double ptimerend (struct timing *t)
{
double local, time;
local = timerend (t);
MPI_Allreduce (&local, &time, 1, MPI_DOUBLE, MPI_MAX, MPI_COMM_WORLD);
return time;
}
int main (int argc, char *argv[])
{
int max_points_per_rank = 100;
int num_time_steps = 100;
REAL time_step = 0.001;
int n, i, rank, size, *ranks;
REAL *point[3], *velo[3];
struct timing t;
double dt[2], gt[2];
if (argc == 1)
{
printf ("SYNOPSIS: test_dynlb max_points_per_rank [num_time_steps = 100] [time_step = 0.001]\n");
return 0;
}
if (argc >= 2) max_points_per_rank = atoi(argv[1]);
if (argc >= 3) num_time_steps = atoi(argv[2]);
if (argc >= 4) time_step = atof(argv[3]);
MPI_Init (&argc, &argv);
MPI_Comm_rank (MPI_COMM_WORLD, &rank);
MPI_Comm_size (MPI_COMM_WORLD, &size);
srand(time(NULL) + rank);
n = rand() % max_points_per_rank + 1;
if (rank == 0) printf ("Generating random points in unit cube...\n");
timerstart (&t);
printf ("Generating %d points on rank %d.\n", n, rank);
ERRMEM (point[0] = malloc (n * sizeof (REAL)));
ERRMEM (point[1] = malloc (n * sizeof (REAL)));
ERRMEM (point[2] = malloc (n * sizeof (REAL)));
ERRMEM (velo[0] = malloc (n * sizeof (REAL)));
ERRMEM (velo[1] = malloc (n * sizeof (REAL)));
ERRMEM (velo[2] = malloc (n * sizeof (REAL)));
ERRMEM (ranks = malloc (n * sizeof (int)));
for (i = 0; i < n; i ++)
{
point[0][i] = DRAND();
point[1][i] = DRAND();
point[2][i] = DRAND();
velo[0][i] = DRAND()-0.5;
velo[1][i] = DRAND()-0.5;
velo[2][i] = DRAND()-0.5;
}
dt[0] = ptimerend (&t);
if (rank == 0) printf ("Generating points took %g sec.\n", dt[0]);
if (rank == 0) printf ("Timing %d simple morton based balancing steps...\n", num_time_steps);
for (i = 0, dt[0] = 0.0, dt[1] = 0.0; i < num_time_steps; i ++)
{
timerstart (&t);
dynlb_morton_balance (n, point, ranks);
dt[0] += timerend (&t);
if (rank == 0) printf ("."), fflush (stdout);
timerstart (&t);
unit_cube_step (0, n, point, velo, time_step);
dt[1] += timerend (&t);
}
MPI_Allreduce (dt, gt, 2, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
gt[0] /= (double)size * (double)num_time_steps;
gt[1] /= (double)size * (double)num_time_steps;
if (rank == 0) printf ("\nMORTON: avg. integration: %g sec. per step, avg. balancing: %g sec. per step; ratio: %g\n", gt[0]+gt[1], gt[1], gt[1]/(gt[0]+gt[1]));
if (rank == 0) printf ("Creating partitioning tree based balancer ...\n");
timerstart (&t);
struct dynlb *lb = dynlb_create (0, n, point, 0, 0.5, DYNLB_RCB_TREE);
dt[0] += ptimerend (&t);
if (rank == 0) printf ("Took %g sec.\nInitial imbalance %g\n", dt[0], lb->imbalance);
if (rank == 0) printf ("Timing %d partitioning tree based balancing steps...\n", num_time_steps);
for (i = 0, dt[0] = 0.0, dt[1] = 0.0; i < num_time_steps; i ++)
{
timerstart (&t);
dynlb_update (lb, n, point);
dt[0] += timerend (&t);
if (rank == 0) printf ("Step %d imbalance %g\n", i, lb->imbalance);
timerstart (&t);
unit_cube_step (0, n, point, velo, time_step);
dt[1] += timerend (&t);
}
MPI_Allreduce (dt, gt, 2, MPI_DOUBLE, MPI_SUM, MPI_COMM_WORLD);
gt[0] /= (double)size * (double)num_time_steps;
gt[1] /= (double)size * (double)num_time_steps;
if (rank == 0) printf ("TREE: avg. integration: %g sec. per step, avg. balancing: %g sec. per step; ratio: %g\n", gt[0]+gt[1], gt[1], gt[1]/(gt[0]+gt[1]));
free (point[0]);
free (point[1]);
free (point[2]);
free (velo[0]);
free (velo[1]);
free (velo[2]);
free (ranks);
MPI_Finalize ();
return 0;
}
int main(int argc, char *argv[])
{
int i, ret,rep=1;
LibRaw RawProcessor;
#ifdef OUT
#undef OUT
#endif
#define OUT RawProcessor.imgdata.params
#define S RawProcessor.imgdata.sizes
if(argc<2)
{
printf(
"postprocessing benchmark: LibRaw %s sample, %d cameras supported\n"
"Measures postprocessing speed with different options\n"
"Usage: %s [-a] [-H N] [-q N] [-h] [-m N] [-n N] [-s N] [-B x y w h] [-R N]\n"
"-a average image for white balance\n"
"-H <num> Highlight mode (0=clip, 1=unclip, 2=blend, 3+=rebuild)\n"
"-q <num> Set the interpolation quality\n"
"-h Half-size color image\n"
"-m <num> Apply a num-passes 3x3 median filter to R-G and B-G\n"
"-n <num> Set threshold for wavelet denoising\n"
"-s <num> Select one raw image from input file\n"
"-B <x y w h> Crop output image\n"
"-R <num> Number of repetitions\n"
"-c Do not use rawspeed\n"
,LibRaw::version(), LibRaw::cameraCount(),
argv[0]);
return 0;
}
char opm,opt,*cp,*sp;
int arg,c;
int shrink = 0;
argv[argc] = (char*)"";
for (arg=1; (((opm = argv[arg][0]) - 2) | 2) == '+'; )
{
char *optstr = argv[arg];
opt = argv[arg++][1];
if ((cp = strchr (sp=(char*)"HqmnsBR", opt))!=0)
for (i=0; i < "1111141"[cp-sp]-'0'; i++)
if (!isdigit(argv[arg+i][0]) && !optstr[2])
{
fprintf (stderr,"Non-numeric argument to \"-%c\"\n", opt);
return 1;
}
switch (opt)
{
case 'a':
OUT.use_auto_wb = 1;
break;
case 'H':
OUT.highlight = atoi(argv[arg++]);
break;
case 'q':
OUT.user_qual = atoi(argv[arg++]);
break;
case 'h':
OUT.half_size = 1;
OUT.four_color_rgb = 1;
shrink = 1;
break;
case 'm':
OUT.med_passes = atoi(argv[arg++]);
break;
case 'n':
OUT.threshold = (float)atof(argv[arg++]);
break;
case 's':
OUT.shot_select = abs(atoi(argv[arg++]));
break;
case 'B':
for(c=0; c<4;c++) OUT.cropbox[c] = atoi(argv[arg++]);
break;
case 'R':
rep = abs(atoi(argv[arg++]));
if(rep<1) rep = 1;
break;
case 'c':
OUT.use_rawspeed = 0;
break;
default:
fprintf (stderr,"Unknown option \"-%c\".\n", opt);
return 1;
}
}
for ( ; arg < argc; arg++)
{
printf("Processing file %s\n",argv[arg]);
timerstart();
if( (ret = RawProcessor.open_file(argv[arg])) != LIBRAW_SUCCESS)
{
fprintf(stderr,"Cannot open_file %s: %s\n",argv[arg],libraw_strerror(ret));
continue; // no recycle b/c open file will recycle itself
}
if( (ret = RawProcessor.unpack() ) != LIBRAW_SUCCESS)
{
fprintf(stderr,"Cannot unpack %s: %s\n",argv[arg],libraw_strerror(ret));
continue;
}
float qsec = timerend();
printf("\n%.1f msec for unpack\n",qsec);
float mpix,rmpix;
timerstart();
for(c=0; c < rep; c++)
{
if( (ret = RawProcessor.dcraw_process() ) != LIBRAW_SUCCESS)
{
fprintf(stderr,"Cannot postprocess %s: %s\n",argv[arg],libraw_strerror(ret));
break;
}
libraw_processed_image_t *p = RawProcessor.dcraw_make_mem_image();
if(p)
RawProcessor.dcraw_clear_mem(p);
RawProcessor.free_image();
}
float msec = timerend()/(float)rep;
if( (ret = RawProcessor.adjust_sizes_info_only() ) != LIBRAW_SUCCESS)
{
fprintf(stderr,"Cannot adjust sizes for %s: %s\n",argv[arg],libraw_strerror(ret));
break;
}
rmpix = (S.iwidth*S.iheight)/1000000.0f;
if(c==rep) // no failure
{
unsigned int crop[4];
for(int i=0;i<4;i++) crop[i] = (OUT.cropbox[i])>>shrink;
if(crop[0]+crop[2]>S.iwidth) crop[2] = S.iwidth-crop[0];
if(crop[1]+crop[3]>S.iheight) crop[3] = S.iheight-crop[1];
mpix = float(crop[2]*crop[3])/1000000.0f;
float mpixsec = mpix*1000.0f/msec;
printf(
"Performance: %.2f Mpix/sec\n"
"File: %s, Frame: %d %.1f total Mpix, %.1f msec\n"
"Params: WB=%s Highlight=%d Qual=%d HalfSize=%s Median=%d Wavelet=%.0f\n"
"Crop: %u-%u:%ux%u, active Mpix: %.2f, %.1f frames/sec\n",
mpixsec
,argv[arg],OUT.shot_select,rmpix,msec,
OUT.use_auto_wb?"auto":"default",OUT.highlight,OUT.user_qual,OUT.half_size?"YES":"No",
OUT.med_passes,OUT.threshold,
crop[0],crop[1],crop[2],crop[3],mpix,1000.0f/msec);
}
}
