int main(int argc, char **argv)
{
if (argc != 4) {
printf("usage: %s <server-address> <server-port> <phase[-1 - 6]>\n",
argv[0]);
return EXIT_FAILURE;
}
int start_phase = atoi(argv[3]);
bool res;
test_state_t *state = NULL;
rvm_cfg_t *cfg;
if(start_phase >= 0) {
/* Try to recover from server */
cfg = initialize_rvm(argv[1], argv[2], true,
create_rmem_layer);
/* Recover the state (if any) */
state = (test_state_t*)rvm_get_usr_data(cfg);
} else {
/* Starting from scratch */
cfg = initialize_rvm(argv[1], argv[2], false,
create_rmem_layer);
CHECK_ERROR(cfg == NULL, ("Failed to initialize rvm\n"));
state = NULL;
}
rvm_txid_t txid;
/*====================================================================
* TX 0 - Allocate and Initialize Arrays
*===================================================================*/
/* If state is NULL then we are starting from scratch or recovering from
an early error */
if(state == NULL) {
LOG(1,("Phase 0:\n"));
TX_START;
/* Allocate a "state" structure to test pointers */
state = (test_state_t*)rvm_alloc(cfg, sizeof(test_state_t));
CHECK_ERROR(state == NULL, ("FAILURE: Couldn't allocate state\n"));
/* Initialize the arrays */
res = phase0(cfg, state);
CHECK_ERROR(!res, ("FAILURE: Phase 0 Failure\n"));
if(start_phase == -1) {
LOG(1, ("SUCCESS: Phase 0, simulating failure\n"));
return EXIT_SUCCESS;
}
/* End of first txn */
state->phase = PHASE1;
TX_COMMIT;
}
switch(state->phase) {
case PHASE1:
/*====================================================================
* TX 1 Increment arrays, don't mess with LL
*===================================================================*/
LOG(1, ("Phase 1:\n"));
TX_START;
res = phase1(cfg, state);
CHECK_ERROR(!res, ("FAILURE: Phase 1 failed\n"));
/* Simulate Failure */
if(start_phase == 0) {
LOG(1, ("SUCCESS: Phase 1, simulating failure\n"));
return EXIT_SUCCESS;
}
state->phase = PHASE2;
TX_COMMIT;
case PHASE2: //Fallthrough
/*====================================================================
* TX 2 Free Arrays
*===================================================================*/
LOG(1, ("Phase 2:\n"));
TX_START;
res = phase2(cfg, state);
CHECK_ERROR(!res, ("FAILURE: Phase 2 failed\n"));
/* Simulate Failure */
if(start_phase == 1) {
LOG(1, ("SUCCESS: Phase 2, simulating failure\n"));
return EXIT_SUCCESS;
}
state->phase = PHASE3;
TX_COMMIT;
case PHASE3: //Fallthrough
/*====================================================================
* TX 3 Fill in Linked list
*===================================================================*/
LOG(1, ("Phase 3:\n"));
TX_START;
res = phase3(cfg, state);
CHECK_ERROR(!res, ("FAILURE: Phase 3 failed\n"));
/* Simulate Failure */
if(start_phase == 2) {
LOG(1, ("SUCCESS: Phase 3, simulating failure\n"));
return EXIT_SUCCESS;
}
state->phase = PHASE4;
TX_COMMIT;
case PHASE4:
/*====================================================================
* TX 4 Free Half the linked list
*===================================================================*/
LOG(1, ("Phase 4:\n"));
TX_START;
res = phase4(cfg, state);
CHECK_ERROR(!res, ("FAILURE: Phase 4 failed\n"));
/* Simulate Failure */
if(start_phase == 3) {
LOG(1, ("SUCCESS: Phase 4, simulating failure\n"));
return EXIT_SUCCESS;
}
state->phase = PHASE5;
TX_COMMIT;
case PHASE5:
/*====================================================================
* TX 5 Re-allocate half of the linked list
*===================================================================*/
LOG(1, ("Phase 5:\n"));
TX_START;
res = phase5(cfg, state);
CHECK_ERROR(!res, ("FAILURE: Phase 5 failed\n"));
/* Simulate Failure */
if(start_phase == 4) {
LOG(1, ("SUCCESS: Phase5, simulating failure\n"));
return EXIT_SUCCESS;
}
state->phase = PHASE6;
TX_COMMIT;
case PHASE6:
/*====================================================================
* TX 6 Free whole linked list
*===================================================================*/
LOG(1, ("Phase 6:\n"));
TX_START;
res = phase6(cfg, state);
CHECK_ERROR(!res, ("FAILURE: Phase 6 failed\n"));
/* Simulate Failure */
if(start_phase == 5) {
LOG(1, ("SUCCESS: Phase 6, simulating failure\n"));
return EXIT_SUCCESS;
}
state->phase = DONE;
TX_COMMIT;
case DONE:
res = rvm_cfg_destroy(cfg);
CHECK_ERROR(!res, ("FAILURE: Failed to destroy rvm state\n"));
LOG(1, ("SUCCESS: Got through all phases\n"));
break;
default:
LOG(1, ("FAILURE: Corrupted State, tried phase %d\n", state->phase));
return EXIT_FAILURE;
}
return EXIT_SUCCESS;
}
void
psrs(int size, int rank, int nInts, int *toSort, char *fname)
{
char hname[256];
double start, end, total = 0, algStart;
int i, *privateInts, *regularSamples, *collectedSamples, *pivots,
*partitionIndices, *localPartitionSizes, *incomingPartitionSizes,
**partitions, *mergedPartitions, *partitionSizes, *sortedArray;
FILE *fptr = NULL;
if (rank == MASTER) fptr = fopen(fname, "a");
memset (hname, '\0', sizeof(unsigned char)*256);
gethostname(hname, 255);
DBPRINT(("%d of %d running on pid %d %s\n", rank, size, getpid(),
hname));
if (rank == MASTER) {
collectedSamples = calloc(1, sizeof(int)*size*size);
if (!collectedSamples) {
err(1,"Failed to allocate memory for collected samples "
"array for master process");
MPI_Finalize();
exit(1);
}
}
MPI_Barrier(MPI_COMM_WORLD);
if (nInts < VALIDATION_THRESHOLD) validateResults = 1;
/*
* "Phase 0" in which the array is split into contiguous chunks
* distributed amongst the processes.
*/
phase0(size, rank, nInts, toSort, &privateInts);
/* Phase 1 */
algStart = MPI_Wtime();
START_TIMER((start));
phase1(size, rank, nInts, toSort, privateInts, ®ularSamples);
MPI_Barrier(MPI_COMM_WORLD);
STOP_TIMER((1), (start), (end), (total), (rank), (fptr));
/* Phase 2 */
START_TIMER((start));
phase2(size, rank, regularSamples, &collectedSamples, &pivots);
MPI_Barrier(MPI_COMM_WORLD);
STOP_TIMER((2), (start), (end), (total), (rank), (fptr));
/* Phase 3 */
START_TIMER((start));
phase3(size, rank, nInts, privateInts, pivots, &localPartitionSizes,
&partitionIndices, &incomingPartitionSizes, &partitions);
MPI_Barrier(MPI_COMM_WORLD);
STOP_TIMER((3), (start), (end), (total), (rank), (fptr));
/* Phase 4 */
START_TIMER((start));
phase4(size, incomingPartitionSizes, partitions, &mergedPartitions);
MPI_Barrier(MPI_COMM_WORLD);
STOP_TIMER((4), (start), (end), (total), (rank), (fptr));
if (rank == MASTER) {
total = end - algStart;
fprintf(fptr, "The algorithm took %f seconds in total\n", total);
}
if (!validateResults) return;
/* Run validation test */
/* "Phase 5" concatenate the lists back at the master */
phase5(size, rank, nInts, incomingPartitionSizes,
mergedPartitions, &partitionSizes, &sortedArray);
/*
* Assert that the array is equivalent to the sorted original array
* where we sort the original array using a known, proven, sequential,
* method.
*/
if (rank == MASTER) {
qsort(toSort, nInts, sizeof(int), intComp);
}
for (i = 0; i < nInts; i++) {
if (rank == MASTER) {
if (toSort[i] != sortedArray[i]) {
printf("OH NO, got %d at pos %d, expected "
"%d\n", sortedArray[i], i, toSort[i]);
}
}
}
if (rank == MASTER) fclose(fptr);
}