nifty_gridder.cc 48 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
32
#ifdef __GNUC__
#define RESTRICT __restrict__
Martin Reinecke's avatar
Martin Reinecke committed
33
#define NOINLINE __attribute__ ((noinline))
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
34
35
36
37
#else
#define RESTRICT
#endif

Martin Reinecke's avatar
import  
Martin Reinecke committed
38
39
40
41
42
43
using namespace std;

namespace py = pybind11;

namespace {

Martin Reinecke's avatar
Martin Reinecke committed
44
45
auto None = py::none();

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
46
47
48
49
//
// basic utilities
//

50
51
52
53
54
55
void myassert(bool cond, const char *msg)
  {
  if (cond) return;
  throw runtime_error(msg);
  }

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
56
57
58
59
60
61
62
63
64
65
66
/*! 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;
  }

67
68
69
70
71
72
73
74
75
76
77
78
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_)
  {
79
  myassert(nthreads_>=1, "nthreads must be >= 1");
80
81
82
83
84
85
86
87
88
89
90
  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
91
  { return nthreads; }
92

Martin Reinecke's avatar
Martin Reinecke committed
93
94
95
96
//
// Utilities for Gauss-Legendre quadrature
//

Martin Reinecke's avatar
Martin Reinecke committed
97
98
99
100
101
102
103
104
105
106
107
108
109
110
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.);

111
#pragma omp parallel num_threads(nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
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
138
139
140
141
142
143
144
{
  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;
145
      myassert(++j<100, "convergence problem");
Martin Reinecke's avatar
Martin Reinecke committed
146
147
148
149
150
151
152
153
      }

    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
154
155
156
157
//
// Start of real gridder functionality
//

158
159
template<typename T>
  using pyarr = py::array_t<T>;
160
// The "_c" suffix here stands for "C memory order, contiguous"
161
162
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
163

Martin Reinecke's avatar
merge    
Martin Reinecke committed
164
template<typename T> pyarr_c<T> makeArray(const vector<size_t> &shape)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
165
166
  { return pyarr_c<T>(shape); }

Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
167
size_t get_w(double epsilon)
Martin Reinecke's avatar
Martin Reinecke committed
168
169
170
171
172
173
174
175
176
177
178
  {
  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
179

Martin Reinecke's avatar
merge    
Martin Reinecke committed
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
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");
      }
  }

198
template<typename T> pyarr<T> provideArray(py::object &in,
Martin Reinecke's avatar
merge    
Martin Reinecke committed
199
200
  const vector<size_t> &shape)
  {
201
  if (in.is_none())
Martin Reinecke's avatar
merge    
Martin Reinecke committed
202
203
204
205
206
207
208
209
    {
    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_;
    }
210
  auto tmp_ = in.cast<pyarr<T>>();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
211
212
213
214
  checkArray(tmp_, "temporary", shape);
  return tmp_;
  }

Martin Reinecke's avatar
Martin Reinecke committed
215
template<typename T> pyarr_c<T> provideCArray(py::object &in,
Martin Reinecke's avatar
merge    
Martin Reinecke committed
216
217
  const vector<size_t> &shape)
  {
218
  if (in.is_none())
Martin Reinecke's avatar
merge    
Martin Reinecke committed
219
220
221
222
223
224
225
226
227
228
229
230
231
    {
    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
232
template<typename T> pyarr_c<T> complex2hartley
233
  (const pyarr_c<complex<T>> &grid_, py::object &grid_in)
Martin Reinecke's avatar
Martin Reinecke committed
234
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
235
  checkArray(grid_, "grid", {0,0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
236
  size_t nu = size_t(grid_.shape(0)), nv = size_t(grid_.shape(1));
Martin Reinecke's avatar
Martin Reinecke committed
237
238
  auto grid = grid_.data();

239
  auto res = provideCArray<T>(grid_in, {nu, nv});
Martin Reinecke's avatar
Martin Reinecke committed
240
  auto grid2 = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
241
242
  {
  py::gil_scoped_release release;
243
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
244
  for (size_t u=0; u<nu; ++u)
Martin Reinecke's avatar
Martin Reinecke committed
245
    {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
246
247
    size_t xu = (u==0) ? 0 : nu-u;
    for (size_t v=0; v<nv; ++v)
Martin Reinecke's avatar
Martin Reinecke committed
248
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
249
250
251
      size_t xv = (v==0) ? 0 : nv-v;
      size_t i1 = u*nv+v;
      size_t i2 = xu*nv+xv;
252
253
      grid2[i1] += T(0.5)*(grid[i1].real()+grid[i1].imag()+
                           grid[i2].real()-grid[i2].imag());
Martin Reinecke's avatar
Martin Reinecke committed
254
255
      }
    }
Martin Reinecke's avatar
Martin Reinecke committed
256
  }
Martin Reinecke's avatar
Martin Reinecke committed
257
258
259
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
260
261
template<typename T> pyarr_c<complex<T>> hartley2complex
  (const pyarr_c<T> &grid_)
Martin Reinecke's avatar
Martin Reinecke committed
262
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
263
  checkArray(grid_, "grid", {0, 0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
264
  size_t nu = size_t(grid_.shape(0)), nv = size_t(grid_.shape(1));
Martin Reinecke's avatar
Martin Reinecke committed
265
266
  auto grid = grid_.data();

Martin Reinecke's avatar
merge    
Martin Reinecke committed
267
  auto res=makeArray<complex<T>>({nu, nv});
Martin Reinecke's avatar
Martin Reinecke committed
268
  auto grid2 = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
269
270
  {
  py::gil_scoped_release release;
271
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
272
  for (size_t u=0; u<nu; ++u)
Martin Reinecke's avatar
Martin Reinecke committed
273
    {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
274
275
    size_t xu = (u==0) ? 0 : nu-u;
    for (size_t v=0; v<nv; ++v)
Martin Reinecke's avatar
Martin Reinecke committed
276
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
277
278
279
      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
280
281
282
      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
283
284
      }
    }
Martin Reinecke's avatar
Martin Reinecke committed
285
  }
Martin Reinecke's avatar
Martin Reinecke committed
286
287
288
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
289
290
291
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
292
293
294
  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
295
  auto ptmp = out.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
296
297
  {
  py::gil_scoped_release release;
298
299
300
  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
301
302
303
304
305
306
307
308
309
310
311
312
313
  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
314
  }
Martin Reinecke's avatar
Martin Reinecke committed
315

Martin Reinecke's avatar
Martin Reinecke committed
316
317
/* 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
318
vector<double> correction_factors (size_t n, size_t nval, size_t w)
Martin Reinecke's avatar
Martin Reinecke committed
319
320
321
322
323
324
325
326
327
328
  {
  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
329
  vector<double> res(nval);
330
#pragma omp parallel for schedule(static) num_threads(nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
331
332
333
334
335
  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
336
    res[k] = 1./(w*tmp);
Martin Reinecke's avatar
Martin Reinecke committed
337
338
339
340
    }
  return res;
  }

Martin Reinecke's avatar
Martin Reinecke committed
341
template<typename T> struct UVW
Martin Reinecke's avatar
updates    
Martin Reinecke committed
342
  {
Martin Reinecke's avatar
Martin Reinecke committed
343
  T u, v, w;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
344
  UVW () {}
Martin Reinecke's avatar
Martin Reinecke committed
345
346
  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
347
348
    { return UVW(u*fct, v*fct, w*fct); }
  };
Martin Reinecke's avatar
Martin Reinecke committed
349

350
351
352
353
354
355
356
357
358
359
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
360
template<typename T> class Baselines
Martin Reinecke's avatar
Martin Reinecke committed
361
362
  {
  private:
Martin Reinecke's avatar
Martin Reinecke committed
363
    vector<UVW<T>> coord;
364
    vector<T> f_over_c;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
365
    size_t nrows, nchan;
Martin Reinecke's avatar
Martin Reinecke committed
366
367

  public:
368
    Baselines(const pyarr<T> &coord_, const pyarr<T> &freq_)
Martin Reinecke's avatar
Martin Reinecke committed
369
      {
370
      constexpr double speedOfLight = 299792458.;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
371
      checkArray(coord_, "coord", {0, 3});
372
      checkArray(freq_, "freq", {0});
Martin Reinecke's avatar
Martin Reinecke committed
373
      nrows = coord_.shape(0);
374
      nchan = freq_.shape(0);
Martin Reinecke's avatar
Martin Reinecke committed
375
      myassert(nrows*nchan<(size_t(1)<<32), "too many entries in MS");
376
377
      auto freq = freq_.template unchecked<1>();
      auto cood = coord_.template unchecked<2>();
Martin Reinecke's avatar
Martin Reinecke committed
378
379
      {
      py::gil_scoped_release release;
380
      f_over_c.resize(nchan);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
381
      for (size_t i=0; i<nchan; ++i)
382
        f_over_c[i] = freq(i)/speedOfLight;
Martin Reinecke's avatar
Martin Reinecke committed
383
384
      coord.resize(nrows);
      for (size_t i=0; i<coord.size(); ++i)
385
        coord[i] = UVW<T>(cood(i,0), cood(i,1), cood(i,2));
Martin Reinecke's avatar
Martin Reinecke committed
386
      }
Martin Reinecke's avatar
Martin Reinecke committed
387
388
      }

Martin Reinecke's avatar
Martin Reinecke committed
389
    UVW<T> effectiveCoord(uint32_t index) const
390
391
392
      {
      size_t irow = index/nchan;
      size_t ichan = index-nchan*irow;
393
      return coord[irow]*f_over_c[ichan];
394
395
      }
    UVW<T> effectiveCoord(size_t irow, size_t ichan) const
396
      { return coord[irow]*f_over_c[ichan]; }
Martin Reinecke's avatar
Martin Reinecke committed
397
    size_t Nrows() const { return nrows; }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
398
399
    size_t Nchannels() const { return nchan; }

400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
    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
    )""";
415
    template<typename T2> pyarr_c<T2> ms2vis(const pyarr<T2> &ms_,
Martin Reinecke's avatar
updates    
Martin Reinecke committed
416
417
      const pyarr_c<uint32_t> &idx_) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
418
419
      checkArray(idx_, "idx", {0});
      checkArray(ms_, "ms", {nrows, nchan});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
420
      size_t nvis = size_t(idx_.shape(0));
421
422
      auto idx = idx_.template unchecked<1>();
      auto ms = ms_.template unchecked<2>();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
423

Martin Reinecke's avatar
merge    
Martin Reinecke committed
424
      auto res=makeArray<T2>({nvis});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
425
      auto vis = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
426
427
      {
      py::gil_scoped_release release;
428
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
429
      for (size_t i=0; i<nvis; ++i)
430
431
432
433
434
435
        {
        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
436
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
437
438
439
      return res;
      }

440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
    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)
    )""";
457
458
    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
459
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
460
      checkArray(vis_, "vis", {0});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
461
      size_t nvis = size_t(vis_.shape(0));
Martin Reinecke's avatar
merge    
Martin Reinecke committed
462
      checkArray(idx_, "idx", {nvis});
463
464
      auto idx = idx_.template unchecked<1>();
      auto vis = vis_.template unchecked<1>();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
465

Martin Reinecke's avatar
merge    
Martin Reinecke committed
466
      auto res = provideArray<T2>(ms_in, {nrows, nchan});
467
      auto ms = res.template mutable_unchecked<2>();
Martin Reinecke's avatar
Martin Reinecke committed
468
469
      {
      py::gil_scoped_release release;
470
#pragma omp parallel for num_threads(nthreads)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
471
      for (size_t i=0; i<nvis; ++i)
472
473
474
475
        {
        auto t = idx(i);
        auto row = t/nchan;
        auto chan = t-row*nchan;
Martin Reinecke's avatar
Martin Reinecke committed
476
        ms(row, chan) += vis(i);
477
        }
Martin Reinecke's avatar
Martin Reinecke committed
478
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
479
480
      return res;
      }
Martin Reinecke's avatar
Martin Reinecke committed
481
482
  };

Martin Reinecke's avatar
Martin Reinecke committed
483
484
constexpr int logsquare=4;

485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
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
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
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
)""";

529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
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
546
template<typename T> class GridderConfig
Martin Reinecke's avatar
Martin Reinecke committed
547
548
549
  {
  private:
    size_t nx_dirty, ny_dirty;
Martin Reinecke's avatar
Martin Reinecke committed
550
    double eps, psx, psy;
Martin Reinecke's avatar
Martin Reinecke committed
551
    size_t w, nsafe, nu, nv;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
552
    T beta;
Martin Reinecke's avatar
Martin Reinecke committed
553
    vector<T> cfu, cfv;
Martin Reinecke's avatar
Martin Reinecke committed
554
555
556

  public:
    GridderConfig(size_t nxdirty, size_t nydirty, double epsilon,
Martin Reinecke's avatar
Martin Reinecke committed
557
      double pixsize_x, double pixsize_y)
Martin Reinecke's avatar
Martin Reinecke committed
558
559
      : nx_dirty(nxdirty), ny_dirty(nydirty), eps(epsilon),
        psx(pixsize_x), psy(pixsize_y),
Martin Reinecke's avatar
Martin Reinecke committed
560
561
        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
562
        beta(2.3*w),
Martin Reinecke's avatar
updates    
Martin Reinecke committed
563
        cfu(nx_dirty), cfv(ny_dirty)
Martin Reinecke's avatar
Martin Reinecke committed
564
      {
Martin Reinecke's avatar
Martin Reinecke committed
565
566
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
Martin Reinecke committed
567
568
569
      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
570
571
      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
572
573
574
575
576
577
578
579
580
581
582

      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
583
      }
Martin Reinecke's avatar
Martin Reinecke committed
584
      }
Martin Reinecke's avatar
Martin Reinecke committed
585
586
587
588
589
    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
590
591
592
    size_t Nu() const { return nu; }
    size_t Nv() const { return nv; }
    size_t W() const { return w; }
593
    size_t Nsafe() const { return nsafe; }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
594
    T Beta() const { return beta; }
595

Martin Reinecke's avatar
Martin Reinecke committed
596
    pyarr_c<T> grid2dirty(const pyarr_c<T> &grid) const
Martin Reinecke's avatar
Martin Reinecke committed
597
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
598
599
      checkArray(grid, "grid", {nu, nv});
      auto tmp = makeArray<T>({nu, nv});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
600
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
601
      hartley2_2D<T>(grid, tmp);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
602
      auto res = makeArray<T>({nx_dirty, ny_dirty});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
603
      auto pout = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
604
605
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
606
607
608
609
610
611
612
613
614
      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
615
      }
Martin Reinecke's avatar
Martin Reinecke committed
616
617
      return res;
      }
Martin Reinecke's avatar
Martin Reinecke committed
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
    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;
      }
637
638
    pyarr_c<complex<T>> grid2dirty_c(const pyarr_c<complex<T>> &grid) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
639
640
      checkArray(grid, "grid", {nu, nv});
      auto tmp = makeArray<complex<T>>({nu, nv});
641
642
643
      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,
644
        grid.data(), tmp.mutable_data(), T(1), nthreads);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
645
      auto res = makeArray<complex<T>>({nx_dirty, ny_dirty});
646
      auto pout = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
647
648
      {
      py::gil_scoped_release release;
649
650
651
652
653
654
655
656
657
      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
658
      }
659
660
      return res;
      }
661

Philipp Arras's avatar
Philipp Arras committed
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
    pyarr_c<T> grid2dirty_wstacking(const pyarr_c<complex<T>> &grid, double w) const
      {
      checkArray(grid, "grid", {nu, nv});
      auto tmp = makeArray<complex<T>>({nu, nv});
      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,
        grid.data(), tmp.mutable_data(), T(1), nthreads);
      auto res = makeArray<T>({nx_dirty, ny_dirty});
      auto pout = res.mutable_data();
      {
        py::gil_scoped_release release;
#pragma omp parallel num_threads(nthreads)
{
#pragma omp for schedule(static)
        for (size_t i=0; i<nx_dirty; ++i)
          {
            double fx = (i-nx_dirty/2.)*Pixsize_x();
            fx *= fx;
            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;
                double fy = (j-ny_dirty/2.)*Pixsize_y();
                auto ws = wscreen(fx, fy*fy, w, true);
                pout[ny_dirty*i + j] = std::real(ptmp[nv*i2+j2]*cfu[i]*cfv[j]*ws);
              }
          }
 }
      }
      return res;
      }

Martin Reinecke's avatar
Martin Reinecke committed
697
    pyarr_c<T> dirty2grid(const pyarr_c<T> &dirty) const
Martin Reinecke's avatar
updates    
Martin Reinecke committed
698
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
699
      checkArray(dirty, "dirty", {nx_dirty, ny_dirty});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
700
      auto pdirty = dirty.data();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
701
      auto tmp = makeArray<T>({nu, nv});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
702
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
703
704
      {
      py::gil_scoped_release release;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
705
706
707
708
709
710
711
712
713
714
715
      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
716
      }
Martin Reinecke's avatar
Martin Reinecke committed
717
      hartley2_2D<T>(tmp, tmp);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
718
719
      return tmp;
      }
720
721
    pyarr_c<complex<T>> dirty2grid_c(const pyarr_c<complex<T>> &dirty) const
      {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
722
      checkArray(dirty, "dirty", {nx_dirty, ny_dirty});
723
      auto pdirty = dirty.data();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
724
      auto tmp = makeArray<complex<T>>({nu, nv});
725
      auto ptmp = tmp.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
726
727
728
      pocketfft::stride_t strides{tmp.strides(0),tmp.strides(1)};
      {
      py::gil_scoped_release release;
729
730
731
732
733
734
735
736
737
738
739
      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
740
      pocketfft::c2c({nu,nv}, strides, strides, {0,1}, pocketfft::FORWARD,
741
        ptmp, ptmp, T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
742
      }
743
744
      return tmp;
      }
Philipp Arras's avatar
Philipp Arras committed
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
    inline complex<T> wscreen(double x, double y, double w, bool adjoint) const
    {
      constexpr double pi = 3.141592653589793238462643383279502884197;
      constexpr complex<double> imagunit {0,1};
      double n = cos(sqrt(x+y));
      complex<T> phase = 2*pi*w*imagunit*(n-1);
      if (adjoint) phase *= -1;
      return exp(phase)/n;
    }
    pyarr_c<complex<T>> dirty2grid_wstacking(const pyarr_c<T> &dirty, double w) const
    {
      checkArray(dirty, "dirty", {nx_dirty, ny_dirty});
      auto pdirty = dirty.data();
      auto tmp = makeArray<complex<T>>({nu, nv});
      auto ptmp = tmp.mutable_data();
      pocketfft::stride_t strides{tmp.strides(0),tmp.strides(1)};
      {
        py::gil_scoped_release release;
        for (size_t i=0; i<nu*nv; ++i)
          ptmp[i] = 0.;
#pragma omp parallel num_threads(nthreads)
        {
#pragma omp for schedule(static)
        for (size_t i=0; i<nx_dirty; ++i)
          {
            double fx = (i-nx_dirty/2.)*Pixsize_x();
            fx *= fx;
            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;
                double fy = (j-ny_dirty/2.)*Pixsize_y();
                auto ws = wscreen(fx, fy*fy, w, false);
                ptmp[nv*i2+j2] = std::real(pdirty[ny_dirty*i + j])*ws*cfu[i]*cfv[j];
              }
          }
 }
        pocketfft::c2c({nu,nv}, strides, strides, {0,1}, pocketfft::FORWARD,
                       ptmp, ptmp, T(1), nthreads);
      }
      return tmp;
    }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
789
790
    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
791
      u=fmodulo(u_in*psx, T(1))*nu,
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
792
793
      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
794
      v=fmodulo(v_in*psy, T(1))*nv;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
795
796
797
      iv0 = int(v-w*0.5 + 1 + nv) - nv;
      if (iv0+w>nv+nsafe) iv0 = nv+nsafe-w;
      }
Martin Reinecke's avatar
Martin Reinecke committed
798
799
  };

800
template<typename T, typename T2=complex<T>> class Helper
Martin Reinecke's avatar
import  
Martin Reinecke committed
801
  {
Martin Reinecke's avatar
Martin Reinecke committed
802
  private:
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
803
    const GridderConfig<T> &gconf;
Martin Reinecke's avatar
Martin Reinecke committed
804
805
    int nu, nv, nsafe, w;
    T beta;
806
807
    const T2 *grid_r;
    T2 *grid_w;
808
    int su, sv;
Martin Reinecke's avatar
Martin Reinecke committed
809
810
811
    int iu0, iv0; // start index of the current visibility
    int bu0, bv0; // start index of the current buffer

812
    vector<T2> rbuf, wbuf;
Martin Reinecke's avatar
import  
Martin Reinecke committed
813

Martin Reinecke's avatar
Martin Reinecke committed
814
    void dump() const
Martin Reinecke's avatar
import  
Martin Reinecke committed
815
      {
Martin Reinecke's avatar
Martin Reinecke committed
816
      if (bu0<-nsafe) return; // nothing written into buffer yet
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
817

Martin Reinecke's avatar
merge    
Martin Reinecke committed
818
#pragma omp critical (gridder_writing_to_grid)
Martin Reinecke's avatar
import  
Martin Reinecke committed
819
{
Martin Reinecke's avatar
Martin Reinecke committed
820
821
822
      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
823
824
        {
        int idxv = idxv0;
Martin Reinecke's avatar
Martin Reinecke committed
825
        for (int iv=0; iv<sv; ++iv)
Martin Reinecke's avatar
import  
Martin Reinecke committed
826
          {
827
          grid_w[idxu*nv + idxv] += wbuf[iu*sv + iv];
Martin Reinecke's avatar
import  
Martin Reinecke committed
828
829
830
831
832
833
834
835
836
          if (++idxv>=nv) idxv=0;
          }
        if (++idxu>=nu) idxu=0;
        }
}
      }

    void load()
      {
Martin Reinecke's avatar
Martin Reinecke committed
837
838
839
      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
840
841
        {
        int idxv = idxv0;
Martin Reinecke's avatar
Martin Reinecke committed
842
        for (int iv=0; iv<sv; ++iv)
Martin Reinecke's avatar
import  
Martin Reinecke committed
843
          {
844
          rbuf[iu*sv + iv] = grid_r[idxu*nv + idxv];
Martin Reinecke's avatar
import  
Martin Reinecke committed
845
846
847
848
849
850
851
          if (++idxv>=nv) idxv=0;
          }
        if (++idxu>=nu) idxu=0;
        }
      }

  public:
852
853
    const T2 *p0r;
    T2 *p0w;
Martin Reinecke's avatar
Martin Reinecke committed
854
    vector<T> kernel;
Martin Reinecke's avatar
import  
Martin Reinecke committed
855

856
    Helper(const GridderConfig<T> &gconf_, const T2 *grid_r_, T2 *grid_w_)
857
858
      : 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
859
        su(2*nsafe+(1<<logsquare)), sv(2*nsafe+(1<<logsquare)),
860
861
862
863
        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
864
      {}
865
866
867
    ~Helper() { if (grid_w) dump(); }

    int lineJump() const { return sv; }
Martin Reinecke's avatar
Martin Reinecke committed
868
869

    void prep(T u_in, T v_in)
Martin Reinecke's avatar
import  
Martin Reinecke committed
870
      {
Martin Reinecke's avatar
Martin Reinecke committed
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
      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
887
        {
888
        if (grid_w) { dump(); fill(wbuf.begin(), wbuf.end(), T(0)); }
Martin Reinecke's avatar
Martin Reinecke committed
889
890
        bu0=((((iu0+nsafe)>>logsquare)<<logsquare))-nsafe;
        bv0=((((iv0+nsafe)>>logsquare)<<logsquare))-nsafe;
891
        if (grid_r) load();
Martin Reinecke's avatar
import  
Martin Reinecke committed
892
        }
893
894
      p0r = rbuf.data() + sv*(iu0-bu0) + iv0-bv0;
      p0w = wbuf.data() + sv*(iu0-bu0) + iv0-bv0;
Martin Reinecke's avatar
import  
Martin Reinecke committed
895
896
897
      }
  };

898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
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.

Returns
=======
np.array((nu,nv), dtype=np.complex128):
    the gridded visibilities
)""";
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
920
921
template<typename T> pyarr_c<complex<T>> vis2grid_c(
  const Baselines<T> &baselines, const GridderConfig<T> &gconf,
Martin Reinecke's avatar
Martin Reinecke committed
922
923
  const pyarr<uint32_t> &idx_, const pyarr<complex<T>> &vis_,
  py::object &grid_in)
924
  {
Martin Reinecke's avatar
merge    
Martin Reinecke committed
925
926
927
  checkArray(vis_, "vis", {0});
  size_t nvis = size_t(vis_.shape(0));
  checkArray(idx_, "idx", {nvis});
928
929
  auto vis=vis_.template unchecked<1>();
  auto idx = idx_.template unchecked<1>();
930

931
  size_t nu=gconf.Nu(), nv=gconf.Nv();
Martin Reinecke's avatar
Martin Reinecke committed
932
  auto res = provideCArray<complex<T>>(grid_in, {nu, nv});
933
  auto grid = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
934
935
  {
  py::gil_scoped_release release;
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
936
937
  T beta = gconf.Beta();
  size_t w = gconf.W();
938

939
#pragma omp parallel num_threads(nthreads)
940
{
941
  Helper<T> hlp(gconf, nullptr, grid);
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
942
  T emb = exp(-2*beta);
943
  int jump = hlp.lineJump();
944
  const T * RESTRICT ku = hlp.kernel.data();
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
945
  const T * RESTRICT kv = hlp.kernel.data()+w;
946
947
948
949
950

  // Loop over sampling points
#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
951
    UVW<T> coord = baselines.effectiveCoord(idx(ipart));
Martin Reinecke's avatar
Martin Reinecke committed
952
    hlp.prep(coord.u, coord.v);
953
    auto * RESTRICT ptr = hlp.p0w;
954
    auto v(vis(ipart)*emb);
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
955
    for (size_t cu=0; cu<w; ++cu)
956
957
      {
      complex<T> tmp(v*ku[cu]);
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
958
      for (size_t cv=0; cv<w; ++cv)
959
        ptr[cv] += tmp*kv[cv];
960
      ptr+=jump;
961
962
963
      }
    }
} // end of parallel region
Martin Reinecke's avatar
Martin Reinecke committed
964
  }
965
966
967
  return res;
  }

968
969
970
971
972
973
974
975
976
977
978
979
980
981
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
982
983
grid_in: np.array((nu,nv), dtype=np.float64), optional
    If present, the result is added to this array.
984
985
986
987
988
989

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
990
template<typename T> pyarr_c<T> vis2grid(const Baselines<T> &baselines,
991
  const GridderConfig<T> &gconf, const pyarr<uint32_t> &idx_,
992
993
  const pyarr<complex<T>> &vis_, py::object &grid_in)
  { return complex2hartley(vis2grid_c(baselines, gconf, idx_, vis_, None), grid_in); }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
994

995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
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.

Returns
=======
np.array((nu,nv), dtype=np.complex128):
    the gridded visibilities
)""";
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
1017
1018
template<typename T> pyarr_c<complex<T>> ms2grid_c(
  const Baselines<T> &baselines, const GridderConfig<T> &gconf,
Martin Reinecke's avatar
Martin Reinecke committed
1019
1020
  const pyarr<uint32_t> &idx_, const pyarr<complex<T>> &ms_,
  py::object &grid_in)
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1021
1022
1023
1024
1025
1026
  {
  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));
1027
1028
  auto ms = ms_.template unchecked<2>();
  auto idx = idx_.template unchecked<1>();
Simon Perkins's avatar
Simon Perkins committed
1029

Martin Reinecke's avatar
merge    
Martin Reinecke committed
1030
  size_t nu=gconf.Nu(), nv=gconf.Nv();
Martin Reinecke's avatar
Martin Reinecke committed
1031
  auto res = provideCArray<complex<T>>(grid_in, {nu, nv});
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1032
  auto grid = res.mutable_data();
Martin Reinecke's avatar
Martin Reinecke committed
1033
1034
  {
  py::gil_scoped_release release;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1035
1036
1037
  T beta = gconf.Beta();
  size_t w = gconf.W();

1038
#pragma omp parallel num_threads(nthreads)
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1039
{
1040
  Helper<T> hlp(gconf, nullptr, grid);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1041
  T emb = exp(-2*beta);
1042
  int jump = hlp.lineJump();
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1043
1044
1045
1046
1047
1048
1049
  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)
    {
1050
1051
1052
1053
    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
1054
    hlp.prep(coord.u, coord.v);
1055
    auto * RESTRICT ptr = hlp.p0w;
1056
    auto v(ms(row,chan)*emb);
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1057
1058
1059
1060
1061
    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];
1062
      ptr+=jump;
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1063
1064
1065
      }
    }
} // end of parallel region
Martin Reinecke's avatar
Martin Reinecke committed
1066
  }
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1067
1068
1069
1070
  return res;
  }

template<typename T> pyarr_c<T> ms2grid(const Baselines<T> &baselines,
1071
  const GridderConfig<T> &gconf, const pyarr<uint32_t> &idx_,
1072
1073
  const pyarr<complex<T>> &ms_, py::object &grid_in)
  { return complex2hartley(ms2grid_c(baselines, gconf, idx_, ms_, None), grid_in); }
Martin Reinecke's avatar
merge    
Martin Reinecke committed
1074

1075
1076
1077
template<typename T> pyarr_c<complex<T>> ms2grid_c_wgt(
  const Baselines<T> &baselines, const GridderConfig<T> &gconf,
  const pyarr<uint32_t> &idx_, const pyarr<complex<T>> &ms_,
Martin Reinecke's avatar
Martin Reinecke committed
1078
  const pyarr<T> &wgt_, py::object &grid_in)
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
  {
  auto nrows = baselines.Nrows();
  auto nchan = baselines.Nchannels();
  checkArray(wgt_, "wgt", {nrows, nchan});
  checkArray(ms_, "ms", {nrows, nchan});
  checkArray(idx_, "idx", {0});
  size_t nvis = size_t(idx_.shape(0));
  auto ms = ms_.template unchecked<2>();
  auto wgt = wgt_.template unchecked<2>();
  auto idx = idx_.template unchecked<1>();

  size_t nu=gconf.Nu(), nv=gconf.Nv();
Martin Reinecke's avatar
Martin Reinecke committed
1091
  auto res = provideArray<complex<T>>(grid_in, {nu, nv});
1092
1093
1094
1095
1096
1097
  auto grid = res.mutable_data();
  {
  py::gil_scoped_release release;
  T beta = gconf.Beta();
  size_t w = gconf.W();

1098
#pragma omp parallel num_threads(nthreads)
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
{
  Helper<T> hlp(gconf, nullptr, grid);
  T emb = exp(-2*beta);
  int jump = hlp.lineJump();
  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)
    {
    auto tidx = idx(ipart);
    auto row = tidx/nchan;
    auto chan = tidx-row*nchan;
    UVW<T> coord = baselines.effectiveCoord(tidx);
    hlp.prep(coord.u, coord.v);
    auto * RESTRICT ptr = hlp.p0w;
    auto v(ms(row,chan)*(emb*wgt(row, chan)));
    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];
      ptr+=jump;
      }
    }
} // end of parallel region
  }
  return res;
  }

template<typename T> pyarr_c<T> ms2grid_wgt(const Baselines<T> &baselines,
  const GridderConfig<T> &gconf, const pyarr<uint32_t> &idx_,
1132
1133
1134
  const pyarr<complex<T>> &ms_, const pyarr<T> &wgt_,
  py::object &grid_in)
  { return complex2hartley(ms2grid_c_wgt(baselines, gconf, idx_, ms_, wgt_, None), grid_in); }
1135