void oskar_dftw(
int normalise,
int num_in,
double wavenumber,
const oskar_Mem* weights_in,
const oskar_Mem* x_in,
const oskar_Mem* y_in,
const oskar_Mem* z_in,
int offset_coord_out,
int num_out,
const oskar_Mem* x_out,
const oskar_Mem* y_out,
const oskar_Mem* z_out,
const oskar_Mem* data,
int offset_out,
oskar_Mem* output,
int* status)
{
if (*status) return;
const int location = oskar_mem_location(output);
const int type = oskar_mem_precision(output);
const int is_dbl = oskar_mem_is_double(output);
const int is_3d = (z_in != NULL && z_out != NULL);
const int is_data = (data != NULL);
const int is_matrix = oskar_mem_is_matrix(output);
if (!oskar_mem_is_complex(output) || !oskar_mem_is_complex(weights_in) ||
oskar_mem_is_matrix(weights_in))
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
if (oskar_mem_location(weights_in) != location ||
oskar_mem_location(x_in) != location ||
oskar_mem_location(y_in) != location ||
oskar_mem_location(x_out) != location ||
oskar_mem_location(y_out) != location)
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
if (oskar_mem_precision(weights_in) != type ||
oskar_mem_type(x_in) != type ||
oskar_mem_type(y_in) != type ||
oskar_mem_type(x_out) != type ||
oskar_mem_type(y_out) != type)
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if (is_data)
{
if (oskar_mem_location(data) != location)
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
if (!oskar_mem_is_complex(data) ||
oskar_mem_type(data) != oskar_mem_type(output) ||
oskar_mem_precision(data) != type)
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
}
if (is_3d)
{
if (oskar_mem_location(z_in) != location ||
oskar_mem_location(z_out) != location)
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
if (oskar_mem_type(z_in) != type || oskar_mem_type(z_out) != type)
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
}
oskar_mem_ensure(output, (size_t) offset_out + num_out, status);
if (*status) return;
if (location == OSKAR_CPU)
{
if (is_data)
{
if (is_matrix)
{
if (is_3d)
{
if (is_dbl)
dftw_m2m_3d_double(num_in, wavenumber,
oskar_mem_double2_const(weights_in, status),
oskar_mem_double_const(x_in, status),
oskar_mem_double_const(y_in, status),
oskar_mem_double_const(z_in, status),
offset_coord_out, num_out,
oskar_mem_double_const(x_out, status),
oskar_mem_double_const(y_out, status),
oskar_mem_double_const(z_out, status),
oskar_mem_double4c_const(data, status),
offset_out,
oskar_mem_double4c(output, status), 0);
else
dftw_m2m_3d_float(num_in, (float)wavenumber,
oskar_mem_float2_const(weights_in, status),
oskar_mem_float_const(x_in, status),
oskar_mem_float_const(y_in, status),
oskar_mem_float_const(z_in, status),
offset_coord_out, num_out,
oskar_mem_float_const(x_out, status),
oskar_mem_float_const(y_out, status),
oskar_mem_float_const(z_out, status),
oskar_mem_float4c_const(data, status),
offset_out,
oskar_mem_float4c(output, status), 0);
}
else
{
if (is_dbl)
dftw_m2m_2d_double(num_in, wavenumber,
oskar_mem_double2_const(weights_in, status),
oskar_mem_double_const(x_in, status),
oskar_mem_double_const(y_in, status), 0,
offset_coord_out, num_out,
oskar_mem_double_const(x_out, status),
oskar_mem_double_const(y_out, status), 0,
oskar_mem_double4c_const(data, status),
offset_out,
oskar_mem_double4c(output, status), 0);
else
dftw_m2m_2d_float(num_in, (float)wavenumber,
oskar_mem_float2_const(weights_in, status),
oskar_mem_float_const(x_in, status),
oskar_mem_float_const(y_in, status), 0,
offset_coord_out, num_out,
oskar_mem_float_const(x_out, status),
oskar_mem_float_const(y_out, status), 0,
oskar_mem_float4c_const(data, status),
offset_out,
oskar_mem_float4c(output, status), 0);
}
}
else
{
if (is_3d)
{
if (is_dbl)
dftw_c2c_3d_double(num_in, wavenumber,
oskar_mem_double2_const(weights_in, status),
oskar_mem_double_const(x_in, status),
oskar_mem_double_const(y_in, status),
oskar_mem_double_const(z_in, status),
offset_coord_out, num_out,
oskar_mem_double_const(x_out, status),
oskar_mem_double_const(y_out, status),
oskar_mem_double_const(z_out, status),
oskar_mem_double2_const(data, status),
offset_out,
oskar_mem_double2(output, status), 0);
else
dftw_c2c_3d_float(num_in, (float)wavenumber,
oskar_mem_float2_const(weights_in, status),
oskar_mem_float_const(x_in, status),
oskar_mem_float_const(y_in, status),
oskar_mem_float_const(z_in, status),
offset_coord_out, num_out,
oskar_mem_float_const(x_out, status),
oskar_mem_float_const(y_out, status),
oskar_mem_float_const(z_out, status),
oskar_mem_float2_const(data, status),
offset_out,
oskar_mem_float2(output, status), 0);
}
else
{
if (is_dbl)
dftw_c2c_2d_double(num_in, wavenumber,
oskar_mem_double2_const(weights_in, status),
oskar_mem_double_const(x_in, status),
oskar_mem_double_const(y_in, status), 0,
offset_coord_out, num_out,
oskar_mem_double_const(x_out, status),
oskar_mem_double_const(y_out, status), 0,
oskar_mem_double2_const(data, status),
offset_out,
oskar_mem_double2(output, status), 0);
else
dftw_c2c_2d_float(num_in, (float)wavenumber,
oskar_mem_float2_const(weights_in, status),
oskar_mem_float_const(x_in, status),
oskar_mem_float_const(y_in, status), 0,
offset_coord_out, num_out,
oskar_mem_float_const(x_out, status),
oskar_mem_float_const(y_out, status), 0,
oskar_mem_float2_const(data, status),
offset_out,
oskar_mem_float2(output, status), 0);
}
}
}
else
{
if (is_3d)
{
if (is_dbl)
dftw_o2c_3d_double(num_in, wavenumber,
oskar_mem_double2_const(weights_in, status),
oskar_mem_double_const(x_in, status),
oskar_mem_double_const(y_in, status),
oskar_mem_double_const(z_in, status),
offset_coord_out, num_out,
oskar_mem_double_const(x_out, status),
oskar_mem_double_const(y_out, status),
oskar_mem_double_const(z_out, status),
0, offset_out,
oskar_mem_double2(output, status), 0);
else
dftw_o2c_3d_float(num_in, (float)wavenumber,
oskar_mem_float2_const(weights_in, status),
oskar_mem_float_const(x_in, status),
oskar_mem_float_const(y_in, status),
oskar_mem_float_const(z_in, status),
offset_coord_out, num_out,
oskar_mem_float_const(x_out, status),
oskar_mem_float_const(y_out, status),
oskar_mem_float_const(z_out, status),
0, offset_out,
oskar_mem_float2(output, status), 0);
}
else
{
if (is_dbl)
dftw_o2c_2d_double(num_in, wavenumber,
oskar_mem_double2_const(weights_in, status),
oskar_mem_double_const(x_in, status),
oskar_mem_double_const(y_in, status), 0,
offset_coord_out, num_out,
oskar_mem_double_const(x_out, status),
oskar_mem_double_const(y_out, status), 0,
0, offset_out,
oskar_mem_double2(output, status), 0);
else
dftw_o2c_2d_float(num_in, (float)wavenumber,
oskar_mem_float2_const(weights_in, status),
oskar_mem_float_const(x_in, status),
oskar_mem_float_const(y_in, status), 0,
offset_coord_out, num_out,
oskar_mem_float_const(x_out, status),
oskar_mem_float_const(y_out, status), 0,
0, offset_out,
oskar_mem_float2(output, status), 0);
}
}
}
else
{
size_t local_size[] = {256, 1, 1}, global_size[] = {1, 1, 1};
const void* np = 0;
const char* k = 0;
int max_in_chunk;
float wavenumber_f = (float) wavenumber;
/* Select the kernel. */
switch (is_dbl * DBL | is_3d * D3 | is_data * DAT | is_matrix * MAT)
{
case D2 | FLT: k = "dftw_o2c_2d_float"; break;
case D2 | DBL: k = "dftw_o2c_2d_double"; break;
case D3 | FLT: k = "dftw_o2c_3d_float"; break;
case D3 | DBL: k = "dftw_o2c_3d_double"; break;
case D2 | FLT | DAT: k = "dftw_c2c_2d_float"; break;
case D2 | DBL | DAT: k = "dftw_c2c_2d_double"; break;
case D3 | FLT | DAT: k = "dftw_c2c_3d_float"; break;
case D3 | DBL | DAT: k = "dftw_c2c_3d_double"; break;
case D2 | FLT | DAT | MAT: k = "dftw_m2m_2d_float"; break;
case D2 | DBL | DAT | MAT: k = "dftw_m2m_2d_double"; break;
case D3 | FLT | DAT | MAT: k = "dftw_m2m_3d_float"; break;
case D3 | DBL | DAT | MAT: k = "dftw_m2m_3d_double"; break;
default:
*status = OSKAR_ERR_FUNCTION_NOT_AVAILABLE;
return;
}
if (oskar_device_is_nv(location))
local_size[0] = (size_t) get_block_size(num_out);
oskar_device_check_local_size(location, 0, local_size);
global_size[0] = oskar_device_global_size(
(size_t) num_out, local_size[0]);
/* max_in_chunk must be multiple of 16. */
max_in_chunk = is_3d ? (is_dbl ? 384 : 800) : (is_dbl ? 448 : 896);
if (is_data && is_3d && !is_dbl) max_in_chunk = 768;
const size_t element_size = is_dbl ? sizeof(double) : sizeof(float);
const size_t local_mem_size = max_in_chunk * element_size;
const size_t arg_size_local[] = {
2 * local_mem_size, 2 * local_mem_size,
(is_3d ? local_mem_size : 0)
};
/* Set kernel arguments. */
const oskar_Arg args[] = {
{INT_SZ, &num_in},
{is_dbl ? DBL_SZ : FLT_SZ,
is_dbl ? (void*)&wavenumber : (void*)&wavenumber_f},
{PTR_SZ, oskar_mem_buffer_const(weights_in)},
{PTR_SZ, oskar_mem_buffer_const(x_in)},
{PTR_SZ, oskar_mem_buffer_const(y_in)},
{PTR_SZ, is_3d ? oskar_mem_buffer_const(z_in) : &np},
{INT_SZ, &offset_coord_out},
{INT_SZ, &num_out},
{PTR_SZ, oskar_mem_buffer_const(x_out)},
{PTR_SZ, oskar_mem_buffer_const(y_out)},
{PTR_SZ, is_3d ? oskar_mem_buffer_const(z_out) : &np},
{PTR_SZ, is_data ? oskar_mem_buffer_const(data) : &np},
{INT_SZ, &offset_out},
{PTR_SZ, oskar_mem_buffer(output)},
{INT_SZ, &max_in_chunk}
};
oskar_device_launch_kernel(k, location, 1, local_size, global_size,
sizeof(args) / sizeof(oskar_Arg), args,
sizeof(arg_size_local) / sizeof(size_t), arg_size_local,
status);
}
if (normalise)
oskar_mem_scale_real(output, 1. / num_in, offset_out, num_out, status);
}
int main(int argc, char** argv)
{
// ===== Declare options ==================================================
oskar::OptionParser opt("oskar_vis_to_ascii_table", oskar_version_string());
opt.set_description("Converts an OSKAR visibility binary file to an ASCII "
"table format with the following columns:\n "
"[1] index, [2] baseline-uu, [3] baseline-vv, [4] baseline-ww "
"[5] Real, [6] Imag. "
"The table is written out in baseline-time order where baseline "
"is the fastest varying dimension");
opt.add_required("OSKAR vis file");
opt.add_optional("output file name");
opt.add_flag("-c", "Channel index to write to file. (default = 0)", 1, "0",
false, "--channel");
opt.add_flag("-p", "Polarisation ID to write to file. (default = 0) "
"(0=I, 1=Q, 2=U, 3=V, 4=XX, 5=XY, 6=YX, 7=YY)",
1, "0", false, "--polarisation");
opt.add_flag("-t", "Time index to write to file. (default = all)", 1, "",
false, "--time");
opt.add_flag("-w", "Output baseline coordinates in wavelengths. "
"(default = metres)", false, "--baseline_wavelengths");
opt.add_flag("-h", "Write a summary header in the ASCII table. ");
opt.add_flag("-v", "Verbose mode.");
opt.add_flag("--csv", "Write in CSV format");
opt.add_flag("-s", "Write output table to standard output instead of to file.",
false, "--stdout");
if (!opt.check_options(argc, argv)) return EXIT_FAILURE;
// ===== Read options ====================================================
const char* vis_file = opt.get_arg(0);
std::string txt_file;
if (opt.num_args() == 2)
txt_file = std::string(opt.get_arg(1));
else {
txt_file = std::string(vis_file) + ".txt";
}
int c = 0, p = 0, t = -1;
if (opt.is_set("-c")) c = opt.get_int("-c");
if (opt.is_set("-p")) p = opt.get_int("-p");
if (opt.is_set("-t")) t = opt.get_int("-t");
bool metres = !opt.is_set("-w");
bool write_header = opt.is_set("-h");
bool csv = opt.is_set("--csv");
bool verbose = opt.is_set("-v");
// ===== Write table ======================================================
int status = 0;
oskar_Binary* h = oskar_binary_create(vis_file, 'r', &status);
oskar_Vis* vis = oskar_vis_read(h, &status);
if (status)
{
fprintf(stderr, "ERROR: Unable to read specified visibility file: %s\n",
vis_file);
oskar_vis_free(vis, &status);
oskar_binary_free(h);
return status;
}
oskar_binary_free(h);
int num_chan = oskar_vis_num_channels(vis);
int num_times = oskar_vis_num_times(vis);
int num_baselines = oskar_vis_num_baselines(vis);
int num_pol = oskar_vis_num_pols(vis);
int num_stations = oskar_vis_num_stations(vis);
int total_vis = num_chan * num_times * num_baselines * num_pol;
double freq_start_hz = oskar_vis_freq_start_hz(vis);
double freq_inc_hz = oskar_vis_freq_inc_hz(vis);
double freq_hz = freq_start_hz + c * freq_inc_hz;
double lambda_m = 299792458.0 / freq_hz;
if (t != -1 && t > num_times-1) {
fprintf(stderr, "ERROR: Time index out of range.\n");
return EXIT_FAILURE;
}
if (c > num_chan-1) {
fprintf(stderr, "ERROR: Channel index out of range.\n");
return EXIT_FAILURE;
}
FILE* out;
if (!opt.is_set("-s")) {
out = fopen(txt_file.c_str(), "w");
if (out == NULL) return EXIT_FAILURE;
}
else {
out = stdout;
}
const oskar_Mem* uu = oskar_vis_baseline_uu_metres_const(vis);
const oskar_Mem* vv = oskar_vis_baseline_vv_metres_const(vis);
const oskar_Mem* ww = oskar_vis_baseline_ww_metres_const(vis);
const oskar_Mem* amp = oskar_vis_amplitude_const(vis);
// amplitudes dims: channel x times x baselines x pol
int amp_offset = c * num_times * num_baselines;
if (t != -1) amp_offset += t * num_baselines;
// baseline dims: times x baselines
int baseline_offset = 0;
if (t != -1) baseline_offset = t * num_baselines;
int type = oskar_mem_type(uu);
int num_vis_out = num_baselines;
if (t == -1) num_vis_out *= num_times;
if (verbose) {
write_header_(stdout, total_vis, num_chan, num_times, num_baselines,
num_pol, num_stations, num_vis_out, c, freq_hz, lambda_m, p, t,
metres);
#if 0
fprintf(stdout, "amp_offset = %i\n", amp_offset);
fprintf(stdout, "baseline_offset = %i\n", baseline_offset);
#endif
}
// Write header if specified
if (write_header)
{
write_header_(out, total_vis, num_chan, num_times, num_baselines,
num_pol, num_stations, num_vis_out, c, freq_hz, lambda_m, p, t,
metres);
char pre = '#';
fprintf(out, "%c\n", pre);
fprintf(out, "%c %s %-14s %-15s %-15s %-23s %-15s\n",
pre, "Idx", " uu", " vv", " ww", " Amp. Re.", " Amp. Im.");
}
if (type == OSKAR_DOUBLE)
{
const double* uu_ = oskar_mem_double_const(uu, &status);
const double* vv_ = oskar_mem_double_const(vv, &status);
const double* ww_ = oskar_mem_double_const(ww, &status);
const double4c* amp_ = oskar_mem_double4c_const(amp, &status);
int aIdx = amp_offset;
int bIdx = baseline_offset;
for (int i = 0; i < num_vis_out; ++i, ++bIdx, ++aIdx)
{
double2 a = getPolAmp_<double2, double4c>(amp_[aIdx], p);
double buu = (metres)? uu_[bIdx] : uu_[bIdx]/lambda_m;
double bvv = (metres)? vv_[bIdx] : vv_[bIdx]/lambda_m;
double bww = (metres)? ww_[bIdx] : ww_[bIdx]/lambda_m;
writeData_<double, double2>(i, buu, bvv, bww, a, csv, out);
}
}
else // OSKAR_SINGLE
{
const float* uu_ = oskar_mem_float_const(uu, &status);
const float* vv_ = oskar_mem_float_const(vv, &status);
const float* ww_ = oskar_mem_float_const(ww, &status);
const float4c* amp_ = oskar_mem_float4c_const(amp, &status);
int aIdx = amp_offset;
int bIdx = baseline_offset;
for (int i = 0; i < num_vis_out; ++i, ++bIdx, ++aIdx)
{
float2 a = getPolAmp_<float2, float4c>(amp_[aIdx], p);
float buu = (metres)? uu_[bIdx] : uu_[bIdx]/lambda_m;
float bvv = (metres)? vv_[bIdx] : vv_[bIdx]/lambda_m;
float bww = (metres)? ww_[bIdx] : ww_[bIdx]/lambda_m;
writeData_<float, float2>(i, buu, bvv, bww, a, csv, out);
}
}
fclose(out);
oskar_vis_free(vis, &status);
return status;
}
/* Wrapper. */
void oskar_evaluate_cross_power(int num_sources,
int num_stations, const oskar_Mem* jones, oskar_Mem* out,
int *status)
{
int type, location;
/* Check if safe to proceed. */
if (*status) return;
/* Check type and location. */
type = oskar_mem_type(jones);
location = oskar_mem_location(jones);
if (type != oskar_mem_type(out))
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if (location != oskar_mem_location(out))
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
/* Switch on type and location combination. */
if (type == OSKAR_SINGLE_COMPLEX_MATRIX)
{
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_evaluate_cross_power_cuda_f(num_sources,
num_stations, oskar_mem_float4c_const(jones, status),
oskar_mem_float4c(out, status));
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else if (location == OSKAR_CPU)
{
oskar_evaluate_cross_power_omp_f(num_sources,
num_stations, oskar_mem_float4c_const(jones, status),
oskar_mem_float4c(out, status));
}
}
else if (type == OSKAR_DOUBLE_COMPLEX_MATRIX)
{
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_evaluate_cross_power_cuda_d(num_sources,
num_stations, oskar_mem_double4c_const(jones, status),
oskar_mem_double4c(out, status));
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else if (location == OSKAR_CPU)
{
oskar_evaluate_cross_power_omp_d(num_sources,
num_stations, oskar_mem_double4c_const(jones, status),
oskar_mem_double4c(out, status));
}
}
/* Scalar versions. */
else if (type == OSKAR_SINGLE_COMPLEX)
{
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_evaluate_cross_power_scalar_cuda_f(num_sources,
num_stations, oskar_mem_float2_const(jones, status),
oskar_mem_float2(out, status));
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else if (location == OSKAR_CPU)
{
oskar_evaluate_cross_power_scalar_omp_f(num_sources,
num_stations, oskar_mem_float2_const(jones, status),
oskar_mem_float2(out, status));
}
}
else if (type == OSKAR_DOUBLE_COMPLEX)
{
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_evaluate_cross_power_scalar_cuda_d(num_sources,
num_stations, oskar_mem_double2_const(jones, status),
oskar_mem_double2(out, status));
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else if (location == OSKAR_CPU)
{
oskar_evaluate_cross_power_scalar_omp_d(num_sources,
num_stations, oskar_mem_double2_const(jones, status),
oskar_mem_double2(out, status));
}
}
else
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
}
}
