void printMenu(){
int choice;
do{
choice = 0;
printf("Performance assessment:\n");
printf("-----------------------\n");
printf("1) Enter parameters\n");
printf("2) Print table of parameters\n");
printf("3) Print table of performance\n");
printf("4) Quit\n");
printf("\nEnter selection: ");
scanf("%d", &choice);
switch(choice){
case 1:
enterParameters();
break;
case 2:
printParameters();
break;
case 3:
printPerformance();
break;
case 4:
exit(0);
break;
default:
printf("Invalid selection\n\n");
//Clear the input stream in case of erroneous inputs
while ((choice = getchar()) != '\n' && choice != EOF);
break;
}//Switch
}while(choice != 4);
}//printMenu
void
printReportHeader ( void )
{
char *now;
int nameSize;
char myProcName[MPI_MAX_PROCESSOR_NAME];
if ( rank == 0 )
{
now = getTimeStr ( );
MPI_Get_processor_name ( myProcName, &nameSize );
printSeparator ( );
printf ( "\n com Point-to-Point MPI Bandwidth and Latency Benchmark\n" );
printf ( " Version %d.%d.%d\n", majorVersion, minorVersion,
patchVersion );
printf ( " Run at %s, with rank 0 on %s\n\n", now, myProcName );
free ( now );
printParameters ( );
if ( presta_check_data == 1 )
{
printf
( " *** Verifying operation message buffers. Benchmark results not valid! ***\n" );
}
if ( strlen ( procSrcTitle ) > 0 )
printf ( " Using Task Pair file %s\n", procSrcTitle );
printf ( "\n" );
if ( strcmp ( argStruct.testList, "Latency" ) != 0 )
{
if ( argStruct.sumLocalBW == 1 )
printf ( " Bandwidth calculated as sum of process bandwidths.\n" );
else if ( argStruct.sumLocalBW == 0 )
printf
( " Bandwidth calculated as total volume / longest task communication.\n" );
/* TODO : Not necessary unless providing info about sample time.
printTimingInfo ( );
*/
}
}
if ( argStruct.printHostInfo )
printCommTargets ( argStruct.procsPerNode, argStruct.useNearestRank );
printSeparator ( );
}
int main(int argc, char *argv[])
{
double minTime = 99999.0;
double testTime = 0.0;
double maxBidirBw = 0.0;
double maxUnidirBw = 0.0;
int maxBidirSize = 0;
int maxUnidirSize = 0;
char* procFile = NULL;
int* procList = NULL;
int procListSize;
int rank, wsize, iters, procs;
int messStart, messStop, messFactor, messSize;
int procsPerNode, totalops, procIdx;
int useBarrier, printPairs, useNearestRank, dummy;
double currBw;
char allocPattern;
MPI_Comm activeComm;
command = argv[0];
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &wsize);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if(rank == 0) {
//sleep(5);
//while (!__sync_bool_compare_and_swap(&r_sync, 59, 0)){ sleep(2); printf("sync value %d \n", r_sync); };
int k = 0;
#pragma omp parallel
#pragma omp atomic
k++;
printf ("Done Sync. Number of Threads counted = %i\n",k);
int N = 1048576;
int i = 0;
int j = 0;
double start, diff;
for(j=0; j < 50 ; j++) {
#pragma omp parallel for
for (i=0; i<N; i++) {
a[i] = 1.0;
b[i] = 2.0;
}
start = MPI_Wtime();
#pragma omp parallel for
for (i = 0; i < N; i++) {
a[i]= b[i];
}
diff = MPI_Wtime() - start ;
printf("ELAPSED time to copy : %11.6f N: %d \n",diff, N);
}
}else{
//__sync_fetch_and_add(&r_sync, 1);
printf("OK other ranks \n");
//sleep(100);
}
MPI_Barrier(MPI_COMM_WORLD);
MPI_Finalize();
return 0 ;
if ( !processArgs(argc, argv, rank, wsize, &iters, &dummy,
&messStart, &messStop, &messFactor,
&procFile, &procsPerNode, &allocPattern,
&printPairs, &useBarrier, &useNearestRank) )
{
if ( rank == 0 )
printUse();
MPI_Finalize();
exit(-1);
}
if ( ! getProcList(procFile, wsize, &procList, &procListSize,
procsPerNode, allocPattern) )
{
if ( procFile )
printf("Failed to get process list from file %s.\n", procFile);
else
printf("Failed to allocate process list.\n");
exit(-1);
}
if ( rank == 0 )
printReportHeader();
//Set up intra-node shared memory structures.
if(rank == 0) {
//One rank per node allocates shared send request lists.
req_array = malloc(wsize*sizeof(ReqInfo));
/*if(!IS_SM_BUF((void*)req_array)){
printf("Failed to allocate from SM region :%p \n", req_array);
exit(-1);
}*/
}
//broadcast address of the pointer variable so that each rank has access to it
MPI_Bcast(&req_array , 1, MPI_LONG, 0, MPI_COMM_WORLD);
//printf("Broadcasting of shared mem region complete! rank : %d region_addr : %lu region_ptr : %lu \n", rank, &req_array, req_array);
//printf("Access shm region ! rank : %d region for sync : %lu \n", rank, req_array[rank].sync);
/*
if(rank == 0){
char* comBuf = (char*)malloc(10);
memset(comBuf, 1, 10);
req_array[30].buffer = comBuf ;
comBuf[0] = 'n';
req_array[30].sync = malloc(sizeof(int));
req_array[30].sync[0] = 12;
printf("rank : %d done sending buff ptr : %p sync ptr : %p \n",rank, comBuf, req_array[30].sync );
printf("sleeping ! pid() %d \n", getpid());
sleep(40000);
}else if(rank == 30){
while(req_array[30].sync == NULL){
}
while(req_array[30].sync[0] != 12){
}
printf("rank : %d buffer value: %c sync value : %d buff ptr : %p sync ptr : %p \n",
rank, req_array[30].buffer[0], req_array[30].sync[0], req_array[30].buffer, req_array[30].sync );
printf("sleeping ! pid() %d \n", getpid());
sleep(40000);
}
MPI_Barrier(MPI_COMM_WORLD);
return 0;*/
printPairs = 1 ;
for ( procIdx = 0; procIdx < procListSize; procIdx++ )
{
procs = procList[procIdx];
if ( rank == 0 && printPairs )
{
printActivePairs(procs, rank, wsize, procsPerNode,
allocPattern, useNearestRank);
}
/* Create Communicator of all active processes */
createActiveComm(procs, rank, wsize, procsPerNode,
allocPattern, printPairs, useNearestRank, &activeComm);
// messStart = 8388608;
// messStop = 4096 ;
// messStop = 8388608 ;
for ( messSize = messStart; messSize <= messStop;
messSize *= messFactor )
{
testTime = runUnicomTest(procs, messSize, iters, rank,
wsize, procsPerNode, allocPattern,
useBarrier, useNearestRank, &activeComm);
if ( rank == 0 && testTime > 0 )
{
totalops = iters * (procs/2);
currBw = ((double)totalops*(double)messSize/testTime)/1000000;
if ( currBw > maxUnidirBw )
{
maxUnidirBw = currBw;
maxUnidirSize = messSize;
}
if ( testTime < minTime )
minTime = testTime;
}
}
if ( activeComm != MPI_COMM_NULL )
MPI_Comm_free(&activeComm);
if ( rank == 0 )
printf("\n");
}
for ( procIdx = 0; procIdx < procListSize; procIdx++ )
{
procs = procList[procIdx];
if ( rank == 0 && printPairs )
{
printActivePairs(procs, rank, wsize, procsPerNode,
allocPattern, useNearestRank);
}
// Create Communicator of all active processes
createActiveComm(procs, rank, wsize, procsPerNode,
allocPattern, printPairs, useNearestRank, &activeComm);
for ( messSize = messStart; messSize <= messStop;
messSize *= messFactor )
{
testTime = runBicomTest(procs, messSize, iters, rank,
wsize, procsPerNode, allocPattern,
useBarrier, useNearestRank, &activeComm);
if ( rank == 0 && testTime > 0 )
{
totalops = iters * procs;
currBw = (((double)totalops*(double)messSize)/testTime)/1000000;
if ( currBw > maxBidirBw )
{
maxBidirBw = currBw;
maxBidirSize = messSize;
}
if ( testTime < minTime )
minTime = testTime;
}
}
if ( activeComm != MPI_COMM_NULL )
MPI_Comm_free(&activeComm);
if ( rank == 0 )
printf("\n");
}
if ( rank == 0 )
{
printf("Max Unidirectional Bandwith : %13.2f for message size of %7d bytes\n",
maxUnidirBw, maxUnidirSize);
printf("Max Bidirectional Bandwith : %13.2f for message size of %7d bytes\n",
maxBidirBw, maxBidirSize);
printParameters(iters, messStart, messStop, messFactor,
procFile, procsPerNode, allocPattern, useBarrier);
free(req_array);
}
printReportFooter(minTime, rank, wsize, procsPerNode, useNearestRank);
free(procList);
MPI_Finalize();
exit(0);
}
main(int argc, char** argv)
{
int rank, wsize, iters, i, procs, currtarg, dummy;
double diff = 0.0;
double start, max, mintime = 9999;
MPI_Status stat;
char comBuf;
MPI_Comm activeComm;
char* procFile = NULL;
int* procList = NULL;
int procListSize;
int messStart, messStop, messFactor;
int procsPerNode, procIdx, useBarrier, printPairs, useNearestRank;
char allocPattern;
command = argv[0];
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &wsize);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if ( !processArgs(argc, argv, rank, wsize, &iters,
&dummy, &messStart, &messStop, &messFactor,
&procFile, &procsPerNode, &allocPattern,
&printPairs, &useBarrier, &useNearestRank) )
{
if ( rank == 0 )
printUse();
MPI_Finalize();
exit(-1);
}
if ( ! getProcList(procFile, wsize, &procList, &procListSize,
procsPerNode, allocPattern) )
{
if ( procFile )
printf("Failed to get process list from file %s.\n", procFile);
else
printf("Failed to allocate process list.\n");
exit(-1);
}
if ( rank == 0 )
printReportHeader();
currtarg = getTargetRank(rank, wsize, procsPerNode, useNearestRank);
for ( procIdx = 0; procIdx < procListSize; procIdx++ )
{
procs = procList[procIdx];
if ( printPairs )
{
printActivePairs(procs, rank, wsize, procsPerNode,
allocPattern, useNearestRank);
}
/* Create Communicator of all active processes */
createActiveComm(procs, rank, wsize, procsPerNode,
allocPattern, printPairs, useNearestRank, &activeComm);
if ( isActiveProc(rank, wsize, procsPerNode, procs,
allocPattern, useNearestRank) )
{
if ( rank < currtarg )
{
/* Ensure pair communication has been initialized */
MPI_Recv(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD, &stat);
MPI_Send(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD);
}
else
{
/* Ensure pair communication has been initialized */
MPI_Send(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD);
MPI_Recv(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD, &stat);
}
//generic_barrier(activeComm);
MPI_Barrier(activeComm);
//generic_barrier(activeComm);
MPI_Barrier(activeComm);
if ( rank < currtarg )
{
/* Time operation loop */
start = MPI_Wtime();
for ( i = 0; i < iters; i++ )
{
MPI_Send(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD);
MPI_Recv(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD, &stat);
}
}
else
{
/* Time operation loop */
start = MPI_Wtime();
for ( i = 0; i < iters; i++ )
{
MPI_Recv(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD, &stat);
MPI_Send(&comBuf, 0, MPI_INT, currtarg, 0, MPI_COMM_WORLD);
}
}
if ( useBarrier )
MPI_Barrier(activeComm);
//generic_barrier(activeComm);
diff = MPI_Wtime() - start;
}
if ( activeComm != MPI_COMM_NULL )
MPI_Comm_free(&activeComm);
/* Get maximum sample length */
MPI_Reduce(&diff, &max, 1, MPI_DOUBLE, MPI_MAX, 0, MPI_COMM_WORLD);
if ( rank == 0 )
{
if ( max < mintime )
mintime = max;
printf(outputFormat, procs, max/iters/2*1000000);
}
}
if ( rank == 0 )
{
printParameters(iters, procFile, procsPerNode,
allocPattern, useBarrier);
}
printReportFooter(mintime, rank, wsize, procsPerNode, useNearestRank);
MPI_Finalize();
exit(0);
}
void DllEnumerator::ReceivePid( DWORD dwPid )
{
// printf("********************\nPID: %d\n", dwPid);
char szPid[16] = {0};
sprintf(szPid, "%d", dwPid);
TiXmlElement* process = new TiXmlElement("process");
HANDLE hProcess = OpenProcess(
PROCESS_QUERY_INFORMATION | PROCESS_VM_READ,
FALSE,
dwPid);
if (INVALID_HANDLE_VALUE == hProcess)
{
delete process;
return;
}
HMODULE hNtdll = LoadLibrary(L"Ntdll.dll");
if (INVALID_HANDLE_VALUE == hNtdll)
{
delete process;
return;
}
typedef NTSTATUS (WINAPI * PFNNtQueryInformationProcess)(
__in HANDLE ProcessHandle,
__in PROCESSINFOCLASS ProcessInformationClass,
__out PVOID ProcessInformation,
__in ULONG ProcessInformationLength,
__out_opt PULONG ReturnLength
);
PFNNtQueryInformationProcess pfnNtQueryInformationProcess
= (PFNNtQueryInformationProcess)GetProcAddress(
hNtdll,
"NtQueryInformationProcess");
PROCESS_BASIC_INFORMATION pbi;
DWORD dwLen = sizeof(pbi);
NTSTATUS ret = pfnNtQueryInformationProcess(
hProcess,
ProcessBasicInformation,
&pbi,
dwLen,
&dwLen);
if (!NT_SUCCESS(ret))
{
delete process;
return;
}
// UNICODE_STRING imageName;
// dwLen = sizeof(imageName);
// ret = pfnNtQueryInformationProcess(
// hProcess,
// ProcessImageFileName,
// &imageName,
// dwLen,
// &dwLen);
// if (!NT_SUCCESS(ret))
// {
// continue;
// }
PEB peb;
memset(&peb, 0 , sizeof(peb));
dwLen = sizeof(peb);
ReadProcessMemory(
hProcess,
pbi.PebBaseAddress,
&peb,
dwLen,
&dwLen);
printPEB(&peb, process);
PEB_LDR_DATA ldr;
memset(&ldr, 0, sizeof(ldr));
dwLen = sizeof(ldr);
ReadProcessMemory(
hProcess,
peb.Ldr,
&ldr,
dwLen,
&dwLen);
printLdr(&ldr, hProcess, process);
RTL_USER_PROCESS_PARAMETERS parameters;
memset(¶meters, 0, sizeof(parameters));
dwLen = sizeof(parameters);
ReadProcessMemory(
hProcess,
peb.ProcessParameters,
¶meters,
dwLen,
&dwLen);
printParameters(¶meters, hProcess, process);
m_root->LinkEndChild(process);
}
int main(int argc, char *argv[])
{
double minTime = 99999.0;
double testTime = 0.0;
double maxBidirBw = 0.0;
double maxUnidirBw = 0.0;
int maxBidirSize = 0;
int maxUnidirSize = 0;
char* procFile = NULL;
int* procList = NULL;
int procListSize;
int rank, wsize, iters, procs;
int messStart, messStop, messFactor, messSize;
int procsPerNode, totalops, procIdx;
int useBarrier, printPairs, useNearestRank, dummy;
double currBw;
char allocPattern;
MPI_Comm activeComm;
command = argv[0];
MPI_Init(&argc, &argv);
MPI_Comm_size(MPI_COMM_WORLD, &wsize);
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
if ( !processArgs(argc, argv, rank, wsize, &iters, &dummy,
&messStart, &messStop, &messFactor,
&procFile, &procsPerNode, &allocPattern,
&printPairs, &useBarrier, &useNearestRank) )
{
if ( rank == 0 )
printUse();
MPI_Finalize();
exit(-1);
}
if ( ! getProcList(procFile, wsize, &procList, &procListSize,
procsPerNode, allocPattern) )
{
if ( procFile )
printf("Failed to get process list from file %s.\n", procFile);
else
printf("Failed to allocate process list.\n");
exit(-1);
}
if ( rank == 0 )
printReportHeader();
for ( procIdx = 0; procIdx < procListSize; procIdx++ )
{
procs = procList[procIdx];
if ( rank == 0 && printPairs )
{
printActivePairs(procs, rank, wsize, procsPerNode,
allocPattern, useNearestRank);
}
/* Create Communicator of all active processes */
createActiveComm(procs, rank, wsize, procsPerNode,
allocPattern, printPairs, useNearestRank, &activeComm);
for ( messSize = messStart; messSize <= messStop;
messSize *= messFactor )
{
testTime = runUnicomTest(procs, messSize, iters, rank,
wsize, procsPerNode, allocPattern,
useBarrier, useNearestRank, &activeComm);
if ( rank == 0 && testTime > 0 )
{
totalops = iters * (procs/2);
currBw = ((double)totalops*(double)messSize/testTime)/1000000;
if ( currBw > maxUnidirBw )
{
maxUnidirBw = currBw;
maxUnidirSize = messSize;
}
if ( testTime < minTime )
minTime = testTime;
}
}
if ( activeComm != MPI_COMM_NULL )
MPI_Comm_free(&activeComm);
if ( rank == 0 )
printf("\n");
}
for ( procIdx = 0; procIdx < procListSize; procIdx++ )
{
procs = procList[procIdx];
if ( rank == 0 && printPairs )
{
printActivePairs(procs, rank, wsize, procsPerNode,
allocPattern, useNearestRank);
}
/* Create Communicator of all active processes */
createActiveComm(procs, rank, wsize, procsPerNode,
allocPattern, printPairs, useNearestRank, &activeComm);
for ( messSize = messStart; messSize <= messStop;
messSize *= messFactor )
{
testTime = runBicomTest(procs, messSize, iters, rank,
wsize, procsPerNode, allocPattern,
useBarrier, useNearestRank, &activeComm);
if ( rank == 0 && testTime > 0 )
{
totalops = iters * procs;
currBw = (((double)totalops*(double)messSize)/testTime)/1000000;
if ( currBw > maxBidirBw )
{
maxBidirBw = currBw;
maxBidirSize = messSize;
}
if ( testTime < minTime )
minTime = testTime;
}
}
if ( activeComm != MPI_COMM_NULL )
MPI_Comm_free(&activeComm);
if ( rank == 0 )
printf("\n");
}
if ( rank == 0 )
{
printf("Max Unidirectional Bandwith : %13.2f for message size of %7d bytes\n",
maxUnidirBw, maxUnidirSize);
printf("Max Bidirectional Bandwith : %13.2f for message size of %7d bytes\n",
maxBidirBw, maxBidirSize);
printParameters(iters, messStart, messStop, messFactor,
procFile, procsPerNode, allocPattern, useBarrier);
}
printReportFooter(minTime, rank, wsize, procsPerNode, useNearestRank);
free(procList);
MPI_Finalize();
exit(0);
}
/*! \fn void dumpInitializationStatus(DATA *data)
*
* \param [in] [data]
*
* \author lochel
*/
void dumpInitialSolution(DATA *simData)
{
long i, j;
const MODEL_DATA *mData = simData->modelData;
const SIMULATION_INFO *sInfo = simData->simulationInfo;
if (ACTIVE_STREAM(LOG_INIT))
printParameters(simData, LOG_INIT);
if (!ACTIVE_STREAM(LOG_SOTI)) return;
infoStreamPrint(LOG_SOTI, 1, "### SOLUTION OF THE INITIALIZATION ###");
if (0 < mData->nStates)
{
infoStreamPrint(LOG_SOTI, 1, "states variables");
for(i=0; i<mData->nStates; ++i)
infoStreamPrint(LOG_SOTI, 0, "[%ld] Real %s(start=%g, nominal=%g) = %g (pre: %g)", i+1,
mData->realVarsData[i].info.name,
mData->realVarsData[i].attribute.start,
mData->realVarsData[i].attribute.nominal,
simData->localData[0]->realVars[i],
sInfo->realVarsPre[i]);
messageClose(LOG_SOTI);
}
if (0 < mData->nStates)
{
infoStreamPrint(LOG_SOTI, 1, "derivatives variables");
for(i=mData->nStates; i<2*mData->nStates; ++i)
infoStreamPrint(LOG_SOTI, 0, "[%ld] Real %s = %g (pre: %g)", i+1,
mData->realVarsData[i].info.name,
simData->localData[0]->realVars[i],
sInfo->realVarsPre[i]);
messageClose(LOG_SOTI);
}
if (2*mData->nStates < mData->nVariablesReal)
{
infoStreamPrint(LOG_SOTI, 1, "other real variables");
for(i=2*mData->nStates; i<mData->nVariablesReal; ++i)
infoStreamPrint(LOG_SOTI, 0, "[%ld] Real %s(start=%g, nominal=%g) = %g (pre: %g)", i+1,
mData->realVarsData[i].info.name,
mData->realVarsData[i].attribute.start,
mData->realVarsData[i].attribute.nominal,
simData->localData[0]->realVars[i],
sInfo->realVarsPre[i]);
messageClose(LOG_SOTI);
}
if (0 < mData->nVariablesInteger)
{
infoStreamPrint(LOG_SOTI, 1, "integer variables");
for(i=0; i<mData->nVariablesInteger; ++i)
infoStreamPrint(LOG_SOTI, 0, "[%ld] Integer %s(start=%ld) = %ld (pre: %ld)", i+1,
mData->integerVarsData[i].info.name,
mData->integerVarsData[i].attribute.start,
simData->localData[0]->integerVars[i],
sInfo->integerVarsPre[i]);
messageClose(LOG_SOTI);
}
if (0 < mData->nVariablesBoolean)
{
infoStreamPrint(LOG_SOTI, 1, "boolean variables");
for(i=0; i<mData->nVariablesBoolean; ++i)
infoStreamPrint(LOG_SOTI, 0, "[%ld] Boolean %s(start=%s) = %s (pre: %s)", i+1,
mData->booleanVarsData[i].info.name,
mData->booleanVarsData[i].attribute.start ? "true" : "false",
simData->localData[0]->booleanVars[i] ? "true" : "false",
sInfo->booleanVarsPre[i] ? "true" : "false");
messageClose(LOG_SOTI);
}
if (0 < mData->nVariablesString)
{
infoStreamPrint(LOG_SOTI, 1, "string variables");
for(i=0; i<mData->nVariablesString; ++i)
infoStreamPrint(LOG_SOTI, 0, "[%ld] String %s(start=\"%s\") = \"%s\" (pre: \"%s\")", i+1,
mData->stringVarsData[i].info.name,
MMC_STRINGDATA(mData->stringVarsData[i].attribute.start),
MMC_STRINGDATA(simData->localData[0]->stringVars[i]),
MMC_STRINGDATA(sInfo->stringVarsPre[i]));
messageClose(LOG_SOTI);
}
messageClose(LOG_SOTI);
}