codes-online-comm-wrkld.C 27.5 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13
/*
 * Copyright (C) 2014 University of Chicago
 * See COPYRIGHT notice in top-level directory.
 *
 */

#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <mpi.h>
#include <ross.h>
#include <assert.h>
#include <deque>
mubarak's avatar
mubarak committed
14
#include <iostream>
15
#include <inttypes.h>
mubarak's avatar
mubarak committed
16
#include <fstream>
17 18
#include <boost/property_tree/json_parser.hpp>
#include <boost/property_tree/ptree.hpp>
19 20 21 22 23 24 25
#include "codes/codes-workload.h"
#include "codes/quickhash.h"
#include "codes/codes-jobmap.h"
#include "codes_config.h"
#include "lammps.h"
#include "nekbone_swm_user_code.h"

26 27
#define ALLREDUCE_SHORT_MSG_SIZE 2048

28
//#define DBG_COMM 0
29 30 31 32 33 34 35

using namespace std;

static struct qhash_table *rank_tbl = NULL;
static int rank_tbl_pop = 0;
static int total_rank_cnt = 0;
ABT_thread global_prod_thread = NULL;
36
ABT_xstream self_es;
37 38 39 40 41 42 43 44 45 46 47 48 49
long cpu_freq = 1.0;
long num_allreduce = 0;
long num_isends = 0;
long num_irecvs = 0;
long num_barriers = 0;
long num_sends = 0;
long num_recvs = 0;
long num_sendrecv = 0;
long num_waitalls = 0;

std::map<int64_t, int> send_count;
std::map<int64_t, int> isend_count;
std::map<int64_t, int> allreduce_count;
50 51

struct shared_context {
52
    int my_rank;
53
    uint32_t wait_id;
54 55 56 57 58
    int num_ranks;
    char workload_name[MAX_NAME_LENGTH_WKLD];
    void * swm_obj;
    ABT_thread      producer;
    std::deque<struct codes_workload_op*> fifo;
59 60 61
};

struct rank_mpi_context {
62 63 64
    struct qhash_head hash_link;
    int app_id;
    struct shared_context sctx;
65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82
};

typedef struct rank_mpi_compare {
    int app_id;
    int rank;
} rank_mpi_compare;

/*
 * peer: the receiving peer id 
 * comm_id: the communicator id being used
 * tag: tag id 
 * reqvc: virtual channel being used by the message (to be ignored)
 * rspvc: virtual channel being used by the message (to be ignored)
 * buf: the address of sender's buffer in memory
 * bytes: number of bytes to be sent 
 * reqrt and rsprt: routing types (to be ignored) */

void SWM_Send(SWM_PEER peer,
83 84 85 86 87 88 89 90 91
        SWM_COMM_ID comm_id,
        SWM_TAG tag,
        SWM_VC reqvc,
        SWM_VC rspvc,
        SWM_BUF buf,
        SWM_BYTES bytes,
        SWM_BYTES pktrspbytes,
        SWM_ROUTING_TYPE reqrt,
        SWM_ROUTING_TYPE rsprt)
92 93
{
    /* add an event in the shared queue and then yield */
94
    //    printf("\n Sending to rank %d ", comm_id);
95 96 97 98 99 100 101 102
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_SEND;
    wrkld_per_rank.u.send.tag = tag;
    wrkld_per_rank.u.send.num_bytes = bytes;
    wrkld_per_rank.u.send.dest_rank = peer;

#ifdef DBG_COMM
103 104 105 106 107 108 109 110 111 112 113 114
    if(tag != 1235 && tag != 1234) 
    {
        auto it = send_count.find(bytes);
        if(it == send_count.end())
        {
            send_count.insert(std::make_pair(bytes, 1));
        }
        else
        {
            it->second = it->second + 1;
        }
    }
115 116 117 118 119 120 121 122 123 124 125 126 127
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    wrkld_per_rank.u.send.source_rank = sctx->my_rank;
    sctx->fifo.push_back(&wrkld_per_rank);

    ABT_thread_yield_to(global_prod_thread);
128
    num_sends++;
129 130
}

131 132 133 134 135 136 137
/*
 * @param comm_id: communicator ID (For now, MPI_COMM_WORLD)
 * reqvc and rspvc: virtual channel IDs for request and response (ignore for
 * our purpose)
 * buf: buffer location for the call (ignore for our purpose)
 * reqrt and rsprt: routing types, ignore and use routing from config file instead. 
 * */
138 139 140 141 142 143 144 145 146 147 148
void SWM_Barrier(
        SWM_COMM_ID comm_id,
        SWM_VC reqvc,
        SWM_VC rspvc,
        SWM_BUF buf, 
        SWM_UNKNOWN auto1,
        SWM_UNKNOWN2 auto2,
        SWM_ROUTING_TYPE reqrt, 
        SWM_ROUTING_TYPE rsprt)
{
    /* Add an event in the shared queue and then yield */
149
#if 0
150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_DELAY;
    /* TODO: Check how to convert cycle count into delay? */
    wrkld_per_rank.u.delay.nsecs = 0.1;

#ifdef DBG_COMM
    printf("\n Barrier delay %lf ", wrkld_per_rank.u.delay.nsecs);
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    sctx->fifo.push_back(&wrkld_per_rank);

    ABT_thread_yield_to(global_prod_thread);
170 171 172
#endif
#ifdef DBG_COMM
     printf("\n barrier ");
173 174 175 176 177
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err;
178
    int rank, size, src, dest, mask;
179 180 181 182 183 184 185 186 187 188 189 190

    err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);

    rank = sctx->my_rank;
    size = sctx->num_ranks;
    mask = 0x1;

    while(mask < size) {
191
        dest = (rank + mask) % size;
192 193
        src = (rank - mask + size) % size;

194 195
        SWM_Sendrecv(comm_id, dest, 1234, reqvc, rspvc, 0, 0, 0,
                src,  1234, 0,  reqrt, rsprt);
196 197
        mask <<= 1;
    }
198
    num_barriers++;
199 200 201
}

void SWM_Isend(SWM_PEER peer,
202 203 204 205 206 207 208 209 210 211
        SWM_COMM_ID comm_id,
        SWM_TAG tag,
        SWM_VC reqvc,
        SWM_VC rspvc,
        SWM_BUF buf,
        SWM_BYTES bytes,
        SWM_BYTES pktrspbytes,
        uint32_t * handle,
        SWM_ROUTING_TYPE reqrt,
        SWM_ROUTING_TYPE rsprt)
212 213
{
    /* add an event in the shared queue and then yield */
214
    //    printf("\n Sending to rank %d ", comm_id);
215 216 217 218 219 220 221 222
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_ISEND;
    wrkld_per_rank.u.send.tag = tag;
    wrkld_per_rank.u.send.num_bytes = bytes;
    wrkld_per_rank.u.send.dest_rank = peer;

#ifdef DBG_COMM
223 224 225 226 227 228 229 230 231 232 233 234
    if(tag != 1235 && tag != 1234) 
    {
        auto it = isend_count.find(bytes);
        if(it == isend_count.end())
        {
            isend_count.insert(std::make_pair(bytes, 1));
        }
        else
        {
            it->second = it->second + 1;
        }
    }
235 236 237 238 239 240 241 242 243 244 245 246
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    wrkld_per_rank.u.send.source_rank = sctx->my_rank;
    sctx->fifo.push_back(&wrkld_per_rank);

247 248 249 250
    *handle = sctx->wait_id;
    wrkld_per_rank.u.send.req_id = *handle;
    sctx->wait_id++;

251
    ABT_thread_yield_to(global_prod_thread);
252
    num_isends++;
253 254 255 256 257 258 259 260 261 262 263 264
}
void SWM_Recv(SWM_PEER peer,
        SWM_COMM_ID comm_id,
        SWM_TAG tag,
        SWM_BUF buf)
{
    /* Add an event in the shared queue and then yield */
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_RECV;
    wrkld_per_rank.u.recv.tag = tag;
    wrkld_per_rank.u.recv.source_rank = peer;
265
    wrkld_per_rank.u.recv.num_bytes = 0;
266 267

#ifdef DBG_COMM
268
    //printf("\n recv op tag: %d source: %d ", tag, peer);
269 270 271 272 273 274 275 276 277 278 279 280 281
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    wrkld_per_rank.u.recv.dest_rank = sctx->my_rank;
    sctx->fifo.push_back(&wrkld_per_rank);

    ABT_thread_yield_to(global_prod_thread);
282
    num_recvs++;
283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300
}

/* handle is for the request ID */
void SWM_Irecv(SWM_PEER peer,
        SWM_COMM_ID comm_id,
        SWM_TAG tag,
        SWM_BUF buf, 
        uint32_t* handle)
{
    /* Add an event in the shared queue and then yield */
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_IRECV;
    wrkld_per_rank.u.recv.tag = tag;
    wrkld_per_rank.u.recv.source_rank = peer;
    wrkld_per_rank.u.recv.num_bytes = 0;

#ifdef DBG_COMM
301
//    printf("\n irecv op tag: %d source: %d ", tag, peer);
302 303 304 305 306 307 308 309 310 311 312 313
#endif

    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    wrkld_per_rank.u.recv.dest_rank = sctx->my_rank;
    sctx->fifo.push_back(&wrkld_per_rank);
314 315 316 317
    
    *handle = sctx->wait_id;
    wrkld_per_rank.u.recv.req_id = *handle;
    sctx->wait_id++;
318 319

    ABT_thread_yield_to(global_prod_thread);
320
    num_irecvs++;
321 322 323 324
}

void SWM_Compute(long cycle_count)
{
325 326
    if(!cpu_freq)
        cpu_freq = 4.0e9;
327 328 329 330 331
    /* Add an event in the shared queue and then yield */
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_DELAY;
    /* TODO: Check how to convert cycle count into delay? */
332 333
    wrkld_per_rank.u.delay.nsecs = (cycle_count/cpu_freq);
    wrkld_per_rank.u.delay.seconds = (cycle_count / cpu_freq) / (1000.0 * 1000.0 * 1000.0);
334 335 336 337 338 339 340 341 342 343 344 345
#ifdef DBG_COMM
    printf("\n compute op delay: %ld ", cycle_count);
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    sctx->fifo.push_back(&wrkld_per_rank);
346
	
347
    ABT_thread_yield_to(global_prod_thread);
348

349 350 351 352 353 354 355 356 357 358 359 360
}

void SWM_Wait(uint32_t req_id)
{
    /* Add an event in the shared queue and then yield */
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_WAIT;
    /* TODO: Check how to convert cycle count into delay? */
    wrkld_per_rank.u.wait.req_id = req_id;

#ifdef DBG_COMM
361
//    printf("\n wait op req_id: %"PRIu32"\n", req_id);
362
//      printf("\n wait ");
363 364 365 366 367 368 369 370 371 372
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    sctx->fifo.push_back(&wrkld_per_rank);
373

374 375 376 377 378
    ABT_thread_yield_to(global_prod_thread);
}

void SWM_Waitall(int len, uint32_t * req_ids)
{
379
    num_waitalls++;
380 381 382 383 384 385 386 387 388 389 390 391
    /* Add an event in the shared queue and then yield */
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_WAITALL;
    /* TODO: Check how to convert cycle count into delay? */
    wrkld_per_rank.u.waits.count = len;
    wrkld_per_rank.u.waits.req_ids = (unsigned int*)calloc(len, sizeof(int));    

    for(int i = 0; i < len; i++)
        wrkld_per_rank.u.waits.req_ids[i] = req_ids[i];

#ifdef DBG_COMM
392
//    for(int i = 0; i < len; i++)
393
//        printf("\n wait op len %d req_id: %"PRIu32"\n", len, req_ids[i]);
394 395 396 397 398 399 400 401 402 403
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    sctx->fifo.push_back(&wrkld_per_rank);
404

405 406 407 408
    ABT_thread_yield_to(global_prod_thread);
}

void SWM_Sendrecv(
409 410 411 412 413 414 415 416 417 418 419 420 421
        SWM_COMM_ID comm_id,
        SWM_PEER sendpeer,
        SWM_TAG sendtag,
        SWM_VC sendreqvc,
        SWM_VC sendrspvc,
        SWM_BUF sendbuf,
        SWM_BYTES sendbytes,
        SWM_BYTES pktrspbytes,
        SWM_PEER recvpeer,
        SWM_TAG recvtag,
        SWM_BUF recvbuf,
        SWM_ROUTING_TYPE reqrt,
        SWM_ROUTING_TYPE rsprt)
422
{
423
    //    printf("\n Sending to %d receiving from %d ", sendpeer, recvpeer);
424 425 426 427 428 429 430 431 432 433 434 435 436
    struct codes_workload_op send_op;

    send_op.op_type = CODES_WK_SEND;
    send_op.u.send.tag = sendtag;
    send_op.u.send.num_bytes = sendbytes;
    send_op.u.send.dest_rank = sendpeer;

    /* Add an event in the shared queue and then yield */
    struct codes_workload_op recv_op;

    recv_op.op_type = CODES_WK_RECV;
    recv_op.u.recv.tag = recvtag;
    recv_op.u.recv.source_rank = recvpeer;
437
    recv_op.u.recv.num_bytes = 0;
438

439
#ifdef DBG_COMM
440 441 442 443 444 445 446 447 448 449 450 451
    if(sendtag != 1235 && sendtag != 1234) 
    {
        auto it = send_count.find(sendbytes);
        if(it == send_count.end())
        {
            send_count.insert(std::make_pair(sendbytes, 1));
        }
        else
        {
            it->second = it->second + 1;
        }
    }
452 453 454 455 456 457 458 459 460 461 462 463 464 465 466
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    recv_op.u.recv.dest_rank = sctx->my_rank;
    send_op.u.send.source_rank = sctx->my_rank;
    sctx->fifo.push_back(&send_op);
    sctx->fifo.push_back(&recv_op);

    ABT_thread_yield_to(global_prod_thread);
467
    num_sendrecv++;
468 469
}

470
/* @param count: number of bytes in Allreduce
471 472 473 474 475 476 477 478 479
 * @param respbytes: number of bytes to be sent in response (ignore for our
 * purpose)
 * $params comm_id: communicator ID (MPI_COMM_WORLD for our case)
 * @param sendreqvc: virtual channel of the sender request (ignore for our
 * purpose)
 * @param sendrspvc: virtual channel of the response request (ignore for our
 * purpose)
 * @param sendbuf and rcvbuf: buffers for send and receive calls (ignore for
 * our purpose) */
480
void SWM_Allreduce(
481
        SWM_BYTES count,
482 483 484 485 486 487 488
        SWM_BYTES respbytes,
        SWM_COMM_ID comm_id,
        SWM_VC sendreqvc,
        SWM_VC sendrspvc,
        SWM_BUF sendbuf,
        SWM_BUF rcvbuf)
{
489
#if 0
490
    /* TODO: For now, simulate a constant delay for ALlreduce*/
491
    //    printf("\n Allreduce bytes %d ", bytes);
492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512
    /* Add an event in the shared queue and then yield */
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_DELAY;
    /* TODO: Check how to convert cycle count into delay? */
    wrkld_per_rank.u.delay.nsecs = bytes + 0.1;

#ifdef DBG_COMM
    printf("\n Allreduce delay %lf ", wrkld_per_rank.u.delay.nsecs);
#endif
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    sctx->fifo.push_back(&wrkld_per_rank);

    ABT_thread_yield_to(global_prod_thread);
513 514
#endif

515 516 517 518 519 520 521 522 523 524 525
#ifdef DBG_COMM
        auto it = allreduce_count.find(count);
        if(it == allreduce_count.end())
        {
            allreduce_count.insert(std::make_pair(count, 1));
        }
        else
        {
            it->second = it->second + 1;
        }
#endif
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
    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);

    int comm_size, i, send_idx, recv_idx, last_idx, send_cnt, recv_cnt;
    int pof2, mask, rem, newrank, newdst, dst, *cnts, *disps;
    int rank = sctx->my_rank;
    comm_size = sctx->num_ranks;

    cnts = disps = NULL;

    pof2 = 1;
    while (pof2 <= comm_size) pof2 <<= 1;
    pof2 >>=1;

    rem = comm_size - pof2;

    /* In the non-power-of-two case, all even-numbered
       processes of rank < 2*rem send their data to
       (rank+1). These even-numbered processes no longer
       participate in the algorithm until the very end. The
       remaining processes form a nice power-of-two. */
    if (rank < 2*rem) {
        if (rank % 2 == 0) { /* even */
555
            SWM_Send(rank+1, comm_id, 1235, sendreqvc, sendrspvc, 0, count, 1, 0, 0);
556 557
            newrank = -1;
        } else { /* odd */
558
            SWM_Recv(rank-1, comm_id, 1235, 0);
559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580
            newrank = rank / 2;
        }
    } else {
        newrank = rank - rem;
    }

    /* If op is user-defined or count is less than pof2, use
       recursive doubling algorithm. Otherwise do a reduce-scatter
       followed by allgather. (If op is user-defined,
       derived datatypes are allowed and the user could pass basic
       datatypes on one process and derived on another as long as
       the type maps are the same. Breaking up derived
       datatypes to do the reduce-scatter is tricky, therefore
       using recursive doubling in that case.) */
    if (newrank != -1) {
        if ((count <= ALLREDUCE_SHORT_MSG_SIZE) || (count < pof2)) {

            mask = 0x1;
            while (mask < pof2) {
                newdst = newrank ^ mask;
                dst = (newdst < rem) ? newdst*2 + 1 : newdst + rem;

581 582
                SWM_Sendrecv(comm_id, dst, 1235, sendreqvc, sendrspvc, 0,
                        count, 1, dst, 1235, 0, 0, 0);
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

                mask <<= 1;
            }
        } else {
            /* do a reduce-scatter followed by allgather */
            /* for the reduce-scatter, calculate the count that
               each process receives and the displacement within
               the buffer */

            cnts = (int*)malloc(pof2*sizeof(int));
            disps = (int*)malloc(pof2*sizeof(int));

            for (i=0; i<(pof2-1); i++)
                cnts[i] = count/pof2;
            cnts[pof2-1] = count - (count/pof2)*(pof2-1);

            disps[0] = 0;
            for (i=1; i<pof2; i++)
                disps[i] = disps[i-1] + cnts[i-1];

            mask = 0x1;
            send_idx = recv_idx = 0;
            last_idx = pof2;
            while (mask < pof2) {
                newdst = newrank ^ mask;
                dst = (newdst < rem) ? newdst*2 + 1 : newdst + rem;
                send_cnt = recv_cnt = 0;
                if (newrank < newdst) {
                    send_idx = recv_idx + pof2/(mask*2);
                    for (i=send_idx; i<last_idx; i++)
                        send_cnt += cnts[i];
                    for (i=recv_idx; i<send_idx; i++)
                        recv_cnt += cnts[i];
                } else {
                    recv_idx = send_idx + pof2/(mask*2);
                    for (i=send_idx; i<recv_idx; i++)
                        send_cnt += cnts[i];
                    for (i=recv_idx; i<last_idx; i++)
                        recv_cnt += cnts[i];
                }

624 625
                SWM_Sendrecv(comm_id, dst, 1235, sendreqvc, sendrspvc, 0,
                        send_cnt, 1, dst, 1235, 0, 0, 0);
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

                send_idx = recv_idx;
                mask <<= 1;

                if(mask < pof2)
                    last_idx = recv_idx + pof2/mask;
            }

            /* now do the allgather */
            mask >>= 1;
            while (mask > 0) {
                newdst = newrank ^ mask;
                /* find real rank of dest */
                dst = (newdst < rem) ? newdst*2 + 1 : newdst + rem;

                send_cnt = recv_cnt = 0;
                if (newrank < newdst) {
                    if (mask != pof2/2)
                        last_idx = last_idx + pof2/(mask*2);

                    recv_idx = send_idx + pof2/(mask*2);
                    for (i=send_idx; i<recv_idx; i++)
                        send_cnt += cnts[i];
                    for (i=recv_idx; i<last_idx; i++)
                        recv_cnt += cnts[i];
                } else {
                    recv_idx = send_idx - pof2/(mask*2);
                    for (i=send_idx; i<last_idx; i++)
                        send_cnt += cnts[i];
                    for (i=recv_idx; i<send_idx; i++)
                        recv_cnt += cnts[i];
                }

659 660
                SWM_Sendrecv(comm_id, dst, 1235, sendreqvc, sendrspvc, 0,
                        send_cnt, 1, dst, 1235, 0, 0, 0);
661 662 663 664 665 666 667 668 669 670

                if (newrank > newdst) send_idx = recv_idx;

                mask >>= 1;
            }
        }
    }

    if(rank < 2*rem) {
        if(rank % 2) {/* odd */
671
            SWM_Send(rank-1, comm_id, 1235, sendreqvc, sendrspvc, 0, count, 1, 0, 0);
672
        } else {
673
            SWM_Recv(rank+1, comm_id, 1235, 0);
674 675 676 677 678
        }
    }

    if(cnts) free(cnts);
    if(disps) free(disps);
679 680

    num_allreduce++;
681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715
}

void SWM_Allreduce(
        SWM_BYTES bytes,
        SWM_BYTES respbytes,
        SWM_COMM_ID comm_id,
        SWM_VC sendreqvc,
        SWM_VC sendrspvc,
        SWM_BUF sendbuf,
        SWM_BUF rcvbuf,
        SWM_UNKNOWN auto1,
        SWM_UNKNOWN2 auto2,
        SWM_ROUTING_TYPE reqrt,
        SWM_ROUTING_TYPE rsprt)
{
    SWM_Allreduce(bytes, respbytes, comm_id, sendreqvc, sendrspvc, sendbuf, rcvbuf);
}

void SWM_Finalize()
{
    /* Add an event in the shared queue and then yield */
    struct codes_workload_op wrkld_per_rank;

    wrkld_per_rank.op_type = CODES_WK_END;

    /* Retreive the shared context state */
    ABT_thread prod;
    void * arg;
    int err = ABT_thread_self(&prod);
    assert(err == ABT_SUCCESS);
    err =  ABT_thread_get_arg(prod, &arg);
    assert(err == ABT_SUCCESS);
    struct shared_context * sctx = static_cast<shared_context*>(arg);
    sctx->fifo.push_back(&wrkld_per_rank);

716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733
#ifdef DBG_COMM 
    auto it = allreduce_count.begin();
    for(; it != allreduce_count.end(); it++)
    {
        cout << "\n Allreduce " << it->first << " " << it->second;
    }
    
    it = send_count.begin();
    for(; it != send_count.end(); it++)
    {
        cout << "\n Send " << it->first << " " << it->second;
    }
    
    it = isend_count.begin();
    for(; it != isend_count.end(); it++)
    {
        cout << "\n isend " << it->first << " " << it->second;
    }
734
#endif
735 736 737 738
//#ifdef DBG_COMM
//    printf("\n finalize workload for rank %d ", sctx->my_rank);
    printf("\n finalize workload for rank %d num_sends %d num_recvs %d num_isends %d num_irecvs %d num_allreduce %d num_barrier %d num_waitalls %d", sctx->my_rank, num_sends, num_recvs, num_isends, num_irecvs, num_allreduce, num_barriers, num_waitalls);
//#endif
739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755
    ABT_thread_yield_to(global_prod_thread);
}

static int hash_rank_compare(void *key, struct qhash_head *link)
{
    rank_mpi_compare *in = (rank_mpi_compare*)key;
    rank_mpi_context *tmp;

    tmp = qhash_entry(link, rank_mpi_context, hash_link);
    if (tmp->sctx.my_rank == in->rank && tmp->app_id == in->app_id)
        return 1;
    return 0;
}
static void workload_caller(void * arg)
{
    shared_context* sctx = static_cast<shared_context*>(arg);

756
    printf("\n workload name %s ", sctx->workload_name);
757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772
    if(strcmp(sctx->workload_name, "lammps") == 0)
    {
        LAMMPS_SWM * lammps_swm = static_cast<LAMMPS_SWM*>(sctx->swm_obj);
        lammps_swm->call();
    }
    else if(strcmp(sctx->workload_name, "nekbone") == 0) 
    {
        NEKBONESWMUserCode * nekbone_swm = static_cast<NEKBONESWMUserCode*>(sctx->swm_obj);
        nekbone_swm->call();
    }
}
static int comm_online_workload_load(const char * params, int app_id, int rank)
{
    /* LOAD parameters from JSON file*/
    online_comm_params * o_params = (online_comm_params*)params;
    int nprocs = o_params->nprocs;
773

774 775
    rank_mpi_context *my_ctx = new rank_mpi_context;
    //my_ctx = (rank_mpi_context*)caloc(1, sizeof(rank_mpi_context));  
776 777
    assert(my_ctx); 
    my_ctx->sctx.my_rank = rank; 
778
    my_ctx->sctx.num_ranks = nprocs;
779
    my_ctx->sctx.wait_id = 0;
780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802
    my_ctx->app_id = app_id;

    void** generic_ptrs;
    int array_len = 1;
    generic_ptrs = (void**)calloc(array_len,  sizeof(void*));
    generic_ptrs[0] = (void*)&rank;

    strcpy(my_ctx->sctx.workload_name, o_params->workload_name);
    boost::property_tree::ptree root;
    string path;
    path.append(SWM_DATAROOTDIR);

    if(strcmp(o_params->workload_name, "lammps") == 0)
    {
        path.append("/lammps_workload.json");
    }
    else if(strcmp(o_params->workload_name, "nekbone") == 0)
    {
        path.append("/workload.json"); 
    }
    else
        tw_error(TW_LOC, "\n Undefined workload type %s ", o_params->workload_name);

803 804
    printf("\n path %s ", path.c_str());
    try {
805
        std::ifstream jsonFile(path.c_str());
806 807
        boost::property_tree::json_parser::read_json(jsonFile, root);
        uint32_t process_cnt = root.get<uint32_t>("jobs.size", 1);
808
        cpu_freq = root.get<double>("jobs.cfg.cpu_freq"); 
809 810 811 812 813 814
    }
    catch(std::exception & e)
    {
        printf("%s \n", e.what());
        return -1;
    }
815 816 817 818 819 820 821
    if(strcmp(o_params->workload_name, "lammps") == 0)
    {
        LAMMPS_SWM * lammps_swm = new LAMMPS_SWM(root, generic_ptrs);
        my_ctx->sctx.swm_obj = (void*)lammps_swm;
    }
    else if(strcmp(o_params->workload_name, "nekbone") == 0)
    {
822 823
        NEKBONESWMUserCode * nekbone_swm = new NEKBONESWMUserCode(root, generic_ptrs);
        my_ctx->sctx.swm_obj = (void*)nekbone_swm;
824 825 826
    }

    if(global_prod_thread == NULL)
827 828
    {
        ABT_xstream_self(&self_es);
829
        ABT_thread_self(&global_prod_thread);
830
    }
831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852
    ABT_thread_create_on_xstream(self_es, 
            &workload_caller, (void*)&(my_ctx->sctx),
            ABT_THREAD_ATTR_NULL, &(my_ctx->sctx.producer));

    rank_mpi_compare cmp;
    cmp.app_id = app_id;
    cmp.rank = rank;

    if(!rank_tbl)
    {
        rank_tbl = qhash_init(hash_rank_compare, quickhash_64bit_hash, nprocs);
        if(!rank_tbl)
            return -1;
    }
    qhash_add(rank_tbl, &cmp, &(my_ctx->hash_link));
    rank_tbl_pop++;

    return 0;
}

static void comm_online_workload_get_next(int app_id, int rank, struct codes_workload_op * op)
{
853 854 855
    /* At this point, we will use the "call" function. The send/receive/wait
     * definitions will be replaced by our own function definitions that will do a
     * yield to argobots if an event is not available. */
856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876
    /* if shared queue is empty then yield */

    rank_mpi_context * temp_data;
    struct qhash_head * hash_link = NULL;
    rank_mpi_compare cmp;
    cmp.rank = rank;
    cmp.app_id = app_id;
    hash_link = qhash_search(rank_tbl, &cmp);
    if(!hash_link)
    {
        printf("\n not found for rank id %d , %d", rank, app_id);
        op->op_type = CODES_WK_END;
        return;
    }
    temp_data = qhash_entry(hash_link, rank_mpi_context, hash_link);
    assert(temp_data);
    while(temp_data->sctx.fifo.empty())
    {
        ABT_thread_yield_to(temp_data->sctx.producer); 
    }
    struct codes_workload_op * front_op = temp_data->sctx.fifo.front();
877
    assert(front_op);
878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908
    *op = *front_op;
    temp_data->sctx.fifo.pop_front();
    return;
}
static int comm_online_workload_get_rank_cnt(const char *params, int app_id)
{
    online_comm_params * o_params = (online_comm_params*)params;
    int nprocs = o_params->nprocs;
    return nprocs;
}

static int comm_online_workload_finalize(const char* params, int app_id, int rank)
{
    rank_mpi_context * temp_data;
    struct qhash_head * hash_link = NULL;
    rank_mpi_compare cmp;
    cmp.rank = rank;
    cmp.app_id = app_id;
    hash_link = qhash_search(rank_tbl, &cmp);
    if(!hash_link)
    {
        printf("\n not found for rank id %d , %d", rank, app_id);
        return -1;
    }
    temp_data = qhash_entry(hash_link, rank_mpi_context, hash_link);
    assert(temp_data);

    ABT_thread_join(temp_data->sctx.producer);    
    ABT_thread_free(&(temp_data->sctx.producer));
    return 0;
}
mubarak's avatar
mubarak committed
909
extern "C" {
910 911 912
/* workload method name and function pointers for the CODES workload API */
struct codes_workload_method online_comm_workload_method =
{
mubarak's avatar
mubarak committed
913 914 915 916 917 918 919 920 921 922 923 924 925 926
    //.method_name =
    (char*)"online_comm_workload",
    //.codes_workload_read_config = 
    NULL,
    //.codes_workload_load = 
    comm_online_workload_load,
    //.codes_workload_get_next = 
    comm_online_workload_get_next,
    // .codes_workload_get_next_rc2 = 
    NULL,
    // .codes_workload_get_rank_cnt
    comm_online_workload_get_rank_cnt,
    // .codes_workload_finalize = 
    comm_online_workload_finalize
927
};
mubarak's avatar
mubarak committed
928
} // closing brace for extern "C"
929