TEST(Mem, set_value_real_single_complex_matrix)
{
// Single precision complex matrix.
int n = 100, status = 0;
oskar_Mem *mem, *mem2;
mem = oskar_mem_create(OSKAR_SINGLE_COMPLEX_MATRIX, OSKAR_GPU, n,
&status);
oskar_mem_set_value_real(mem, 6.5, 0, 0, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
mem2 = oskar_mem_create_copy(mem, OSKAR_CPU, &status);
float4c* v = oskar_mem_float4c(mem2, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
for (int i = 0; i < n; ++i)
{
EXPECT_FLOAT_EQ(v[i].a.x, 6.5);
EXPECT_FLOAT_EQ(v[i].a.y, 0.0);
EXPECT_FLOAT_EQ(v[i].b.x, 0.0);
EXPECT_FLOAT_EQ(v[i].b.y, 0.0);
EXPECT_FLOAT_EQ(v[i].c.x, 0.0);
EXPECT_FLOAT_EQ(v[i].c.y, 0.0);
EXPECT_FLOAT_EQ(v[i].d.x, 6.5);
EXPECT_FLOAT_EQ(v[i].d.y, 0.0);
}
oskar_mem_free(mem, &status);
oskar_mem_free(mem2, &status);
}
void oskar_evaluate_station_beam_gaussian(oskar_Mem* beam,
int num_points, const oskar_Mem* l, const oskar_Mem* m,
const oskar_Mem* horizon_mask, double fwhm_rad, int* status)
{
int type, location;
double fwhm_lm, std;
/* Check if safe to proceed. */
if (*status) return;
/* Get type and check consistency. */
type = oskar_mem_precision(beam);
if (type != oskar_mem_type(l) || type != oskar_mem_type(m))
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if (type != OSKAR_SINGLE && type != OSKAR_DOUBLE)
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
if (!oskar_mem_is_complex(beam))
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
if (fwhm_rad == 0.0)
{
*status = OSKAR_ERR_SETTINGS_TELESCOPE;
return;
}
/* Get location and check consistency. */
location = oskar_mem_location(beam);
if (location != oskar_mem_location(l) ||
location != oskar_mem_location(m))
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
/* Check that length of input arrays are consistent. */
if ((int)oskar_mem_length(l) < num_points ||
(int)oskar_mem_length(m) < num_points)
{
*status = OSKAR_ERR_DIMENSION_MISMATCH;
return;
}
/* Resize output array if needed. */
if ((int)oskar_mem_length(beam) < num_points)
oskar_mem_realloc(beam, num_points, status);
/* Check if safe to proceed. */
if (*status) return;
/* Compute Gaussian standard deviation from FWHM. */
fwhm_lm = sin(fwhm_rad);
std = fwhm_lm / (2.0 * sqrt(2.0 * log(2.0)));
if (type == OSKAR_DOUBLE)
{
const double *l_, *m_;
l_ = oskar_mem_double_const(l, status);
m_ = oskar_mem_double_const(m, status);
if (location == OSKAR_CPU)
{
if (oskar_mem_is_scalar(beam))
{
oskar_gaussian_d(oskar_mem_double2(beam, status),
num_points, l_, m_, std);
}
else
{
oskar_gaussian_md(oskar_mem_double4c(beam, status),
num_points, l_, m_, std);
}
}
else
{
#ifdef OSKAR_HAVE_CUDA
if (oskar_mem_is_scalar(beam))
{
oskar_gaussian_cuda_d(oskar_mem_double2(beam, status),
num_points, l_, m_, std);
}
else
{
oskar_gaussian_cuda_md(oskar_mem_double4c(beam, status),
num_points, l_, m_, std);
}
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
}
else /* type == OSKAR_SINGLE */
{
const float *l_, *m_;
l_ = oskar_mem_float_const(l, status);
m_ = oskar_mem_float_const(m, status);
if (location == OSKAR_CPU)
{
if (oskar_mem_is_scalar(beam))
{
oskar_gaussian_f(oskar_mem_float2(beam, status), num_points,
l_, m_, (float)std);
}
else
{
oskar_gaussian_mf(oskar_mem_float4c(beam, status), num_points,
l_, m_, (float)std);
}
}
else
{
#ifdef OSKAR_HAVE_CUDA
if (oskar_mem_is_scalar(beam))
{
oskar_gaussian_cuda_f(oskar_mem_float2(beam, status),
num_points, l_, m_, (float)std);
}
else
{
oskar_gaussian_cuda_mf(oskar_mem_float4c(beam, status),
num_points, l_, m_, (float)std);
}
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
}
/* Blank (zero) sources below the horizon. */
oskar_blank_below_horizon(beam, horizon_mask, num_points, status);
}
void oskar_gaussian_circular(int num_points,
const oskar_Mem* l, const oskar_Mem* m, double std,
oskar_Mem* out, int* status)
{
if (*status) return;
const int location = oskar_mem_location(out);
const double inv_2_var = 1.0 / (2.0 * std * std);
const float inv_2_var_f = (float)inv_2_var;
if (oskar_mem_precision(out) != oskar_mem_type(l) ||
oskar_mem_precision(out) != oskar_mem_type(m))
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if (location != oskar_mem_location(l) || location != oskar_mem_location(m))
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
if ((int)oskar_mem_length(l) < num_points ||
(int)oskar_mem_length(m) < num_points)
{
*status = OSKAR_ERR_DIMENSION_MISMATCH;
return;
}
oskar_mem_ensure(out, num_points, status);
if (*status) return;
if (location == OSKAR_CPU)
{
switch (oskar_mem_type(out))
{
case OSKAR_SINGLE_COMPLEX:
gaussian_circular_complex_f(num_points,
oskar_mem_float_const(l, status),
oskar_mem_float_const(m, status), inv_2_var_f,
oskar_mem_float2(out, status));
break;
case OSKAR_DOUBLE_COMPLEX:
gaussian_circular_complex_d(num_points,
oskar_mem_double_const(l, status),
oskar_mem_double_const(m, status), inv_2_var,
oskar_mem_double2(out, status));
break;
case OSKAR_SINGLE_COMPLEX_MATRIX:
gaussian_circular_matrix_f(num_points,
oskar_mem_float_const(l, status),
oskar_mem_float_const(m, status), inv_2_var_f,
oskar_mem_float4c(out, status));
break;
case OSKAR_DOUBLE_COMPLEX_MATRIX:
gaussian_circular_matrix_d(num_points,
oskar_mem_double_const(l, status),
oskar_mem_double_const(m, status), inv_2_var,
oskar_mem_double4c(out, status));
break;
default:
*status = OSKAR_ERR_BAD_DATA_TYPE;
break;
}
}
else
{
size_t local_size[] = {256, 1, 1}, global_size[] = {1, 1, 1};
const char* k = 0;
const int is_dbl = oskar_mem_is_double(out);
switch (oskar_mem_type(out))
{
case OSKAR_SINGLE_COMPLEX:
k = "gaussian_circular_complex_float"; break;
case OSKAR_DOUBLE_COMPLEX:
k = "gaussian_circular_complex_double"; break;
case OSKAR_SINGLE_COMPLEX_MATRIX:
k = "gaussian_circular_matrix_float"; break;
case OSKAR_DOUBLE_COMPLEX_MATRIX:
k = "gaussian_circular_matrix_double"; break;
default:
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
oskar_device_check_local_size(location, 0, local_size);
global_size[0] = oskar_device_global_size(
(size_t) num_points, local_size[0]);
const oskar_Arg args[] = {
{INT_SZ, &num_points},
{PTR_SZ, oskar_mem_buffer_const(l)},
{PTR_SZ, oskar_mem_buffer_const(m)},
{is_dbl ? DBL_SZ : FLT_SZ, is_dbl ?
(const void*)&inv_2_var : (const void*)&inv_2_var_f},
{PTR_SZ, oskar_mem_buffer(out)}
};
oskar_device_launch_kernel(k, location, 1, local_size, global_size,
sizeof(args) / sizeof(oskar_Arg), args, 0, 0, status);
}
}
void oskar_evaluate_vla_beam_pbcor(oskar_Mem* beam, int num_sources,
const oskar_Mem* l, const oskar_Mem* m, double frequency_hz,
int* status)
{
int index, precision, type, location;
double f, p1, p2, p3;
/* Check if safe to proceed. */
if (*status) return;
/* Check type and location. */
precision = oskar_mem_precision(beam);
type = oskar_mem_type(beam);
location = oskar_mem_location(beam);
if (precision != oskar_mem_type(l) || precision != oskar_mem_type(m))
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if (location != oskar_mem_location(l) || location != oskar_mem_location(m))
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
/* Find the nearest frequency at which data exists. */
index = oskar_find_closest_match_d(frequency_hz / 1.0e9,
sizeof(freqs_ghz) / sizeof(double), freqs_ghz);
f = frequency_hz / 1.0e9;
p1 = p1s[index];
p2 = p2s[index];
p3 = p3s[index];
/* Switch on precision. */
if (precision == OSKAR_SINGLE)
{
const float *l_, *m_;
l_ = oskar_mem_float_const(l, status);
m_ = oskar_mem_float_const(m, status);
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
if (type == OSKAR_SINGLE)
{
oskar_evaluate_vla_beam_pbcor_cuda_f(
oskar_mem_float(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_SINGLE_COMPLEX)
{
oskar_evaluate_vla_beam_pbcor_complex_cuda_f(
oskar_mem_float2(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_SINGLE_COMPLEX_MATRIX)
{
oskar_evaluate_vla_beam_pbcor_matrix_cuda_f(
oskar_mem_float4c(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else if (location == OSKAR_CPU)
{
if (type == OSKAR_SINGLE)
{
oskar_evaluate_vla_beam_pbcor_f(
oskar_mem_float(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_SINGLE_COMPLEX)
{
oskar_evaluate_vla_beam_pbcor_complex_f(
oskar_mem_float2(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_SINGLE_COMPLEX_MATRIX)
{
oskar_evaluate_vla_beam_pbcor_matrix_f(
oskar_mem_float4c(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
}
}
else if (precision == OSKAR_DOUBLE)
{
const double *l_, *m_;
l_ = oskar_mem_double_const(l, status);
m_ = oskar_mem_double_const(m, status);
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
if (type == OSKAR_DOUBLE)
{
oskar_evaluate_vla_beam_pbcor_cuda_d(
oskar_mem_double(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_DOUBLE_COMPLEX)
{
oskar_evaluate_vla_beam_pbcor_complex_cuda_d(
oskar_mem_double2(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_DOUBLE_COMPLEX_MATRIX)
{
oskar_evaluate_vla_beam_pbcor_matrix_cuda_d(
oskar_mem_double4c(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else if (location == OSKAR_CPU)
{
if (type == OSKAR_DOUBLE)
{
oskar_evaluate_vla_beam_pbcor_d(
oskar_mem_double(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_DOUBLE_COMPLEX)
{
oskar_evaluate_vla_beam_pbcor_complex_d(
oskar_mem_double2(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
else if (type == OSKAR_DOUBLE_COMPLEX_MATRIX)
{
oskar_evaluate_vla_beam_pbcor_matrix_d(
oskar_mem_double4c(beam, status),
num_sources, l_, m_, f, p1, p2, p3);
}
}
}
else
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
}
}
void oskar_auto_correlate(oskar_Mem* vis, int n_sources, const oskar_Jones* J,
const oskar_Sky* sky, int* status)
{
int jones_type, base_type, location, matrix_type, n_stations;
/* Check if safe to proceed. */
if (*status) return;
/* Get the data dimensions. */
n_stations = oskar_jones_num_stations(J);
/* Check data locations. */
location = oskar_sky_mem_location(sky);
if (oskar_jones_mem_location(J) != location ||
oskar_mem_location(vis) != location)
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
/* Check for consistent data types. */
jones_type = oskar_jones_type(J);
base_type = oskar_sky_precision(sky);
matrix_type = oskar_type_is_matrix(jones_type) &&
oskar_mem_is_matrix(vis);
if (oskar_mem_precision(vis) != base_type ||
oskar_type_precision(jones_type) != base_type)
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if (oskar_mem_type(vis) != jones_type)
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
/* If neither single or double precision, return error. */
if (base_type != OSKAR_SINGLE && base_type != OSKAR_DOUBLE)
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
/* Check the input dimensions. */
if (oskar_jones_num_sources(J) < n_sources)
{
*status = OSKAR_ERR_DIMENSION_MISMATCH;
return;
}
/* Select kernel. */
if (base_type == OSKAR_DOUBLE)
{
const double *I_, *Q_, *U_, *V_;
I_ = oskar_mem_double_const(oskar_sky_I_const(sky), status);
Q_ = oskar_mem_double_const(oskar_sky_Q_const(sky), status);
U_ = oskar_mem_double_const(oskar_sky_U_const(sky), status);
V_ = oskar_mem_double_const(oskar_sky_V_const(sky), status);
if (matrix_type)
{
double4c *vis_;
const double4c *J_;
vis_ = oskar_mem_double4c(vis, status);
J_ = oskar_jones_double4c_const(J, status);
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_auto_correlate_cuda_d(n_sources, n_stations,
J_, I_, Q_, U_, V_, vis_);
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else /* CPU */
{
oskar_auto_correlate_omp_d(n_sources, n_stations,
J_, I_, Q_, U_, V_, vis_);
}
}
else /* Scalar version. */
{
double2 *vis_;
const double2 *J_;
vis_ = oskar_mem_double2(vis, status);
J_ = oskar_jones_double2_const(J, status);
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_auto_correlate_scalar_cuda_d(n_sources, n_stations,
J_, I_, vis_);
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else /* CPU */
{
oskar_auto_correlate_scalar_omp_d(n_sources, n_stations,
J_, I_, vis_);
}
}
}
else /* Single precision. */
{
const float *I_, *Q_, *U_, *V_;
I_ = oskar_mem_float_const(oskar_sky_I_const(sky), status);
Q_ = oskar_mem_float_const(oskar_sky_Q_const(sky), status);
U_ = oskar_mem_float_const(oskar_sky_U_const(sky), status);
V_ = oskar_mem_float_const(oskar_sky_V_const(sky), status);
if (matrix_type)
{
float4c *vis_;
const float4c *J_;
vis_ = oskar_mem_float4c(vis, status);
J_ = oskar_jones_float4c_const(J, status);
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_auto_correlate_cuda_f(n_sources, n_stations,
J_, I_, Q_, U_, V_, vis_);
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else /* CPU */
{
oskar_auto_correlate_omp_f(n_sources, n_stations,
J_, I_, Q_, U_, V_, vis_);
}
}
else /* Scalar version. */
{
float2 *vis_;
const float2 *J_;
vis_ = oskar_mem_float2(vis, status);
J_ = oskar_jones_float2_const(J, status);
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
oskar_auto_correlate_scalar_cuda_f(n_sources, n_stations,
J_, I_, vis_);
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else /* CPU */
{
oskar_auto_correlate_scalar_omp_f(n_sources, n_stations,
J_, I_, vis_);
}
}
}
}
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);
}
void oskar_blank_below_horizon(int offset_mask, int num_sources,
const oskar_Mem* mask, int offset_out, oskar_Mem* data, int* status)
{
if (*status) return;
const int location = oskar_mem_location(data);
if (oskar_mem_location(mask) != location)
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
if (oskar_mem_type(mask) != oskar_mem_precision(data))
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if ((int)oskar_mem_length(data) < num_sources)
{
*status = OSKAR_ERR_DIMENSION_MISMATCH;
return;
}
if (location == OSKAR_CPU)
{
switch (oskar_mem_type(data))
{
case OSKAR_DOUBLE_COMPLEX_MATRIX:
blank_below_horizon_matrix_d(offset_mask, num_sources,
oskar_mem_double_const(mask, status), offset_out,
oskar_mem_double4c(data, status));
break;
case OSKAR_DOUBLE_COMPLEX:
blank_below_horizon_scalar_d(offset_mask, num_sources,
oskar_mem_double_const(mask, status), offset_out,
oskar_mem_double2(data, status));
break;
case OSKAR_SINGLE_COMPLEX_MATRIX:
blank_below_horizon_matrix_f(offset_mask, num_sources,
oskar_mem_float_const(mask, status), offset_out,
oskar_mem_float4c(data, status));
break;
case OSKAR_SINGLE_COMPLEX:
blank_below_horizon_scalar_f(offset_mask, num_sources,
oskar_mem_float_const(mask, status), offset_out,
oskar_mem_float2(data, status));
break;
default:
*status = OSKAR_ERR_BAD_DATA_TYPE;
break;
}
}
else
{
size_t local_size[] = {256, 1, 1}, global_size[] = {1, 1, 1};
const char* k = 0;
switch (oskar_mem_type(data))
{
case OSKAR_DOUBLE_COMPLEX_MATRIX:
k = "blank_below_horizon_matrix_double"; break;
case OSKAR_DOUBLE_COMPLEX:
k = "blank_below_horizon_scalar_double"; break;
case OSKAR_SINGLE_COMPLEX_MATRIX:
k = "blank_below_horizon_matrix_float"; break;
case OSKAR_SINGLE_COMPLEX:
k = "blank_below_horizon_scalar_float"; break;
default:
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
oskar_device_check_local_size(location, 0, local_size);
global_size[0] = oskar_device_global_size(
(size_t) num_sources, local_size[0]);
const oskar_Arg args[] = {
{INT_SZ, &offset_mask},
{INT_SZ, &num_sources},
{PTR_SZ, oskar_mem_buffer_const(mask)},
{INT_SZ, &offset_out},
{PTR_SZ, oskar_mem_buffer(data)}
};
oskar_device_launch_kernel(k, location, 1, local_size, global_size,
sizeof(args) / sizeof(oskar_Arg), args, 0, 0, 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;
}
}
/* Wrapper. */
void oskar_evaluate_jones_R(oskar_Jones* R, int num_sources,
const oskar_Mem* ra_rad, const oskar_Mem* dec_rad,
const oskar_Telescope* telescope, double gast, int* status)
{
int i, n, num_stations, jones_type, base_type, location;
double latitude, lst;
oskar_Mem *R_station;
/* Check if safe to proceed. */
if (*status) return;
/* Get the Jones matrix block meta-data. */
jones_type = oskar_jones_type(R);
base_type = oskar_type_precision(jones_type);
location = oskar_jones_mem_location(R);
num_stations = oskar_jones_num_stations(R);
n = (oskar_telescope_allow_station_beam_duplication(telescope) ? 1 : num_stations);
/* Check that the data dimensions are OK. */
if (num_sources > (int)oskar_mem_length(ra_rad) ||
num_sources > (int)oskar_mem_length(dec_rad) ||
num_sources > oskar_jones_num_sources(R) ||
num_stations != oskar_telescope_num_stations(telescope))
{
*status = OSKAR_ERR_DIMENSION_MISMATCH;
return;
}
/* Check that the data is in the right location. */
if (location != oskar_mem_location(ra_rad) ||
location != oskar_mem_location(dec_rad))
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
/* Check that the data is of the right type. */
if (!oskar_type_is_matrix(jones_type))
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
if (base_type != oskar_mem_precision(ra_rad) ||
base_type != oskar_mem_precision(dec_rad))
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
/* Evaluate Jones matrix for each source for appropriate stations. */
R_station = oskar_mem_create_alias(0, 0, 0, status);
if (location == OSKAR_GPU)
{
#ifdef OSKAR_HAVE_CUDA
for (i = 0; i < n; ++i)
{
const oskar_Station* station;
/* Get station data. */
station = oskar_telescope_station_const(telescope, i);
latitude = oskar_station_lat_rad(station);
lst = gast + oskar_station_lon_rad(station);
oskar_jones_get_station_pointer(R_station, R, i, status);
/* Evaluate source parallactic angles. */
if (base_type == OSKAR_SINGLE)
{
oskar_evaluate_jones_R_cuda_f(
oskar_mem_float4c(R_station, status), num_sources,
oskar_mem_float_const(ra_rad, status),
oskar_mem_float_const(dec_rad, status),
(float)latitude, (float)lst);
}
else if (base_type == OSKAR_DOUBLE)
{
oskar_evaluate_jones_R_cuda_d(
oskar_mem_double4c(R_station, status), num_sources,
oskar_mem_double_const(ra_rad, status),
oskar_mem_double_const(dec_rad, status),
latitude, lst);
}
}
oskar_device_check_error(status);
#else
*status = OSKAR_ERR_CUDA_NOT_AVAILABLE;
#endif
}
else if (location == OSKAR_CPU)
{
for (i = 0; i < n; ++i)
{
const oskar_Station* station;
/* Get station data. */
station = oskar_telescope_station_const(telescope, i);
latitude = oskar_station_lat_rad(station);
lst = gast + oskar_station_lon_rad(station);
oskar_jones_get_station_pointer(R_station, R, i, status);
/* Evaluate source parallactic angles. */
if (base_type == OSKAR_SINGLE)
{
oskar_evaluate_jones_R_f(
oskar_mem_float4c(R_station, status), num_sources,
oskar_mem_float_const(ra_rad, status),
oskar_mem_float_const(dec_rad, status),
(float)latitude, (float)lst);
}
else if (base_type == OSKAR_DOUBLE)
{
oskar_evaluate_jones_R_d(
oskar_mem_double4c(R_station, status), num_sources,
oskar_mem_double_const(ra_rad, status),
oskar_mem_double_const(dec_rad, status),
latitude, lst);
}
}
}
/* Copy data for station 0 to stations 1 to n, if using a common sky. */
if (oskar_telescope_allow_station_beam_duplication(telescope))
{
oskar_Mem* R0;
R0 = oskar_mem_create_alias(0, 0, 0, status);
oskar_jones_get_station_pointer(R0, R, 0, status);
for (i = 1; i < num_stations; ++i)
{
oskar_jones_get_station_pointer(R_station, R, i, status);
oskar_mem_copy_contents(R_station, R0, 0, 0,
oskar_mem_length(R0), status);
}
oskar_mem_free(R0, status);
}
oskar_mem_free(R_station, status);
}
