nifty_gridder.cc 45.8 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

Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
31
#ifdef __GNUC__
32
33
//#define RESTRICT __restrict__
#define RESTRICT
Martin Reinecke's avatar
Martin Reinecke committed
34
#define NOINLINE __attribute__ ((noinline))
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
35
36
37
38
#else
#define RESTRICT
#endif

Martin Reinecke's avatar
import  
Martin Reinecke committed
39
40
41
42
43
44
using namespace std;

namespace py = pybind11;

namespace {

Martin Reinecke's avatar
Martin Reinecke committed
45
46
auto None = py::none();

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
47
48
49
50
//
// basic utilities
//

51
52
53
54
55
56
void myassert(bool cond, const char *msg)
  {
  if (cond) return;
  throw runtime_error(msg);
  }

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
57
58
59
60
61
62
63
64
65
66
67
/*! 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;
  }

68
69
70
71
72
73
74
75
76
77
78
79
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_)
  {
80
  myassert(nthreads_>=1, "nthreads must be >= 1");
81
82
83
84
85
86
87
88
89
90
91
  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
92
  { return nthreads; }
93

Martin Reinecke's avatar
Martin Reinecke committed
94
95
96
97
//
// Utilities for Gauss-Legendre quadrature
//

Martin Reinecke's avatar
Martin Reinecke committed
98
99
100
101
102
103
104
105
106
107
108
109
110
111
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.);

112
#pragma omp parallel num_threads(nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
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
138
139
140
141
142
143
144
145
{
  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;
146
      myassert(++j<100, "convergence problem");
Martin Reinecke's avatar
Martin Reinecke committed
147
148
149
150
151
152
153
154
      }

    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
155
156
157
158
//
// Start of real gridder functionality
//

159
160
template<typename T>
  using pyarr = py::array_t<T>;
161
// The "_c" suffix here stands for "C memory order, contiguous"
162
163
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
164

Martin Reinecke's avatar
merge    
Martin Reinecke committed
165
template<typename T> pyarr_c<T> makeArray(const vector<size_t> &shape)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
166
167
  { return pyarr_c<T>(shape); }

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
168
size_t get_w(double epsilon)
Martin Reinecke's avatar
Martin Reinecke committed
169
170
171
172
173
174
175
176
177
178
179
  {
  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
180

Martin Reinecke's avatar
merge    
Martin Reinecke committed
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
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
199
template<typename T> pyarr<T> provideArray(const py::object &in,
Martin Reinecke's avatar
merge    
Martin Reinecke committed
200
201
  const vector<size_t> &shape)
  {
202
  if (in.is_none())
Martin Reinecke's avatar
merge    
Martin Reinecke committed
203
204
205
206
207
208
209
210
    {
    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_;
    }
211
  auto tmp_ = in.cast<pyarr<T>>();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
212
213
214
215
  checkArray(tmp_, "temporary", shape);
  return tmp_;
  }

Martin Reinecke's avatar
Martin Reinecke committed
216
217
218
219
220
221
222
223
224
225
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
226
template<typename T> pyarr_c<T> provideCArray(py::object &in,
Martin Reinecke's avatar
merge    
Martin Reinecke committed
227
228
  const vector<size_t> &shape)
  {
229
  if (in.is_none())
Martin Reinecke's avatar
merge    
Martin Reinecke committed
230
231
232
233
234
235
236
237
238
239
240
241
242
    {
    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
243
template<typename T> pyarr_c<T> complex2hartley
244
  (const pyarr_c<complex<T>> &grid_, py::object &grid_in)
Martin Reinecke's avatar
Martin Reinecke committed
245
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
246
  checkArray(grid_, "grid", {0,0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
247
  size_t nu = size_t(grid_.shape(0)), nv = size_t(grid_.shape(1));
Martin Reinecke's avatar
Martin Reinecke committed
248
249
  auto grid = grid_.data();

250
  auto res = provideCArray<T>(grid_in, {nu, nv});
Martin Reinecke's avatar
Martin Reinecke committed
251
  auto grid2 = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
252
253
  {
  py::gil_scoped_release release;
254
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
255
  for (size_t u=0; u<nu; ++u)
Martin Reinecke's avatar
Martin Reinecke committed
256
    {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
257
258
    size_t xu = (u==0) ? 0 : nu-u;
    for (size_t v=0; v<nv; ++v)
Martin Reinecke's avatar
Martin Reinecke committed
259
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
260
261
262
      size_t xv = (v==0) ? 0 : nv-v;
      size_t i1 = u*nv+v;
      size_t i2 = xu*nv+xv;
263
264
      grid2[i1] += T(0.5)*(grid[i1].real()+grid[i1].imag()+
                           grid[i2].real()-grid[i2].imag());
Martin Reinecke's avatar
Martin Reinecke committed
265
266
      }
    }
Martin Reinecke's avatar
Martin Reinecke committed
267
  }
Martin Reinecke's avatar
Martin Reinecke committed
268
269
270
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
271
272
template<typename T> pyarr_c<complex<T>> hartley2complex
  (const pyarr_c<T> &grid_)
Martin Reinecke's avatar
Martin Reinecke committed
273
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
274
  checkArray(grid_, "grid", {0, 0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
275
  size_t nu = size_t(grid_.shape(0)), nv = size_t(grid_.shape(1));
Martin Reinecke's avatar
Martin Reinecke committed
276
277
  auto grid = grid_.data();

Martin Reinecke's avatar
merge    
Martin Reinecke committed
278
  auto res=makeArray<complex<T>>({nu, nv});
Martin Reinecke's avatar
Martin Reinecke committed
279
  auto grid2 = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
280
281
  {
  py::gil_scoped_release release;
282
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
283
  for (size_t u=0; u<nu; ++u)
Martin Reinecke's avatar
Martin Reinecke committed
284
    {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
285
286
    size_t xu = (u==0) ? 0 : nu-u;
    for (size_t v=0; v<nv; ++v)
Martin Reinecke's avatar
Martin Reinecke committed
287
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
288
289
290
      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
291
292
293
      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
294
295
      }
    }
Martin Reinecke's avatar
Martin Reinecke committed
296
  }
Martin Reinecke's avatar
Martin Reinecke committed
297
298
299
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
300
301
302
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
303
304
305
  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
306
  auto ptmp = out.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
307
308
  {
  py::gil_scoped_release release;
309
310
311
  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
312
313
314
315
316
317
318
319
320
321
322
323
324
  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
325
  }
Martin Reinecke's avatar
Martin Reinecke committed
326

Martin Reinecke's avatar
Martin Reinecke committed
327
328
/* Compute correction factors for the ES gridding kernel
   This implementation follows eqs. (3.8) to (3.10) of Barnett et al. 2018 */
Martin Reinecke's avatar
updates    
Martin Reinecke committed
329
vector<double> correction_factors (size_t n, size_t nval, size_t w)
Martin Reinecke's avatar
Martin Reinecke committed
330
331
332
333
334
335
336
337
338
339
  {
  constexpr double pi = 3.141592653589793238462643383279502884197;
  auto beta = 2.3*w;
  auto p = int(1.5*w+2);
  double alpha = pi*w/n;
  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
340
  vector<double> res(nval);
341
#pragma omp parallel for schedule(static) num_threads(nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
342
343
344
345
346
  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]);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
347
    res[k] = 1./(w*tmp);
Martin Reinecke's avatar
Martin Reinecke committed
348
349
350
351
    }
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
352
template<typename T> struct UVW
Martin Reinecke's avatar
updates    
Martin Reinecke committed
353
  {
Martin Reinecke's avatar
Martin Reinecke committed
354
  T u, v, w;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
355
  UVW () {}
Martin Reinecke's avatar
Martin Reinecke committed
356
357
  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
358
359
    { return UVW(u*fct, v*fct, w*fct); }
  };
Martin Reinecke's avatar
Martin Reinecke committed
360

361
362
363
364
365
366
367
368
369
370
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
371
template<typename T> class Baselines
Martin Reinecke's avatar
Martin Reinecke committed
372
373
  {
  private:
Martin Reinecke's avatar
Martin Reinecke committed
374
    vector<UVW<T>> coord;
375
    vector<T> f_over_c;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
376
    size_t nrows, nchan;
Martin Reinecke's avatar
Martin Reinecke committed
377
378

  public:
379
    Baselines(const pyarr<T> &coord_, const pyarr<T> &freq_)
Martin Reinecke's avatar
Martin Reinecke committed
380
      {
381
      constexpr double speedOfLight = 299792458.;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
382
      checkArray(coord_, "coord", {0, 3});
383
      checkArray(freq_, "freq", {0});
Martin Reinecke's avatar
Martin Reinecke committed
384
      nrows = coord_.shape(0);
385
      nchan = freq_.shape(0);
Martin Reinecke's avatar
Martin Reinecke committed
386
      myassert(nrows*nchan<(size_t(1)<<32), "too many entries in MS");
387
388
      auto freq = freq_.template unchecked<1>();
      auto cood = coord_.template unchecked<2>();
Martin Reinecke's avatar
Martin Reinecke committed
389
390
      {
      py::gil_scoped_release release;
391
      f_over_c.resize(nchan);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
392
      for (size_t i=0; i<nchan; ++i)
393
        f_over_c[i] = freq(i)/speedOfLight;
Martin Reinecke's avatar
Martin Reinecke committed
394
395
      coord.resize(nrows);
      for (size_t i=0; i<coord.size(); ++i)
396
        coord[i] = UVW<T>(cood(i,0), cood(i,1), cood(i,2));
Martin Reinecke's avatar
Martin Reinecke committed
397
      }
Martin Reinecke's avatar
Martin Reinecke committed
398
399
      }

Martin Reinecke's avatar
Martin Reinecke committed
400
    UVW<T> effectiveCoord(uint32_t index) const
401
402
403
      {
      size_t irow = index/nchan;
      size_t ichan = index-nchan*irow;
404
      return coord[irow]*f_over_c[ichan];
405
406
      }
    UVW<T> effectiveCoord(size_t irow, size_t ichan) const
407
      { return coord[irow]*f_over_c[ichan]; }
Martin Reinecke's avatar
Martin Reinecke committed
408
    size_t Nrows() const { return nrows; }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
409
410
    size_t Nchannels() const { return nchan; }

411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
    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
    )""";
426

Philipp Arras's avatar
Philipp Arras committed
427
    pyarr_c<T> effectiveuvw(const pyarr_c<uint32_t> &idx_) const
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
    {
      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_;
    }

444
    template<typename T2> pyarr_c<T2> ms2vis(const pyarr<T2> &ms_,
Martin Reinecke's avatar
updates    
Martin Reinecke committed
445
446
      const pyarr_c<uint32_t> &idx_) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
447
448
      checkArray(idx_, "idx", {0});
      checkArray(ms_, "ms", {nrows, nchan});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
449
      size_t nvis = size_t(idx_.shape(0));
450
451
      auto idx = idx_.template unchecked<1>();
      auto ms = ms_.template unchecked<2>();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
452

Martin Reinecke's avatar
merge    
Martin Reinecke committed
453
      auto res=makeArray<T2>({nvis});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
454
      auto vis = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
455
456
      {
      py::gil_scoped_release release;
457
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
458
      for (size_t i=0; i<nvis; ++i)
459
460
461
462
463
464
        {
        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
465
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
466
467
468
      return res;
      }

469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
    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)
    )""";
486
487
    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
488
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
489
      checkArray(vis_, "vis", {0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
490
      size_t nvis = size_t(vis_.shape(0));
Martin Reinecke's avatar
merge    
Martin Reinecke committed
491
      checkArray(idx_, "idx", {nvis});
492
493
      auto idx = idx_.template unchecked<1>();
      auto vis = vis_.template unchecked<1>();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
494

Martin Reinecke's avatar
merge    
Martin Reinecke committed
495
      auto res = provideArray<T2>(ms_in, {nrows, nchan});
496
      auto ms = res.template mutable_unchecked<2>();
Martin Reinecke's avatar
Martin Reinecke committed
497
498
      {
      py::gil_scoped_release release;
499
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
500
      for (size_t i=0; i<nvis; ++i)
501
502
503
504
        {
        auto t = idx(i);
        auto row = t/nchan;
        auto chan = t-row*nchan;
Martin Reinecke's avatar
Martin Reinecke committed
505
        ms(row, chan) += vis(i);
506
        }
Martin Reinecke's avatar
Martin Reinecke committed
507
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
508
509
      return res;
      }
Martin Reinecke's avatar
Martin Reinecke committed
510
511
  };

Martin Reinecke's avatar
Martin Reinecke committed
512
513
constexpr int logsquare=4;

514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
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
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
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
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
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
)""";

576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
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
593
template<typename T> class GridderConfig
Martin Reinecke's avatar
Martin Reinecke committed
594
595
596
  {
  private:
    size_t nx_dirty, ny_dirty;
Martin Reinecke's avatar
Martin Reinecke committed
597
    double eps, psx, psy;
Martin Reinecke's avatar
Martin Reinecke committed
598
    size_t w, nsafe, nu, nv;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
599
    T beta;
Martin Reinecke's avatar
Martin Reinecke committed
600
    vector<T> cfu, cfv;
Martin Reinecke's avatar
Martin Reinecke committed
601

Martin Reinecke's avatar
Martin Reinecke committed
602
603
604
605
606
607
608
609
610
    complex<T> wscreen(double x, double y, double w, bool adjoint) const
      {
      constexpr double pi = 3.141592653589793238462643383279502884197;
      double n = cos(sqrt(x+y)), xn = 1./n;
      double phase = 2*pi*w*(n-1);
      if (adjoint) phase *= -1;
      return complex<T>(cos(phase)*xn, sin(phase)*xn);
      }

Martin Reinecke's avatar
Martin Reinecke committed
611
612
  public:
    GridderConfig(size_t nxdirty, size_t nydirty, double epsilon,
Martin Reinecke's avatar
Martin Reinecke committed
613
      double pixsize_x, double pixsize_y)
Martin Reinecke's avatar
Martin Reinecke committed
614
615
      : nx_dirty(nxdirty), ny_dirty(nydirty), eps(epsilon),
        psx(pixsize_x), psy(pixsize_y),
Martin Reinecke's avatar
Martin Reinecke committed
616
617
        w(get_w(epsilon)), nsafe((w+1)/2),
        nu(max(2*nsafe,2*nx_dirty)), nv(max(2*nsafe,2*ny_dirty)),
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
618
        beta(2.3*w),
Martin Reinecke's avatar
updates    
Martin Reinecke committed
619
        cfu(nx_dirty), cfv(ny_dirty)
Martin Reinecke's avatar
Martin Reinecke committed
620
      {
Martin Reinecke's avatar
Martin Reinecke committed
621
622
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
Martin Reinecke committed
623
624
625
      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
626
627
      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
628
629
630
631
632
633
634
635
636
637
638

      auto tmp = correction_factors(nu, nx_dirty/2+1, w);
      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];
      tmp = correction_factors(nv, ny_dirty/2+1, w);
      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
639
      }
Martin Reinecke's avatar
Martin Reinecke committed
640
      }
Martin Reinecke's avatar
Martin Reinecke committed
641
642
643
644
645
    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
646
647
648
    size_t Nu() const { return nu; }
    size_t Nv() const { return nv; }
    size_t W() const { return w; }
649
    size_t Nsafe() const { return nsafe; }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
650
    T Beta() const { return beta; }
651

Martin Reinecke's avatar
Martin Reinecke committed
652
    pyarr_c<T> grid2dirty(const pyarr_c<T> &grid) const
Martin Reinecke's avatar
Martin Reinecke committed
653
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
654
655
      checkArray(grid, "grid", {nu, nv});
      auto tmp = makeArray<T>({nu, nv});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
656
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
657
      hartley2_2D<T>(grid, tmp);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
658
      auto res = makeArray<T>({nx_dirty, ny_dirty});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
659
      auto pout = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
660
661
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
662
663
664
665
666
667
668
669
670
      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
671
      }
Martin Reinecke's avatar
Martin Reinecke committed
672
673
      return res;
      }
Martin Reinecke's avatar
Martin Reinecke committed
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
    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;
      }
693
694
    pyarr_c<complex<T>> grid2dirty_c(const pyarr_c<complex<T>> &grid) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
695
696
      checkArray(grid, "grid", {nu, nv});
      auto tmp = makeArray<complex<T>>({nu, nv});
697
698
699
      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,
700
        grid.data(), tmp.mutable_data(), T(1), nthreads);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
701
      auto res = makeArray<complex<T>>({nx_dirty, ny_dirty});
702
      auto pout = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
703
704
      {
      py::gil_scoped_release release;
705
706
707
708
709
710
711
712
713
      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
714
      }
715
716
      return res;
      }
717

Martin Reinecke's avatar
Martin Reinecke committed
718
    pyarr_c<T> dirty2grid(const pyarr_c<T> &dirty) const
Martin Reinecke's avatar
updates    
Martin Reinecke committed
719
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
720
      checkArray(dirty, "dirty", {nx_dirty, ny_dirty});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
721
      auto pdirty = dirty.data();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
722
      auto tmp = makeArray<T>({nu, nv});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
723
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
724
725
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
726
727
728
729
730
731
732
733
734
735
736
      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
737
      }
Martin Reinecke's avatar
Martin Reinecke committed
738
      hartley2_2D<T>(tmp, tmp);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
739
740
      return tmp;
      }
741
742
    pyarr_c<complex<T>> dirty2grid_c(const pyarr_c<complex<T>> &dirty) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
743
      checkArray(dirty, "dirty", {nx_dirty, ny_dirty});
744
      auto pdirty = dirty.data();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
745
      auto tmp = makeArray<complex<T>>({nu, nv});
746
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
747
748
749
      pocketfft::stride_t strides{tmp.strides(0),tmp.strides(1)};
      {
      py::gil_scoped_release release;
750
751
752
753
754
755
756
757
758
759
760
      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
761
      pocketfft::c2c({nu,nv}, strides, strides, {0,1}, pocketfft::FORWARD,
762
        ptmp, ptmp, T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
763
      }
764
765
      return tmp;
      }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
766
767
    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
768
      u=fmodulo(u_in*psx, T(1))*nu,
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
769
770
      iu0 = int(u-w*0.5 + 1 + nu) - nu;
      if (iu0+w>nu+nsafe) iu0 = nu+nsafe-w;
Martin Reinecke's avatar
Martin Reinecke committed
771
      v=fmodulo(v_in*psy, T(1))*nv;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
772
773
774
      iv0 = int(v-w*0.5 + 1 + nv) - nv;
      if (iv0+w>nv+nsafe) iv0 = nv+nsafe-w;
      }
Martin Reinecke's avatar
test1    
Martin Reinecke committed
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
    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
796
797
          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
798
799
          if ((i>0)&&(i<i2))
            {
Martin Reinecke's avatar
Martin Reinecke committed
800
            res[ny_dirty*i2+j] = dirty[ny_dirty*i2+j]*ws; // lower right
Martin Reinecke's avatar
Martin Reinecke committed
801
            if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
802
              res[ny_dirty*i2+j2] = dirty[ny_dirty*i2+j2]*ws; // upper right
Martin Reinecke's avatar
Martin Reinecke committed
803
804
            }
          if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
805
            res[ny_dirty*i+j2] = dirty[ny_dirty*i+j2]*ws; // upper left
Martin Reinecke's avatar
test1    
Martin Reinecke committed
806
807
808
809
810
811
          }
        }
}
      }
    return res_;
    }
Martin Reinecke's avatar
Martin Reinecke committed
812
813
  };

814
template<typename T, typename T2=complex<T>> class Helper
Martin Reinecke's avatar
import  
Martin Reinecke committed
815
  {
Martin Reinecke's avatar
Martin Reinecke committed
816
  private:
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
817
    const GridderConfig<T> &gconf;
Martin Reinecke's avatar
Martin Reinecke committed
818
819
    int nu, nv, nsafe, w;
    T beta;
820
821
    const T2 *grid_r;
    T2 *grid_w;
822
    int su, sv;
Martin Reinecke's avatar
Martin Reinecke committed
823
824
825
    int iu0, iv0; // start index of the current visibility
    int bu0, bv0; // start index of the current buffer

826
    vector<T2> rbuf, wbuf;
Martin Reinecke's avatar
import  
Martin Reinecke committed
827

Martin Reinecke's avatar
Martin Reinecke committed
828
    void dump() const
Martin Reinecke's avatar
import  
Martin Reinecke committed
829
      {
Martin Reinecke's avatar
Martin Reinecke committed
830
      if (bu0<-nsafe) return; // nothing written into buffer yet
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
831

Martin Reinecke's avatar
merge    
Martin Reinecke committed
832
#pragma omp critical (gridder_writing_to_grid)
Martin Reinecke's avatar
import  
Martin Reinecke committed
833
{
Martin Reinecke's avatar
Martin Reinecke committed
834
835
836
      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
837
838
        {
        int idxv = idxv0;
Martin Reinecke's avatar
Martin Reinecke committed
839
        for (int iv=0; iv<sv; ++iv)
Martin Reinecke's avatar
import  
Martin Reinecke committed
840
          {
841
          grid_w[idxu*nv + idxv] += wbuf[iu*sv + iv];
Martin Reinecke's avatar
import  
Martin Reinecke committed
842
843
844
845
846
847
848
849
850
          if (++idxv>=nv) idxv=0;
          }
        if (++idxu>=nu) idxu=0;
        }
}
      }

    void load()
      {
Martin Reinecke's avatar
Martin Reinecke committed
851
852
853
      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
854
855
        {
        int idxv = idxv0;
Martin Reinecke's avatar
Martin Reinecke committed
856
        for (int iv=0; iv<sv; ++iv)
Martin Reinecke's avatar
import  
Martin Reinecke committed
857
          {
858
          rbuf[iu*sv + iv] = grid_r[idxu*nv + idxv];
Martin Reinecke's avatar
import  
Martin Reinecke committed
859
860
861
862
863
864
865
          if (++idxv>=nv) idxv=0;
          }
        if (++idxu>=nu) idxu=0;
        }
      }

  public:
866
867
    const T2 *p0r;
    T2 *p0w;
Martin Reinecke's avatar
Martin Reinecke committed
868
    vector<T> kernel;
Martin Reinecke's avatar
import  
Martin Reinecke committed
869

870
    Helper(const GridderConfig<T> &gconf_, const T2 *grid_r_, T2 *grid_w_)
871
872
      : gconf(gconf_), nu(gconf.Nu()), nv(gconf.Nv()), nsafe(gconf.Nsafe()),
        w(gconf.W()), beta(gconf.Beta()), grid_r(grid_r_), grid_w(grid_w_),
Martin Reinecke's avatar
Martin Reinecke committed
873
        su(2*nsafe+(1<<logsquare)), sv(2*nsafe+(1<<logsquare)),
874
875
876
877
        bu0(-1000000), bv0(-1000000),
        rbuf(su*sv*(grid_r!=nullptr),T(0)),
        wbuf(su*sv*(grid_w!=nullptr),T(0)),
        kernel(2*w)
Martin Reinecke's avatar
Martin Reinecke committed
878
      {}
879
880
881
    ~Helper() { if (grid_w) dump(); }

    int lineJump() const { return sv; }
Martin Reinecke's avatar
Martin Reinecke committed
882
883

    void prep(T u_in, T v_in)
Martin Reinecke's avatar
import  
Martin Reinecke committed
884
      {
Martin Reinecke's avatar
Martin Reinecke committed
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
      T u, v;
      gconf.getpix(u_in, v_in, u, v, iu0, iv0);
      T xw=T(2)/w;
      auto x0 = xw*(iu0-u);
      auto y0 = xw*(iv0-v);
      for (int i=0; i<w; ++i)
        {
        auto x = x0+i*xw;
        kernel[i  ] = beta*sqrt(T(1)-x*x);
        auto y = y0+i*xw;
        kernel[i+w] = beta*sqrt(T(1)-y*y);
        }
      for (auto &k : kernel)
        k = exp(k);

      if ((iu0<bu0) || (iv0<bv0) || (iu0+w>bu0+su) || (iv0+w>bv0+sv))
Martin Reinecke's avatar
import  
Martin Reinecke committed
901
        {
902
        if (grid_w) { dump(); fill(wbuf.begin(), wbuf.end(), T(0)); }
Martin Reinecke's avatar
Martin Reinecke committed
903
904
        bu0=((((iu0+nsafe)>>logsquare)<<logsquare))-nsafe;
        bv0=((((iv0+nsafe)>>logsquare)<<logsquare))-nsafe;
905
        if (grid_r) load();
Martin Reinecke's avatar
import  
Martin Reinecke committed
906
        }
907
908
      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
909
910
911
      }
  };

912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
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
928
929
wgt: np.array((nvis,), dtype=np.float64), optional
    If present, visibilities are multiplied by the corresponding entries.
930
931
932
933
934
935

Returns
=======
np.array((nu,nv), dtype=np.complex128):
    the gridded visibilities
)""";
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
936
937
template<typename T> pyarr_c<complex<T>> vis2grid_c(
  const Baselines<T> &baselines, const GridderConfig<T> &gconf,
Martin Reinecke's avatar
Martin Reinecke committed
938
  const pyarr<uint32_t> &idx_, const pyarr<complex<T>> &vis_,
Martin Reinecke's avatar
Martin Reinecke committed
939
  py::object &grid_in, const py::object &wgt_)
940
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
941
942
943
  checkArray(vis_, "vis", {0});
  size_t nvis = size_t(vis_.shape(0));
  checkArray(idx_, "idx", {nvis});
944
945
  auto vis=vis_.template unchecked<1>();
  auto idx = idx_.template unchecked<1>();
Martin Reinecke's avatar
Martin Reinecke committed
946
947
948
  pyarr<T> wgt2 = providePotentialArray<T>(wgt_, {nvis});
  bool have_wgt = wgt2.size()>0;
  auto wgt = wgt2.template unchecked<1>();
949

950
  size_t nu=gconf.Nu(), nv=gconf.Nv();
Martin Reinecke's avatar
Martin Reinecke committed
951
  auto res = provideCArray<complex<T>>(grid_in, {nu, nv});
952
  auto grid = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
953
954
  {
  py::gil_scoped_release release;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
955
956
  T beta = gconf.Beta();
  size_t w = gconf.W();
957

958
#pragma omp parallel num_threads(nthreads)
959
{
960
  Helper<T> hlp(gconf, nullptr, grid);
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
961
  T emb = exp(-2*beta);
962
  int jump = hlp.lineJump();
963
  const T * RESTRICT ku = hlp.kernel.data();
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
964
  const T * RESTRICT kv = hlp.kernel.data()+w;
965
966
967
968
969

  // Loop over sampling points
#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
970
    UVW<T> coord = baselines.effectiveCoord(idx(ipart));
Martin Reinecke's avatar
Martin Reinecke committed
971
    hlp.prep(coord.u, coord.v);
972
    auto * RESTRICT ptr = hlp.p0w;
973
    auto v(vis(ipart)*emb);
Martin Reinecke's avatar
Martin Reinecke committed
974
975
    if (have_wgt)
      v*=wgt(ipart);
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
976
    for (size_t cu=0; cu<w; ++cu)
977
978
      {
      complex<T> tmp(v*ku[cu]);
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
979
      for (size_t cv=0; cv<w; ++cv)
980
        ptr[cv] += tmp*kv[cv];
981
      ptr+=jump;
982
983
984
      }
    }
} // end of parallel region
Martin Reinecke's avatar
Martin Reinecke committed
985
  }
986
987
988
  return res;
  }

989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
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
1003
1004
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
1005
1006
wgt: np.array((nvis,), dtype=np.float64), optional
    If present, visibilities are multiplied by the corresponding entries.
1007
1008
1009
1010
1011
1012

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
1013
template<typename T> pyarr_c<T> vis2grid(const Baselines<T> &baselines,
1014
  const GridderConfig<T> &gconf, const pyarr<uint32_t> &idx_,
Martin Reinecke's avatar
Martin Reinecke committed
1015
1016
  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
1017

1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
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
1034
1035
wgt: np.array((nrows, nchannels), dtype=np.float64), optional
    If present, visibilities are multiplied by the corresponding entries.
1036
1037
1038
1039
1040
1041

Returns
=======
np.array((nu,nv), dtype=np.complex128):
    the gridded visibilities
)""";
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
1042
1043
template<typename T> pyarr_c<complex<T>> ms2grid_c(
  const Baselines<T> &baselines, const GridderConfig<T> &gconf,
Martin Reinecke's avatar
Martin Reinecke committed
1044
  const pyarr<uint32_t> &idx_, const pyarr<complex<T>> &ms_,
Martin Reinecke's avatar
Martin Reinecke committed
1045
  py::object &grid_in, const py::object &wgt_)
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1046
1047
1048
1049
1050
1051
  {
  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));
1052
1053
  auto ms = ms_.template unchecked<2>();
  auto idx = idx_.template unchecked<1>();
Martin Reinecke's avatar
Martin Reinecke committed
1054
1055
1056
  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
1057

Martin Reinecke's avatar
merge    
Martin Reinecke committed
1058
  size_t nu=gconf.Nu(), nv=gconf.Nv();
Martin Reinecke's avatar
Martin Reinecke committed
1059
  auto res = provideCArray<complex<T>>(grid_in, {nu, nv});
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1060
  auto grid = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
1061
1062
  {
  py::gil_scoped_release release;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1063
1064
1065
  T beta = gconf.Beta();
  size_t w = gconf.W();

1066
#pragma omp parallel num_threads(nthreads)
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1067
{
1068
  Helper<T> hlp(gconf, nullptr, grid);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1069
  T emb = exp(-2*beta);
1070
  int jump = hlp.lineJump();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1071
1072
1073
1074
1075
1076
1077
  const T * RESTRICT ku = hlp.kernel.data();
  const T * RESTRICT kv = hlp.kernel.data()+w;

  // Loop over sampling points
#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
1078
1079
1080
1081
    auto tidx = idx(ipart);
    auto row = tidx/nchan;
    auto chan = tidx-row*nchan;
    UVW<T> coord = baselines.effectiveCoord(tidx);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1082
    hlp.prep(coord.u, coord.v);
1083
    auto * RESTRICT ptr = hlp.p0w;
1084
    auto v(ms(row,chan)*emb);
Martin Reinecke's avatar
Martin Reinecke committed
1085
1086
    if (have_wgt)
      v*=wgt(row, chan);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1087
1088
1089
1090
1091
    for (size_t cu=0; cu<w; ++cu)
      {
      complex<T> tmp(v*ku[cu]);
      for (size_t cv=0; cv<w; ++cv)
        ptr[cv] += tmp*kv[cv];
1092
      ptr+=jump;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1093
1094
1095
      }
    }
} // end of parallel region
Martin Reinecke's avatar
Martin Reinecke committed
1096
  }
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1097
1098
1099
1100
  return res;
  }

template<typename T> pyarr_c<T> ms2grid(const Baselines<T> &baselines,