// Load the tile into the texture image and draw it within given coordinates
static void imodvDrawTImage(Ipoint *p1, Ipoint *p2, Ipoint *p3, Ipoint *p4,
Ipoint *clamp,
unsigned char *data,
int width, int height)
{
float xclamp, yclamp, clamp0 = 1. / sTexImageSize;
double wallStart = wallTime();
xclamp = clamp->x;
yclamp = clamp->y;
glTexSubImage2D(GL_TEXTURE_2D, 0, 0, 0, width, height, GL_BGRA, GL_UNSIGNED_BYTE, data);
/*GLenum error = glGetError();
if (error)
imodPrintStderr("GL error on glTexSubImage2D: %s\n", gluErrorString(error)); */
sWallLoad += wallTime() - wallStart;
wallStart = wallTime();
glBegin(GL_QUADS);
glTexCoord2f(clamp0, clamp0); glVertex3fv((GLfloat *)p1);
glTexCoord2f(xclamp, clamp0); glVertex3fv((GLfloat *)p2);
glTexCoord2f(xclamp, yclamp); glVertex3fv((GLfloat *)p3);
glTexCoord2f(clamp0, yclamp); glVertex3fv((GLfloat *)p4);
glEnd();
glFlush();
sWallDraw += wallTime() - wallStart;
}
double Time::stop()
{
if (isRunning_) {
totalTime_ += ( wallTime() - startTime_ );
isRunning_ = false;
startTime_ = 0;
}
return totalTime_;
}
static void sharp_execute_job_mpi (sharp_job *job, MPI_Comm comm)
{
int ntasks;
MPI_Comm_size(comm, &ntasks);
if (ntasks==1) /* fall back to scalar implementation */
{ sharp_execute_job (job); return; }
MPI_Barrier(comm);
double timer=wallTime();
job->opcnt=0;
sharp_mpi_info minfo;
sharp_make_mpi_info(comm, job, &minfo);
if (minfo.npairtotal>minfo.ntasks*300)
{
int nsub=(minfo.npairtotal+minfo.ntasks*200-1)/(minfo.ntasks*200);
for (int isub=0; isub<nsub; ++isub)
{
sharp_job ljob=*job;
// When creating a_lm, every sub-job produces a complete set of
// coefficients; they need to be added up.
if ((isub>0)&&(job->type==SHARP_MAP2ALM)) ljob.flags|=SHARP_ADD;
sharp_geom_info lginfo;
lginfo.pair=RALLOC(sharp_ringpair,(job->ginfo->npairs/nsub)+1);
lginfo.npairs=0;
lginfo.nphmax = job->ginfo->nphmax;
while (lginfo.npairs*nsub+isub<job->ginfo->npairs)
{
lginfo.pair[lginfo.npairs]=job->ginfo->pair[lginfo.npairs*nsub+isub];
++lginfo.npairs;
}
ljob.ginfo=&lginfo;
sharp_execute_job_mpi (&ljob,comm);
job->opcnt+=ljob.opcnt;
DEALLOC(lginfo.pair);
}
}
else
{
int lmax = job->ainfo->lmax;
job->norm_l = sharp_Ylmgen_get_norm (lmax, job->spin);
/* clear output arrays if requested */
init_output (job);
alloc_phase_mpi (job,job->ainfo->nm,job->ginfo->npairs,minfo.mmax+1,
minfo.npairtotal);
double *cth = RALLOC(double,minfo.npairtotal),
*sth = RALLOC(double,minfo.npairtotal);
int *mlim = RALLOC(int,minfo.npairtotal);
for (int i=0; i<minfo.npairtotal; ++i)
{
cth[i] = cos(minfo.theta[i]);
sth[i] = sin(minfo.theta[i]);
mlim[i] = sharp_get_mlim(lmax, job->spin, sth[i], cth[i]);
}
/* map->phase where necessary */
map2phase (job, minfo.mmax, 0, job->ginfo->npairs);
map2alm_comm (job, &minfo);
#pragma omp parallel if ((job->flags&SHARP_NO_OPENMP)==0)
{
sharp_job ljob = *job;
sharp_Ylmgen_C generator;
sharp_Ylmgen_init (&generator,lmax,minfo.mmax,ljob.spin);
alloc_almtmp(&ljob,lmax);
#pragma omp for schedule(dynamic,1)
for (int mi=0; mi<job->ainfo->nm; ++mi)
{
/* alm->alm_tmp where necessary */
alm2almtmp (&ljob, lmax, mi);
/* inner conversion loop */
inner_loop (&ljob, minfo.ispair, cth, sth, 0, minfo.npairtotal,
&generator, mi, mlim);
/* alm_tmp->alm where necessary */
almtmp2alm (&ljob, lmax, mi);
}
sharp_Ylmgen_destroy(&generator);
dealloc_almtmp(&ljob);
#pragma omp critical
job->opcnt+=ljob.opcnt;
} /* end of parallel region */
alm2map_comm (job, &minfo);
/* phase->map where necessary */
phase2map (job, minfo.mmax, 0, job->ginfo->npairs);
DEALLOC(mlim);
DEALLOC(cth);
DEALLOC(sth);
DEALLOC(job->norm_l);
dealloc_phase (job);
}
sharp_destroy_mpi_info(&minfo);
job->time=wallTime()-timer;
}
int main(int argc, char** argv)
{
bookType* bt;
int bookCnt = 0, i;
char nullString[]="\0";
char inputFile[FILE_LEN] = {0};
char* inputFolder = nullString;
char* timeFile = nullString;
char* outputFile = nullString;
double* simMat;
int nprocs, myrank, error;
double timeStart, timeMid, timeEnd, timeElapsed;
double timeParaElapsed, timeSerialElapsed, timeParaEnd, timeParaStart;
int initSecond;
int ntag = 100;
MPI_Status status;
MPI_Init(&argc, &argv);
struct timeval init_tv;
gettimeofday(&init_tv, (struct timezone*)0);
initSecond = init_tv.tv_sec;
// record the start time
wallTime( &timeStart, &initSecond);
//////////////////////////
// Argument process
commandLineFlagType flag[] =
{
{"inputFolder", _string, &inputFolder },
{"timeFile", _string, &timeFile },
{"outputFile", _string, &outputFile },
};
// set the number of processes
int numFlags = sizeof ( flag ) / sizeof ( commandLineFlagType );
usageErrorType parseErrorCode = parseArgs ( argc, argv, numFlags, flag ) ;
if ( parseErrorCode == argError || parseErrorCode == parseError )
{
usage( argv[0] );
return ( 1 );
}
// make sure the "inputFolder" contains a "/" in the end
if (inputFolder[strlen(inputFolder) - 1] != '/')
{
char* tmpIF = (char*) malloc(strlen(inputFolder)+2);
strcpy(tmpIF, inputFolder);
strcat(tmpIF, "/");
inputFolder = tmpIF;
}
error = MPI_Comm_size(MPI_COMM_WORLD, &nprocs);
error = MPI_Comm_rank(MPI_COMM_WORLD, &myrank);
wallTime( &timeParaStart, &initSecond);
// word load shared among nprocs nodes, all nodes work parallelly
struct dirent *dir;
char *filename;
DIR* d;
d = opendir(inputFolder);
if (d)
{
while ((dir = readdir(d)) != NULL)
{
if (dir->d_type == DT_REG)
{
bookCnt ++;
i = bookCnt % nprocs;
if (i == myrank)
{
strcpy(inputFile, dir->d_name);
bt = (bookType*) malloc(sizeof(bookType));
memset(bt, 0, sizeof(bookType));
bt->fileName = (char*) malloc(strlen(inputFolder) + strlen(inputFile) + 1);
strcpy(bt->fileName, inputFolder);
strcat(bt->fileName, inputFile);
// 2. fill the structs with the desired content (P)
CreateFileObjects(bt);
// 3. sort the word list according to the word frequency (P)
SortWordForBook(bt);
// DebugPrintList(bt->wordList);
RemoveStopWords(bt);
char wordFile[FILE_LEN] = {0};
sprintf(wordFile, "words/%s.dat", inputFile);
DebugPrintList(bt->wordList, wordFile);
}
// MPI_Send(inputFile, FILE_LEN, MPI_CHAR, i, ntag, MPI_COMM_WORLD);
// printf("Node %d sent %s to %d\n", myrank, inputFile, i);
}
}
}
else
{
perror("Folder read error");
return ERROR_READ;
}
closedir(d);
wallTime( &timeParaEnd, &initSecond);
printf("rank=%d np=%d\n", myrank, nprocs);
// tell node 0 that the work is done
if (myrank != 0)
{
MPI_Send(&myrank, 1, MPI_INT, 0, ntag, MPI_COMM_WORLD);
}
if (myrank == 0)
{
int numSum = 0;
int num = 0;
// wait until all other nodes report
for (i = 1; i < nprocs; i++)
{
MPI_Recv(&num, 1, MPI_INT, i, ntag, MPI_COMM_WORLD, &status);
numSum += num;
}
printf("numSum=%d nprocs=%d\n", numSum, nprocs);
if (numSum != (nprocs - 1)*(nprocs)/2)
{
printf("Some node didn't report\n");
return 1;
}
wallTime( &timeEnd, &initSecond);
timeElapsed = timeEnd - timeStart;
timeParaElapsed = timeParaEnd - timeParaStart;
timeSerialElapsed = timeElapsed - timeParaElapsed;
printf("SaveTime(%d): %1f %1f %1f %1f \n", nprocs, timeStart, timeParaStart, timeParaEnd, timeEnd);
pSaveRunStat(timeStart, timeEnd, timeElapsed, timeSerialElapsed, timeFile);
}
// all nodes receive file name
// MPI_Recv(inputFile, FILE_LEN, MPI_CHAR, 0, ntag, MPI_COMM_WORLD, &status);
// printf("Node %d recv %s (%d)\n", myrank, inputFile, i);
MPI_Finalize () ;
return 0;
}
void Time::start(bool reset_in)
{
isRunning_ = true;
if (reset_in) totalTime_ = 0;
startTime_ = wallTime();
}
double Time::totalElapsedTime(bool readCurrentTime) const
{
if(readCurrentTime)
return wallTime() - startTime_ + totalTime_;
return totalTime_;
}
