gridder_cxx.h 37.4 KB
Newer Older
Martin Reinecke's avatar
Martin Reinecke committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
#ifndef GRIDDER_CXX_H
#define GRIDDER_CXX_H

/*
 *  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
 *  along with nifty_gridder; if not, write to the Free Software
 *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
 */

22
/* Copyright (C) 2019-2020 Max-Planck-Society
Martin Reinecke's avatar
Martin Reinecke committed
23
24
25
26
27
28
29
30
31
   Author: Martin Reinecke */

#include <iostream>
#include <algorithm>
#include <cstdlib>
#include <cmath>
#include <vector>
#include <array>

32
33
34
35
36
#include "mr_util/infra/error_handling.h"
#include "mr_util/math/fft.h"
#include "mr_util/infra/threading.h"
#include "mr_util/infra/useful_macros.h"
#include "mr_util/infra/mav.h"
37
#include "mr_util/math/es_kernel.h"
Martin Reinecke's avatar
Martin Reinecke committed
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66

#if defined(__GNUC__)
#define ALIGNED(align) __attribute__ ((aligned(align)))
#else
#define ALIGNED(align)
#endif

namespace gridder {

namespace detail {

using namespace std;
using namespace mr;

template<size_t ndim> void checkShape
  (const array<size_t, ndim> &shp1, const array<size_t, ndim> &shp2)
  {
  for (size_t i=0; i<ndim; ++i)
    MR_assert(shp1[i]==shp2[i], "shape mismatch");
  }

template<typename T> inline T fmod1 (T v)
  { return v-floor(v); }

//
// Start of real gridder functionality
//

template<typename T> void complex2hartley
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
67
  (const mav<complex<T>, 2> &grid, mav<T,2> &grid2, size_t nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
68
69
70
71
72
73
74
75
76
77
78
79
  {
  checkShape(grid.shape(), grid2.shape());
  size_t nu=grid.shape(0), nv=grid.shape(1);

  execStatic(nu, nthreads, 0, [&](Scheduler &sched)
    {
    while (auto rng=sched.getNext()) for(auto u=rng.lo; u<rng.hi; ++u)
      {
      size_t xu = (u==0) ? 0 : nu-u;
      for (size_t v=0; v<nv; ++v)
        {
        size_t xv = (v==0) ? 0 : nv-v;
Martin Reinecke's avatar
Martin Reinecke committed
80
81
        grid2.v(u,v) = T(0.5)*(grid( u, v).real()+grid( u, v).imag()+
                               grid(xu,xv).real()-grid(xu,xv).imag());
Martin Reinecke's avatar
Martin Reinecke committed
82
83
84
85
86
87
        }
      }
    });
  }

template<typename T> void hartley2complex
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
88
  (const mav<T,2> &grid, mav<complex<T>,2> &grid2, size_t nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
89
90
91
92
93
94
95
96
97
98
99
100
101
102
  {
  checkShape(grid.shape(), grid2.shape());
  size_t nu=grid.shape(0), nv=grid.shape(1);

  execStatic(nu, nthreads, 0, [&](Scheduler &sched)
    {
    while (auto rng=sched.getNext()) for(auto u=rng.lo; u<rng.hi; ++u)
      {
      size_t xu = (u==0) ? 0 : nu-u;
      for (size_t v=0; v<nv; ++v)
        {
        size_t xv = (v==0) ? 0 : nv-v;
        T v1 = T(0.5)*grid( u, v);
        T v2 = T(0.5)*grid(xu,xv);
Martin Reinecke's avatar
Martin Reinecke committed
103
        grid2.v(u,v) = std::complex<T>(v1+v2, v1-v2);
Martin Reinecke's avatar
Martin Reinecke committed
104
105
106
107
108
        }
      }
    });
  }

Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
109
110
template<typename T> void hartley2_2D(const mav<T,2> &in,
  mav<T,2> &out, size_t nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
111
112
113
  {
  checkShape(in.shape(), out.shape());
  size_t nu=in.shape(0), nv=in.shape(1);
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
114
115
  fmav<T> fin(in), fout(out);
  r2r_separable_hartley(fin, fout, {0,1}, T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
116
117
118
119
120
  execStatic((nu+1)/2-1, nthreads, 0, [&](Scheduler &sched)
    {
    while (auto rng=sched.getNext()) for(auto i=rng.lo+1; i<rng.hi+1; ++i)
      for(size_t j=1; j<(nv+1)/2; ++j)
         {
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
121
122
123
124
         T a = out(i,j);
         T b = out(nu-i,j);
         T c = out(i,nv-j);
         T d = out(nu-i,nv-j);
Martin Reinecke's avatar
Martin Reinecke committed
125
126
127
128
         out.v(i,j) = T(0.5)*(a+b+c-d);
         out.v(nu-i,j) = T(0.5)*(a+b+d-c);
         out.v(i,nv-j) = T(0.5)*(a+c+d-b);
         out.v(nu-i,nv-j) = T(0.5)*(b+c+d-a);
Martin Reinecke's avatar
Martin Reinecke committed
129
130
131
132
133
134
135
136
137
138
         }
     });
  }

/* Compute correction factors for the ES gridding kernel
   This implementation follows eqs. (3.8) to (3.10) of Barnett et al. 2018 */
vector<double> correction_factors(size_t n, double ofactor, size_t nval, size_t supp,
  size_t nthreads)
  {
  ES_Kernel kernel(supp, ofactor, nthreads);
139
  return kernel.correction_factors(n, nval, nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
  }

using idx_t = uint32_t;

struct RowChan
  {
  idx_t row, chan;
  };

struct UVW
  {
  double u, v, w;
  UVW() {}
  UVW(double u_, double v_, double w_) : u(u_), v(v_), w(w_) {}
  UVW operator* (double fct) const
    { return UVW(u*fct, v*fct, w*fct); }
  void Flip() { u=-u; v=-v; w=-w; }
  bool FixW()
    {
    bool flip = w<0;
    if (flip) Flip();
    return flip;
    }
  };

class Baselines
  {
  protected:
    vector<UVW> coord;
    vector<double> f_over_c;
    idx_t nrows, nchan;
    idx_t shift, mask;

  public:
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
174
175
    template<typename T> Baselines(const mav<T,2> &coord_,
      const mav<T,1> &freq, bool negate_v=false)
Martin Reinecke's avatar
Martin Reinecke committed
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
      {
      constexpr double speedOfLight = 299792458.;
      MR_assert(coord_.shape(1)==3, "dimension mismatch");
      auto hugeval = size_t(~(idx_t(0)));
      MR_assert(coord_.shape(0)<hugeval, "too many entries in MS");
      MR_assert(coord_.shape(1)<hugeval, "too many entries in MS");
      MR_assert(coord_.size()<hugeval, "too many entries in MS");
      nrows = coord_.shape(0);
      nchan = freq.shape(0);
      shift=0;
      while((idx_t(1)<<shift)<nchan) ++shift;
      mask=(idx_t(1)<<shift)-1;
      MR_assert(nrows*(mask+1)<hugeval, "too many entries in MS");
      f_over_c.resize(nchan);
      for (size_t i=0; i<nchan; ++i)
        {
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
192
        MR_assert(freq(i)>0, "negative channel frequency encountered");
Martin Reinecke's avatar
Martin Reinecke committed
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
        f_over_c[i] = freq(i)/speedOfLight;
        }
      coord.resize(nrows);
      if (negate_v)
        for (size_t i=0; i<coord.size(); ++i)
          coord[i] = UVW(coord_(i,0), -coord_(i,1), coord_(i,2));
      else
        for (size_t i=0; i<coord.size(); ++i)
          coord[i] = UVW(coord_(i,0), coord_(i,1), coord_(i,2));
      }

    RowChan getRowChan(idx_t index) const
      { return RowChan{index>>shift, index&mask}; }

    UVW effectiveCoord(const RowChan &rc) const
      { return coord[rc.row]*f_over_c[rc.chan]; }
    UVW effectiveCoord(idx_t index) const
      { return effectiveCoord(getRowChan(index)); }
    size_t Nrows() const { return nrows; }
    size_t Nchannels() const { return nchan; }
    idx_t getIdx(idx_t irow, idx_t ichan) const
      { return ichan+(irow<<shift); }
  };

class GridderConfig
  {
  protected:
    size_t nx_dirty, ny_dirty, nu, nv;
    double ofactor, eps, psx, psy;
    size_t supp, nsafe;
    double beta;
    size_t nthreads;
    double ushift, vshift;
    int maxiu0, maxiv0;

    template<typename T> complex<T> wscreen(T x, T y, T w, bool adjoint) const
      {
      constexpr T pi = T(3.141592653589793238462643383279502884197);
      T tmp = 1-x-y;
      if (tmp<=0) return 1; // no phase factor beyond the horizon
      T nm1 = (-x-y)/(sqrt(tmp)+1); // more accurate form of sqrt(1-x-y)-1
      T phase = 2*pi*w*nm1;
      if (adjoint) phase *= -1;
      return complex<T>(cos(phase), sin(phase));
      }

  public:
    GridderConfig(size_t nxdirty, size_t nydirty, size_t nu_, size_t nv_,
      double epsilon, double pixsize_x, double pixsize_y, size_t nthreads_)
      : nx_dirty(nxdirty), ny_dirty(nydirty), nu(nu_), nv(nv_),
        ofactor(min(double(nu)/nxdirty, double(nv)/nydirty)),
        eps(epsilon),
        psx(pixsize_x), psy(pixsize_y),
        supp(ES_Kernel::get_supp(epsilon, ofactor)), nsafe((supp+1)/2),
        beta(ES_Kernel::get_beta(supp, ofactor)*supp),
        nthreads(nthreads_),
        ushift(supp*(-0.5)+1+nu), vshift(supp*(-0.5)+1+nv),
        maxiu0((nu+nsafe)-supp), maxiv0((nv+nsafe)-supp)
      {
      MR_assert(nu>=2*nsafe, "nu too small");
      MR_assert(nv>=2*nsafe, "nv too small");
      MR_assert((nx_dirty&1)==0, "nx_dirty must be even");
      MR_assert((ny_dirty&1)==0, "ny_dirty must be even");
      MR_assert((nu&1)==0, "nu must be even");
      MR_assert((nv&1)==0, "nv must be even");
      MR_assert(epsilon>0, "epsilon must be positive");
      MR_assert(pixsize_x>0, "pixsize_x must be positive");
      MR_assert(pixsize_y>0, "pixsize_y must be positive");
Martin Reinecke's avatar
Martin Reinecke committed
261
262
      MR_assert(ofactor>=1.175,
        "oversampling factor too small (>=1.2 recommended)");
Martin Reinecke's avatar
Martin Reinecke committed
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
      }
    GridderConfig(size_t nxdirty, size_t nydirty,
      double epsilon, double pixsize_x, double pixsize_y, size_t nthreads_)
      : GridderConfig(nxdirty, nydirty, max<size_t>(30,2*nxdirty),
                      max<size_t>(30,2*nydirty), epsilon, pixsize_x,
                      pixsize_y, nthreads_) {}
    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; }
    size_t Nu() const { return nu; }
    size_t Nv() const { return nv; }
    size_t Supp() const { return supp; }
    size_t Nsafe() const { return nsafe; }
    double Beta() const { return beta; }
    size_t Nthreads() const { return nthreads; }
    double Ofactor() const{ return ofactor; }

Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
282
283
    template<typename T> void grid2dirty_post(mav<T,2> &tmav,
      mav<T,2> &dirty) const
Martin Reinecke's avatar
Martin Reinecke committed
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
      {
      checkShape(dirty.shape(), {nx_dirty,ny_dirty});
      auto cfu = correction_factors(nu, ofactor, nx_dirty/2+1, supp, nthreads);
      auto cfv = correction_factors(nv, ofactor, ny_dirty/2+1, supp, nthreads);
      execStatic(nx_dirty, nthreads, 0, [&](Scheduler &sched)
        {
        while (auto rng=sched.getNext()) for(auto i=rng.lo; i<rng.hi; ++i)
          {
          int icfu = abs(int(nx_dirty/2)-int(i));
          for (size_t j=0; j<ny_dirty; ++j)
            {
            int icfv = abs(int(ny_dirty/2)-int(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;
            // FIXME: for some reason g++ warns about double-to-float conversion
            // here, even though there is an explicit cast...
Martin Reinecke's avatar
Martin Reinecke committed
302
            dirty.v(i,j) = tmav(i2,j2)*T(cfu[icfu]*cfv[icfv]);
Martin Reinecke's avatar
Martin Reinecke committed
303
304
305
306
307
            }
          }
        });
      }
    template<typename T> void grid2dirty_post2(
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
308
      mav<complex<T>,2> &tmav, mav<T,2> &dirty, T w) const
Martin Reinecke's avatar
Martin Reinecke committed
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
      {
      checkShape(dirty.shape(), {nx_dirty,ny_dirty});
      double x0 = -0.5*nx_dirty*psx,
             y0 = -0.5*ny_dirty*psy;
      execStatic(nx_dirty/2+1, nthreads, 0, [&](Scheduler &sched)
        {
        while (auto rng=sched.getNext()) for(auto i=rng.lo; i<rng.hi; ++i)
          {
          T fx = T(x0+i*psx);
          fx *= fx;
          for (size_t j=0; j<=ny_dirty/2; ++j)
            {
            T fy = T(y0+j*psy);
            auto ws = wscreen(fx, fy*fy, w, true);
            size_t ix = nu-nx_dirty/2+i;
            if (ix>=nu) ix-=nu;
            size_t jx = nv-ny_dirty/2+j;
            if (jx>=nv) jx-=nv;
Martin Reinecke's avatar
Martin Reinecke committed
327
            dirty.v(i,j) += (tmav(ix,jx)*ws).real(); // lower left
Martin Reinecke's avatar
Martin Reinecke committed
328
329
330
331
332
333
334
            size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
            size_t ix2 = nu-nx_dirty/2+i2;
            if (ix2>=nu) ix2-=nu;
            size_t jx2 = nv-ny_dirty/2+j2;
            if (jx2>=nv) jx2-=nv;
            if ((i>0)&&(i<i2))
              {
Martin Reinecke's avatar
Martin Reinecke committed
335
              dirty.v(i2,j) += (tmav(ix2,jx)*ws).real(); // lower right
Martin Reinecke's avatar
Martin Reinecke committed
336
              if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
337
                dirty.v(i2,j2) += (tmav(ix2,jx2)*ws).real(); // upper right
Martin Reinecke's avatar
Martin Reinecke committed
338
339
              }
            if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
340
              dirty.v(i,j2) += (tmav(ix,jx2)*ws).real(); // upper left
Martin Reinecke's avatar
Martin Reinecke committed
341
342
343
344
345
            }
          }
        });
      }

Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
346
347
    template<typename T> void grid2dirty(const mav<T,2> &grid,
      mav<T,2> &dirty) const
Martin Reinecke's avatar
Martin Reinecke committed
348
349
      {
      checkShape(grid.shape(), {nu,nv});
350
      mav<T,2> tmav({nu,nv});
Martin Reinecke's avatar
Martin Reinecke committed
351
352
353
354
355
      hartley2_2D<T>(grid, tmav, nthreads);
      grid2dirty_post(tmav, dirty);
      }

    template<typename T> void grid2dirty_c_overwrite_wscreen_add
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
356
      (mav<complex<T>,2> &grid, mav<T,2> &dirty, T w) const
Martin Reinecke's avatar
Martin Reinecke committed
357
358
      {
      checkShape(grid.shape(), {nu,nv});
Martin Reinecke's avatar
Martin Reinecke committed
359
360
      fmav<complex<T>> inout(grid);
      c2c(inout, inout, {0,1}, BACKWARD, T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
361
362
363
      grid2dirty_post2(grid, dirty, w);
      }

Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
364
365
    template<typename T> void dirty2grid_pre(const mav<T,2> &dirty,
      mav<T,2> &grid) const
Martin Reinecke's avatar
Martin Reinecke committed
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
      {
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(grid.shape(), {nu, nv});
      auto cfu = correction_factors(nu, ofactor, nx_dirty/2+1, supp, nthreads);
      auto cfv = correction_factors(nv, ofactor, ny_dirty/2+1, supp, nthreads);
      grid.fill(0);
      execStatic(nx_dirty, nthreads, 0, [&](Scheduler &sched)
        {
        while (auto rng=sched.getNext()) for(auto i=rng.lo; i<rng.hi; ++i)
          {
          int icfu = abs(int(nx_dirty/2)-int(i));
          for (size_t j=0; j<ny_dirty; ++j)
            {
            int icfv = abs(int(ny_dirty/2)-int(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;
Martin Reinecke's avatar
Martin Reinecke committed
384
            grid.v(i2,j2) = dirty(i,j)*T(cfu[icfu]*cfv[icfv]);
Martin Reinecke's avatar
Martin Reinecke committed
385
386
387
388
            }
          }
        });
      }
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
389
390
    template<typename T> void dirty2grid_pre2(const mav<T,2> &dirty,
      mav<complex<T>,2> &grid, T w) const
Martin Reinecke's avatar
Martin Reinecke committed
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
      {
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(grid.shape(), {nu, nv});
      grid.fill(0);

      double x0 = -0.5*nx_dirty*psx,
             y0 = -0.5*ny_dirty*psy;
      execStatic(nx_dirty/2+1, nthreads, 0, [&](Scheduler &sched)
        {
        while (auto rng=sched.getNext()) for(auto i=rng.lo; i<rng.hi; ++i)
          {
          T fx = T(x0+i*psx);
          fx *= fx;
          for (size_t j=0; j<=ny_dirty/2; ++j)
            {
            T fy = T(y0+j*psy);
            auto ws = wscreen(fx, fy*fy, w, false);
            size_t ix = nu-nx_dirty/2+i;
            if (ix>=nu) ix-=nu;
            size_t jx = nv-ny_dirty/2+j;
            if (jx>=nv) jx-=nv;
Martin Reinecke's avatar
Martin Reinecke committed
412
            grid.v(ix,jx) = dirty(i,j)*ws; // lower left
Martin Reinecke's avatar
Martin Reinecke committed
413
414
415
416
417
418
419
            size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
            size_t ix2 = nu-nx_dirty/2+i2;
            if (ix2>=nu) ix2-=nu;
            size_t jx2 = nv-ny_dirty/2+j2;
            if (jx2>=nv) jx2-=nv;
            if ((i>0)&&(i<i2))
              {
Martin Reinecke's avatar
Martin Reinecke committed
420
              grid.v(ix2,jx) = dirty(i2,j)*ws; // lower right
Martin Reinecke's avatar
Martin Reinecke committed
421
              if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
422
                grid.v(ix2,jx2) = dirty(i2,j2)*ws; // upper right
Martin Reinecke's avatar
Martin Reinecke committed
423
424
              }
            if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
425
              grid.v(ix,jx2) = dirty(i,j2)*ws; // upper left
Martin Reinecke's avatar
Martin Reinecke committed
426
427
428
429
430
            }
          }
        });
      }

Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
431
432
    template<typename T> void dirty2grid(const mav<T,2> &dirty,
      mav<T,2> &grid) const
Martin Reinecke's avatar
Martin Reinecke committed
433
434
      {
      dirty2grid_pre(dirty, grid);
Martin Reinecke's avatar
Martin Reinecke committed
435
      hartley2_2D<T>(grid, grid, nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
436
437
      }

Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
438
439
    template<typename T> void dirty2grid_c_wscreen(const mav<T,2> &dirty,
      mav<complex<T>,2> &grid, T w) const
Martin Reinecke's avatar
Martin Reinecke committed
440
441
      {
      dirty2grid_pre2(dirty, grid, w);
Martin Reinecke's avatar
Martin Reinecke committed
442
443
      fmav<complex<T>> inout(grid);
      c2c(inout, inout, {0,1}, FORWARD, T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
444
445
446
447
448
449
450
451
452
453
      }

    void getpix(double u_in, double v_in, double &u, double &v, int &iu0, int &iv0) const
      {
      u=fmod1(u_in*psx)*nu;
      iu0 = min(int(u+ushift)-int(nu), maxiu0);
      v=fmod1(v_in*psy)*nv;
      iv0 = min(int(v+vshift)-int(nv), maxiv0);
      }

Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
454
    template<typename T> void apply_wscreen(const mav<complex<T>,2> &dirty,
Martin Reinecke's avatar
Martin Reinecke committed
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
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
513
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
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
      mav<complex<T>,2> &dirty2, double w, bool adjoint) const
      {
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(dirty2.shape(), {nx_dirty, ny_dirty});

      double x0 = -0.5*nx_dirty*psx,
             y0 = -0.5*ny_dirty*psy;
      execStatic(nx_dirty/2+1, nthreads, 0, [&](Scheduler &sched)
        {
        while (auto rng=sched.getNext()) for(auto i=rng.lo; i<rng.hi; ++i)
          {
          T fx = T(x0+i*psx);
          fx *= fx;
          for (size_t j=0; j<=ny_dirty/2; ++j)
            {
            T fy = T(y0+j*psy);
            auto ws = wscreen(fx, fy*fy, T(w), adjoint);
            dirty2(i,j) = dirty(i,j)*ws; // lower left
            size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
            if ((i>0)&&(i<i2))
              {
              dirty2(i2,j) = dirty(i2,j)*ws; // lower right
              if ((j>0)&&(j<j2))
                dirty2(i2,j2) = dirty(i2,j2)*ws; // upper right
              }
            if ((j>0)&&(j<j2))
              dirty2(i,j2) = dirty(i,j2)*ws; // upper left
            }
          }
        });
      }
  };

constexpr int logsquare=4;

template<typename T, typename T2=complex<T>> class Helper
  {
  private:
    const GridderConfig &gconf;
    int nu, nv, nsafe, supp;
    T beta;
    const T2 *grid_r;
    T2 *grid_w;
    int su, sv;
    int iu0, iv0; // start index of the current visibility
    int bu0, bv0; // start index of the current buffer

    vector<T2> rbuf, wbuf;
    bool do_w_gridding;
    double w0, xdw;
    size_t nexp;
    size_t nvecs;
    vector<std::mutex> &locks;

    void dump() const
      {
      if (bu0<-nsafe) return; // nothing written into buffer yet

      int idxu = (bu0+nu)%nu;
      int idxv0 = (bv0+nv)%nv;
      for (int iu=0; iu<su; ++iu)
        {
        int idxv = idxv0;
        {
        std::lock_guard<std::mutex> lock(locks[idxu]);
        for (int iv=0; iv<sv; ++iv)
          {
          grid_w[idxu*nv + idxv] += wbuf[iu*sv + iv];
          if (++idxv>=nv) idxv=0;
          }
        }
        if (++idxu>=nu) idxu=0;
        }
      }

    void load()
      {
      int idxu = (bu0+nu)%nu;
      int idxv0 = (bv0+nv)%nv;
      for (int iu=0; iu<su; ++iu)
        {
        int idxv = idxv0;
        for (int iv=0; iv<sv; ++iv)
          {
          rbuf[iu*sv + iv] = grid_r[idxu*nv + idxv];
          if (++idxv>=nv) idxv=0;
          }
        if (++idxu>=nu) idxu=0;
        }
      }

  public:
    const T2 *p0r;
    T2 *p0w;
    T kernel[64] ALIGNED(64);
    static constexpr size_t vlen=native_simd<T>::size();

    Helper(const GridderConfig &gconf_, const T2 *grid_r_, T2 *grid_w_,
      vector<std::mutex> &locks_, double w0_=-1, double dw_=-1)
      : gconf(gconf_), nu(gconf.Nu()), nv(gconf.Nv()), nsafe(gconf.Nsafe()),
        supp(gconf.Supp()), beta(T(gconf.Beta())), grid_r(grid_r_),
        grid_w(grid_w_), su(2*nsafe+(1<<logsquare)), sv(2*nsafe+(1<<logsquare)),
        bu0(-1000000), bv0(-1000000),
        rbuf(su*sv*(grid_r!=nullptr),T(0)),
        wbuf(su*sv*(grid_w!=nullptr),T(0)),
        do_w_gridding(dw_>0),
        w0(w0_),
        xdw(T(1)/dw_),
        nexp(2*supp + do_w_gridding),
        nvecs(vlen*((nexp+vlen-1)/vlen)),
        locks(locks_)
      {}
    ~Helper() { if (grid_w) dump(); }

    int lineJump() const { return sv; }
    T Wfac() const { return kernel[2*supp]; }
    void prep(const UVW &in)
      {
      double u, v;
      gconf.getpix(in.u, in.v, u, v, iu0, iv0);
      double xsupp=2./supp;
      double x0 = xsupp*(iu0-u);
      double y0 = xsupp*(iv0-v);
      for (int i=0; i<supp; ++i)
        {
        kernel[i  ] = T(x0+i*xsupp);
        kernel[i+supp] = T(y0+i*xsupp);
        }
      if (do_w_gridding)
        kernel[2*supp] = min(T(1), T(xdw*xsupp*abs(w0-in.w)));
      for (size_t i=nexp; i<nvecs; ++i)
        kernel[i]=0;
      for (size_t i=0; i<nvecs; ++i)
Martin Reinecke's avatar
Martin Reinecke committed
588
589
590
591
592
593
        {
        kernel[i] = T(1) - kernel[i]*kernel[i];
        kernel[i] = (kernel[i]<0) ? T(-200.) : beta*(sqrt(kernel[i])-T(1));
        }
      for (size_t i=0; i<nvecs; ++i)
        kernel[i] = exp(kernel[i]);
Martin Reinecke's avatar
Martin Reinecke committed
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
      if ((iu0<bu0) || (iv0<bv0) || (iu0+supp>bu0+su) || (iv0+supp>bv0+sv))
        {
        if (grid_w) { dump(); fill(wbuf.begin(), wbuf.end(), T(0)); }
        bu0=((((iu0+nsafe)>>logsquare)<<logsquare))-nsafe;
        bv0=((((iv0+nsafe)>>logsquare)<<logsquare))-nsafe;
        if (grid_r) load();
        }
      p0r = grid_r ? rbuf.data() + sv*(iu0-bu0) + iv0-bv0 : nullptr;
      p0w = grid_w ? wbuf.data() + sv*(iu0-bu0) + iv0-bv0 : nullptr;
      }
  };

template<class T, class Serv> class SubServ
  {
  private:
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
609
610
    Serv &srv;
    mav<idx_t,1> subidx;
Martin Reinecke's avatar
Martin Reinecke committed
611
612

  public:
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
613
    SubServ(Serv &orig, const mav<idx_t,1> &subidx_)
Martin Reinecke's avatar
Martin Reinecke committed
614
615
616
617
      : srv(orig), subidx(subidx_){}
    size_t Nvis() const { return subidx.size(); }
    const Baselines &getBaselines() const { return srv.getBaselines(); }
    UVW getCoord(size_t i) const
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
618
      { return srv.getCoord(subidx(i)); }
Martin Reinecke's avatar
Martin Reinecke committed
619
    complex<T> getVis(size_t i) const
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
620
621
622
623
624
625
      { return srv.getVis(subidx(i)); }
    idx_t getIdx(size_t i) const { return srv.getIdx(subidx(i)); }
    void setVis (size_t i, const complex<T> &v)
      { srv.setVis(subidx(i), v); }
    void addVis (size_t i, const complex<T> &v)
      { srv.addVis(subidx(i), v); }
Martin Reinecke's avatar
Martin Reinecke committed
626
627
628
629
630
631
  };

template<class T, class T2> class MsServ
  {
  private:
    const Baselines &baselines;
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
632
    mav<idx_t,1> idx;
Martin Reinecke's avatar
Martin Reinecke committed
633
    T2 ms;
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
634
    mav<T,2> wgt;
Martin Reinecke's avatar
Martin Reinecke committed
635
636
637
638
639
640
641
    size_t nvis;
    bool have_wgt;

  public:
    using Tsub = SubServ<T, MsServ>;

    MsServ(const Baselines &baselines_,
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
642
    const mav<idx_t,1> &idx_, T2 ms_, const mav<T,2> &wgt_)
Martin Reinecke's avatar
Martin Reinecke committed
643
644
645
646
647
648
      : baselines(baselines_), idx(idx_), ms(ms_), wgt(wgt_),
        nvis(idx.shape(0)), have_wgt(wgt.size()!=0)
      {
      checkShape(ms.shape(), {baselines.Nrows(), baselines.Nchannels()});
      if (have_wgt) checkShape(wgt.shape(), ms.shape());
      }
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
649
    Tsub getSubserv(const mav<idx_t,1> &subidx)
Martin Reinecke's avatar
Martin Reinecke committed
650
651
652
653
654
655
656
657
658
659
660
      { return Tsub(*this, subidx); }
    size_t Nvis() const { return nvis; }
    const Baselines &getBaselines() const { return baselines; }
    UVW getCoord(size_t i) const
      { return baselines.effectiveCoord(idx(i)); }
    complex<T> getVis(size_t i) const
      {
      auto rc = baselines.getRowChan(idx(i));
      return have_wgt ? ms(rc.row, rc.chan)*wgt(rc.row, rc.chan)
                      : ms(rc.row, rc.chan);
      }
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
661
662
    idx_t getIdx(size_t i) const { return idx(i); }
    void setVis (size_t i, const complex<T> &v)
Martin Reinecke's avatar
Martin Reinecke committed
663
664
      {
      auto rc = baselines.getRowChan(idx(i));
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
665
      ms.w()(rc.row, rc.chan) = have_wgt ? v*wgt(rc.row, rc.chan) : v;
Martin Reinecke's avatar
Martin Reinecke committed
666
      }
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
667
    void addVis (size_t i, const complex<T> &v)
Martin Reinecke's avatar
Martin Reinecke committed
668
669
      {
      auto rc = baselines.getRowChan(idx(i));
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
670
      ms.v(rc.row, rc.chan) += have_wgt ? v*wgt(rc.row, rc.chan) : v;
Martin Reinecke's avatar
Martin Reinecke committed
671
672
673
674
      }
  };
template<class T, class T2> MsServ<T, T2> makeMsServ
  (const Baselines &baselines,
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
675
   const mav<idx_t,1> &idx, T2 &ms, const mav<T,2> &wgt)
Martin Reinecke's avatar
Martin Reinecke committed
676
677
678
  { return MsServ<T, T2>(baselines, idx, ms, wgt); }

template<typename T, typename Serv> void x2grid_c
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
679
  (const GridderConfig &gconf, Serv &srv, mav<complex<T>,2> &grid,
Martin Reinecke's avatar
Martin Reinecke committed
680
681
682
683
684
685
686
687
688
689
690
691
  double w0=-1, double dw=-1)
  {
  checkShape(grid.shape(), {gconf.Nu(), gconf.Nv()});
  MR_assert(grid.contiguous(), "grid is not contiguous");
  size_t supp = gconf.Supp();
  size_t nthreads = gconf.Nthreads();
  bool do_w_gridding = dw>0;
  vector<std::mutex> locks(gconf.Nu());

  size_t np = srv.Nvis();
  execGuided(np, nthreads, 100, 0.2, [&](Scheduler &sched)
    {
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
692
    Helper<T> hlp(gconf, nullptr, grid.vdata(), locks, w0, dw);
Martin Reinecke's avatar
Martin Reinecke committed
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
    int jump = hlp.lineJump();
    const T * MRUTIL_RESTRICT ku = hlp.kernel;
    const T * MRUTIL_RESTRICT kv = hlp.kernel+supp;

    while (auto rng=sched.getNext()) for(auto ipart=rng.lo; ipart<rng.hi; ++ipart)
      {
      UVW coord = srv.getCoord(ipart);
      auto flip = coord.FixW();
      hlp.prep(coord);
      auto * MRUTIL_RESTRICT ptr = hlp.p0w;
      auto v(srv.getVis(ipart));
      if (do_w_gridding) v*=hlp.Wfac();
      if (flip) v=conj(v);
      for (size_t cu=0; cu<supp; ++cu)
        {
        complex<T> tmp(v*ku[cu]);
        size_t cv=0;
        for (; cv+3<supp; cv+=4)
          {
          ptr[cv  ] += tmp*kv[cv  ];
          ptr[cv+1] += tmp*kv[cv+1];
          ptr[cv+2] += tmp*kv[cv+2];
          ptr[cv+3] += tmp*kv[cv+3];
          }
        for (; cv<supp; ++cv)
          ptr[cv] += tmp*kv[cv];
        ptr+=jump;
        }
      }
    });
  }

template<typename T, typename Serv> void grid2x_c
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
726
727
  (const GridderConfig &gconf, const mav<complex<T>,2> &grid,
  Serv &srv, double w0=-1, double dw=-1)
Martin Reinecke's avatar
Martin Reinecke committed
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
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
  {
  checkShape(grid.shape(), {gconf.Nu(), gconf.Nv()});
  MR_assert(grid.contiguous(), "grid is not contiguous");
  size_t supp = gconf.Supp();
  size_t nthreads = gconf.Nthreads();
  bool do_w_gridding = dw>0;
  vector<std::mutex> locks(gconf.Nu());

  // Loop over sampling points
  size_t np = srv.Nvis();
  execGuided(np, nthreads, 1000, 0.5, [&](Scheduler &sched)
    {
    Helper<T> hlp(gconf, grid.data(), nullptr, locks, w0, dw);
    int jump = hlp.lineJump();
    const T * MRUTIL_RESTRICT ku = hlp.kernel;
    const T * MRUTIL_RESTRICT kv = hlp.kernel+supp;

    while (auto rng=sched.getNext()) for(auto ipart=rng.lo; ipart<rng.hi; ++ipart)
      {
      UVW coord = srv.getCoord(ipart);
      auto flip = coord.FixW();
      hlp.prep(coord);
      complex<T> r = 0;
      const auto * MRUTIL_RESTRICT ptr = hlp.p0r;
      for (size_t cu=0; cu<supp; ++cu)
        {
        complex<T> tmp(0);
        size_t cv=0;
        for (; cv+3<supp; cv+=4)
          tmp += ptr[cv  ]*kv[cv  ]
               + ptr[cv+1]*kv[cv+1]
               + ptr[cv+2]*kv[cv+2]
               + ptr[cv+3]*kv[cv+3];
        for (; cv<supp; ++cv)
          tmp += ptr[cv] * kv[cv];
        r += tmp*ku[cu];
        ptr += jump;
        }
      if (flip) r=conj(r);
      if (do_w_gridding) r*=hlp.Wfac();
      srv.addVis(ipart, r);
      }
    });
  }

template<typename T> void apply_global_corrections(const GridderConfig &gconf,
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
774
  mav<T,2> &dirty, const ES_Kernel &kernel, double dw, bool divide_by_n)
Martin Reinecke's avatar
Martin Reinecke committed
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
  {
  auto nx_dirty=gconf.Nxdirty();
  auto ny_dirty=gconf.Nydirty();
  size_t nthreads = gconf.Nthreads();
  auto psx=gconf.Pixsize_x();
  auto psy=gconf.Pixsize_y();
  double x0 = -0.5*nx_dirty*psx,
         y0 = -0.5*ny_dirty*psy;
  auto cfu = correction_factors(gconf.Nu(), gconf.Ofactor(),
                                nx_dirty/2+1, gconf.Supp(), nthreads);
  auto cfv = correction_factors(gconf.Nv(), gconf.Ofactor(),
                                ny_dirty/2+1, gconf.Supp(), nthreads);
  execStatic(nx_dirty/2+1, nthreads, 0, [&](Scheduler &sched)
    {
    while (auto rng=sched.getNext()) for(auto i=rng.lo; i<rng.hi; ++i)
      {
      auto fx = T(x0+i*psx);
      fx *= fx;
      for (size_t j=0; j<=ny_dirty/2; ++j)
        {
        auto fy = T(y0+j*psy);
        fy*=fy;
        T fct = 0;
        auto tmp = 1-fx-fy;
        if (tmp>=0)
          {
          auto nm1 = (-fx-fy)/(sqrt(tmp)+1); // accurate form of sqrt(1-x-y)-1
          fct = T(kernel.corfac(nm1*dw));
          if (divide_by_n)
            fct /= nm1+1;
          }
        else // beyond the horizon, don't really know what to do here
          {
          if (divide_by_n)
            fct=0;
          else
            {
            auto nm1 = sqrt(-tmp)-1;
            fct = T(kernel.corfac(nm1*dw));
            }
          }
        fct *= T(cfu[nx_dirty/2-i]*cfv[ny_dirty/2-j]);
        size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
Martin Reinecke's avatar
Martin Reinecke committed
818
        dirty.v(i,j)*=fct;
Martin Reinecke's avatar
Martin Reinecke committed
819
820
        if ((i>0)&&(i<i2))
          {
Martin Reinecke's avatar
Martin Reinecke committed
821
          dirty.v(i2,j)*=fct;
Martin Reinecke's avatar
Martin Reinecke committed
822
          if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
823
            dirty.v(i2,j2)*=fct;
Martin Reinecke's avatar
Martin Reinecke committed
824
825
          }
        if ((j>0)&&(j<j2))
Martin Reinecke's avatar
Martin Reinecke committed
826
          dirty.v(i,j2)*=fct;
Martin Reinecke's avatar
Martin Reinecke committed
827
828
829
830
831
832
833
834
        }
      }
    });
  }

template<typename Serv> class WgridHelper
  {
  private:
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
835
    Serv &srv;
Martin Reinecke's avatar
Martin Reinecke committed
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
    double wmin, dw;
    size_t nplanes, supp;
    vector<vector<idx_t>> minplane;
    size_t verbosity;

    int curplane;
    vector<idx_t> subidx;

    static void wminmax(const Serv &srv, double &wmin, double &wmax)
      {
      size_t nvis = srv.Nvis();

      wmin= 1e38;
      wmax=-1e38;
      // FIXME maybe this can be done more intelligently
      for (size_t ipart=0; ipart<nvis; ++ipart)
        {
        auto wval = abs(srv.getCoord(ipart).w);
        wmin = min(wmin,wval);
        wmax = max(wmax,wval);
        }
      }

    template<typename T> static void update_idx(vector<T> &v, const vector<T> &add,
      const vector<T> &del)
      {
      MR_assert(v.size()>=del.size(), "must not happen");
      vector<T> res;
      res.reserve((v.size()+add.size())-del.size());
      auto iin=v.begin(), ein=v.end();
      auto iadd=add.begin(), eadd=add.end();
      auto irem=del.begin(), erem=del.end();

      while(iin!=ein)
        {
        if ((irem!=erem) && (*iin==*irem))
          {  ++irem; ++iin; } // skip removed entry
        else if ((iadd!=eadd) && (*iadd<*iin))
           res.push_back(*(iadd++)); // add new entry
        else
          res.push_back(*(iin++));
        }
      MR_assert(irem==erem, "must not happen");
      while(iadd!=eadd)
        res.push_back(*(iadd++));
      MR_assert(res.size()==(v.size()+add.size())-del.size(), "must not happen");
      v.swap(res);
      }

  public:
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
886
    WgridHelper(const GridderConfig &gconf, Serv &srv_, size_t verbosity_)
Martin Reinecke's avatar
Martin Reinecke committed
887
888
889
890
891
892
893
894
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
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
      : srv(srv_), verbosity(verbosity_), curplane(-1)
      {
      size_t nvis = srv.Nvis();
      size_t nthreads = gconf.Nthreads();
      double wmax;

      wminmax(srv, wmin, wmax);
      if (verbosity>0) cout << "Using " << nthreads << " thread"
                            << ((nthreads!=1) ? "s" : "") << endl;
      if (verbosity>0) cout << "W range: " << wmin << " to " << wmax << endl;

      double x0 = -0.5*gconf.Nxdirty()*gconf.Pixsize_x(),
             y0 = -0.5*gconf.Nydirty()*gconf.Pixsize_y();
      double nmin = sqrt(max(1.-x0*x0-y0*y0,0.))-1.;
      if (x0*x0+y0*y0>1.)
        nmin = -sqrt(abs(1.-x0*x0-y0*y0))-1.;
      dw = 0.25/abs(nmin);
      nplanes = size_t((wmax-wmin)/dw+2);
      dw = (1.+1e-13)*(wmax-wmin)/(nplanes-1);

      supp = gconf.Supp();
      wmin -= (0.5*supp-1)*dw;
      wmax += (0.5*supp-1)*dw;
      nplanes += supp-2;
      if (verbosity>0) cout << "Kernel support: " << supp << endl;
      if (verbosity>0) cout << "nplanes: " << nplanes << endl;

      minplane.resize(nplanes);
#if 0
      // extra short, but potentially inefficient version:
      for (size_t ipart=0; ipart<nvis; ++ipart)
        {
        int plane0 = max(0,int(1+(abs(srv.getCoord(ipart).w)-(0.5*supp*dw)-wmin)/dw));
        minplane[plane0].push_back(idx_t(ipart));
        }
#else
      // more efficient: precalculate final vector sizes and avoid reallocations
      vector<size_t> cnt(nplanes,0);
      for(size_t ipart=0; ipart<nvis; ++ipart)
        {
        int plane0 = max(0,int(1+(abs(srv.getCoord(ipart).w)-(0.5*supp*dw)-wmin)/dw));
        ++cnt[plane0];
        }

      // fill minplane
      for (size_t j=0; j<nplanes; ++j)
        minplane[j].resize(cnt[j]);
      vector<size_t> ofs(nplanes, 0);
      for (size_t ipart=0; ipart<nvis; ++ipart)
        {
        int plane0 = max(0,int(1+(abs(srv.getCoord(ipart).w)-(0.5*supp*dw)-wmin)/dw));
        minplane[plane0][ofs[plane0]++]=idx_t(ipart);
        }
#endif
      }

    typename Serv::Tsub getSubserv() const
      {
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
945
      auto subidx2 = mav<idx_t, 1>(subidx.data(), {subidx.size()});
Martin Reinecke's avatar
Martin Reinecke committed
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
      return srv.getSubserv(subidx2);
      }
    double W() const { return wmin+curplane*dw; }
    size_t Nvis() const { return subidx.size(); }
    double DW() const { return dw; }
    bool advance()
      {
      if (++curplane>=int(nplanes)) return false;
      update_idx(subidx, minplane[curplane], curplane>=int(supp) ? minplane[curplane-supp] : vector<idx_t>());
      if (verbosity>1)
        cout << "Working on plane " << curplane << " containing " << subidx.size()
             << " visibilities" << endl;
      return true;
      }
  };

template<typename T, typename Serv> void x2dirty(
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
963
  const GridderConfig &gconf, Serv &srv, mav<T,2> &dirty,
Martin Reinecke's avatar
Martin Reinecke committed
964
965
966
967
968
969
970
971
972
  bool do_wstacking, size_t verbosity)
  {
  if (do_wstacking)
    {
    size_t nthreads = gconf.Nthreads();
    if (verbosity>0) cout << "Gridding using improved w-stacking" << endl;
    WgridHelper<Serv> hlp(gconf, srv, verbosity);
    double dw = hlp.DW();
    dirty.fill(0);
973
    mav<complex<T>,2> grid({gconf.Nu(),gconf.Nv()});
Martin Reinecke's avatar
Martin Reinecke committed
974
975
976
977
    while(hlp.advance())  // iterate over w planes
      {
      if (hlp.Nvis()==0) continue;
      grid.fill(0);
Martin Reinecke's avatar
stage 2    
Martin Reinecke committed
978
979
      auto serv = hlp.getSubserv();
      x2grid_c(gconf, serv, grid, hlp.W(), dw);
Martin Reinecke's avatar
Martin Reinecke committed
980
981
982
983
984
985
986
987
988
989
990
991
      gconf.grid2dirty_c_overwrite_wscreen_add(grid, dirty, T(hlp.W()));
      }
    // correct for w gridding etc.
    apply_global_corrections(gconf, dirty, ES_Kernel(gconf.Supp(), gconf.Ofactor(), nthreads), dw, true);
    }
  else
    {
    if (verbosity>0)
      cout << "Gridding without w-stacking: " << srv.Nvis()
           << " visibilities" << endl;
    if (verbosity>0) cout << "Using " << gconf.Nthreads() << " threads" << endl;

992
993
    mav<complex<T>,2> grid({gconf.Nu(), gconf.Nv()});
    grid.fill(0.);
Martin Reinecke's avatar
Martin Reinecke committed
994
    x2grid_c(gconf, srv, grid);
995
    mav<T,2> rgrid(grid.shape());
Martin Reinecke's avatar
Martin Reinecke committed
996
997
    complex2hartley(grid, rgrid, gconf.Nthreads());
    gconf.grid2dirty(rgrid, dirty);
Martin Reinecke's avatar
Martin Reinecke committed
998
999
1000
    }
  }