static void write_pixels(oskar_Mem* data, const char* filename,
int width, int height, int num_planes, int i_plane, int* status)
{
long naxes[3], firstpix[3], num_pix;
int dims_ok = 0;
fitsfile* f = 0;
if (*status) return;
if (oskar_file_exists(filename))
{
int naxis = 0, imagetype = 0;
fits_open_file(&f, filename, READWRITE, status);
fits_get_img_param(f, 3, &imagetype, &naxis, naxes, status);
if (naxis == 3 &&
naxes[0] == width &&
naxes[1] == height &&
naxes[2] == num_planes)
{
dims_ok = 1;
}
else
{
*status = 0;
fits_close_file(f, status);
remove(filename);
f = 0;
}
}
if (!dims_ok)
{
naxes[0] = width;
naxes[1] = height;
naxes[2] = num_planes;
fits_create_file(&f, filename, status);
fits_create_img(f, oskar_mem_is_double(data) ? DOUBLE_IMG : FLOAT_IMG,
3, naxes, status);
}
if (*status || !f)
{
if (f) fits_close_file(f, status);
*status = OSKAR_ERR_FILE_IO;
return;
}
num_pix = width * height;
firstpix[0] = 1;
firstpix[1] = 1;
firstpix[2] = 1 + i_plane;
if (i_plane < 0)
{
firstpix[2] = 1;
num_pix *= num_planes;
}
fits_write_pix(f, oskar_mem_is_double(data) ? TDOUBLE : TFLOAT,
firstpix, num_pix, oskar_mem_void(data), status);
fits_close_file(f, status);
}
static PyObject* auto_correlations(PyObject* self, PyObject* args)
{
oskar_VisBlock* h = 0;
oskar_Mem* m = 0;
PyObject *capsule = 0;
PyArrayObject *array = 0;
npy_intp dims[4];
if (!PyArg_ParseTuple(args, "O", &capsule)) return 0;
if (!(h = (oskar_VisBlock*) get_handle(capsule, name))) return 0;
/* Check that auto-correlations exist. */
if (!oskar_vis_block_has_auto_correlations(h))
{
PyErr_SetString(PyExc_RuntimeError, "No auto-correlations in block.");
return 0;
}
/* Return an array reference to Python. */
m = oskar_vis_block_auto_correlations(h);
dims[0] = oskar_vis_block_num_times(h);
dims[1] = oskar_vis_block_num_channels(h);
dims[2] = oskar_vis_block_num_stations(h);
dims[3] = oskar_vis_block_num_pols(h);
array = (PyArrayObject*)PyArray_SimpleNewFromData(4, dims,
(oskar_mem_is_double(m) ? NPY_CDOUBLE : NPY_CFLOAT),
oskar_mem_void(m));
return Py_BuildValue("N", array); /* Don't increment refcount. */
}
static void check_values(const oskar_Mem* approx, const oskar_Mem* accurate)
{
int status = 0;
double min_rel_error, max_rel_error, avg_rel_error, std_rel_error, tol;
oskar_mem_evaluate_relative_error(approx, accurate, &min_rel_error,
&max_rel_error, &avg_rel_error, &std_rel_error, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
tol = oskar_mem_is_double(approx) &&
oskar_mem_is_double(accurate) ? 1e-11 : 2e-3;
EXPECT_LT(max_rel_error, tol) << std::setprecision(5) <<
"RELATIVE ERROR" <<
" MIN: " << min_rel_error << " MAX: " << max_rel_error <<
" AVG: " << avg_rel_error << " STD: " << std_rel_error;
tol = oskar_mem_is_double(approx) &&
oskar_mem_is_double(accurate) ? 1e-12 : 1e-5;
EXPECT_LT(avg_rel_error, tol) << std::setprecision(5) <<
"RELATIVE ERROR" <<
" MIN: " << min_rel_error << " MAX: " << max_rel_error <<
" AVG: " << avg_rel_error << " STD: " << std_rel_error;
}
TEST(Mem, type_check_double_complex)
{
int status = 0;
oskar_Mem *mem;
mem = oskar_mem_create(OSKAR_DOUBLE_COMPLEX, OSKAR_CPU, 0,
&status);
EXPECT_EQ((int)OSKAR_TRUE, oskar_mem_is_double(mem));
EXPECT_EQ((int)OSKAR_TRUE, oskar_mem_is_complex(mem));
EXPECT_EQ((int)OSKAR_TRUE, oskar_mem_is_scalar(mem));
EXPECT_EQ((int)OSKAR_TRUE, oskar_type_is_double(OSKAR_DOUBLE_COMPLEX));
EXPECT_EQ((int)OSKAR_TRUE, oskar_type_is_complex(OSKAR_DOUBLE_COMPLEX));
EXPECT_EQ((int)OSKAR_TRUE, oskar_type_is_scalar(OSKAR_DOUBLE_COMPLEX));
oskar_mem_free(mem, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
}
TEST(Mem, type_check_single)
{
int status = 0;
oskar_Mem *mem;
mem = oskar_mem_create(OSKAR_SINGLE, OSKAR_CPU, 0, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
EXPECT_EQ((int)OSKAR_FALSE, oskar_mem_is_double(mem));
EXPECT_EQ((int)OSKAR_FALSE, oskar_mem_is_complex(mem));
EXPECT_EQ((int)OSKAR_TRUE, oskar_mem_is_scalar(mem));
EXPECT_EQ((int)OSKAR_FALSE, oskar_type_is_double(OSKAR_SINGLE));
EXPECT_EQ((int)OSKAR_FALSE, oskar_type_is_complex(OSKAR_SINGLE));
EXPECT_EQ((int)OSKAR_TRUE, oskar_type_is_scalar(OSKAR_SINGLE));
oskar_mem_free(mem, &status);
ASSERT_EQ(0, status) << oskar_get_error_string(status);
}
void write_plane(oskar_Imager* h, oskar_Mem* plane,
int t, int c, int p, int* status)
{
int datatype, num_pixels;
long firstpix[4];
if (*status) return;
if (!h->fits_file[p]) return;
datatype = (oskar_mem_is_double(plane) ? TDOUBLE : TFLOAT);
firstpix[0] = 1;
firstpix[1] = 1;
firstpix[2] = 1 + c;
firstpix[3] = 1 + t;
num_pixels = h->image_size * h->image_size;
fits_write_pix(h->fits_file[p], datatype, firstpix, num_pixels,
oskar_mem_void(plane), 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_element_weights_dft(int num_elements,
const oskar_Mem* x, const oskar_Mem* y, const oskar_Mem* z,
double wavenumber, double x_beam, double y_beam, double z_beam,
oskar_Mem* weights, int* status)
{
if (*status) return;
const int location = oskar_mem_location(weights);
const int type = oskar_mem_type(weights);
const int precision = oskar_mem_precision(weights);
if (oskar_mem_location(x) != location ||
oskar_mem_location(y) != location ||
oskar_mem_location(z) != location)
{
*status = OSKAR_ERR_LOCATION_MISMATCH;
return;
}
if (oskar_mem_type(x) != precision || oskar_mem_type(y) != precision ||
oskar_mem_type(z) != precision)
{
*status = OSKAR_ERR_TYPE_MISMATCH;
return;
}
if ((int)oskar_mem_length(weights) < num_elements ||
(int)oskar_mem_length(x) < num_elements ||
(int)oskar_mem_length(y) < num_elements ||
(int)oskar_mem_length(z) < num_elements)
{
*status = OSKAR_ERR_DIMENSION_MISMATCH;
return;
}
if (location == OSKAR_CPU)
{
if (type == OSKAR_DOUBLE_COMPLEX)
evaluate_element_weights_dft_d(num_elements,
oskar_mem_double_const(x, status),
oskar_mem_double_const(y, status),
oskar_mem_double_const(z, status),
wavenumber, x_beam, y_beam, z_beam,
oskar_mem_double2(weights, status));
else if (type == OSKAR_SINGLE_COMPLEX)
evaluate_element_weights_dft_f(num_elements,
oskar_mem_float_const(x, status),
oskar_mem_float_const(y, status),
oskar_mem_float_const(z, status),
wavenumber, x_beam, y_beam, z_beam,
oskar_mem_float2(weights, status));
else
*status = OSKAR_ERR_BAD_DATA_TYPE;
}
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(weights);
if (type == OSKAR_DOUBLE_COMPLEX)
k = "evaluate_element_weights_dft_double";
else if (type == OSKAR_SINGLE_COMPLEX)
k = "evaluate_element_weights_dft_float";
else
{
*status = OSKAR_ERR_BAD_DATA_TYPE;
return;
}
const float w = (float) wavenumber;
const float x1 = (float) x_beam;
const float y1 = (float) y_beam;
const float z1 = (float) z_beam;
oskar_device_check_local_size(location, 0, local_size);
global_size[0] = oskar_device_global_size(
(size_t) num_elements, local_size[0]);
const oskar_Arg args[] = {
{INT_SZ, &num_elements},
{PTR_SZ, oskar_mem_buffer_const(x)},
{PTR_SZ, oskar_mem_buffer_const(y)},
{PTR_SZ, oskar_mem_buffer_const(z)},
{is_dbl ? DBL_SZ : FLT_SZ,
is_dbl ? (const void*)&wavenumber : (const void*)&w},
{is_dbl ? DBL_SZ : FLT_SZ,
is_dbl ? (const void*)&x_beam : (const void*)&x1},
{is_dbl ? DBL_SZ : FLT_SZ,
is_dbl ? (const void*)&y_beam : (const void*)&y1},
{is_dbl ? DBL_SZ : FLT_SZ,
is_dbl ? (const void*)&z_beam : (const void*)&z1},
{PTR_SZ, oskar_mem_buffer(weights)}
};
oskar_device_launch_kernel(k, location, 1, local_size, global_size,
sizeof(args) / sizeof(oskar_Arg), args, 0, 0, status);
}
}
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);
}
