conc-latencyall.c 51.3 KB
Newer Older
1
/**********************************************************************
2
 * This file was generated by coNCePTuaL on Mon Jan  7 23:10:00 2019
3
 * using the c_mpi backend (C + MPI).
4
 * Do not modify this file; modify /Users/xin/macworkspace/codes-dev/codes/scripts/conceptual_benchmarks/latency-all.ncptl instead.
5 6 7
 *
 * Entire source program
 * ---------------------
8 9
 *   # An all-pairs ping-pong latency test written in coNCePTuaL
 *   # By Scott Pakin <pakin@lanl.gov>
10 11 12 13
 *   
 *   Require language version "1.5".
 *   
 *   # Parse the command line.
14 15
 *   reps is "Number of repetitions of each message size" and comes from "--reps" or "-r" with default 1000.
 *   maxbytes is "Maximum number of bytes to transmit" and comes from "--maxbytes" or "-m" with default 1M.
16 17 18 19 20 21 22
 *   
 *   # Ensure that we have a peer with whom to communicate.
 *   Assert that "the latency test requires at least two tasks" with num_tasks>=2.
 *   
 *   # Perform the benchmark.
 *   For each msgsize in {0}, {1, 2, 4, ..., maxbytes} {
 *     for reps repetitions {
23 24 25 26 27
 *       tasks ev such that ev is even reset their counters then
 *       tasks ev such that ev is even send a msgsize byte message to task ev+1 then
 *       tasks od such that od is odd send a msgsize byte message to task od-1 then
 *       tasks ev such that ev is even log the msgsize as "Bytes" and
 *                                         the median of elapsed_usecs/2 as "1/2 RTT (usecs)"
28
 *     } then
29
 *     tasks ev such that ev is even compute aggregates
30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336
 *   }
 **********************************************************************/

#include "codes/codes-conc-addon.h"
/*****************
 * Include files *
 *****************/

/* Header files needed by all C-based backends */
#include <stdio.h>
#include <string.h>
#include <ncptl/ncptl.h>

/* Header files specific to the c_mpi backend */
#include <mpi.h>
#include <stdarg.h>

/**********
 * Macros *
 **********/

/* Define the maximum loop trip count that we're willing to unroll fully. */
#define CONC_MAX_UNROLL 5

/* Specify the minimum number of trial iterations in each FOR <time> loop. */
#define CONC_FOR_TIME_TRIALS 1

/* Define a macro that rounds a double to a ncptl_int. */
#define CONC_DBL2INT(D) ((ncptl_int)((D)+0.5))

/* Define a macro that increments a buffer pointer by a byte offset. */
#define CONC_GETBUFPTR(S) ((void *)((char *)thisev->s.S.buffer + thisev->s.S.bufferofs))

/* Implement ncptl_func_task_of in terms of ncptl_physical_to_virtual. */
#define ncptl_func_task_of(P) ((ncptl_int)(P) < 0LL || (ncptl_int)(P) >= var_num_tasks ? -1LL : ncptl_physical_to_virtual (procmap, (ncptl_int)(P)))
#define ncptl_dfunc_task_of(P) ((double) ncptl_func_task_of(P))

/* Implement ncptl_func_processor_of in terms of ncptl_virtual_to_physical. */
#define ncptl_func_processor_of(V) ((ncptl_int)(V) < 0LL || (ncptl_int)(V) >= var_num_tasks ? -1LL : ncptl_virtual_to_physical (procmap, (ncptl_int)(V)))
#define ncptl_dfunc_processor_of(V) ((double) ncptl_func_processor_of(V))

/* Define a macro that increments REDUCE's alternate buffer pointer by a byte offset. */
#define CONC_GETALTBUFPTR(S) ((void *)((char *)thisev->s.S.altbuffer + thisev->s.S.bufferofs))

/* Estimate the number of unique communicators that this program will need.
 * (The tradeoff is one of initialization time versus memory consumption.) */
#define ESTIMATED_COMMUNICATORS 128

/* Specify an operation to use for all reduction operations. */
#define REDUCE_OPERATION MPI_SUM
#define REDUCE_OPERATION_NAME "MPI_SUM"


/*********************
 * Type declarations *
 *********************/

/* Enumerate the various mechanisms used to implement MULTICAST statements. */
typedef enum {
CONC_MCAST_MPI_BCAST,       /* One to many */
CONC_MCAST_MPI_ALLTOALL,    /* Many to many, same data to all */ 
CONC_MCAST_MPI_ALLTOALLV,   /* General many to many */
CONC_MCAST_MPI_NUM_FUNCS    /* Number of the above */
} CONC_MCAST_MPI_FUNC;

/* Define the type of event to perform. */
typedef enum {
EV_SEND,     /* Synchronous send */
EV_ASEND,    /* Asynchronous send */
EV_RECV,     /* Synchronous receive */
EV_ARECV,    /* Asynchronous receive */
EV_WAIT,     /* Wait for all asynchronous sends/receives to complete */
EV_DELAY,    /* Spin or sleep */
EV_TOUCH,    /* Touch a region of memory */
EV_SYNC,     /* Barrier synchronization */
EV_RESET,    /* Reset counters */
EV_STORE,    /* Store all counters' current values */
EV_RESTORE,  /* Restore the previously pushed counter values */
EV_FLUSH,    /* Compute aggregate functions for log-file columns */
EV_MCAST,    /* Synchronous multicast */
EV_REDUCE,   /* Reduction with or without a subsequent multicast */
EV_BTIME,    /* Beginning of a timed loop */
EV_ETIME,    /* Ending of a timed loop */
EV_REPEAT,   /* Repeatedly process the next N events */
EV_SUPPRESS, /* Suppress writing to the log and standard output */
EV_NEWSTMT,  /* Beginning of a new top-level statement */
EV_CODE,     /* None of the above */
NUM_EVS      /* Number of event types in CONC_EVENT_TYPE */
} CONC_EVENT_TYPE;

/* Describe a synchronous or asynchronous send event. */
typedef struct {
ncptl_int dest;         /* Destination task */
ncptl_int size;         /* Number of bytes to send */
ncptl_int alignment;    /* Message alignment (in bytes) */
ncptl_int pendingsends; /* # of outstanding sends */
ncptl_int pendingrecvs; /* # of outstanding receives */
ncptl_int buffernum;    /* Buffer # to send from */
ncptl_int bufferofs;    /* Byte offset into the message buffer */
ncptl_int tag;          /* Tag to use for selective receives */
int misaligned;         /* 1=misaligned from a page; 0=align as specified */
int touching;           /* 1=touch every word before sending */
int verification;       /* 1=fill message buffer with known contents */
void *buffer;           /* Pointer to message memory */
MPI_Request * handle;   /* MPI handle representing an asynchronous send */
} CONC_SEND_EVENT;

/* Describe a synchronous or asynchronous receive event. */
typedef struct {
ncptl_int source;       /* Source task */
ncptl_int size;         /* Number of bytes to receive */
ncptl_int alignment;    /* Message alignment (in bytes) */
ncptl_int pendingsends; /* # of outstanding sends */
ncptl_int pendingrecvs; /* # of outstanding receives */
ncptl_int buffernum;    /* Buffer # to receive into */
ncptl_int bufferofs;    /* Byte offset into the message buffer */
ncptl_int tag;          /* Tag to use for selective receives */
int misaligned;         /* 1=misaligned from a page; 0=align as specified */
int touching;           /* 1=touch every word after reception */
int verification;       /* 1=verify that all bits are correct */
void *buffer;           /* Pointer to message memory */
MPI_Request * handle;   /* MPI handle representing an asynchronous receive */
} CONC_RECV_EVENT;

/* Describe a wait-for-asynchronous-completions event. */
typedef struct {
ncptl_int numsends;     /* # of sends we expect to complete. */
ncptl_int numrecvs;     /* # of receives we expect to complete. */
ncptl_int numrecvbytes; /* # of bytes we expect to receive-complete */
ncptl_int *touchedlist;    /* List of receives that need to be touched */
ncptl_int numtouches;         /* # of elements in the above */
} CONC_WAIT_EVENT;

/* Describe a spin or sleep delay. */
typedef struct {
uint64_t microseconds;  /* Length of delay in microseconds */
int spin0sleep1;        /* 0=spin; 1=sleep */
} CONC_DELAY_EVENT;

/* Describe a barrier synchronization event. */
typedef struct {
MPI_Comm communicator;   /* Set of tasks to synchronize */
} CONC_SYNC_EVENT;

/* Describe a walk over a memory-region. */
typedef struct {
ncptl_int regionbytes;  /* Size in bytes of the region to touch */
ncptl_int bytestride;   /* Stride in bytes to touch */
ncptl_int numaccesses;  /* Number of words to touch */
ncptl_int wordsize;     /* Size in bytes of each touch */
ncptl_int firstbyte;    /* Byte offset of the first byte to touch */
} CONC_TOUCH_EVENT;

/* Describe a synchronous multicast event. */
typedef struct {
ncptl_int source;       /* Source task, -1 in the many-to-many case */
ncptl_int size;         /* Number of bytes to send */
ncptl_int alignment;    /* Message alignment (in bytes) */
ncptl_int pendingsends; /* # of outstanding sends */
ncptl_int pendingrecvs; /* # of outstanding receives */
ncptl_int buffernum;    /* Buffer # to send/receive from */
ncptl_int bufferofs;    /* Byte offset into the message buffer */
ncptl_int tag;          /* Tag to use for selective receives */
int misaligned;         /* 1=misaligned from a page; 0=align as specified */
int touching;           /* 1=touch every word before sending */
int verification;       /* 1=verify that all bits are correct */
void *buffer;           /* Pointer to message memory */
ncptl_int size2;   /* Number of bytes to receive in the many-to-many case */
ncptl_int bufferofs2;   /* Byte offset into the message buffer in the many-to-many case */
void * buffer2;   /* Pointer to receive-message memory in the many-to-many case */
MPI_Comm communicator;   /* Set of tasks to multicast to/from */
int root;   /* source's rank within communicator */
int * sndvol;   /* Volume of data to send to each rank in the communicator */
int * snddisp;   /* Offset from buffer of each message to send */
int * rcvvol;   /* Volume of data to receive from each rank in the communicator */
int * rcvdisp;   /* Offset from buffer2 of each message to receive */
CONC_MCAST_MPI_FUNC mpi_func;   /* MPI function to use to perform the multicast */
} CONC_MCAST_EVENT;

/* Describe a reduction event. */
typedef struct {
ncptl_int numitems;     /* # of items to reduce */
ncptl_int itemsize;     /* # of bytes per item */
ncptl_int alignment;    /* Message alignment (in bytes) */
ncptl_int pendingsends; /* # of outstanding sends */
ncptl_int pendingrecvs; /* # of outstanding receives */
ncptl_int buffernum;    /* Buffer # to send/receive from */
ncptl_int bufferofs;    /* Byte offset into the message buffer */
ncptl_int tag;          /* Tag to use for selective receives */
int misaligned;         /* 1=misaligned from a page; 0=align as specified */
int touching;           /* 1=touch every word before sending/after receiving */
int sending;            /* 1=we're a sender */
int receiving;          /* 1=we're a receiver */
void *buffer;           /* Pointer to message memory */
void * altbuffer;   /* Pointer to additional message memory */
MPI_Comm sendcomm;   /* Set of tasks to reduce from */
MPI_Comm recvcomm;   /* Set of tasks to reduce to */
MPI_Datatype datatype;   /* MPI datatype to reduce */
int reducetype;   /* 0=reduce; 1=allreduce; 2=reduce+bcast */
ncptl_int reduceroot;   /* Root task of the reduction if reducetype is 0 or 2 */
ncptl_int bcastroot;   /* Root task of the multicast if reducetype is 2 */
} CONC_REDUCE_EVENT;

/* Describe an event representing the beginning of a timed loop. */
typedef struct {
uint64_t usecs;         /* Requested loop duration */
uint64_t warmup_usecs;  /* Requested duration of warmup loops */
uint64_t starttime;     /* Time at which the loop state last changed */
uint64_t itersleft;     /* # of iterations remaining */
uint64_t previters;     /* # of iterations we performed last time */
int prev_quiet;         /* Previous value of suppress_output */
int timing_trial;       /* 1=performing a timing trial; 0=running for real */
volatile int finished;  /* 1=time has expired; 0=still ticking */
} CONC_BTIME_EVENT;

/* Describe an event representing the end of a timed loop. */
typedef struct {
ncptl_int begin_event;  /* Index into eventlist[] of the corresponding BTIME event */
} CONC_ETIME_EVENT;

/* Describe an event representing repetitions of subsequent events. */
typedef struct {
ncptl_int end_event;    /* Index into eventlist[] of the last event to repeat */
ncptl_int numreps;      /* # of repetitions to perform */
} CONC_REPEAT_EVENT;

/* Describe an event representing output suppression (either on or off). */
typedef struct conc_suppress_event {
int quiet;              /* 0=allow output; 1=suppress it */
int prev_quiet;         /* Previous value of suppress_output */
ncptl_int matching_event;  /* Event ID of the "suppression on" event */
uint64_t stop_elapsed_usecs;   /* Time at which we suppressed output */
} CONC_SUPPRESS_EVENT;

/* Describe an event representing arbitrary code to execute at run time. */
typedef struct {
ncptl_int number;       /* Unique number corresponding to a specific piece of code */NCPTL_VIRT_PHYS_MAP *procmap;  /* Current mapping between tasks and processors */
ncptl_int var_msgsize;   /* Copy of var_msgsize to use within a piece of code */
} CONC_CODE_EVENT;

/* Describe an arbitrary coNCePTuaL event. */
typedef struct {
CONC_EVENT_TYPE type;          /* Type of event */
union {
CONC_SEND_EVENT send;          /* Send state */
CONC_RECV_EVENT recv;          /* Receive state */
CONC_WAIT_EVENT wait;          /* Wait-for-completions state */
CONC_DELAY_EVENT delay;        /* State for spins and sleeps */
CONC_TOUCH_EVENT touch;        /* State for memory touching */
CONC_SYNC_EVENT sync;          /* Synchronization state */
CONC_MCAST_EVENT mcast;        /* Multicast state */
CONC_REDUCE_EVENT reduce;      /* Reduction state */
CONC_BTIME_EVENT btime;        /* Timed-loop state */
CONC_ETIME_EVENT etime;        /* Additional timed-loop state */
CONC_REPEAT_EVENT rep;         /* Repeated-events state */
CONC_SUPPRESS_EVENT suppress;  /* State for suppressing output */
CONC_CODE_EVENT code;          /* State for arbitrary code */
} s;
} CONC_EVENT;

/* Fully specify an arbitrary for() loop (used by FOR EACH). */
typedef struct {
NCPTL_QUEUE *list_comp;  /* NULL=ordinary list; other=list comprehension values */
int integral;        /* 1=integral values; 0=floating-point values */
enum {               /* Comparison of loop variable to end variable */
CONC_LEQ,                /* Increasing progression */
CONC_GEQ                 /* Decreasing progression */
} comparator;
enum {               /* How to increment the loop variable */
CONC_ADD,                /* Arithmetically */
CONC_MULT,               /* Geometrically increasing */
CONC_DIV                 /* Geometrically decreasing */
} increment;
union {
struct {
ncptl_int loopvar;   /* Loop variable */
ncptl_int prev_loopvar; /* Previous value of loop variable */
ncptl_int startval;  /* Initial value of loop variable */
ncptl_int endval;    /* Value not to exceed */
ncptl_int incval;    /* Loop-variable increment */
} i;
struct {
double loopvar;      /* Loop variable */
double prev_loopvar; /* Previous value of loop variable */
double startval;     /* Initial value of loop variable */
double endval;       /* Value not to exceed */
double incval;       /* Loop-variable increment */
} d;
} u;
} LOOPBOUNDS;

/********************
 * Global variables *
 ********************/

/* Variables exported to coNCePTuaL programs */
static ncptl_int var_bytes_received = 0;   /* Total number of bytes received */
static ncptl_int var_msgs_received = 0;   /* Total number of messages received */
static ncptl_int var_bit_errors = 0;   /* Total number of bit errors observed */
static ncptl_int var_total_msgs = 0;   /* Sum of messages sent and messages received */
static ncptl_int var_msgs_sent = 0;   /* Total number of messages sent */
static ncptl_int var_bytes_sent = 0;   /* Total number of bytes sent */
static ncptl_int var_num_tasks = 1;   /* Number of tasks running the program */
static ncptl_int var_elapsed_usecs = 0;   /* Elapsed time in microseconds */
static ncptl_int var_total_bytes = 0;   /* Sum of bytes sent and bytes received */

/* Dummy variable to help mark other variables as used */
337
static union {
338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377
ncptl_int ni;
int i;
void *vp;
} conc_dummy_var;

/* Variables used internally by boilerplate code */
static uint64_t starttime;   /* Time the clock was last reset (microseconds) */
static ncptl_int pendingrecvs = 0;   /* Current # of outstanding receives */
static ncptl_int pendingrecvbytes = 0; /* Current # of bytes in outstanding receives */
static NCPTL_QUEUE *touchedqueue;      /* Queue of asynchronous receives to touch */
static ncptl_int pendingsends = 0;   /* Current # of outstanding sends */
static NCPTL_QUEUE *eventqueue;   /* List of coNCePTuaL events to perform */
static int within_time_loop = 0;   /* 1=we're within a FOR <time> loop */
static int suppress_output = 0;    /* 1=suppress output to stdout and the log file */
static void *touch_region = NULL;   /* Memory region to touch */
static ncptl_int touch_region_size = 0;   /* # of bytes in the above */
static int virtrank;    /* This task's virtual rank in the computation */
static int physrank;    /* This task's physical rank in the computation */
static NCPTL_VIRT_PHYS_MAP *procmap;  /* Virtual to physical rank mapping */
static NCPTL_LOG_FILE_STATE *logstate;   /* Opaque object representing all log-file state */
static char *logfile_uuid;   /* Execution UUID to write to every log file */
static char *logfiletmpl;   /* Template for the log file's name */
//static char *logfiletmpl_default;   /* Default value of the above */

/* Global variables specific to the c_mpi backend */
static ncptl_int mpi_is_running = 0;   /* 1=MPI has been initialized */
static NCPTL_QUEUE * recvreqQ;   /* List of MPI receive requests */
static MPI_Request * recvrequests;   /* List version of recvreqQ */
static NCPTL_QUEUE * recvstatQ;   /* List of MPI receive statuses */
static MPI_Status * recvstatuses;   /* List version of recvstatQ */
static NCPTL_QUEUE * sendreqQ;   /* List of MPI send requests */
static MPI_Request * sendrequests;   /* List version of sendreqQ */
static NCPTL_QUEUE * sendstatQ;   /* List of MPI send statuses */
static MPI_Status * sendstatuses;   /* List version of sendstatQ */
static NCPTL_SET * communicators;   /* Map from an array of processor flags to an MPI communicator */
static MPI_Errhandler mpi_error_handler;   /* Handle to handle_MPI_error() */
static ncptl_int mpi_tag_ub;   /* Upper bound on an MPI tag value */
static ncptl_int conc_mcast_tallies[CONC_MCAST_MPI_NUM_FUNCS] = {0};   /* Tallies of (static) multicast implementation functions */

/* Program-specific variables */
378 379
static ncptl_int var_reps;   /* Number of repetitions of each message size (command-line argument) */
static ncptl_int var_maxbytes;   /* Maximum number of bytes to transmit (command-line argument) */
380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554

/*************************
 * Function declarations *
 *************************/

/* Make MPI errors invoke ncptl_fatal(). */
static void handle_MPI_error (MPI_Comm *comm, int *errcode, ...)
{
va_list args;
char errstring[MPI_MAX_ERROR_STRING];
int errstrlen;

va_start (args, errcode);
if (MPI_Error_string (*errcode, errstring, &errstrlen) == MPI_SUCCESS)
ncptl_fatal ("MPI run-time error: %s", errstring);
else
ncptl_fatal ("MPI aborted with unrecognized error code %d", *errcode);
conc_dummy_var.vp = (void *) comm;   /* Prevent the compiler from complaining that comm is unused. */
va_end (args);
}

/* Perform the equivalent of CODES_MPI_Comm_rank() for an arbitrary process. */
static int rank_in_MPI_communicator (MPI_Comm subcomm, int global_rank)
{
  MPI_Group world_group;   /* Group associated with MPI_COMM_WORLD */
  MPI_Group subgroup;      /* Group associate with subcomm */
  int subrank;             /* global_rank's rank within subcomm */

  MPI_Comm_group (MPI_COMM_WORLD, &world_group);
  MPI_Comm_group (subcomm, &subgroup);
  MPI_Group_translate_ranks (world_group, 1, &global_rank, subgroup, &subrank);
  return subrank;
}

/* Map an arbitrary tag to within MPI's valid range of [0, mpi_tag_ub]. */
static ncptl_int map_tag_into_MPI_range (ncptl_int tag)
{
if (tag == NCPTL_INT_MIN)
 /* Avoid taking the absolute value of NCPTL_INT_MIN. */
tag = 555666773LL;   /* Arbitrary value */
tag = ncptl_func_abs (tag);   /* Only nonnegatives values are allowed. */
if (mpi_tag_ub < NCPTL_INT_MAX)
tag %= mpi_tag_ub + 1;
return tag;
}

/* Given an array of task in/out booleans return an MPI
 * communicator that represents the "in" tasks. */
static MPI_Comm define_MPI_communicator (char *procflags)
{
MPI_Comm *existing_comm;    /* Previously defined MPI communicator */
MPI_Comm new_comm;          /* Newly defined MPI communicator */

existing_comm = (MPI_Comm *) ncptl_set_find (communicators, (void *)procflags);
if (existing_comm)
return *existing_comm;
(void) MPI_Comm_split (MPI_COMM_WORLD, (int)procflags[physrank], physrank, &new_comm);
//(void) MPI_Errhandler_set (new_comm, mpi_error_handler);
ncptl_set_insert (communicators, (void *)procflags, (void *)&new_comm);
return define_MPI_communicator (procflags);
}

/* Return 1 if a sequence loop will take at least one trip. */
static int conc_seq_nonempty (LOOPBOUNDS *seq)
{
ncptl_int startval;   /* Integer version of seq's startval element */
ncptl_int endval;   /* Integer version of seq's endval element */

if (seq->integral) {
startval = seq->u.i.startval;
endval = seq->u.i.endval;
}
else {
startval = CONC_DBL2INT (seq->u.d.startval);
endval = CONC_DBL2INT (seq->u.d.endval);
}
switch (seq->comparator) {
case CONC_LEQ:
return startval <= endval;

case CONC_GEQ:
return startval >= endval;

default:
ncptl_fatal ("Internal error -- unknown comparator");
}
return -1;     /* Appease idiotic compilers. */
}

/* Initialize a sequence loop. */
static void conc_seq_init (LOOPBOUNDS *seq)
{
if (seq->integral) {
seq->u.i.loopvar = seq->u.i.startval;
seq->u.i.prev_loopvar = seq->u.i.loopvar - 1;
}
else {
seq->u.d.loopvar = seq->u.d.startval;
seq->u.d.prev_loopvar = seq->u.d.loopvar - 1.0;
}
}

/* Return 1 if a sequence loop should continue, 0 when finished. */
static int conc_seq_continue (LOOPBOUNDS *seq)
{
LOOPBOUNDS seq_int;   /* Integer equivalent of *seq */
if (seq->integral)
seq_int = *seq;
else {
seq_int.u.i.loopvar = CONC_DBL2INT (seq->u.d.loopvar);
seq_int.u.i.prev_loopvar = CONC_DBL2INT (seq->u.d.prev_loopvar);
seq_int.u.i.endval = CONC_DBL2INT (seq->u.d.endval);
}

if (seq_int.u.i.loopvar == seq_int.u.i.prev_loopvar)
return 0;
switch (seq->comparator) {
case CONC_LEQ:
return seq_int.u.i.loopvar <= seq_int.u.i.endval;

case CONC_GEQ:
return seq_int.u.i.loopvar >= seq_int.u.i.endval;

default:
ncptl_fatal ("Internal error -- unknown comparator");
}
return -1;     /* Appease idiotic compilers. */
}

/* Proceed to the next iteration of a sequence loop. */
static void conc_seq_next (LOOPBOUNDS *seq)
{
if (seq->integral) {
seq->u.i.prev_loopvar = seq->u.i.loopvar;
switch (seq->increment) {
case CONC_ADD:
seq->u.i.loopvar += seq->u.i.incval;
break;

case CONC_MULT:
seq->u.i.loopvar *= seq->u.i.incval;
break;

case CONC_DIV:
seq->u.i.loopvar /= seq->u.i.incval;
break;

default:
ncptl_fatal ("Internal error -- unknown incrementer");
}
}
else {
seq->u.d.prev_loopvar = seq->u.d.loopvar;
switch (seq->increment) {
case CONC_ADD:
seq->u.d.loopvar += seq->u.d.incval;
break;

case CONC_MULT:
seq->u.d.loopvar *= seq->u.d.incval;
break;

case CONC_DIV:
seq->u.d.loopvar /= seq->u.d.incval;
break;

default:
ncptl_fatal ("Internal error -- unknown incrementer");
}
}
}

/* Inhibit the compiler from complaining that
 * certain variables are defined but not used.
 * This function should never be called. */
555
static void conc_mark_variables_used (void)
556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609
{
conc_dummy_var.ni = var_bytes_received;
conc_dummy_var.ni = var_msgs_received;
conc_dummy_var.ni = var_bit_errors;
conc_dummy_var.ni = var_total_msgs;
conc_dummy_var.ni = var_msgs_sent;
conc_dummy_var.ni = var_bytes_sent;
conc_dummy_var.ni = var_num_tasks;
conc_dummy_var.ni = var_elapsed_usecs;
conc_dummy_var.ni = var_total_bytes;
conc_dummy_var.ni = pendingrecvbytes;
conc_dummy_var.ni = touch_region_size;
conc_dummy_var.vp = touch_region;
conc_dummy_var.i  = within_time_loop;
conc_dummy_var.i  = suppress_output;
rank_in_MPI_communicator (MPI_COMM_WORLD, 0);
}

/* Allocate a new event of a given type and return a pointer to it. */
static CONC_EVENT *conc_allocate_event (CONC_EVENT_TYPE type)
{
CONC_EVENT *newevent = (CONC_EVENT *) ncptl_queue_allocate (eventqueue);

newevent->type = type;
return newevent;
}

/* Declare an exit handler that gets called automatically when the
 * program terminates, whether successfully or not. */
static void conc_exit_handler (void)
{
if (mpi_is_running)
MPI_Abort (MPI_COMM_WORLD, 1);
}


/* Initialize coNCePTuaL, the messaging layer, and this program itself. */
static void conc_initialize (int argc, char *argv[])
{
 /* Variables needed by all C-based backends */
CONC_EVENT * eventlist;   /* List of events to execute */
ncptl_int numevents;   /* Number of entries in eventlist[] */
int help_only = 0;   /* 1=User specified --help; save time by skipping ncptl_init() */
char * argv0 = strrchr(argv[0], '/') ? strrchr(argv[0], '/')+1 : argv[0];   /* Base name of the executable program */
int i;   /* Generic loop variable */

 /* Declare all of our command-line arguments. */
NCPTL_CMDLINE arguments[] = {
{ NCPTL_TYPE_STRING, NULL, "logfile", 'L', "Log-file template", {0}},
{ NCPTL_TYPE_INT, NULL, "reps", 'r', "Number of repetitions of each message size", {0}},
{ NCPTL_TYPE_INT, NULL, "maxbytes", 'm', "Maximum number of bytes to transmit", {0}}
};

 /* Incorporate the complete coNCePTuaL source code as an array
610
//  * for use by ncptl_log_write_prologue(). */
611
char *sourcecode[] = {
612 613
"# An all-pairs ping-pong latency test written in coNCePTuaL",
"# By Scott Pakin <pakin@lanl.gov>",
614 615 616 617
"",
"Require language version \"1.5\".",
"",
"# Parse the command line.",
618 619
"reps is \"Number of repetitions of each message size\" and comes from \"--reps\" or \"-r\" with default 1000.",
"maxbytes is \"Maximum number of bytes to transmit\" and comes from \"--maxbytes\" or \"-m\" with default 1M.",
620 621 622 623 624 625 626
"",
"# Ensure that we have a peer with whom to communicate.",
"Assert that \"the latency test requires at least two tasks\" with num_tasks>=2.",
"",
"# Perform the benchmark.",
"For each msgsize in {0}, {1, 2, 4, ..., maxbytes} {",
"  for reps repetitions {",
627 628 629 630 631
"    tasks ev such that ev is even reset their counters then",
"    tasks ev such that ev is even send a msgsize byte message to task ev+1 then",
"    tasks od such that od is odd send a msgsize byte message to task od-1 then",
"    tasks ev such that ev is even log the msgsize as \"Bytes\" and",
"                                      the median of elapsed_usecs/2 as \"1/2 RTT (usecs)\"",
632
"  } then",
633
"  tasks ev such that ev is even compute aggregates",
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
"}",
NULL
};

 /* Variables specific to the c_mpi backend */
int num_tasks;   /* int version of var_num_tasks needed by CODES_MPI_Comm_size() */
char * procflags;   /* Array of 1s representing an all-task MPI communicator */
MPI_Comm comm_world = MPI_COMM_WORLD;   /* Copy of MPI_COMM_WORLD that we can take the address of */
void * attr_val;   /* Pointed to the value of MPI_TAG_UB */
int attr_flag = 0;   /* true=MPI_TAG_UB was extracted; false=not extracted */
char log_key_str[128];   /* String representing the range of valid MPI tags */

 /* As a special case, if the command line contains --help, then skip
  * the coNCePTuaL initialization step. */
for (i=1; i<argc; i++)
if (!strcmp(argv[i], "--"))
break;
else
if (!strcmp(argv[i], "--help") || !strcmp(argv[i], "-?")) {
argv[1] = "-?";   /* Guaranteed to work, even with getopt() */
help_only = 1;
break;
}

 /* Perform various initializations specific to the c_mpi backend. */
 /* Initialize MPI. */
//(void) MPI_Init(&argc, &argv);
mpi_is_running = 1;

 /* Initialize the coNCePTuaL run-time library. */
if (!help_only)
665 666
//ncptl_init (NCPTL_RUN_TIME_VERSION, argv[0]);
//(void) atexit (conc_exit_handler);
667 668 669 670 671 672 673

 /* Initialize the communication routines needed by the c_mpi backend. */
//(void) MPI_Errhandler_create ((MPI_Handler_function *)handle_MPI_error, &mpi_error_handler);
//(void) MPI_Errhandler_set (MPI_COMM_WORLD, mpi_error_handler);
(void) CODES_MPI_Comm_rank(MPI_COMM_WORLD, &physrank);
(void) CODES_MPI_Comm_size(MPI_COMM_WORLD, &num_tasks);
var_num_tasks = (ncptl_int) num_tasks;
674
//(void) MPI_Comm_get_attr(MPI_COMM_WORLD, MPI_TAG_UB, &attr_val, &attr_flag);
675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709
mpi_tag_ub = (ncptl_int) (attr_flag ? *(int *)attr_val : 32767);

 /* Generate and broadcast a UUID. */
//logfile_uuid = ncptl_log_generate_uuid();
//(void) CODES_MPI_Bcast ((void *)logfile_uuid, 37, MPI_CHAR, 0, MPI_COMM_WORLD);

 /* Plug variables and default values into the NCPTL_CMDLINE structure. */
arguments[0].variable = (CMDLINE_VALUE *) &logfiletmpl;
arguments[1].variable = (CMDLINE_VALUE *) &var_reps;
arguments[1].defaultvalue.intval = 1000LL;
arguments[2].variable = (CMDLINE_VALUE *) &var_maxbytes;
arguments[2].defaultvalue.intval = 1048576LL;
//logfiletmpl_default = (char *) ncptl_malloc (strlen(argv0) + 15, 0);
//sprintf (logfiletmpl_default, "%s-%%p.log", argv0);
//arguments[0].defaultvalue.stringval = logfiletmpl_default;

 /* Parse the command line. */
mpi_is_running = 0;   /* Don't invoke MPI_Abort() after --help. */
ncptl_parse_command_line (argc, argv, arguments, sizeof(arguments)/sizeof(NCPTL_CMDLINE));
if (help_only)
ncptl_fatal ("Internal error in the c_generic backend: failed to exit after giving help");
mpi_is_running = 1;

 /* Establish a mapping from (virtual) task IDs to (physical) ranks. */
procmap = ncptl_allocate_task_map (var_num_tasks);
virtrank = ncptl_physical_to_virtual (procmap, physrank);

 /* Perform initializations specific to the c_mpi backend. */
//ncptl_log_add_comment ("MPI send routines", "MPI_Send() and CODES_MPI_Isend()");
//ncptl_log_add_comment ("MPI reduction operation", REDUCE_OPERATION_NAME);
//sprintf (log_key_str, "[0, %" NICS "]", mpi_tag_ub);
//ncptl_log_add_comment ("MPI tag range", log_key_str);
//
// /* Open the log file and write some standard prologue information to it. */
//logstate = ncptl_log_open (logfiletmpl, physrank);
710
////ncptl_log_write_prologue (logstate, argv[0], logfile_uuid, "c_mpi", "C + MPI",
711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843
//var_num_tasks,
//arguments, sizeof(arguments)/sizeof(NCPTL_CMDLINE),
//sourcecode);
//ncptl_free (logfile_uuid);

 /* Allocate a variety of dynamically growing queues. */
eventqueue = ncptl_queue_init (sizeof (CONC_EVENT));
touchedqueue = ncptl_queue_init (sizeof (ncptl_int));

 /* Perform initializations specific to the c_mpi backend. */
sendreqQ = ncptl_queue_init (sizeof (MPI_Request));
sendstatQ = ncptl_queue_init (sizeof (MPI_Status));
recvreqQ = ncptl_queue_init (sizeof (MPI_Request));
recvstatQ = ncptl_queue_init (sizeof (MPI_Status));
communicators = ncptl_set_init (ESTIMATED_COMMUNICATORS, var_num_tasks*sizeof(char), sizeof(MPI_Comm));procflags = (char *) ncptl_malloc (var_num_tasks*sizeof(char), 0);
for (i=0; i<var_num_tasks; i++)
procflags[i] = 1;
ncptl_set_insert (communicators, (void *)procflags, (void *)&comm_world);
ncptl_free (procflags);

 /****************************************************
  * Generated, program-specific initialization code. *
  ****************************************************/

 /* ASSERT THAT the latency test requires at least two tasks. */
if (!((var_num_tasks)>=(2LL)))
ncptl_fatal ("Assertion failure: the latency test requires at least two tasks");
 /* FOR EACH var_msgsize IN [(['0LL'], None), (['1LL', '2LL', '4LL'], 'var_maxbytes')]... */
{
LOOPBOUNDS loopbounds[2];   /* List of range descriptions */
ncptl_int rangenum;   /* Current offset into loopbounds[] */
ncptl_int initial_vals[3];   /* Cache of the initial, enumerated values */
ncptl_int final_val;   /* Cache of the final value */

 /* Write range 0's loop bounds, "next" function, and termination function to loopbounds[0]. */
loopbounds[0].list_comp = NULL;
initial_vals[0] = 0LL;
final_val = 0LL;
loopbounds[0].integral = 1;
loopbounds[0].u.i.startval = initial_vals[0];
loopbounds[0].u.i.endval = final_val;
loopbounds[0].comparator = CONC_LEQ;
loopbounds[0].increment = CONC_ADD;
loopbounds[0].u.i.incval = 1LL;

 /* Write range 1's loop bounds, "next" function, and termination function to loopbounds[1]. */
loopbounds[1].list_comp = NULL;
initial_vals[0] = 1LL;
initial_vals[1] = 2LL;
initial_vals[2] = 4LL;
final_val = var_maxbytes;
loopbounds[1].integral = 1;
loopbounds[1].u.i.startval = initial_vals[0];
loopbounds[1].u.i.endval = final_val;
if (initial_vals[1]-initial_vals[0]==initial_vals[2]-initial_vals[1]) {
 /* Arithmetic progression */
loopbounds[1].comparator = initial_vals[0]<=initial_vals[1] ? CONC_LEQ : CONC_GEQ;
loopbounds[1].increment = CONC_ADD;
loopbounds[1].u.i.incval = initial_vals[1]-initial_vals[0];
if (!loopbounds[1].u.i.incval)
loopbounds[1].u.i.incval = 1LL;   /* Handle {x,x,x,...,x} case (constant value) */
}
else
if (initial_vals[0] && initial_vals[1] && initial_vals[0]<initial_vals[1] && initial_vals[0]*(initial_vals[1]/initial_vals[0])==initial_vals[1] && initial_vals[1]*(initial_vals[2]/initial_vals[1])==initial_vals[2] && initial_vals[1]/initial_vals[0]==initial_vals[2]/initial_vals[1]) {
 /* Geometric progression (increasing, integral multiplier) */
loopbounds[1].comparator = CONC_LEQ;
loopbounds[1].increment = CONC_MULT;
loopbounds[1].u.i.incval = initial_vals[1]/initial_vals[0];
}
else
if (initial_vals[1] && initial_vals[2] && initial_vals[0]>initial_vals[1] && initial_vals[1]*(initial_vals[0]/initial_vals[1])==initial_vals[0] && initial_vals[2]*(initial_vals[1]/initial_vals[2])==initial_vals[1] && initial_vals[0]/initial_vals[1]==initial_vals[1]/initial_vals[2]) {
 /* Geometric progression (decreasing, integral multiplier) */
loopbounds[1].comparator = CONC_GEQ;
loopbounds[1].increment = CONC_DIV;
loopbounds[1].u.i.incval = initial_vals[0]/initial_vals[1];
}
else
if (initial_vals[0] && initial_vals[1]) {
double initial_vals_d[3];   /* Cache of the initial, enumerated values, but in floating-point context */
double avg_factor;   /* Average multiplier for terms in the sequence */
initial_vals_d[0] = 1.0;
initial_vals_d[1] = 2.0;
initial_vals_d[2] = 4.0;
avg_factor = (initial_vals_d[1]/initial_vals_d[0] + initial_vals_d[2]/initial_vals_d[1]) / 2.0;
if (CONC_DBL2INT(initial_vals_d[0]*avg_factor)==initial_vals[1] && CONC_DBL2INT(initial_vals_d[1]*avg_factor)==initial_vals[2]) {
 /* Geometric progression (decreasing or non-integral multiplier) */
loopbounds[1].comparator = initial_vals[0]<initial_vals[1] ? CONC_LEQ : CONC_GEQ;
loopbounds[1].integral = 0;
loopbounds[1].increment = CONC_MULT;
loopbounds[1].u.d.startval = initial_vals_d[0];
loopbounds[1].u.d.endval = (double)var_maxbytes;
loopbounds[1].u.d.incval = avg_factor;
}
else
ncptl_fatal ("Unable to find an arithmetic or geometric pattern to {%" NICS ", %" NICS ", %" NICS ", ..., %" NICS "}", initial_vals[0], initial_vals[1], initial_vals[2], final_val);
}
else
ncptl_fatal ("Unable to find an arithmetic or geometric pattern to {%" NICS ", %" NICS ", %" NICS ", ..., %" NICS "}", initial_vals[0], initial_vals[1], initial_vals[2], final_val);

 /* Now that we've defined all of our ranges we iterate over each range
  * and each element within each range. */
for (rangenum=0LL; rangenum<2; rangenum++) {
LOOPBOUNDS * thisrange = &loopbounds[rangenum];   /* Current range */
if (conc_seq_nonempty (thisrange))
for (conc_seq_init (thisrange);
conc_seq_continue (thisrange);
conc_seq_next (thisrange)) {
ncptl_int var_msgsize;
if (thisrange->list_comp == NULL)
 /* The loopvar field contains the value for var_msgsize. */
var_msgsize = thisrange->integral ? thisrange->u.i.loopvar : CONC_DBL2INT(thisrange->u.d.loopvar);
else
 /* The queue of list-comprehension values provides the value for var_msgsize. */
var_msgsize = *(ncptl_int *)ncptl_queue_pop(thisrange->list_comp);
{
 /* FOR var_reps REPETITIONS... */
{
ncptl_int numreps = var_reps;   /* Total # of repetitions to perform */
int unroll_loop = numreps <= CONC_MAX_UNROLL;   /* 1=unroll loop; 0=use a REPEAT event */
ncptl_int repnum;   /* Current repetition number */
CONC_EVENT * repevent = NULL;   /* Event designating repetition */
ncptl_int repeventnum = -1LL;   /* Event number corresponding to repevent */
 /* Conditionally unroll the loop. */
for (repnum=0; repnum<(unroll_loop?numreps:1); repnum++) {
 /* Allocate a repeat event if we have more than one repetition. */
if (!unroll_loop && numreps > 1LL) {
repeventnum = ncptl_queue_length (eventqueue);
repevent =  conc_allocate_event (EV_REPEAT);
repevent->s.rep.numreps = numreps;
}

 /* Output a loop body if we have at least one repetition. */
if (unroll_loop || numreps > 0LL) {
844 845 846 847
 /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 RESET THEIR COUNTERS */
{
ncptl_int var_ev = virtrank;
if (((var_ev)&1)==0) {   /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 */
848 849
 /* The current coNCePTuaL statement applies to our task. */
(void) conc_allocate_event (EV_RESET);
850
}
851 852
}
 /* THEN... */
853 854 855 856 857 858 859 860 861
 /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 RECEIVES FROM TASK (var_ev)+(1LL) */
{
ncptl_int var_ev;
 /* Loop over all tasks to see which will send to us. */
for (var_ev=0; var_ev<var_num_tasks; var_ev++)
if (((var_ev)&1)==0) {
 /* var_ev now represents one of the tasks that will send to us. */
ncptl_int ivar_a_task = var_ev;
if (virtrank == ((var_ev)+(1LL))) {
862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893
 /* In this scope, we must be a message recipient. */
 /* Prepare to receive 1LL messages. */
ncptl_int numreps = 1LL;   /* Number of messages */
if (numreps > 1LL) {
CONC_EVENT * repeatev;   /* Event specifying the number of repetitions to perform */
repeatev = conc_allocate_event (EV_REPEAT);
repeatev->s.rep.end_event = ncptl_queue_length(eventqueue);
repeatev->s.rep.numreps = numreps;
}

 /* Ensure we have at least one message to receive. */
if (numreps > 0LL) {
CONC_EVENT *thisev = conc_allocate_event (EV_RECV);

 /* Fill in all of the fields of a receive-event structure. */
thisev->s.recv.source = ncptl_virtual_to_physical (procmap, ivar_a_task);
thisev->s.recv.size = var_msgsize;
thisev->s.recv.alignment = 0LL;
thisev->s.recv.misaligned = 0;
thisev->s.recv.touching = 0;
thisev->s.recv.verification = 0;
thisev->s.recv.tag = 0LL;
thisev->s.recv.pendingsends = pendingsends;
thisev->s.recv.pendingrecvs = pendingrecvs;
thisev->s.recv.bufferofs = 0LL;
thisev->s.recv.buffernum = thisev->s.recv.verification ? pendingsends+pendingrecvs : 0;
(void) ncptl_malloc_message (thisev->s.recv.size+thisev->s.recv.bufferofs, thisev->s.recv.alignment, thisev->s.recv.buffernum, thisev->s.recv.misaligned);
thisev->s.recv.buffer = NULL;
thisev->s.recv.tag = map_tag_into_MPI_range (thisev->s.recv.tag);
}
}
}
894 895 896 897 898
}
 /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 SENDS TO TASK (var_ev)+(1LL) */
{
ncptl_int var_ev = virtrank;
if (((var_ev)&1)==0) {   /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 */
899
 /* The current coNCePTuaL statement applies to our task. */
900
ncptl_int virtdest = (var_ev)+(1LL);
901
if (virtdest>=0 && virtdest<var_num_tasks) {
902
 /* In this scope, (var_ev)+(1LL) represents a single receiver. */
903 904 905 906 907 908 909 910 911 912 913 914
 /* Prepare to send 1LL messages. */
ncptl_int numreps = 1LL;   /* Number of messages */
if (numreps > 1LL) {
CONC_EVENT * repeatev;   /* Event specifying the number of repetitions to perform */
repeatev = conc_allocate_event (EV_REPEAT);
repeatev->s.rep.end_event = ncptl_queue_length(eventqueue);
repeatev->s.rep.numreps = numreps;
}

 /* Ensure we have at least one message to send. */
if (numreps > 0LL) {
CONC_EVENT *thisev = conc_allocate_event (EV_SEND);
915 916
if (virtrank == ((var_ev)+(1LL)))
ncptl_fatal ("Send-to-self deadlock encountered on task %d in line 17 of the source code", virtrank);
917 918

 /* Fill in all of the fields of a send-event structure. */
919
thisev->s.send.dest = ncptl_virtual_to_physical (procmap, (var_ev)+(1LL));
920 921 922 923 924 925 926 927 928 929 930 931 932 933 934
thisev->s.send.size = var_msgsize;
thisev->s.send.alignment = 0LL;
thisev->s.send.misaligned = 0;
thisev->s.send.touching = 0;
thisev->s.send.verification = 0;
thisev->s.send.tag = 0LL;
thisev->s.send.pendingsends = pendingsends;
thisev->s.send.pendingrecvs = pendingrecvs;
thisev->s.send.bufferofs = 0LL;
thisev->s.send.buffernum = thisev->s.send.verification ? pendingsends+pendingrecvs : 0;
(void) ncptl_malloc_message (thisev->s.send.size+thisev->s.send.bufferofs, thisev->s.send.alignment, thisev->s.send.buffernum, thisev->s.send.misaligned);
thisev->s.send.buffer = NULL;
thisev->s.send.tag = map_tag_into_MPI_range (thisev->s.send.tag);
}
}
935
}
936 937
}
 /* THEN... */
938 939 940 941 942 943 944 945 946
 /* TASKS var_od SUCH THAT ((var_od)&1)==1 RECEIVES FROM TASK (var_od)-(1LL) */
{
ncptl_int var_od;
 /* Loop over all tasks to see which will send to us. */
for (var_od=0; var_od<var_num_tasks; var_od++)
if (((var_od)&1)==1) {
 /* var_od now represents one of the tasks that will send to us. */
ncptl_int ivar_b_task = var_od;
if (virtrank == ((var_od)-(1LL))) {
947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978
 /* In this scope, we must be a message recipient. */
 /* Prepare to receive 1LL messages. */
ncptl_int numreps = 1LL;   /* Number of messages */
if (numreps > 1LL) {
CONC_EVENT * repeatev;   /* Event specifying the number of repetitions to perform */
repeatev = conc_allocate_event (EV_REPEAT);
repeatev->s.rep.end_event = ncptl_queue_length(eventqueue);
repeatev->s.rep.numreps = numreps;
}

 /* Ensure we have at least one message to receive. */
if (numreps > 0LL) {
CONC_EVENT *thisev = conc_allocate_event (EV_RECV);

 /* Fill in all of the fields of a receive-event structure. */
thisev->s.recv.source = ncptl_virtual_to_physical (procmap, ivar_b_task);
thisev->s.recv.size = var_msgsize;
thisev->s.recv.alignment = 0LL;
thisev->s.recv.misaligned = 0;
thisev->s.recv.touching = 0;
thisev->s.recv.verification = 0;
thisev->s.recv.tag = 0LL;
thisev->s.recv.pendingsends = pendingsends;
thisev->s.recv.pendingrecvs = pendingrecvs;
thisev->s.recv.bufferofs = 0LL;
thisev->s.recv.buffernum = thisev->s.recv.verification ? pendingsends+pendingrecvs : 0;
(void) ncptl_malloc_message (thisev->s.recv.size+thisev->s.recv.bufferofs, thisev->s.recv.alignment, thisev->s.recv.buffernum, thisev->s.recv.misaligned);
thisev->s.recv.buffer = NULL;
thisev->s.recv.tag = map_tag_into_MPI_range (thisev->s.recv.tag);
}
}
}
979 980 981 982 983
}
 /* TASKS var_od SUCH THAT ((var_od)&1)==1 SENDS TO TASK (var_od)-(1LL) */
{
ncptl_int var_od = virtrank;
if (((var_od)&1)==1) {   /* TASKS var_od SUCH THAT ((var_od)&1)==1 */
984
 /* The current coNCePTuaL statement applies to our task. */
985
ncptl_int virtdest = (var_od)-(1LL);
986
if (virtdest>=0 && virtdest<var_num_tasks) {
987
 /* In this scope, (var_od)-(1LL) represents a single receiver. */
988 989 990 991 992 993 994 995 996 997 998 999
 /* Prepare to send 1LL messages. */
ncptl_int numreps = 1LL;   /* Number of messages */
if (numreps > 1LL) {
CONC_EVENT * repeatev;   /* Event specifying the number of repetitions to perform */
repeatev = conc_allocate_event (EV_REPEAT);
repeatev->s.rep.end_event = ncptl_queue_length(eventqueue);
repeatev->s.rep.numreps = numreps;
}

 /* Ensure we have at least one message to send. */
if (numreps > 0LL) {
CONC_EVENT *thisev = conc_allocate_event (EV_SEND);
1000 1001
if (virtrank == ((var_od)-(1LL)))
ncptl_fatal ("Send-to-self deadlock encountered on task %d in line 18 of the source code", virtrank);
1002 1003

 /* Fill in all of the fields of a send-event structure. */
1004
thisev->s.send.dest = ncptl_virtual_to_physical (procmap, (var_od)-(1LL));
1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019
thisev->s.send.size = var_msgsize;
thisev->s.send.alignment = 0LL;
thisev->s.send.misaligned = 0;
thisev->s.send.touching = 0;
thisev->s.send.verification = 0;
thisev->s.send.tag = 0LL;
thisev->s.send.pendingsends = pendingsends;
thisev->s.send.pendingrecvs = pendingrecvs;
thisev->s.send.bufferofs = 0LL;
thisev->s.send.buffernum = thisev->s.send.verification ? pendingsends+pendingrecvs : 0;
(void) ncptl_malloc_message (thisev->s.send.size+thisev->s.send.bufferofs, thisev->s.send.alignment, thisev->s.send.buffernum, thisev->s.send.misaligned);
thisev->s.send.buffer = NULL;
thisev->s.send.tag = map_tag_into_MPI_range (thisev->s.send.tag);
}
}
1020
}
1021 1022
}
 /* THEN... */
1023 1024 1025 1026
 /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 LOGS "Bytes" AND "1/2 RTT (usecs)" */
{
ncptl_int var_ev = virtrank;
if (((var_ev)&1)==0) {   /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 */
1027 1028 1029 1030 1031 1032 1033
 /* The current coNCePTuaL statement applies to our task. */
CONC_EVENT *thisev = conc_allocate_event (EV_CODE);
thisev->s.code.number = 0;
thisev->s.code.procmap = NULL;
thisev->s.code.var_msgsize = var_msgsize;
}
}
1034
}
1035 1036 1037 1038 1039 1040 1041 1042 1043
}

 /* Assign the number of events to repeat, now that we know that number. */
if (!unroll_loop && numreps > 1LL) {
repevent = repeventnum + (CONC_EVENT *) ncptl_queue_contents (eventqueue, 0);
repevent->s.rep.end_event = ncptl_queue_length (eventqueue) - 1;
}
}
 /* THEN... */
1044 1045 1046 1047
 /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 COMPUTES AGGREGATES */
{
ncptl_int var_ev = virtrank;
if (((var_ev)&1)==0) {   /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 */
1048 1049 1050 1051 1052 1053
 /* The current coNCePTuaL statement applies to our task. */
(void) conc_allocate_event (EV_FLUSH);
}
}
}
}
1054
}
1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182
}
 /* Begin a new top-level statement. */
(void) conc_allocate_event (EV_NEWSTMT);

 /*************************
  * More boilerplate code *
  *************************/

 /* Abort if the program will terminate with pending messages. */
if (pendingsends && pendingrecvs)
ncptl_fatal("Neglected to await the completion of %" NICS " asynchronous %s and %" NICS " asynchronous %s",
pendingsends, pendingsends==1LL ? "send" : "sends",
pendingrecvs, pendingrecvs==1LL ? "receive" : "receives");
else
if (pendingsends)
ncptl_fatal("Neglected to await the completion of %" NICS " asynchronous %s",
pendingsends, pendingsends==1LL ? "send" : "sends");
else
if (pendingrecvs)
ncptl_fatal("Neglected to await the completion of %" NICS " asynchronous %s",
pendingrecvs, pendingrecvs==1LL ? "receive" : "receives");

 /* Allocate memory for non-unique messages and asynchronous
  * message handles now that we know how much memory we need
  * to allocate. */
eventlist = (CONC_EVENT *) ncptl_queue_contents (eventqueue, 0);
numevents = ncptl_queue_length (eventqueue);
sendrequests = (MPI_Request *) ncptl_queue_contents (sendreqQ, 0);
recvrequests = (MPI_Request *) ncptl_queue_contents (recvreqQ, 0);
for (i=0; i<numevents; i++) {
CONC_EVENT *thisev = &eventlist[i];   /* Cache of the current event */
switch (thisev->type) {
case EV_SEND:
if (!thisev->s.send.buffer)
thisev->s.send.buffer = ncptl_malloc_message (thisev->s.send.bufferofs + thisev->s.send.size,
thisev->s.send.alignment,
thisev->s.send.buffernum,
thisev->s.send.misaligned);
if (thisev->s.send.verification)
ncptl_fill_buffer (CONC_GETBUFPTR(send), thisev->s.send.bufferofs + thisev->s.send.size, -1);
break;

case EV_RECV:
if (!thisev->s.recv.buffer)
thisev->s.recv.buffer = ncptl_malloc_message (thisev->s.recv.bufferofs + thisev->s.recv.size,
thisev->s.recv.alignment,
thisev->s.recv.buffernum,
thisev->s.recv.misaligned);
if (thisev->s.recv.verification)
ncptl_fill_buffer (CONC_GETBUFPTR(recv), thisev->s.recv.bufferofs + thisev->s.recv.size, -1);
break;

default:
break;
}
}
}

/* Process a subset of the events in a given event list. */
static void conc_process_events (CONC_EVENT *eventlist,
ncptl_int firstev, ncptl_int lastev, ncptl_int numreps)
{
CONC_EVENT * thisev;   /* Cache of the current event */
CONC_EVENT * thisev_first = &eventlist[firstev];   /* Cache of the first event */
ncptl_int i;   /* Iterate over events. */
ncptl_int j;   /* Iterate over repetitions. */

 /* Declarations specific to the c_mpi backend */
MPI_Status status;   /* Not needed but required by CODES_MPI_Recv() */

 /* Process from event firstev to event lastev (both inclusive). */
for (j=numreps; j>0; j--)
for (i=firstev, thisev=thisev_first; i<=lastev; i++, thisev++) {
 /* Declare variables needed by all C-based backends. */

 /* Process a single event. */
switch (thisev->type) {
case EV_SEND:
 /* Synchronous send */
(void) CODES_MPI_Send (CONC_GETBUFPTR(send),
(int)thisev->s.send.size, MPI_BYTE,
(int)thisev->s.send.dest, (int)thisev->s.send.tag, MPI_COMM_WORLD);
break;

case EV_RECV:
 /* Synchronous receive */
(void) CODES_MPI_Recv (CONC_GETBUFPTR(recv),
(int)thisev->s.recv.size, MPI_BYTE,
(int)thisev->s.recv.source, (int)thisev->s.recv.tag,
MPI_COMM_WORLD, &status);
break;

case EV_RESET:
 /* Reset all of the counters exported to coNCePTuaL programs. */
starttime = ncptl_time();
break;

case EV_FLUSH:
 /* Force all aggregate functions to produce a result. */
if (!suppress_output) {
uint64_t stop_elapsed_usecs = ncptl_time();
var_elapsed_usecs = stop_elapsed_usecs - starttime;
//ncptl_log_compute_aggregates (logstate);
starttime += ncptl_time() - stop_elapsed_usecs;
}
break;

case EV_REPEAT:
 /* Repeatedly perform the next batch of events. */
conc_process_events (eventlist, i+1, thisev->s.rep.end_event, thisev->s.rep.numreps);
i = thisev->s.rep.end_event;
thisev = &eventlist[i];
break;

case EV_NEWSTMT:
 /* Begin a new table in the log file. */
if (!suppress_output) {
uint64_t stop_elapsed_usecs = ncptl_time();
var_elapsed_usecs = stop_elapsed_usecs - starttime;
//ncptl_log_commit_data (logstate);
starttime += ncptl_time() - stop_elapsed_usecs;
}
break;

case EV_CODE:
 /* Execute an arbitrary piece of code. */
switch (thisev->s.code.number) {
case 0:
1183
 /* TASKS var_ev SUCH THAT ((var_ev)&1)==0 LOGS "Bytes" AND "1/2 RTT (usecs)" */
1184 1185 1186
if (!suppress_output) {
uint64_t stop_elapsed_usecs = ncptl_time();
var_elapsed_usecs = stop_elapsed_usecs - starttime;
1187 1188
//ncptl_log_write (logstate, 0, "Bytes", NCPTL_FUNC_ONLY, 0.0, (double)thisev->s.code.var_msgsize);
//ncptl_log_write (logstate, 1, "1/2 RTT (usecs)", NCPTL_FUNC_MEDIAN, 0.0, ((double)var_elapsed_usecs)/(2.0));
1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241
starttime += ncptl_time() - stop_elapsed_usecs;
}
break;

default:
 /* The C code generation module must be broken. */
ncptl_fatal ("Internal error: unknown EV_CODE block %" NICS, thisev->s.code.number);
break;
}
break;

default:
 /* The c_generic backend or the c_mpi backend must be broken. */
ncptl_fatal ("Internal error: unknown event type %d", thisev->type);
break;
}
}
}

/* Finish up cleanly and return a status code. */
static int conc_finalize (void)
{
int exitcode = 0;   /* Program exit code (to pass to exit()) */

 /* Declarations specific to the c_mpi backend */
//int mpiresult;   /* Return code from MPI_Finalize() */
//char log_key_str[128];   /* String representing the range of valid MPI tags */
//
// /* Finalization code specific to the c_mpi backend */
//log_key_str[0] = '\0';
//if (conc_mcast_tallies[CONC_MCAST_MPI_BCAST] > 0) {
//char onefuncstr[50];
//sprintf (onefuncstr, "%sCODES_MPI_Bcast()*%" NICS,
//log_key_str[0] == '\0' ? "" : " ", conc_mcast_tallies[CONC_MCAST_MPI_BCAST]);
//strcat (log_key_str, onefuncstr);
//}
//if (conc_mcast_tallies[CONC_MCAST_MPI_ALLTOALL] > 0) {
//char onefuncstr[50];
//sprintf (onefuncstr, "%sCODES_MPI_Alltoall()*%" NICS,
//log_key_str[0] == '\0' ? "" : " ", conc_mcast_tallies[CONC_MCAST_MPI_ALLTOALL]);
//strcat (log_key_str, onefuncstr);
//}
//if (conc_mcast_tallies[CONC_MCAST_MPI_ALLTOALLV] > 0) {
//char onefuncstr[50];
//sprintf (onefuncstr, "%sCODES_MPI_Alltoallv()*%" NICS,
//log_key_str[0] == '\0' ? "" : " ", conc_mcast_tallies[CONC_MCAST_MPI_ALLTOALLV]);
//strcat (log_key_str, onefuncstr);
//}
//if (log_key_str[0] != '\0')
//ncptl_log_add_comment ("Multicast functions used (statically)", log_key_str);
//
// /* Write a standard epilogue to the log file. */
////ncptl_log_commit_data (logstate);
1242
////ncptl_log_write_epilogue (logstate);
1243 1244 1245
//ncptl_log_close (logstate);

 /* Inform the run-time library that it's no longer needed. */
1246 1247 1248
//ncptl_queue_empty (eventqueue);
//ncptl_free (eventqueue);
//ncptl_finalize();
1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263

 /* Finalization code specific to the c_mpi backend */
CODES_MPI_Finalize();
mpi_is_running = 0;
exitcode = 0;

 /* Return an exit status code. */
return exitcode;
}

/*************************************************************************/
/***************************** MAIN ROUTINE ******************************/
/*************************************************************************/

/* Program execution starts here. */
1264
static int latencyall_main (int argc, char *argv[])
1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286
{
 /* Declare variables needed by all C-based backends. */
CONC_EVENT * eventlist;   /* List of events to execute */
ncptl_int numevents;   /* Number of entries in eventlist[] */

 /* ----- Initialization ----- */
conc_initialize (argc, argv);
eventlist = (CONC_EVENT *) ncptl_queue_contents (eventqueue, 0);
numevents = ncptl_queue_length (eventqueue);
sendrequests = (MPI_Request *) ncptl_queue_contents (sendreqQ, 0);
sendstatuses = (MPI_Status *) ncptl_queue_contents (sendstatQ, 0);
recvrequests = (MPI_Request *) ncptl_queue_contents (recvreqQ, 0);
recvstatuses = (MPI_Status *) ncptl_queue_contents (recvstatQ, 0);
starttime = ncptl_time();

 /* ----- Event-list processing ----- */
conc_process_events (eventlist, 0, numevents-1, 1);

 /* ----- Finalization ----- */
return conc_finalize();
}

1287 1288 1289 1290 1291 1292
/* fill in function pointers for this method */
struct codes_conceptual_bench latencyall_bench = 
{
.program_name = "latencyall",
.conceptual_main = latencyall_main,
};