_lp, poffset+offset, sender);

    model_net_wrap_msg *m = tw_event_data(e);
    m->event_type = MN_BASE_NEW_MSG;
    m->magic = model_net_base_magic;

    // set the request struct 
    model_net_request *r = &m->msg.m_base.u.req;
    r->net_id = net_id;
    r->final_dest_lp = final_dest_lp;
255
    r->src_lp = sender->gid;
256 257 258
    r->msg_size = message_size;
    r->remote_event_size = remote_event_size;
    r->self_event_size = self_event_size;
259
    r->is_pull = is_pull;
260 261 262 263 264 265 266 267 268 269 270
    strncpy(r->category, category, CATEGORY_NAME_MAX-1);
    r->category[CATEGORY_NAME_MAX-1]='\0';
    
    void *e_msg = (m+1);
    if (remote_event_size > 0){
        memcpy(e_msg, remote_event, remote_event_size);
        e_msg = (char*)e_msg + remote_event_size; 
    }
    if (self_event_size > 0){
        memcpy(e_msg, self_event, self_event_size);
    }
271

272 273 274 275 276
    //print_base(m);
    tw_event_send(e);
}
static void model_net_event_impl_base_rc(tw_lp *sender){
    codes_local_latency_reverse(sender);
277
}
278 279 280 281 282 283 284 285 286 287 288 289 290

void model_net_event(
    int net_id,
    char* category, 
    tw_lpid final_dest_lp, 
    uint64_t message_size, 
    tw_stime offset,
    int remote_event_size,
    const void* remote_event,
    int self_event_size,
    const void* self_event,
    tw_lp *sender)
{
291 292 293
    model_net_event_impl_base(net_id, category, final_dest_lp, message_size,
            0, offset, remote_event_size, remote_event, self_event_size,
            self_event, sender);
294 295 296 297 298 299 300 301 302 303 304 305 306
}

void model_net_pull_event(
        int net_id,
        char *category,
        tw_lpid final_dest_lp,
        uint64_t message_size,
        tw_stime offset,
        int self_event_size,
        const void *self_event,
        tw_lp *sender){
    /* NOTE: for a pull, we are filling the *remote* event - it will be remote
     * from the destination's POV */
307 308
    model_net_event_impl_base(net_id, category, final_dest_lp, message_size,
            1, offset, self_event_size, self_event, 0, NULL, sender);
309 310 311 312 313 314
}

void model_net_event_rc(
        int net_id,
        tw_lp *sender,
        uint64_t message_size){
315
    model_net_event_impl_base_rc(sender);
316 317 318 319 320
}

void model_net_pull_event_rc(
        int net_id,
        tw_lp *sender) {
321
    model_net_event_impl_base_rc(sender);
322 323
}

324
/* returns the message size, can be either simplenet, dragonfly or torus message size*/
325
int model_net_get_msg_sz(int net_id)
326 327 328
{
   // TODO: Add checks on network name
   // TODO: Add dragonfly and torus network models
329 330
   return sizeof(model_net_wrap_msg);
#if 0
331
   if(net_id < 0 || net_id >= MAX_NETS)
332 333 334 335 336 337
     {
      printf("%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
      exit(-1);
     }

       return method_array[net_id]->mn_get_msg_sz();
338
#endif
339 340 341
}

/* returns the packet size in the modelnet struct */
342
uint64_t model_net_get_packet_size(int net_id)
343
{
344
  if(net_id < 0 || net_id >= MAX_NETS)
345 346 347 348 349 350 351
     {
       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
       exit(-1);
     }
  return method_array[net_id]->packet_size; // TODO: where to set the packet size?
}

352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373
/* This event does a collective operation call for model-net */
void model_net_event_collective(int net_id, char* category, int message_size, int remote_event_size, const void* remote_event, tw_lp* sender)
{
  if(net_id < 0 || net_id > MAX_NETS)
     {
       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
       exit(-1);
     }
  return method_array[net_id]->mn_collective_call(category, message_size, remote_event_size, remote_event, sender);
}

/* reverse event of the collective operation call */
void model_net_event_collective_rc(int net_id, int message_size, tw_lp* sender)
{
  if(net_id < 0 || net_id > MAX_NETS)
     {
       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
       exit(-1);
     }
  return method_array[net_id]->mn_collective_call_rc(message_size, sender);
}

374 375 376
/* returns lp type for modelnet */
const tw_lptype* model_net_get_lp_type(int net_id)
{
377
    if(net_id < 0 || net_id >= MAX_NETS)
378 379 380 381 382 383 384 385 386 387
     {
       fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
       exit(-1);
     }

   // TODO: ADd checks by network names
   // Add dragonfly and torus network models
   return method_array[net_id]->mn_get_lp_type();
}

388 389
void model_net_report_stats(int net_id)
{
390
  if(net_id < 0 || net_id >= MAX_NETS)
391 392 393 394 395 396 397
  {
    fprintf(stderr, "%s Error: Uninitializied modelnet network, call modelnet_init first\n", __FUNCTION__);
    exit(-1);
   }

     // TODO: ADd checks by network names
     //    // Add dragonfly and torus network models
398 399
   method_array[net_id]->mn_report_stats();
   return;
400
}
401 402 403 404 405 406

tw_lpid model_net_find_local_device(int net_id, tw_lp *sender)
{
    return(method_array[net_id]->model_net_method_find_local_device(sender));
}

407 408 409 410 411 412 413 414
/*
 * Local variables:
 *  c-indent-level: 4
 *  c-basic-offset: 4
 * End:
 *
 * vim: ft=c ts=8 sts=4 sw=4 expandtab
 */