void oskar_vis_block_clear(oskar_VisBlock* vis, int* status)
{
/* Check if safe to proceed. */
if (*status) return;
oskar_mem_clear_contents(vis->auto_correlations, status);
oskar_mem_clear_contents(vis->cross_correlations, status);
oskar_mem_clear_contents(vis->baseline_uu_metres, status);
oskar_mem_clear_contents(vis->baseline_vv_metres, status);
oskar_mem_clear_contents(vis->baseline_ww_metres, status);
}
static void set_up_device_data(oskar_BeamPattern* h, int* status)
{
int i, beam_type, max_src, max_size, auto_power, cross_power, raw_data;
if (*status) return;
/* Get local variables. */
max_src = h->max_chunk_size;
max_size = h->num_active_stations * max_src;
beam_type = h->prec | OSKAR_COMPLEX;
if (h->pol_mode == OSKAR_POL_MODE_FULL)
beam_type |= OSKAR_MATRIX;
raw_data = h->ixr_txt || h->ixr_fits ||
h->voltage_raw_txt || h->voltage_amp_txt || h->voltage_phase_txt ||
h->voltage_amp_fits || h->voltage_phase_fits;
auto_power = h->auto_power_fits || h->auto_power_txt;
cross_power = h->cross_power_raw_txt ||
h->cross_power_amp_fits || h->cross_power_phase_fits ||
h->cross_power_amp_txt || h->cross_power_phase_txt;
/* Expand the number of devices to the number of selected GPUs,
* if required. */
if (h->num_devices < h->num_gpus)
oskar_beam_pattern_set_num_devices(h, h->num_gpus);
for (i = 0; i < h->num_devices; ++i)
{
int dev_loc, i_stokes;
DeviceData* d = &h->d[i];
if (*status) break;
/* Select the device. */
if (i < h->num_gpus)
{
oskar_device_set(h->gpu_ids[i], status);
dev_loc = OSKAR_GPU;
}
else
{
dev_loc = OSKAR_CPU;
}
/* Device memory. */
d->previous_chunk_index = -1;
if (!d->tel)
{
d->jones_data = oskar_mem_create(beam_type, dev_loc, max_size,
status);
d->x = oskar_mem_create(h->prec, dev_loc, 1 + max_src, status);
d->y = oskar_mem_create(h->prec, dev_loc, 1 + max_src, status);
d->z = oskar_mem_create(h->prec, dev_loc, 1 + max_src, status);
d->tel = oskar_telescope_create_copy(h->tel, dev_loc, status);
d->work = oskar_station_work_create(h->prec, dev_loc, status);
}
/* Host memory. */
if (!d->jones_data_cpu[0] && raw_data)
{
d->jones_data_cpu[0] = oskar_mem_create(beam_type, OSKAR_CPU,
max_size, status);
d->jones_data_cpu[1] = oskar_mem_create(beam_type, OSKAR_CPU,
max_size, status);
}
/* Auto-correlation beam output arrays. */
for (i_stokes = 0; i_stokes < 4; ++i_stokes)
{
if (!h->stokes[i_stokes]) continue;
if (!d->auto_power[i_stokes] && auto_power)
{
/* Device memory. */
d->auto_power[i_stokes] = oskar_mem_create(beam_type, dev_loc,
max_size, status);
/* Host memory. */
d->auto_power_cpu[i_stokes][0] = oskar_mem_create(
beam_type, OSKAR_CPU, max_size, status);
d->auto_power_cpu[i_stokes][1] = oskar_mem_create(
beam_type, OSKAR_CPU, max_size, status);
if (h->average_single_axis == 'T')
d->auto_power_time_avg[i_stokes] = oskar_mem_create(
beam_type, OSKAR_CPU, max_size, status);
if (h->average_single_axis == 'C')
d->auto_power_channel_avg[i_stokes] = oskar_mem_create(
beam_type, OSKAR_CPU, max_size, status);
if (h->average_time_and_channel)
d->auto_power_channel_and_time_avg[i_stokes] =
oskar_mem_create(beam_type, OSKAR_CPU,
max_size, status);
}
/* Cross-correlation beam output arrays. */
if (!d->cross_power[i_stokes] && cross_power)
{
if (h->num_active_stations < 2)
{
oskar_log_error(h->log, "Cannot create cross-power beam "
"using less than two active stations.");
*status = OSKAR_ERR_INVALID_ARGUMENT;
break;
}
/* Device memory. */
d->cross_power[i_stokes] = oskar_mem_create(
beam_type, dev_loc, max_src, status);
/* Host memory. */
d->cross_power_cpu[i_stokes][0] = oskar_mem_create(
beam_type, OSKAR_CPU, max_src, status);
d->cross_power_cpu[i_stokes][1] = oskar_mem_create(
beam_type, OSKAR_CPU, max_src, status);
if (h->average_single_axis == 'T')
d->cross_power_time_avg[i_stokes] = oskar_mem_create(
beam_type, OSKAR_CPU, max_src, status);
if (h->average_single_axis == 'C')
d->cross_power_channel_avg[i_stokes] = oskar_mem_create(
beam_type, OSKAR_CPU, max_src, status);
if (h->average_time_and_channel)
d->cross_power_channel_and_time_avg[i_stokes] =
oskar_mem_create(beam_type, OSKAR_CPU,
max_src, status);
}
if (d->auto_power[i_stokes])
oskar_mem_clear_contents(d->auto_power[i_stokes], status);
if (d->cross_power[i_stokes])
oskar_mem_clear_contents(d->cross_power[i_stokes], status);
}
/* Timers. */
if (!d->tmr_compute)
d->tmr_compute = oskar_timer_create(OSKAR_TIMER_NATIVE);
}
}
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
}
void runTest(int prec1, int prec2, int loc1, int loc2, int matrix)
{
int status = 0, type;
oskar_Mem *beam1, *beam2;
oskar_Timer *timer1, *timer2;
double time1, time2;
// 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.
type = prec1 | OSKAR_COMPLEX;
if (matrix) type |= OSKAR_MATRIX;
beam1 = oskar_mem_create(type, loc1, num_sources, &status);
oskar_mem_clear_contents(beam1, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
createTestData(prec1, loc1, matrix);
oskar_timer_start(timer1);
oskar_evaluate_cross_power(num_sources, num_stations,
jones, 0, beam1, &status);
time1 = oskar_timer_elapsed(timer1);
destroyTestData();
ASSERT_EQ(0, status) << oskar_get_error_string(status);
// Run second part.
type = prec2 | OSKAR_COMPLEX;
if (matrix) type |= OSKAR_MATRIX;
beam2 = oskar_mem_create(type, loc2, num_sources, &status);
oskar_mem_clear_contents(beam2, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
createTestData(prec2, loc2, matrix);
oskar_timer_start(timer2);
oskar_evaluate_cross_power(num_sources, num_stations,
jones, 0, beam2, &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(beam1, beam2);
// Free memory.
oskar_mem_free(beam1, &status);
oskar_mem_free(beam2, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
// Record properties for test.
RecordProperty("JonesType", matrix ? "Matrix" : "Scalar");
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.\n", matrix ? "Matrix" : "Scalar");
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
}