int benchmark(int num_stations, int num_sources, int type,
int jones_type, int loc, int use_extended, int use_time_ave, int niter,
std::vector<double>& times)
{
int status = 0;
oskar_Timer* timer;
timer = oskar_timer_create(loc == OSKAR_GPU ?
OSKAR_TIMER_CUDA : OSKAR_TIMER_OMP);
// Set up a test sky model, telescope model and Jones matrices.
oskar_Telescope* tel = oskar_telescope_create(type, loc,
num_stations, &status);
oskar_Sky* sky = oskar_sky_create(type, loc, num_sources, &status);
oskar_Jones* J = oskar_jones_create(jones_type, loc, num_stations,
num_sources, &status);
oskar_telescope_set_channel_bandwidth(tel, 1e6);
oskar_telescope_set_time_average(tel, (double) use_time_ave);
oskar_sky_set_use_extended(sky, use_extended);
// Memory for visibility coordinates and output visibility slice.
oskar_Mem *vis, *u, *v, *w;
vis = oskar_mem_create(jones_type, loc, oskar_telescope_num_baselines(tel),
&status);
u = oskar_mem_create(type, loc, num_stations, &status);
v = oskar_mem_create(type, loc, num_stations, &status);
w = oskar_mem_create(type, loc, num_stations, &status);
// Run benchmark.
times.resize(niter);
for (int i = 0; i < niter; ++i)
{
oskar_timer_start(timer);
oskar_cross_correlate(vis, oskar_sky_num_sources(sky), J, sky, tel, u, v, w,
0.0, 100e6, &status);
times[i] = oskar_timer_elapsed(timer);
}
// Free memory.
oskar_mem_free(u, &status);
oskar_mem_free(v, &status);
oskar_mem_free(w, &status);
oskar_mem_free(vis, &status);
oskar_jones_free(J, &status);
oskar_telescope_free(tel, &status);
oskar_sky_free(sky, &status);
oskar_timer_free(timer);
return status;
}
static void sim_baselines(oskar_Simulator* h, DeviceData* d, oskar_Sky* sky,
int channel_index_block, int time_index_block,
int time_index_simulation, int* status)
{
int num_baselines, num_stations, num_src, num_times_block, num_channels;
double dt_dump_days, t_start, t_dump, gast, frequency, ra0, dec0;
const oskar_Mem *x, *y, *z;
oskar_Mem* alias = 0;
/* Get dimensions. */
num_baselines = oskar_telescope_num_baselines(d->tel);
num_stations = oskar_telescope_num_stations(d->tel);
num_src = oskar_sky_num_sources(sky);
num_times_block = oskar_vis_block_num_times(d->vis_block);
num_channels = oskar_vis_block_num_channels(d->vis_block);
/* Return if there are no sources in the chunk,
* or if block time index requested is outside the valid range. */
if (num_src == 0 || time_index_block >= num_times_block) return;
/* Get the time and frequency of the visibility slice being simulated. */
dt_dump_days = h->time_inc_sec / 86400.0;
t_start = h->time_start_mjd_utc;
t_dump = t_start + dt_dump_days * (time_index_simulation + 0.5);
gast = oskar_convert_mjd_to_gast_fast(t_dump);
frequency = h->freq_start_hz + channel_index_block * h->freq_inc_hz;
/* Scale source fluxes with spectral index and rotation measure. */
oskar_sky_scale_flux_with_frequency(sky, frequency, status);
/* Evaluate station u,v,w coordinates. */
ra0 = oskar_telescope_phase_centre_ra_rad(d->tel);
dec0 = oskar_telescope_phase_centre_dec_rad(d->tel);
x = oskar_telescope_station_true_x_offset_ecef_metres_const(d->tel);
y = oskar_telescope_station_true_y_offset_ecef_metres_const(d->tel);
z = oskar_telescope_station_true_z_offset_ecef_metres_const(d->tel);
oskar_convert_ecef_to_station_uvw(num_stations, x, y, z, ra0, dec0, gast,
d->u, d->v, d->w, status);
/* Set dimensions of Jones matrices. */
if (d->R)
oskar_jones_set_size(d->R, num_stations, num_src, status);
if (d->Z)
oskar_jones_set_size(d->Z, num_stations, num_src, status);
oskar_jones_set_size(d->J, num_stations, num_src, status);
oskar_jones_set_size(d->E, num_stations, num_src, status);
oskar_jones_set_size(d->K, num_stations, num_src, status);
/* Evaluate station beam (Jones E: may be matrix). */
oskar_timer_resume(d->tmr_E);
oskar_evaluate_jones_E(d->E, num_src, OSKAR_RELATIVE_DIRECTIONS,
oskar_sky_l(sky), oskar_sky_m(sky), oskar_sky_n(sky), d->tel,
gast, frequency, d->station_work, time_index_simulation, status);
oskar_timer_pause(d->tmr_E);
#if 0
/* Evaluate ionospheric phase (Jones Z: scalar) and join with Jones E.
* NOTE this is currently only a CPU implementation. */
if (d->Z)
{
oskar_evaluate_jones_Z(d->Z, num_src, sky, d->tel,
&settings->ionosphere, gast, frequency, &(d->workJonesZ),
status);
oskar_timer_resume(d->tmr_join);
oskar_jones_join(d->E, d->Z, d->E, status);
oskar_timer_pause(d->tmr_join);
}
#endif
/* Evaluate parallactic angle (Jones R: matrix), and join with Jones Z*E.
* TODO Move this into station beam evaluation instead. */
if (d->R)
{
oskar_timer_resume(d->tmr_E);
oskar_evaluate_jones_R(d->R, num_src, oskar_sky_ra_rad_const(sky),
oskar_sky_dec_rad_const(sky), d->tel, gast, status);
oskar_timer_pause(d->tmr_E);
oskar_timer_resume(d->tmr_join);
oskar_jones_join(d->R, d->E, d->R, status);
oskar_timer_pause(d->tmr_join);
}
/* Evaluate interferometer phase (Jones K: scalar). */
oskar_timer_resume(d->tmr_K);
oskar_evaluate_jones_K(d->K, num_src, oskar_sky_l_const(sky),
oskar_sky_m_const(sky), oskar_sky_n_const(sky), d->u, d->v, d->w,
frequency, oskar_sky_I_const(sky),
h->source_min_jy, h->source_max_jy, status);
oskar_timer_pause(d->tmr_K);
/* Join Jones K with Jones Z*E. */
oskar_timer_resume(d->tmr_join);
oskar_jones_join(d->J, d->K, d->R ? d->R : d->E, status);
oskar_timer_pause(d->tmr_join);
/* Create alias for auto/cross-correlations. */
oskar_timer_resume(d->tmr_correlate);
alias = oskar_mem_create_alias(0, 0, 0, status);
/* Auto-correlate for this time and channel. */
if (oskar_vis_block_has_auto_correlations(d->vis_block))
{
oskar_mem_set_alias(alias,
oskar_vis_block_auto_correlations(d->vis_block),
num_stations *
(num_channels * time_index_block + channel_index_block),
num_stations, status);
oskar_auto_correlate(alias, num_src, d->J, sky, status);
}
/* Cross-correlate for this time and channel. */
if (oskar_vis_block_has_cross_correlations(d->vis_block))
{
oskar_mem_set_alias(alias,
oskar_vis_block_cross_correlations(d->vis_block),
num_baselines *
(num_channels * time_index_block + channel_index_block),
num_baselines, status);
oskar_cross_correlate(alias, num_src, d->J, sky, d->tel,
d->u, d->v, d->w, gast, frequency, status);
}
/* Free alias for auto/cross-correlations. */
oskar_mem_free(alias, status);
oskar_timer_pause(d->tmr_correlate);
}
void runTest(int prec1, int prec2, int loc1, int loc2, int matrix,
int extended, double time_average)
{
int num_baselines, status = 0, type;
oskar_Mem *vis1, *vis2;
oskar_Timer *timer1, *timer2;
double time1, time2, frequency = 100e6;
// Create the timers.
timer1 = oskar_timer_create(loc1 == OSKAR_GPU ?
OSKAR_TIMER_CUDA : OSKAR_TIMER_NATIVE);
timer2 = oskar_timer_create(loc2 == OSKAR_GPU ?
OSKAR_TIMER_CUDA : OSKAR_TIMER_NATIVE);
// Run first part.
createTestData(prec1, loc1, matrix);
num_baselines = oskar_telescope_num_baselines(tel);
type = prec1 | OSKAR_COMPLEX;
if (matrix) type |= OSKAR_MATRIX;
vis1 = oskar_mem_create(type, loc1, num_baselines, &status);
oskar_mem_clear_contents(vis1, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
oskar_sky_set_use_extended(sky, extended);
oskar_telescope_set_channel_bandwidth(tel, bandwidth);
oskar_telescope_set_time_average(tel, time_average);
oskar_timer_start(timer1);
oskar_cross_correlate(vis1, oskar_sky_num_sources(sky), jones, sky,
tel, u_, v_, w_, 1.0, frequency, &status);
time1 = oskar_timer_elapsed(timer1);
destroyTestData();
ASSERT_EQ(0, status) << oskar_get_error_string(status);
// Run second part.
createTestData(prec2, loc2, matrix);
num_baselines = oskar_telescope_num_baselines(tel);
type = prec2 | OSKAR_COMPLEX;
if (matrix) type |= OSKAR_MATRIX;
vis2 = oskar_mem_create(type, loc2, num_baselines, &status);
oskar_mem_clear_contents(vis2, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
oskar_sky_set_use_extended(sky, extended);
oskar_telescope_set_channel_bandwidth(tel, bandwidth);
oskar_telescope_set_time_average(tel, time_average);
oskar_timer_start(timer2);
oskar_cross_correlate(vis2, oskar_sky_num_sources(sky), jones, sky,
tel, u_, v_, w_, 1.0, frequency, &status);
time2 = oskar_timer_elapsed(timer2);
destroyTestData();
ASSERT_EQ(0, status) << oskar_get_error_string(status);
// Destroy the timers.
oskar_timer_free(timer1);
oskar_timer_free(timer2);
// Compare results.
check_values(vis1, vis2);
// Free memory.
oskar_mem_free(vis1, &status);
oskar_mem_free(vis2, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
// Record properties for test.
RecordProperty("SourceType", extended ? "Gaussian" : "Point");
RecordProperty("JonesType", matrix ? "Matrix" : "Scalar");
RecordProperty("TimeSmearing", time_average == 0.0 ? "off" : "on");
RecordProperty("Prec1", prec1 == OSKAR_SINGLE ? "Single" : "Double");
RecordProperty("Loc1", loc1 == OSKAR_CPU ? "CPU" : "GPU");
RecordProperty("Time1_ms", int(time1 * 1000));
RecordProperty("Prec2", prec2 == OSKAR_SINGLE ? "Single" : "Double");
RecordProperty("Loc2", loc2 == OSKAR_CPU ? "CPU" : "GPU");
RecordProperty("Time2_ms", int(time2 * 1000));
#ifdef ALLOW_PRINTING
// Print times.
printf(" > %s. %s sources. Time smearing %s.\n",
matrix ? "Matrix" : "Scalar",
extended ? "Gaussian" : "Point",
time_average == 0.0 ? "off" : "on");
printf(" %s precision %s: %.2f ms, %s precision %s: %.2f ms\n",
prec1 == OSKAR_SINGLE ? "Single" : "Double",
loc1 == OSKAR_CPU ? "CPU" : "GPU",
time1 * 1000.0,
prec2 == OSKAR_SINGLE ? "Single" : "Double",
loc2 == OSKAR_CPU ? "CPU" : "GPU",
time2 * 1000.0);
#endif
}
