nifty_gridder.cc 49.9 KB
Newer Older
Martin Reinecke's avatar
Martin Reinecke committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
/*
 *  This file is part of nifty_gridder.
 *
 *  nifty_gridder is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  nifty_gridder is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
Martin Reinecke's avatar
Martin Reinecke committed
15
 *  along with nifty_gridder; if not, write to the Free Software
Martin Reinecke's avatar
Martin Reinecke committed
16
17
18
 *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */

Martin Reinecke's avatar
Martin Reinecke committed
19
20
21
/* Copyright (C) 2019 Max-Planck-Society
   Author: Martin Reinecke */

Martin Reinecke's avatar
import  
Martin Reinecke committed
22
23
24
25
#include <pybind11/pybind11.h>
#include <pybind11/numpy.h>
#include <iostream>
#include <algorithm>
26
27
#include <cstdlib>
#include <cmath>
Martin Reinecke's avatar
Martin Reinecke committed
28

Martin Reinecke's avatar
updates    
Martin Reinecke committed
29
#include "pocketfft_hdronly.h"
Martin Reinecke's avatar
import  
Martin Reinecke committed
30
31
32
33
34
35
36

using namespace std;

namespace py = pybind11;

namespace {

Martin Reinecke's avatar
Martin Reinecke committed
37
38
auto None = py::none();

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
39
40
41
42
//
// basic utilities
//

43
44
45
46
47
48
void myassert(bool cond, const char *msg)
  {
  if (cond) return;
  throw runtime_error(msg);
  }

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
49
50
51
52
53
54
55
56
57
58
59
/*! Returns the remainder of the division \a v1/v2.
    The result is non-negative.
    \a v1 can be positive or negative; \a v2 must be positive. */
template<typename T> inline T fmodulo (T v1, T v2)
  {
  if (v1>=0)
    return (v1<v2) ? v1 : fmod(v1,v2);
  T tmp=fmod(v1,v2)+v2;
  return (tmp==v2) ? T(0) : tmp;
  }

60
61
62
63
64
65
66
67
68
69
70
71
static size_t nthreads = 1;

constexpr auto set_nthreads_DS = R"""(
Specifies the number of threads to be used by the module

Parameters
==========
nthreads: int
    the number of threads to be used. Must be >=1.
)""";
void set_nthreads(size_t nthreads_)
  {
72
  myassert(nthreads_>=1, "nthreads must be >= 1");
73
74
75
76
77
78
79
80
81
82
83
  nthreads = nthreads_;
  }

constexpr auto get_nthreads_DS = R"""(
Returns the number of threads used by the module

Returns
=======
int : the number of threads used by the module
)""";
size_t get_nthreads()
Martin Reinecke's avatar
Martin Reinecke committed
84
  { return nthreads; }
85

Martin Reinecke's avatar
Martin Reinecke committed
86
87
88
89
//
// Utilities for Gauss-Legendre quadrature
//

Martin Reinecke's avatar
Martin Reinecke committed
90
91
92
93
94
95
96
97
98
99
100
101
102
103
static inline double one_minus_x2 (double x)
  { return (fabs(x)>0.1) ? (1.+x)*(1.-x) : 1.-x*x; }

void legendre_prep(int n, vector<double> &x, vector<double> &w)
  {
  constexpr double pi = 3.141592653589793238462643383279502884197;
  constexpr double eps = 3e-14;
  int m = (n+1)>>1;
  x.resize(m);
  w.resize(m);

  double t0 = 1 - (1-1./n) / (8.*n*n);
  double t1 = 1./(4.*n+2.);

104
#pragma omp parallel num_threads(nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
{
  int i;
#pragma omp for schedule(dynamic,100)
  for (i=1; i<=m; ++i)
    {
    double x0 = cos(pi * ((i<<2)-1) * t1) * t0;

    int dobreak=0;
    int j=0;
    double dpdx;
    while(1)
      {
      double P_1 = 1.0;
      double P0 = x0;
      double dx, x1;

      for (int k=2; k<=n; k++)
        {
        double P_2 = P_1;
        P_1 = P0;
//        P0 = ((2*k-1)*x0*P_1-(k-1)*P_2)/k;
        P0 = x0*P_1 + (k-1.)/k * (x0*P_1-P_2);
        }

      dpdx = (P_1 - x0*P0) * n / one_minus_x2(x0);

      /* Newton step */
      x1 = x0 - P0/dpdx;
      dx = x0-x1;
      x0 = x1;
      if (dobreak) break;

      if (abs(dx)<=eps) dobreak=1;
138
      myassert(++j<100, "convergence problem");
Martin Reinecke's avatar
Martin Reinecke committed
139
140
141
142
143
144
145
146
      }

    x[m-i] = x0;
    w[m-i] = 2. / (one_minus_x2(x0) * dpdx * dpdx);
    }
} // end of parallel region
  }

Martin Reinecke's avatar
Martin Reinecke committed
147
148
149
150
//
// Start of real gridder functionality
//

151
152
template<typename T>
  using pyarr = py::array_t<T>;
153
// The "_c" suffix here stands for "C memory order, contiguous"
154
155
template<typename T>
  using pyarr_c = py::array_t<T, py::array::c_style | py::array::forcecast>;
Martin Reinecke's avatar
import  
Martin Reinecke committed
156

Martin Reinecke's avatar
merge    
Martin Reinecke committed
157
template<typename T> pyarr_c<T> makeArray(const vector<size_t> &shape)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
158
159
  { return pyarr_c<T>(shape); }

160
size_t get_supp(double epsilon)
Martin Reinecke's avatar
Martin Reinecke committed
161
162
163
164
165
166
167
168
169
170
171
  {
  static const vector<double> maxmaperr { 1e8, 0.32, 0.021, 6.2e-4,
    1.08e-5, 1.25e-7, 8.25e-10, 5.70e-12, 1.22e-13, 2.48e-15, 4.82e-17,
    6.74e-19, 5.41e-21, 4.41e-23, 7.88e-25, 3.9e-26 };

  double epssq = epsilon*epsilon;

  for (size_t i=1; i<maxmaperr.size(); ++i)
    if (epssq>maxmaperr[i]) return i;
  throw runtime_error("requested epsilon too small - minimum is 2e-13");
  }
Martin Reinecke's avatar
Martin Reinecke committed
172

Martin Reinecke's avatar
merge    
Martin Reinecke committed
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
void checkArray(const py::array &arr, const char *aname,
  const vector<size_t> &shape)
  {
  if (size_t(arr.ndim())!=shape.size())
    {
    cerr << "Array '" << aname << "' has " << arr.ndim() << " dimensions; "
            "expected " << shape.size() << endl;
    throw runtime_error("bad dimensionality");
    }
  for (size_t i=0; i<shape.size(); ++i)
    if ((shape[i]!=0) && (size_t(arr.shape(i))!=shape[i]))
      {
      cerr << "Dimension " << i << " of array '" << aname << "' has size "
           << arr.shape(i) << "; expected " << shape[i] << endl;
      throw runtime_error("bad array size");
      }
  }

Martin Reinecke's avatar
Martin Reinecke committed
191
template<typename T> pyarr<T> provideArray(const py::object &in,
Martin Reinecke's avatar
merge    
Martin Reinecke committed
192
193
  const vector<size_t> &shape)
  {
194
  if (in.is_none())
Martin Reinecke's avatar
merge    
Martin Reinecke committed
195
196
197
198
199
200
201
202
    {
    auto tmp_ = makeArray<T>(shape);
    size_t sz = size_t(tmp_.size());
    auto tmp = tmp_.mutable_data();
    for (size_t i=0; i<sz; ++i)
      tmp[i] = T(0);
    return tmp_;
    }
203
  auto tmp_ = in.cast<pyarr<T>>();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
204
205
206
207
  checkArray(tmp_, "temporary", shape);
  return tmp_;
  }

Martin Reinecke's avatar
Martin Reinecke committed
208
209
210
211
212
213
214
215
216
217
template<typename T> pyarr<T> providePotentialArray(const py::object &in,
  const vector<size_t> &shape)
  {
  if (in.is_none())
    return makeArray<T>(vector<size_t>(shape.size(), 0));
  auto tmp_ = in.cast<pyarr<T>>();
  checkArray(tmp_, "temporary", shape);
  return tmp_;
  }

Martin Reinecke's avatar
Martin Reinecke committed
218
template<typename T> pyarr_c<T> provideCArray(py::object &in,
Martin Reinecke's avatar
merge    
Martin Reinecke committed
219
220
  const vector<size_t> &shape)
  {
221
  if (in.is_none())
Martin Reinecke's avatar
merge    
Martin Reinecke committed
222
223
224
225
226
227
228
229
230
231
232
233
234
    {
    auto tmp_ = makeArray<T>(shape);
    size_t sz = size_t(tmp_.size());
    auto tmp = tmp_.mutable_data();
    for (size_t i=0; i<sz; ++i)
      tmp[i] = T(0);
    return tmp_;
    }
  auto tmp_ = in.cast<pyarr_c<T>>();
  checkArray(tmp_, "temporary", shape);
  return tmp_;
  }

Martin Reinecke's avatar
Martin Reinecke committed
235
template<typename T> pyarr_c<T> complex2hartley
236
  (const pyarr_c<complex<T>> &grid_, py::object &grid_in)
Martin Reinecke's avatar
Martin Reinecke committed
237
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
238
  checkArray(grid_, "grid", {0,0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
239
  size_t nu = size_t(grid_.shape(0)), nv = size_t(grid_.shape(1));
Martin Reinecke's avatar
Martin Reinecke committed
240
241
  auto grid = grid_.data();

242
  auto res = provideCArray<T>(grid_in, {nu, nv});
Martin Reinecke's avatar
Martin Reinecke committed
243
  auto grid2 = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
244
245
  {
  py::gil_scoped_release release;
246
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
247
  for (size_t u=0; u<nu; ++u)
Martin Reinecke's avatar
Martin Reinecke committed
248
    {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
249
250
    size_t xu = (u==0) ? 0 : nu-u;
    for (size_t v=0; v<nv; ++v)
Martin Reinecke's avatar
Martin Reinecke committed
251
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
252
253
254
      size_t xv = (v==0) ? 0 : nv-v;
      size_t i1 = u*nv+v;
      size_t i2 = xu*nv+xv;
255
256
      grid2[i1] += T(0.5)*(grid[i1].real()+grid[i1].imag()+
                           grid[i2].real()-grid[i2].imag());
Martin Reinecke's avatar
Martin Reinecke committed
257
258
      }
    }
Martin Reinecke's avatar
Martin Reinecke committed
259
  }
Martin Reinecke's avatar
Martin Reinecke committed
260
261
262
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
263
264
template<typename T> pyarr_c<complex<T>> hartley2complex
  (const pyarr_c<T> &grid_)
Martin Reinecke's avatar
Martin Reinecke committed
265
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
266
  checkArray(grid_, "grid", {0, 0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
267
  size_t nu = size_t(grid_.shape(0)), nv = size_t(grid_.shape(1));
Martin Reinecke's avatar
Martin Reinecke committed
268
269
  auto grid = grid_.data();

Martin Reinecke's avatar
merge    
Martin Reinecke committed
270
  auto res=makeArray<complex<T>>({nu, nv});
Martin Reinecke's avatar
Martin Reinecke committed
271
  auto grid2 = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
272
273
  {
  py::gil_scoped_release release;
274
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
275
  for (size_t u=0; u<nu; ++u)
Martin Reinecke's avatar
Martin Reinecke committed
276
    {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
277
278
    size_t xu = (u==0) ? 0 : nu-u;
    for (size_t v=0; v<nv; ++v)
Martin Reinecke's avatar
Martin Reinecke committed
279
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
280
281
282
      size_t xv = (v==0) ? 0 : nv-v;
      size_t i1 = u*nv+v;
      size_t i2 = xu*nv+xv;
Martin Reinecke's avatar
Martin Reinecke committed
283
284
285
      T v1 = T(0.5)*grid[i1];
      T v2 = T(0.5)*grid[i2];
      grid2[i1] = complex<T>(v1+v2, v1-v2);
Martin Reinecke's avatar
Martin Reinecke committed
286
287
      }
    }
Martin Reinecke's avatar
Martin Reinecke committed
288
  }
Martin Reinecke's avatar
Martin Reinecke committed
289
290
291
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
292
293
294
template<typename T> void hartley2_2D(const pyarr_c<T> &in, pyarr_c<T> &out)
  {
  size_t nu=in.shape(0), nv=in.shape(1);
Martin Reinecke's avatar
Martin Reinecke committed
295
296
297
  pocketfft::stride_t s_i{in.strides(0), in.strides(1)},
                      s_o{out.strides(0), out.strides(1)};
  auto d_i = in.data();
Martin Reinecke's avatar
Martin Reinecke committed
298
  auto ptmp = out.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
299
300
  {
  py::gil_scoped_release release;
301
302
303
  pocketfft::r2r_separable_hartley({nu, nv}, s_i, s_o, {0,1}, d_i, ptmp, T(1),
    nthreads);
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
304
305
306
307
308
309
310
311
312
313
314
315
316
  for(size_t i=1; i<(nu+1)/2; ++i)
    for(size_t j=1; j<(nv+1)/2; ++j)
       {
       T a = ptmp[i*nv+j];
       T b = ptmp[(nu-i)*nv+j];
       T c = ptmp[i*nv+nv-j];
       T d = ptmp[(nu-i)*nv+nv-j];
       ptmp[i*nv+j] = T(0.5)*(a+b+c-d);
       ptmp[(nu-i)*nv+j] = T(0.5)*(a+b+d-c);
       ptmp[i*nv+nv-j] = T(0.5)*(a+c+d-b);
       ptmp[(nu-i)*nv+nv-j] = T(0.5)*(b+c+d-a);
       }
  }
Martin Reinecke's avatar
Martin Reinecke committed
317
  }
Martin Reinecke's avatar
Martin Reinecke committed
318

Martin Reinecke's avatar
Martin Reinecke committed
319
320
/* Compute correction factors for the ES gridding kernel
   This implementation follows eqs. (3.8) to (3.10) of Barnett et al. 2018 */
321
vector<double> correction_factors (size_t n, size_t nval, size_t supp)
Martin Reinecke's avatar
Martin Reinecke committed
322
323
  {
  constexpr double pi = 3.141592653589793238462643383279502884197;
324
325
326
  auto beta = 2.3*supp;
  auto p = int(1.5*supp+2);
  double alpha = pi*supp/n;
Martin Reinecke's avatar
Martin Reinecke committed
327
328
329
330
331
  vector<double> x, wgt;
  legendre_prep(2*p,x,wgt);
  auto psi = x;
  for (auto &v:psi)
    v = exp(beta*(sqrt(1-v*v)-1.));
Martin Reinecke's avatar
updates    
Martin Reinecke committed
332
  vector<double> res(nval);
333
#pragma omp parallel for schedule(static) num_threads(nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
334
335
336
337
338
  for (size_t k=0; k<nval; ++k)
    {
    double tmp=0;
    for (int i=0; i<p; ++i)
      tmp += wgt[i]*psi[i]*cos(alpha*k*x[i]);
339
    res[k] = 1./(supp*tmp);
Martin Reinecke's avatar
Martin Reinecke committed
340
341
342
343
    }
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
344
template<typename T> struct UVW
Martin Reinecke's avatar
updates    
Martin Reinecke committed
345
  {
Martin Reinecke's avatar
Martin Reinecke committed
346
  T u, v, w;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
347
  UVW () {}
Martin Reinecke's avatar
Martin Reinecke committed
348
349
  UVW (T u_, T v_, T w_) : u(u_), v(v_), w(w_) {}
  UVW operator* (T fct) const
Martin Reinecke's avatar
updates    
Martin Reinecke committed
350
351
    { return UVW(u*fct, v*fct, w*fct); }
  };
Martin Reinecke's avatar
Martin Reinecke committed
352

353
354
355
356
357
358
359
360
361
362
constexpr auto Baselines_DS = R"""(
Class storing UVW coordinates and channel information.

Parameters
==========
coord: np.array((nrows, 3), dtype=np.float)
    u, v and w coordinates for each row
freq: np.array((nchannels,), dtype=np.float)
    frequency for each individual channel (in Hz)
)""";
Martin Reinecke's avatar
Martin Reinecke committed
363
template<typename T> class Baselines
Martin Reinecke's avatar
Martin Reinecke committed
364
365
  {
  private:
Martin Reinecke's avatar
Martin Reinecke committed
366
    vector<UVW<T>> coord;
367
    vector<T> f_over_c;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
368
    size_t nrows, nchan;
Martin Reinecke's avatar
Martin Reinecke committed
369
370

  public:
371
    Baselines(const pyarr<T> &coord_, const pyarr<T> &freq_)
Martin Reinecke's avatar
Martin Reinecke committed
372
      {
373
      constexpr double speedOfLight = 299792458.;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
374
      checkArray(coord_, "coord", {0, 3});
375
      checkArray(freq_, "freq", {0});
Martin Reinecke's avatar
Martin Reinecke committed
376
      nrows = coord_.shape(0);
377
      nchan = freq_.shape(0);
Martin Reinecke's avatar
Martin Reinecke committed
378
      myassert(nrows*nchan<(size_t(1)<<32), "too many entries in MS");
379
380
      auto freq = freq_.template unchecked<1>();
      auto cood = coord_.template unchecked<2>();
Martin Reinecke's avatar
Martin Reinecke committed
381
382
      {
      py::gil_scoped_release release;
383
      f_over_c.resize(nchan);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
384
      for (size_t i=0; i<nchan; ++i)
385
        f_over_c[i] = freq(i)/speedOfLight;
Martin Reinecke's avatar
Martin Reinecke committed
386
387
      coord.resize(nrows);
      for (size_t i=0; i<coord.size(); ++i)
388
        coord[i] = UVW<T>(cood(i,0), cood(i,1), cood(i,2));
Martin Reinecke's avatar
Martin Reinecke committed
389
      }
Martin Reinecke's avatar
Martin Reinecke committed
390
391
      }

Martin Reinecke's avatar
Martin Reinecke committed
392
    UVW<T> effectiveCoord(uint32_t index) const
393
394
395
      {
      size_t irow = index/nchan;
      size_t ichan = index-nchan*irow;
396
      return coord[irow]*f_over_c[ichan];
397
398
      }
    UVW<T> effectiveCoord(size_t irow, size_t ichan) const
399
      { return coord[irow]*f_over_c[ichan]; }
Martin Reinecke's avatar
Martin Reinecke committed
400
    size_t Nrows() const { return nrows; }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
401
402
    size_t Nchannels() const { return nchan; }

403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
    static constexpr auto ms2vis_DS = R"""(
    Extracts visibility data from a measurement for the provided indices.

    Parameters
    ==========
    ms: np.array((nrows, nchannels), dtype=np.complex)
        the measurement set's visibility data
    idx: np.array((nvis,), dtype=np.uint32)
        the indices to be extracted

    Returns
    =======
    np.array((nvis,), dtype=np.complex)
        The visibility data for the index array
    )""";
418

Philipp Arras's avatar
Philipp Arras committed
419
    pyarr_c<T> effectiveuvw(const pyarr_c<uint32_t> &idx_) const
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
    {
      checkArray(idx_, "idx", {0});
      size_t nvis = size_t(idx_.shape(0));
      auto idx = idx_.template unchecked<1>();
      auto res_=makeArray<T>({nvis, 3});
      auto res = res_.template mutable_unchecked<2>();
      for (size_t i=0; i<nvis; i++)
        {
          auto uvw = effectiveCoord(idx(i));
          res(i,0) = uvw.u;
          res(i,1) = uvw.v;
          res(i,2) = uvw.w;
        }
      return res_;
    }

436
    template<typename T2> pyarr_c<T2> ms2vis(const pyarr<T2> &ms_,
Martin Reinecke's avatar
updates    
Martin Reinecke committed
437
438
      const pyarr_c<uint32_t> &idx_) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
439
440
      checkArray(idx_, "idx", {0});
      checkArray(ms_, "ms", {nrows, nchan});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
441
      size_t nvis = size_t(idx_.shape(0));
442
443
      auto idx = idx_.template unchecked<1>();
      auto ms = ms_.template unchecked<2>();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
444

Martin Reinecke's avatar
merge    
Martin Reinecke committed
445
      auto res=makeArray<T2>({nvis});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
446
      auto vis = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
447
448
      {
      py::gil_scoped_release release;
449
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
450
      for (size_t i=0; i<nvis; ++i)
451
452
453
454
455
456
        {
        auto t = idx(i);
        auto row = t/nchan;
        auto chan = t-row*nchan;
        vis[i] = ms(row, chan);
        }
Martin Reinecke's avatar
Martin Reinecke committed
457
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
458
459
460
      return res;
      }

461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
    static constexpr auto vis2ms_DS = R"""(
    Produces a new MS with the provided visibilities set.

    Parameters
    ==========
    vis: np.array((nvis,), dtype=np.complex)
        The visibility data for the index array
    idx: np.array((nvis,), dtype=np.uint32)
        the indices to be inserted
    ms_in: np.array((nrows, nchannels), dtype=np.complex), optional
        input measurement set to which the visibilities are added.

    Returns
    =======
    np.array((nrows, nchannels), dtype=np.complex)
        the measurement set's visibility data (0 where not covered by idx)
    )""";
478
479
    template<typename T2> pyarr_c<T2> vis2ms(const pyarr<T2> &vis_,
      const pyarr<uint32_t> &idx_, py::object &ms_in) const
Martin Reinecke's avatar
updates    
Martin Reinecke committed
480
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
481
      checkArray(vis_, "vis", {0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
482
      size_t nvis = size_t(vis_.shape(0));
Martin Reinecke's avatar
merge    
Martin Reinecke committed
483
      checkArray(idx_, "idx", {nvis});
484
485
      auto idx = idx_.template unchecked<1>();
      auto vis = vis_.template unchecked<1>();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
486

Martin Reinecke's avatar
merge    
Martin Reinecke committed
487
      auto res = provideArray<T2>(ms_in, {nrows, nchan});
488
      auto ms = res.template mutable_unchecked<2>();
Martin Reinecke's avatar
Martin Reinecke committed
489
490
      {
      py::gil_scoped_release release;
491
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
492
      for (size_t i=0; i<nvis; ++i)
493
494
495
496
        {
        auto t = idx(i);
        auto row = t/nchan;
        auto chan = t-row*nchan;
Martin Reinecke's avatar
Martin Reinecke committed
497
        ms(row, chan) += vis(i);
498
        }
Martin Reinecke's avatar
Martin Reinecke committed
499
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
500
501
      return res;
      }
Martin Reinecke's avatar
Martin Reinecke committed
502
503
  };

Martin Reinecke's avatar
Martin Reinecke committed
504
505
constexpr int logsquare=4;

506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
constexpr auto grid2dirty_DS = R"""(
Converts from UV grid to dirty image (FFT, cropping, correction)

Parameters
==========
grid: np.array((nu, nv), dtype=np.float64)
    gridded UV data

Returns
=======
nd.array((nxdirty, nydirty), dtype=np.float64)
    the dirty image
)""";

constexpr auto dirty2grid_DS = R"""(
Converts from a dirty image to a UV grid (correction, padding, FFT)

Parameters
==========
dirty: nd.array((nxdirty, nydirty), dtype=np.float64)
    the dirty image

Returns
=======
np.array((nu, nv), dtype=np.float64)
    gridded UV data
)""";

Martin Reinecke's avatar
Martin Reinecke committed
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
constexpr auto apply_taper_DS = R"""(
Applies the taper (or its inverse) to an image

Parameters
==========
img: nd.array((nxdirty, nydirty), dtype=np.float64)
    the image
divide: bool
    if True, the routine dividex by the taper, otherwise it multiplies by it

Returns
=======
np.array((nxdirty, nydirty), dtype=np.float64)
    the image with the taper applied
)""";

Martin Reinecke's avatar
Martin Reinecke committed
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
constexpr auto apply_wscreen_DS = R"""(
Applies the w screen to an image

Parameters
==========
dirty: nd.array((nxdirty, nydirty), dtype=np.complex128)
    the image
w : float
    the w value to use
adjoint: bool
    if True, apply the complex conjugate of the w screen

Returns
=======
np.array((nxdirty, nydirty), dtype=np.complex128)
    the image with the w screen applied
)""";

568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
constexpr auto GridderConfig_DS = R"""(
Class storing information related to the gridding/degridding process.

Parameters
==========
nxdirty: int
    x resolution of the dirty image; must be even
nydirty: int
    y resolution of the dirty image; must be even
epsilon: float
    required accuracy for the gridding/degridding step
    Must be >= 2e-13.
pixsize_x: float
    Pixel size in x direction (radians)
pixsize_y: float
    Pixel size in y direction (radians)
)""";
Martin Reinecke's avatar
Martin Reinecke committed
585
template<typename T> class GridderConfig
Martin Reinecke's avatar
Martin Reinecke committed
586
587
588
  {
  private:
    size_t nx_dirty, ny_dirty;
Martin Reinecke's avatar
Martin Reinecke committed
589
    double eps, psx, psy;
590
    size_t supp, nsafe, nu, nv;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
591
    T beta;
Martin Reinecke's avatar
Martin Reinecke committed
592
    vector<T> cfu, cfv;
Martin Reinecke's avatar
Martin Reinecke committed
593

Martin Reinecke's avatar
Martin Reinecke committed
594
595
596
    complex<T> wscreen(double x, double y, double w, bool adjoint) const
      {
      constexpr double pi = 3.141592653589793238462643383279502884197;
597
598
599
600
601
      double eps = sqrt(x+y);
      double s = sin(eps);
      double nm1 = -s*s/(1.+cos(eps));
      double n = nm1+1., xn = 1./n;
      double phase = 2*pi*w*nm1;
Martin Reinecke's avatar
Martin Reinecke committed
602
603
604
605
      if (adjoint) phase *= -1;
      return complex<T>(cos(phase)*xn, sin(phase)*xn);
      }

Martin Reinecke's avatar
Martin Reinecke committed
606
607
  public:
    GridderConfig(size_t nxdirty, size_t nydirty, double epsilon,
Martin Reinecke's avatar
Martin Reinecke committed
608
      double pixsize_x, double pixsize_y)
Martin Reinecke's avatar
Martin Reinecke committed
609
610
      : nx_dirty(nxdirty), ny_dirty(nydirty), eps(epsilon),
        psx(pixsize_x), psy(pixsize_y),
611
        supp(get_supp(epsilon)), nsafe((supp+1)/2),
Martin Reinecke's avatar
Martin Reinecke committed
612
        nu(max(2*nsafe,2*nx_dirty)), nv(max(2*nsafe,2*ny_dirty)),
613
        beta(2.3*supp),
Martin Reinecke's avatar
updates    
Martin Reinecke committed
614
        cfu(nx_dirty), cfv(ny_dirty)
Martin Reinecke's avatar
Martin Reinecke committed
615
      {
Martin Reinecke's avatar
Martin Reinecke committed
616
617
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
Martin Reinecke committed
618
619
620
      myassert((nx_dirty&1)==0, "nx_dirty must be even");
      myassert((ny_dirty&1)==0, "ny_dirty must be even");
      myassert(epsilon>0, "epsilon must be positive");
Martin Reinecke's avatar
Martin Reinecke committed
621
622
      myassert(pixsize_x>0, "pixsize_x must be positive");
      myassert(pixsize_y>0, "pixsize_y must be positive");
Martin Reinecke's avatar
updates    
Martin Reinecke committed
623

624
      auto tmp = correction_factors(nu, nx_dirty/2+1, supp);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
625
626
627
628
      cfu[nx_dirty/2]=tmp[0];
      cfu[0]=tmp[nx_dirty/2];
      for (size_t i=1; i<nx_dirty/2; ++i)
        cfu[nx_dirty/2-i] = cfu[nx_dirty/2+i] = tmp[i];
629
      tmp = correction_factors(nv, ny_dirty/2+1, supp);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
630
631
632
633
      cfv[ny_dirty/2]=tmp[0];
      cfv[0]=tmp[ny_dirty/2];
      for (size_t i=1; i<ny_dirty/2; ++i)
        cfv[ny_dirty/2-i] = cfv[ny_dirty/2+i] = tmp[i];
Martin Reinecke's avatar
Martin Reinecke committed
634
      }
Martin Reinecke's avatar
Martin Reinecke committed
635
      }
Martin Reinecke's avatar
Martin Reinecke committed
636
637
638
639
640
    size_t Nxdirty() const { return nx_dirty; }
    size_t Nydirty() const { return ny_dirty; }
    double Epsilon() const { return eps; }
    double Pixsize_x() const { return psx; }
    double Pixsize_y() const { return psy; }
Martin Reinecke's avatar
Martin Reinecke committed
641
642
    size_t Nu() const { return nu; }
    size_t Nv() const { return nv; }
643
    size_t Supp() const { return supp; }
644
    size_t Nsafe() const { return nsafe; }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
645
    T Beta() const { return beta; }
646

Martin Reinecke's avatar
Martin Reinecke committed
647
    pyarr_c<T> grid2dirty(const pyarr_c<T> &grid) const
Martin Reinecke's avatar
Martin Reinecke committed
648
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
649
650
      checkArray(grid, "grid", {nu, nv});
      auto tmp = makeArray<T>({nu, nv});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
651
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
652
      hartley2_2D<T>(grid, tmp);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
653
      auto res = makeArray<T>({nx_dirty, ny_dirty});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
654
      auto pout = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
655
656
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
657
658
659
660
661
662
663
664
665
      for (size_t i=0; i<nx_dirty; ++i)
        for (size_t j=0; j<ny_dirty; ++j)
          {
          size_t i2 = nu-nx_dirty/2+i;
          if (i2>=nu) i2-=nu;
          size_t j2 = nv-ny_dirty/2+j;
          if (j2>=nv) j2-=nv;
          pout[ny_dirty*i + j] = ptmp[nv*i2+j2]*cfu[i]*cfv[j];
          }
Martin Reinecke's avatar
Martin Reinecke committed
666
      }
Martin Reinecke's avatar
Martin Reinecke committed
667
668
      return res;
      }
Martin Reinecke's avatar
Martin Reinecke committed
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
    pyarr_c<T> apply_taper(const pyarr_c<T> &img, bool divide) const
      {
      checkArray(img, "img", {nx_dirty, ny_dirty});
      auto pin = img.data();
      auto res = makeArray<T>({nx_dirty, ny_dirty});
      auto pout = res.mutable_data();
      {
      py::gil_scoped_release release;
      if (divide)
        for (size_t i=0; i<nx_dirty; ++i)
          for (size_t j=0; j<ny_dirty; ++j)
            pout[ny_dirty*i + j] = pin[ny_dirty*i + j]/(cfu[i]*cfv[j]);
      else
        for (size_t i=0; i<nx_dirty; ++i)
          for (size_t j=0; j<ny_dirty; ++j)
            pout[ny_dirty*i + j] = pin[ny_dirty*i + j]*cfu[i]*cfv[j];
      }
      return res;
      }
688
689
    pyarr_c<complex<T>> grid2dirty_c(const pyarr_c<complex<T>> &grid) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
690
691
      checkArray(grid, "grid", {nu, nv});
      auto tmp = makeArray<complex<T>>({nu, nv});
692
693
694
      auto ptmp = tmp.mutable_data();
      pocketfft::c2c({nu,nv},{grid.strides(0),grid.strides(1)},
        {tmp.strides(0), tmp.strides(1)}, {0,1}, pocketfft::BACKWARD,
695
        grid.data(), tmp.mutable_data(), T(1), nthreads);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
696
      auto res = makeArray<complex<T>>({nx_dirty, ny_dirty});
697
      auto pout = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
698
699
      {
      py::gil_scoped_release release;
700
701
702
703
704
705
706
707
708
      for (size_t i=0; i<nx_dirty; ++i)
        for (size_t j=0; j<ny_dirty; ++j)
          {
          size_t i2 = nu-nx_dirty/2+i;
          if (i2>=nu) i2-=nu;
          size_t j2 = nv-ny_dirty/2+j;
          if (j2>=nv) j2-=nv;
          pout[ny_dirty*i + j] = ptmp[nv*i2+j2]*cfu[i]*cfv[j];
          }
Martin Reinecke's avatar
Martin Reinecke committed
709
      }
710
711
      return res;
      }
712

Martin Reinecke's avatar
Martin Reinecke committed
713
    pyarr_c<T> dirty2grid(const pyarr_c<T> &dirty) const
Martin Reinecke's avatar
updates    
Martin Reinecke committed
714
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
715
      checkArray(dirty, "dirty", {nx_dirty, ny_dirty});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
716
      auto pdirty = dirty.data();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
717
      auto tmp = makeArray<T>({nu, nv});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
718
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
719
720
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
721
722
723
724
725
726
727
728
729
730
731
      for (size_t i=0; i<nu*nv; ++i)
        ptmp[i] = 0.;
      for (size_t i=0; i<nx_dirty; ++i)
        for (size_t j=0; j<ny_dirty; ++j)
          {
          size_t i2 = nu-nx_dirty/2+i;
          if (i2>=nu) i2-=nu;
          size_t j2 = nv-ny_dirty/2+j;
          if (j2>=nv) j2-=nv;
          ptmp[nv*i2+j2] = pdirty[ny_dirty*i + j]*cfu[i]*cfv[j];
          }
Martin Reinecke's avatar
Martin Reinecke committed
732
      }
Martin Reinecke's avatar
Martin Reinecke committed
733
      hartley2_2D<T>(tmp, tmp);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
734
735
      return tmp;
      }
736
737
    pyarr_c<complex<T>> dirty2grid_c(const pyarr_c<complex<T>> &dirty) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
738
      checkArray(dirty, "dirty", {nx_dirty, ny_dirty});
739
      auto pdirty = dirty.data();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
740
      auto tmp = makeArray<complex<T>>({nu, nv});
741
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
742
743
744
      pocketfft::stride_t strides{tmp.strides(0),tmp.strides(1)};
      {
      py::gil_scoped_release release;
745
746
747
748
749
750
751
752
753
754
755
      for (size_t i=0; i<nu*nv; ++i)
        ptmp[i] = 0.;
      for (size_t i=0; i<nx_dirty; ++i)
        for (size_t j=0; j<ny_dirty; ++j)
          {
          size_t i2 = nu-nx_dirty/2+i;
          if (i2>=nu) i2-=nu;
          size_t j2 = nv-ny_dirty/2+j;
          if (j2>=nv) j2-=nv;
          ptmp[nv*i2+j2] = pdirty[ny_dirty*i + j]*cfu[i]*cfv[j];
          }
Martin Reinecke's avatar
Martin Reinecke committed
756
      pocketfft::c2c({nu,nv}, strides, strides, {0,1}, pocketfft::FORWARD,
757
        ptmp, ptmp, T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
758
      }
759
760
      return tmp;
      }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
761
762
    inline void getpix(T u_in, T v_in, T &u, T &v, int &iu0, int &iv0) const
      {
Martin Reinecke's avatar
Martin Reinecke committed
763
      u=fmodulo(u_in*psx, T(1))*nu,
764
765
      iu0 = int(u-supp*0.5 + 1 + nu) - nu;
      if (iu0+supp>nu+nsafe) iu0 = nu+nsafe-supp;
Martin Reinecke's avatar
Martin Reinecke committed
766
      v=fmodulo(v_in*psy, T(1))*nv;
767
768
      iv0 = int(v-supp*0.5 + 1 + nv) - nv;
      if (iv0+supp>nv+nsafe) iv0 = nv+nsafe-supp;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
769
      }
Martin Reinecke's avatar
test1    
Martin Reinecke committed
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
    pyarr_c<complex<T>> apply_wscreen(const pyarr_c<complex<T>> &dirty_, double w, bool adjoint) const
      {
      checkArray(dirty_, "dirty", {nx_dirty, ny_dirty});
      auto dirty = dirty_.data();
      auto res_ = makeArray<complex<T>>({nx_dirty, ny_dirty});
      auto res = res_.mutable_data();
      double x0 = -0.5*nx_dirty*psx,
             y0 = -0.5*ny_dirty*psy;
      {
      py::gil_scoped_release release;
#pragma omp parallel num_threads(nthreads)
{
#pragma omp for schedule(static)
      for (size_t i=0; i<=nx_dirty/2; ++i)
        {
        double fx = x0+i*psx;
        fx *= fx;
        for (size_t j=0; j<=ny_dirty/2; ++j)
          {
          double fy = y0+j*psy;
          auto ws = wscreen(fx, fy*fy, w, adjoint);
Martin Reinecke's avatar
Martin Reinecke committed
791
792
          res[ny_dirty*i+j] = dirty[ny_dirty*i+j]*ws; // lower left
          size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
Martin Reinecke's avatar
Martin Reinecke committed
793
794
          if ((i>0)&&(i<i2))
            {
Martin Reinecke's avatar
Martin Reinecke committed
795
            res[ny_dirty*i2+j] = dirty[ny_dirty*i2+j]*ws; // lower right
Martin Reinecke's avatar
Martin Reinecke committed
796
            if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
797
              res[ny_dirty*i2+j2] = dirty[ny_dirty*i2+j2]*ws; // upper right
Martin Reinecke's avatar
Martin Reinecke committed
798
799
            }
          if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
800
            res[ny_dirty*i+j2] = dirty[ny_dirty*i+j2]*ws; // upper left
Martin Reinecke's avatar
test1    
Martin Reinecke committed
801
802
803
804
805
806
          }
        }
}
      }
    return res_;
    }
Martin Reinecke's avatar
Martin Reinecke committed
807
808
  };

809
template<typename T, typename T2=complex<T>> class Helper
Martin Reinecke's avatar
import  
Martin Reinecke committed
810
  {
Martin Reinecke's avatar
Martin Reinecke committed
811
  private:
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
812
    const GridderConfig<T> &gconf;
813
    int nu, nv, nsafe, supp;
Martin Reinecke's avatar
Martin Reinecke committed
814
    T beta;
815
816
    const T2 *grid_r;
    T2 *grid_w;
817
    int su, sv;
Martin Reinecke's avatar
Martin Reinecke committed
818
819
820
    int iu0, iv0; // start index of the current visibility
    int bu0, bv0; // start index of the current buffer

821
    vector<T2> rbuf, wbuf;
Martin Reinecke's avatar
import  
Martin Reinecke committed
822

Martin Reinecke's avatar
Martin Reinecke committed
823
    void dump() const
Martin Reinecke's avatar
import  
Martin Reinecke committed
824
      {
Martin Reinecke's avatar
Martin Reinecke committed
825
      if (bu0<-nsafe) return; // nothing written into buffer yet
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
826

Martin Reinecke's avatar
merge    
Martin Reinecke committed
827
#pragma omp critical (gridder_writing_to_grid)
Martin Reinecke's avatar
import  
Martin Reinecke committed
828
{
Martin Reinecke's avatar
Martin Reinecke committed
829
830
831
      int idxu = (bu0+nu)%nu;
      int idxv0 = (bv0+nv)%nv;
      for (int iu=0; iu<su; ++iu)
Martin Reinecke's avatar
import  
Martin Reinecke committed
832
833
        {
        int idxv = idxv0;
Martin Reinecke's avatar
Martin Reinecke committed
834
        for (int iv=0; iv<sv; ++iv)
Martin Reinecke's avatar
import  
Martin Reinecke committed
835
          {
836
          grid_w[idxu*nv + idxv] += wbuf[iu*sv + iv];
Martin Reinecke's avatar
import  
Martin Reinecke committed
837
838
839
840
841
842
843
844
845
          if (++idxv>=nv) idxv=0;
          }
        if (++idxu>=nu) idxu=0;
        }
}
      }

    void load()
      {
Martin Reinecke's avatar
Martin Reinecke committed
846
847
848
      int idxu = (bu0+nu)%nu;
      int idxv0 = (bv0+nv)%nv;
      for (int iu=0; iu<su; ++iu)
Martin Reinecke's avatar
import  
Martin Reinecke committed
849
850
        {
        int idxv = idxv0;
Martin Reinecke's avatar
Martin Reinecke committed
851
        for (int iv=0; iv<sv; ++iv)
Martin Reinecke's avatar
import  
Martin Reinecke committed
852
          {
853
          rbuf[iu*sv + iv] = grid_r[idxu*nv + idxv];
Martin Reinecke's avatar
import  
Martin Reinecke committed
854
855
856
857
858
859
860
          if (++idxv>=nv) idxv=0;
          }
        if (++idxu>=nu) idxu=0;
        }
      }

  public:
861
862
    const T2 *p0r;
    T2 *p0w;
Martin Reinecke's avatar
Martin Reinecke committed
863
    vector<T> kernel;
Martin Reinecke's avatar
import  
Martin Reinecke committed
864

865
    Helper(const GridderConfig<T> &gconf_, const T2 *grid_r_, T2 *grid_w_)
866
      : gconf(gconf_), nu(gconf.Nu()), nv(gconf.Nv()), nsafe(gconf.Nsafe()),
867
868
        supp(gconf.Supp()), beta(gconf.Beta()), grid_r(grid_r_),
        grid_w(grid_w_), su(2*nsafe+(1<<logsquare)), sv(2*nsafe+(1<<logsquare)),
869
870
871
        bu0(-1000000), bv0(-1000000),
        rbuf(su*sv*(grid_r!=nullptr),T(0)),
        wbuf(su*sv*(grid_w!=nullptr),T(0)),
872
        kernel(2*supp)
Martin Reinecke's avatar
Martin Reinecke committed
873
      {}
874
875
876
    ~Helper() { if (grid_w) dump(); }

    int lineJump() const { return sv; }
Martin Reinecke's avatar
Martin Reinecke committed
877
878

    void prep(T u_in, T v_in)
Martin Reinecke's avatar
import  
Martin Reinecke committed
879
      {
Martin Reinecke's avatar
Martin Reinecke committed
880
881
      T u, v;
      gconf.getpix(u_in, v_in, u, v, iu0, iv0);
882
883
884
885
      T xsupp=T(2)/supp;
      auto x0 = xsupp*(iu0-u);
      auto y0 = xsupp*(iv0-v);
      for (int i=0; i<supp; ++i)
Martin Reinecke's avatar
Martin Reinecke committed
886
        {
887
        auto x = x0+i*xsupp;
Martin Reinecke's avatar
Martin Reinecke committed
888
        kernel[i  ] = beta*sqrt(T(1)-x*x);
889
890
        auto y = y0+i*xsupp;
        kernel[i+supp] = beta*sqrt(T(1)-y*y);
Martin Reinecke's avatar
Martin Reinecke committed
891
892
893
894
        }
      for (auto &k : kernel)
        k = exp(k);

895
      if ((iu0<bu0) || (iv0<bv0) || (iu0+supp>bu0+su) || (iv0+supp>bv0+sv))
Martin Reinecke's avatar
import  
Martin Reinecke committed
896
        {
897
        if (grid_w) { dump(); fill(wbuf.begin(), wbuf.end(), T(0)); }
Martin Reinecke's avatar
Martin Reinecke committed
898
899
        bu0=((((iu0+nsafe)>>logsquare)<<logsquare))-nsafe;
        bv0=((((iv0+nsafe)>>logsquare)<<logsquare))-nsafe;
900
        if (grid_r) load();
Martin Reinecke's avatar
import  
Martin Reinecke committed
901
        }
902
903
      p0r = grid_r ? rbuf.data() + sv*(iu0-bu0) + iv0-bv0 : nullptr;
      p0w = grid_w ? wbuf.data() + sv*(iu0-bu0) + iv0-bv0 : nullptr;
Martin Reinecke's avatar
import  
Martin Reinecke committed
904
905
906
      }
  };

907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
constexpr auto vis2grid_c_DS = R"""(
Grids visibilities onto a UV grid

Parameters
==========
baselines: Baselines
    the Baselines object
gconf: GridderConf
    the GridderConf object to be used
    (used to optimize the ordering of the indices)
idx: np.array((nvis,), dtype=np.uint32)
    the indices for the entries to be gridded
vis: np.array((nvis,), dtype=np.complex)
    The visibility data for the index array
grid_in: np.array((nu,nv), dtype=np.complex128), optional
    If present, the result is added to this array.
Martin Reinecke's avatar
Martin Reinecke committed
923
924
wgt: np.array((nvis,), dtype=np.float64), optional
    If present, visibilities are multiplied by the corresponding entries.
925
926
927
928
929
930

Returns
=======
np.array((nu,nv), dtype=np.complex128):
    the gridded visibilities
)""";
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
931
932
template<typename T> pyarr_c<complex<T>> vis2grid_c(
  const Baselines<T> &baselines, const GridderConfig<T> &gconf,
Martin Reinecke's avatar
Martin Reinecke committed
933
  const pyarr<uint32_t> &idx_, const pyarr<complex<T>> &vis_,
Martin Reinecke's avatar
Martin Reinecke committed
934
  py::object &grid_in, const py::object &wgt_)
935
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
936
937
938
  checkArray(vis_, "vis", {0});
  size_t nvis = size_t(vis_.shape(0));
  checkArray(idx_, "idx", {nvis});
939
940
  auto vis=vis_.template unchecked<1>();
  auto idx = idx_.template unchecked<1>();
Martin Reinecke's avatar
Martin Reinecke committed
941
942
943
  pyarr<T> wgt2 = providePotentialArray<T>(wgt_, {nvis});
  bool have_wgt = wgt2.size()>0;
  auto wgt = wgt2.template unchecked<1>();
944

945
  size_t nu=gconf.Nu(), nv=gconf.Nv();
Martin Reinecke's avatar
Martin Reinecke committed
946
  auto res = provideCArray<complex<T>>(grid_in, {nu, nv});
947
  auto grid = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
948
949
  {
  py::gil_scoped_release release;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
950
  T beta = gconf.Beta();
951
  size_t supp = gconf.Supp();
952

953
#pragma omp parallel num_threads(nthreads)
954
{
955
  Helper<T> hlp(gconf, nullptr, grid);
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
956
  T emb = exp(-2*beta);
957
  int jump = hlp.lineJump();
Martin Reinecke's avatar
Martin Reinecke committed
958
  const T * ku = hlp.kernel.data();
959
  const T * kv = hlp.kernel.data()+supp;
960
961
962
963
964

  // Loop over sampling points
#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
965
    UVW<T> coord = baselines.effectiveCoord(idx(ipart));
Martin Reinecke's avatar
Martin Reinecke committed
966
    hlp.prep(coord.u, coord.v);
Martin Reinecke's avatar
Martin Reinecke committed
967
    auto * ptr = hlp.p0w;
968
    auto v(vis(ipart)*emb);
Martin Reinecke's avatar
Martin Reinecke committed
969
970
    if (have_wgt)
      v*=wgt(ipart);
971
    for (size_t cu=0; cu<supp; ++cu)
972
973
      {
      complex<T> tmp(v*ku[cu]);
974
      for (size_t cv=0; cv<supp; ++cv)
975
        ptr[cv] += tmp*kv[cv];
976
      ptr+=jump;
977
978
979
      }
    }
} // end of parallel region
Martin Reinecke's avatar
Martin Reinecke committed
980
  }
981
982
983
  return res;
  }

984
985
986
987
988
989
990
991
992
993
994
995
996
997
constexpr auto vis2grid_DS = R"""(
Grids visibilities onto a UV grid

Parameters
==========
baselines: Baselines
    the Baselines object
gconf: GridderConf
    the GridderConf object to be used
    (used to optimize the ordering of the indices)
idx: np.array((nvis,), dtype=np.uint32)
    the indices for the entries to be gridded
vis: np.array((nvis,), dtype=np.complex)
    The visibility data for the index array
Martin Reinecke's avatar
Martin Reinecke committed
998
999
grid_in: np.array((nu,nv), dtype=np.float64), optional
    If present, the result is added to this array.
Martin Reinecke's avatar
Martin Reinecke committed
1000
1001
wgt: np.array((nvis,), dtype=np.float64), optional
    If present, visibilities are multiplied by the corresponding entries.
1002
1003
1004
1005
1006
1007

Returns
=======
np.array((nu,nv), dtype=np.float64):
    the gridded visibilities (made real by making use of Hermitian symmetry)
)""";
Martin Reinecke's avatar
Martin Reinecke committed
1008
template<typename T> pyarr_c<T> vis2grid(const Baselines<T> &baselines,
1009
  const GridderConfig<T> &gconf, const pyarr<uint32_t> &idx_,
Martin Reinecke's avatar
Martin Reinecke committed
1010
1011
  const pyarr<complex<T>> &vis_, py::object &grid_in, const py::object &wgt_)
  { return complex2hartley(vis2grid_c(baselines, gconf, idx_, vis_, None, wgt_), grid_in); }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
1012

1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
constexpr auto ms2grid_c_DS = R"""(
Grids measurement set data onto a UV grid

Parameters
==========
baselines: Baselines
    the Baselines object
gconf: GridderConf
    the GridderConf object to be used
    (used to optimize the ordering of the indices)
idx: np.array((nvis,), dtype=np.uint32)
    the indices for the entries to be gridded
ms: np.array((nrows, nchannels), dtype=np.complex128)
    the measurement set.
grid_in: np.array((nu,nv), dtype=np.complex128), optional
    If present, the result is added to this array.
Martin Reinecke's avatar
Martin Reinecke committed
1029
1030
wgt: np.array((nrows, nchannels), dtype=np.float64), optional
    If present, visibilities are multiplied by the corresponding entries.
1031
1032
1033
1034
1035
1036

Returns
=======
np.array((nu,nv), dtype=np.complex128):
    the gridded visibilities
)""";
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
1037
1038
template<typename T> pyarr_c<complex<T>> ms2grid_c(
  const Baselines<T> &baselines, const GridderConfig<T> &gconf,
Martin Reinecke's avatar
Martin Reinecke committed
1039
  const pyarr<uint32_t> &idx_, const pyarr<complex<T>> &ms_,
Martin Reinecke's avatar
Martin Reinecke committed
1040
  py::object &grid_in, const py::object &wgt_)
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1041
1042
1043
1044
1045
1046
  {
  auto nrows = baselines.Nrows();
  auto nchan = baselines.Nchannels();
  checkArray(ms_, "ms", {nrows, nchan});
  checkArray(idx_, "idx", {0});
  size_t nvis = size_t(idx_.shape(0));
1047
1048
  auto ms = ms_.template unchecked<2>();
  auto idx = idx_.template unchecked<1>();
Martin Reinecke's avatar
Martin Reinecke committed
1049
1050
1051
  auto wgt2 = providePotentialArray<T>(wgt_, {nrows, nchan});
  bool have_wgt = wgt2.size()>0;
  auto wgt = wgt2.template unchecked<2>();
Simon Perkins's avatar
Simon Perkins committed
1052

Martin Reinecke's avatar
merge    
Martin Reinecke committed
1053
  size_t nu=gconf.Nu(), nv=gconf.Nv();
Martin Reinecke's avatar
Martin Reinecke committed
1054
  auto res = provideCArray<complex<T>>(grid_in, {nu, nv});
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1055
  auto grid = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
1056
1057
  {
  py::gil_scoped_release release;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1058
  T beta = gconf.Beta();
1059
  size_t supp = gconf.Supp();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1060

1061
#pragma omp parallel num_threads(nthreads)
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1062
{
1063
  Helper<T> hlp(gconf, nullptr, grid);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1064
  T emb = exp(-2*beta);
1065
  int jump = hlp.lineJump();
Martin Reinecke's avatar
Martin Reinecke committed
1066
  const T * ku = hlp.kernel.data();