gridder_cxx.h 27.4 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
22
23
24
25
26
27
28
29
30
31
32
33
#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
 */

/* Copyright (C) 2019 Max-Planck-Society
   Author: Martin Reinecke */

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

#include "pocketfft_hdronly.h"
#include <array>

Martin Reinecke's avatar
Martin Reinecke committed
34
35
36
37
38
namespace gridder {

using namespace std;

template<typename T, size_t ndim> class mav
Martin Reinecke's avatar
Martin Reinecke committed
39
40
41
42
43
  {
  static_assert((ndim>0) && (ndim<3), "only supports 1D and 2D arrays");

  private:
    T *d;
Martin Reinecke's avatar
Martin Reinecke committed
44
45
    array<size_t, ndim> shp;
    array<ptrdiff_t, ndim> str;
Martin Reinecke's avatar
Martin Reinecke committed
46
47

  public:
Martin Reinecke's avatar
Martin Reinecke committed
48
49
50
51
    mav(T *d_, const array<size_t,ndim> &shp_,
        const array<ptrdiff_t,ndim> &str_)
      : d(d_), shp(shp_), str(str_) {}
    mav(T *d_, const array<size_t,ndim> &shp_)
Martin Reinecke's avatar
Martin Reinecke committed
52
53
54
55
56
57
      : d(d_), shp(shp_)
      {
      str[ndim-1]=1;
      for (size_t d=2; d<=ndim; ++d)
        str[ndim-d] = str[ndim-d+1]*shp[ndim-d+1];
      }
Martin Reinecke's avatar
Martin Reinecke committed
58
59
    operator mav<const T, ndim>() const
      { return mav<const T, ndim>(d,shp,str); }
Martin Reinecke's avatar
Martin Reinecke committed
60
61
62
63
64
65
66
    T &operator[](size_t i)
      { return operator()(i); }
    const T &operator[](size_t i) const
      { return operator()(i); }
    T &operator()(size_t i)
      {
      static_assert(ndim==1, "ndim must be 1");
Martin Reinecke's avatar
Martin Reinecke committed
67
      return d[str[0]*i];
Martin Reinecke's avatar
Martin Reinecke committed
68
69
70
71
      }
    const T &operator()(size_t i) const
      {
      static_assert(ndim==1, "ndim must be 1");
Martin Reinecke's avatar
Martin Reinecke committed
72
      return d[str[0]*i];
Martin Reinecke's avatar
Martin Reinecke committed
73
74
75
76
      }
    T &operator()(size_t i, size_t j)
      {
      static_assert(ndim==2, "ndim must be 2");
Martin Reinecke's avatar
Martin Reinecke committed
77
      return d[str[0]*i + str[1]*j];
Martin Reinecke's avatar
Martin Reinecke committed
78
79
80
81
      }
    const T &operator()(size_t i, size_t j) const
      {
      static_assert(ndim==2, "ndim must be 2");
Martin Reinecke's avatar
Martin Reinecke committed
82
      return d[str[0]*i + str[1]*j];
Martin Reinecke's avatar
Martin Reinecke committed
83
84
      }
    size_t shape(size_t i) const { return shp[i]; }
Martin Reinecke's avatar
Martin Reinecke committed
85
    const array<size_t,ndim> &shape() const { return shp; }
Martin Reinecke's avatar
Martin Reinecke committed
86
87
88
89
90
91
92
93
    size_t size() const
      {
      size_t res=1;
      for (auto v: shp) res*=v;
      return res;
      }
    size_t stride(size_t i) const { return str[i]; }
    T *data()
Martin Reinecke's avatar
Martin Reinecke committed
94
      { return d; }
Martin Reinecke's avatar
Martin Reinecke committed
95
    const T *data() const
Martin Reinecke's avatar
Martin Reinecke committed
96
      { return d; }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
97
98
99
100
101
102
103
104
105
106
    bool contiguous() const
      {
      ptrdiff_t stride=1;
      for (size_t i=0; i<ndim; ++i)
        {
        if (str[ndim-1-i]!=stride) return false;
        stride *= shp[ndim-1-i];
        }
      return true;
      }
Martin Reinecke's avatar
Martin Reinecke committed
107
108
  };

Martin Reinecke's avatar
updates    
Martin Reinecke committed
109
110
template<typename T, size_t ndim> using const_mav = mav<const T, ndim>;

Martin Reinecke's avatar
Martin Reinecke committed
111
112
113
114
115
116
117
//
// basic utilities
//

void myassert(bool cond, const char *msg)
  {
  if (cond) return;
Martin Reinecke's avatar
Martin Reinecke committed
118
  throw runtime_error(msg);
Martin Reinecke's avatar
Martin Reinecke committed
119
120
121
  }

template<size_t ndim> void checkShape
Martin Reinecke's avatar
Martin Reinecke committed
122
  (const array<size_t, ndim> &shp1, const array<size_t, ndim> &shp2)
Martin Reinecke's avatar
Martin Reinecke committed
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
  {
  for (size_t i=0; i<ndim; ++i)
    myassert(shp1[i]==shp2[i], "shape mismatch");
  }

/*! 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;
  }

static size_t nthreads = 1;

void set_nthreads(size_t nthreads_)
  {
  myassert(nthreads_>=1, "nthreads must be >= 1");
  nthreads = nthreads_;
  }

size_t get_nthreads()
  { return nthreads; }

//
// Utilities for Gauss-Legendre quadrature
//

static inline double one_minus_x2 (double x)
  { return (fabs(x)>0.1) ? (1.+x)*(1.-x) : 1.-x*x; }

Martin Reinecke's avatar
Martin Reinecke committed
157
void legendre_prep(int n, vector<double> &x, vector<double> &w)
Martin Reinecke's avatar
Martin Reinecke committed
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
  {
  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.);

#pragma omp parallel num_threads(nthreads)
{
  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;
      myassert(++j<100, "convergence problem");
      }

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

//
// Start of real gridder functionality
//

size_t get_supp(double epsilon)
  {
Martin Reinecke's avatar
Martin Reinecke committed
217
  static const vector<double> maxmaperr { 1e8, 0.32, 0.021, 6.2e-4,
Martin Reinecke's avatar
Martin Reinecke committed
218
219
220
221
222
223
224
    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;
Martin Reinecke's avatar
Martin Reinecke committed
225
  throw runtime_error("requested epsilon too small - minimum is 2e-13");
Martin Reinecke's avatar
Martin Reinecke committed
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
  }

template<typename T> class EC_Kernel
  {
  protected:
    size_t supp;
    T beta, emb_;

  public:
    EC_Kernel(size_t supp_)
      : supp(supp_), beta(2.3*supp_), emb_(exp(-beta)) {}
    T operator()(T v) const { return exp(beta*(sqrt(T(1)-v*v)-T(1))); }
    T val_no_emb(T v) const { return exp(beta*sqrt(T(1)-v*v)); }
    T emb() const { return emb_; }
  };

template<typename T> class EC_Kernel_with_correction: public EC_Kernel<T>
  {
  protected:
    static constexpr T pi = T(3.141592653589793238462643383279502884197L);
    int p;
Martin Reinecke's avatar
Martin Reinecke committed
247
    vector<T> x, wgt, psi;
Martin Reinecke's avatar
Martin Reinecke committed
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
    using EC_Kernel<T>::supp;

  public:
    using EC_Kernel<T>::operator();
    EC_Kernel_with_correction(size_t supp_)
      : EC_Kernel<T>(supp_), p(int(1.5*supp_+2))
      {
      legendre_prep(2*p,x,wgt);
      psi=x;
      for (auto &v:psi)
        v=operator()(v);
      }
    /* Compute correction factors for the ES gridding kernel
       This implementation follows eqs. (3.8) to (3.10) of Barnett et al. 2018 */
    T corfac(T v) const
      {
      T tmp=0;
      for (int i=0; i<p; ++i)
        tmp += wgt[i]*psi[i]*cos(pi*supp*v*x[i]);
      return T(1)/(supp*tmp);
      }
  };
/* 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
Martin Reinecke committed
272
vector<double> correction_factors (size_t n, size_t nval, size_t supp)
Martin Reinecke's avatar
Martin Reinecke committed
273
274
  {
  EC_Kernel_with_correction<double> kernel(supp);
Martin Reinecke's avatar
Martin Reinecke committed
275
  vector<double> res(nval);
Martin Reinecke's avatar
Martin Reinecke committed
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
  double xn = 1./n;
#pragma omp parallel for schedule(static) num_threads(nthreads)
  for (size_t k=0; k<nval; ++k)
    res[k] = kernel.corfac(k*xn);
  return res;
  }

template<typename T> struct UVW
  {
  T u, v, w;
  UVW () {}
  UVW (T u_, T v_, T w_) : u(u_), v(v_), w(w_) {}
  UVW operator* (T fct) const
    { return UVW(u*fct, v*fct, w*fct); }
  };

template<typename T> class Baselines
  {
  protected:
Martin Reinecke's avatar
Martin Reinecke committed
295
296
    vector<UVW<T>> coord;
    vector<T> f_over_c;
Martin Reinecke's avatar
Martin Reinecke committed
297
298
299
    size_t nrows, nchan;

  public:
Martin Reinecke's avatar
updates    
Martin Reinecke committed
300
    Baselines(const const_mav<T,2> &coord_, const const_mav<T,1> &freq)
Martin Reinecke's avatar
Martin Reinecke committed
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
      {
      constexpr double speedOfLight = 299792458.;
      myassert(coord_.shape(1)==3, "dimension mismatch");
      nrows = coord_.shape(0);
      nchan = freq.shape(0);
      myassert(nrows*nchan<(size_t(1)<<32), "too many entries in MS");
      f_over_c.resize(nchan);
      for (size_t i=0; i<nchan; ++i)
        f_over_c[i] = freq(i)/speedOfLight;
      coord.resize(nrows);
      for (size_t i=0; i<coord.size(); ++i)
        coord[i] = UVW<T>(coord_(i,0), coord_(i,1), coord_(i,2));
      }

    UVW<T> effectiveCoord(uint32_t index) const
      {
      size_t irow = index/nchan;
      size_t ichan = index-nchan*irow;
      return coord[irow]*f_over_c[ichan];
      }
    UVW<T> effectiveCoord(size_t irow, size_t ichan) const
      { return coord[irow]*f_over_c[ichan]; }
    size_t Nrows() const { return nrows; }
    size_t Nchannels() const { return nchan; }

Martin Reinecke's avatar
Martin Reinecke committed
326
    void effectiveUVW(const mav<const uint32_t,1> &idx, mav<T,2> &res) const
Martin Reinecke's avatar
Martin Reinecke committed
327
328
329
330
331
332
333
334
335
336
337
338
339
      {
      size_t nvis = idx.shape(0);
      myassert(res.shape(0)==nvis, "shape mismatch");
      myassert(res.shape(1)==3, "shape mismatch");
      for (size_t i=0; i<nvis; i++)
        {
        auto uvw = effectiveCoord(idx(i));
        res(i,0) = uvw.u;
        res(i,1) = uvw.v;
        res(i,2) = uvw.w;
        }
      }

Martin Reinecke's avatar
Martin Reinecke committed
340
341
    template<typename T2> void ms2vis(const mav<const T2,2> &ms,
      const mav<const uint32_t,1> &idx, mav<T2,1> &vis) const
Martin Reinecke's avatar
Martin Reinecke committed
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
      {
      myassert(ms.shape(0)==nrows, "shape mismatch");
      myassert(ms.shape(1)==nchan, "shape mismatch");
      size_t nvis = idx.shape(0);
      myassert(vis.shape(0)==nvis, "shape mismatch");
#pragma omp parallel for num_threads(nthreads)
      for (size_t i=0; i<nvis; ++i)
        {
        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
357
358
    template<typename T2> void vis2ms(const mav<const T2,1> &vis,
      const mav<const uint32_t,1> &idx, mav<T2,2> &ms) const
Martin Reinecke's avatar
Martin Reinecke committed
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
      {
      size_t nvis = vis.shape(0);
      myassert(idx.shape(0)==nvis, "shape mismatch");
      myassert(ms.shape(0)==nrows, "shape mismatch");
      myassert(ms.shape(1)==nchan, "shape mismatch");
#pragma omp parallel for num_threads(nthreads)
      for (size_t i=0; i<nvis; ++i)
        {
        auto t = idx(i);
        auto row = t/nchan;
        auto chan = t-row*nchan;
        ms(row, chan) += vis(i);
        }
      }
  };

template<typename T> class GridderConfig
  {
  protected:
    size_t nx_dirty, ny_dirty;
    double eps, psx, psy;
    size_t supp, nsafe, nu, nv;
    T beta;
Martin Reinecke's avatar
Martin Reinecke committed
382
    vector<T> cfu, cfv;
Martin Reinecke's avatar
Martin Reinecke committed
383

Martin Reinecke's avatar
Martin Reinecke committed
384
    complex<T> wscreen(double x, double y, double w, bool adjoint) const
Martin Reinecke's avatar
Martin Reinecke committed
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
      {
      constexpr double pi = 3.141592653589793238462643383279502884197;
#if 1
      double eps = sqrt(x+y);
      double s = sin(eps);
      double nm1 = -s*s/(1.+cos(eps));
#else
      double nm1 = (-x-y)/(sqrt(1.-x-y)+1);
#endif
      double n = nm1+1., xn = 1./n;
      double phase = 2*pi*w*nm1;
      if (adjoint) phase *= -1;
      return complex<T>(cos(phase)*xn, sin(phase)*xn);
      }

  public:
    GridderConfig(size_t nxdirty, size_t nydirty, double epsilon,
      double pixsize_x, double pixsize_y)
      : nx_dirty(nxdirty), ny_dirty(nydirty), eps(epsilon),
        psx(pixsize_x), psy(pixsize_y),
        supp(get_supp(epsilon)), nsafe((supp+1)/2),
Martin Reinecke's avatar
Martin Reinecke committed
406
        nu(max(2*nsafe,2*nx_dirty)), nv(max(2*nsafe,2*ny_dirty)),
Martin Reinecke's avatar
Martin Reinecke committed
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
        beta(2.3*supp),
        cfu(nx_dirty), cfv(ny_dirty)
      {
      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");
      myassert(pixsize_x>0, "pixsize_x must be positive");
      myassert(pixsize_y>0, "pixsize_y must be positive");

      auto tmp = correction_factors(nu, nx_dirty/2+1, supp);
      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, supp);
      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];
      }
    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; }
    T Beta() const { return beta; }

Martin Reinecke's avatar
Martin Reinecke committed
438
    void apply_taper(const mav<const T,2> &img, mav<T,2> &img2, bool divide) const
Martin Reinecke's avatar
Martin Reinecke committed
439
      {
Martin Reinecke's avatar
Martin Reinecke committed
440
441
      checkShape(img.shape(), {nx_dirty, ny_dirty});
      checkShape(img2.shape(), {nx_dirty, ny_dirty});
Martin Reinecke's avatar
Martin Reinecke committed
442
443
444
      if (divide)
        for (size_t i=0; i<nx_dirty; ++i)
          for (size_t j=0; j<ny_dirty; ++j)
Martin Reinecke's avatar
Martin Reinecke committed
445
            img2(i,j) = img(i,j)/(cfu[i]*cfv[j]);
Martin Reinecke's avatar
Martin Reinecke committed
446
447
448
      else
        for (size_t i=0; i<nx_dirty; ++i)
          for (size_t j=0; j<ny_dirty; ++j)
Martin Reinecke's avatar
Martin Reinecke committed
449
            img2(i,j) = img(i,j)*cfu[i]*cfv[j];
Martin Reinecke's avatar
Martin Reinecke committed
450
451
      }

Martin Reinecke's avatar
Martin Reinecke committed
452
    void grid2dirty_c(const mav<const complex<T>,2> &grid, mav<complex<T>,2> &dirty) const
Martin Reinecke's avatar
Martin Reinecke committed
453
      {
Martin Reinecke's avatar
Martin Reinecke committed
454
455
456
457
458
459
460
461
      checkShape(grid.shape(), {nu,nv});
      checkShape(dirty.shape(), {nx_dirty,ny_dirty});
      vector<complex<T>> tmpdat(nu*nv);
      auto tmp=mav<complex<T>,2>(tmpdat.data(),{nu,nv},{nv,1});
      constexpr auto sc = sizeof(complex<T>);
      pocketfft::c2c({nu,nv},{grid.stride(0)*sc,grid.stride(1)*sc},
        {tmp.stride(0)*sc, tmp.stride(1)*sc}, {0,1}, pocketfft::BACKWARD,
        grid.data(), tmp.data(), T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
462
463
464
465
466
467
468
      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;
Martin Reinecke's avatar
Martin Reinecke committed
469
          dirty(i,j) = tmp(i2,j2)*cfu[i]*cfv[j];
Martin Reinecke's avatar
Martin Reinecke committed
470
471
          }
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
472
473
474

    void dirty2grid_c(const const_mav<complex<T>,2> &dirty,
      mav<complex<T>,2> &grid) const
Martin Reinecke's avatar
Martin Reinecke committed
475
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
476
477
478
479
480
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(grid.shape(), {nu, nv});
      for (size_t i=0; i<nu; ++i)
        for (size_t j=0; j<nv; ++j)
          grid(i,j)=0.;
Martin Reinecke's avatar
Martin Reinecke committed
481
482
483
484
485
486
487
      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;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
488
          grid(i2,j2) = dirty(i,j)*cfu[i]*cfv[j];
Martin Reinecke's avatar
Martin Reinecke committed
489
          }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
490
491
      constexpr auto sc = sizeof(complex<T>);
      pocketfft::stride_t strides{grid.stride(0)*sc,grid.stride(1)*sc};
Martin Reinecke's avatar
Martin Reinecke committed
492
      pocketfft::c2c({nu,nv}, strides, strides, {0,1}, pocketfft::FORWARD,
Martin Reinecke's avatar
updates    
Martin Reinecke committed
493
        grid.data(), grid.data(), T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
494
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
495

Martin Reinecke's avatar
Martin Reinecke committed
496
497
498
499
500
501
502
503
504
    inline void getpix(T u_in, T v_in, T &u, T &v, int &iu0, int &iv0) const
      {
      u=fmodulo(u_in*psx, T(1))*nu,
      iu0 = int(u-supp*0.5 + 1 + nu) - nu;
      if (iu0+supp>nu+nsafe) iu0 = nu+nsafe-supp;
      v=fmodulo(v_in*psy, T(1))*nv;
      iv0 = int(v-supp*0.5 + 1 + nv) - nv;
      if (iv0+supp>nv+nsafe) iv0 = nv+nsafe-supp;
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
505
506
507

    void apply_wscreen(const const_mav<complex<T>,2> &dirty,
      mav<complex<T>,2> &dirty2, double w, bool adjoint) const
Martin Reinecke's avatar
Martin Reinecke committed
508
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
509
510
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(dirty2.shape(), {nx_dirty, ny_dirty});
Martin Reinecke's avatar
Martin Reinecke committed
511
512
      double x0 = -0.5*nx_dirty*psx,
             y0 = -0.5*ny_dirty*psy;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
513
#pragma omp parallel for num_threads(nthreads) schedule(static)
Martin Reinecke's avatar
Martin Reinecke committed
514
515
516
517
518
519
520
521
      for (size_t i=0; i<=nx_dirty/2; ++i)
        {
        double fx = x0+i*psx;
        fx *= fx;
        for (size_t j=0; j<=ny_dirty/2; ++j)
          {
          double fy = y0+j*psy;
          auto ws = wscreen(fx, fy*fy, w, adjoint);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
522
          dirty2(i,j) = dirty(i,j)*ws; // lower left
Martin Reinecke's avatar
Martin Reinecke committed
523
524
525
          size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
          if ((i>0)&&(i<i2))
            {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
526
            dirty2(i2,j) = dirty(i2,j)*ws; // lower right
Martin Reinecke's avatar
Martin Reinecke committed
527
            if ((j>0)&&(j<j2))
Martin Reinecke's avatar
updates    
Martin Reinecke committed
528
              dirty2(i2,j2) = dirty(i2,j2)*ws; // upper right
Martin Reinecke's avatar
Martin Reinecke committed
529
530
            }
          if ((j>0)&&(j<j2))
Martin Reinecke's avatar
updates    
Martin Reinecke committed
531
            dirty2(i,j2) = dirty(i,j2)*ws; // upper left
Martin Reinecke's avatar
Martin Reinecke committed
532
533
534
          }
        }
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
535

Martin Reinecke's avatar
Martin Reinecke committed
536
537
538
539
  };

constexpr int logsquare=4;

Martin Reinecke's avatar
Martin Reinecke committed
540
template<typename T, typename T2=complex<T>> class Helper
Martin Reinecke's avatar
Martin Reinecke committed
541
542
543
544
545
546
547
548
549
550
551
  {
  private:
    const GridderConfig<T> &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

Martin Reinecke's avatar
Martin Reinecke committed
552
    vector<T2> rbuf, wbuf;
Martin Reinecke's avatar
Martin Reinecke committed
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
588
589
590
591
592
593

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

#pragma omp critical (gridder_writing_to_grid)
{
      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)
          {
          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;
Martin Reinecke's avatar
Martin Reinecke committed
594
    vector<T> kernel;
Martin Reinecke's avatar
Martin Reinecke committed
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637

    Helper(const GridderConfig<T> &gconf_, const T2 *grid_r_, T2 *grid_w_)
      : gconf(gconf_), nu(gconf.Nu()), nv(gconf.Nv()), nsafe(gconf.Nsafe()),
        supp(gconf.Supp()), beta(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)),
        kernel(2*supp)
      {}
    ~Helper() { if (grid_w) dump(); }

    int lineJump() const { return sv; }

    void prep(T u_in, T v_in)
      {
      T u, v;
      gconf.getpix(u_in, v_in, u, v, iu0, iv0);
      T xsupp=T(2)/supp;
      auto x0 = xsupp*(iu0-u);
      auto y0 = xsupp*(iv0-v);
      for (int i=0; i<supp; ++i)
        {
        auto x = x0+i*xsupp;
        kernel[i  ] = beta*sqrt(T(1)-x*x);
        auto y = y0+i*xsupp;
        kernel[i+supp] = beta*sqrt(T(1)-y*y);
        }
      for (auto &k : kernel)
        k = exp(k);

      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;
      }
  };

Martin Reinecke's avatar
updates    
Martin Reinecke committed
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
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
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
726
727
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
774
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
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
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
886
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
945
946
template<typename T> void vis2grid_c
  (const Baselines<T> &baselines, const GridderConfig<T> &gconf,
  const const_mav<uint32_t,1> &idx, const const_mav<complex<T>,1> &vis,
  mav<complex<T>,2> &grid, const const_mav<T,1> &wgt)
  {
  size_t nvis = vis.shape(0);
  checkShape(idx.shape(), {nvis});
  bool have_wgt = wgt.size()!=0;
  if (have_wgt) checkShape(wgt.shape(), {nvis});

  size_t nu=gconf.Nu(), nv=gconf.Nv();
  checkShape(grid.shape(), {nu, nv});
  myassert(grid.contiguous(), "grid is not contiguous");
  T beta = gconf.Beta();
  size_t supp = gconf.Supp();

#pragma omp parallel num_threads(nthreads)
{
  Helper<T> hlp(gconf, nullptr, grid.data());
  T emb = exp(-2*beta);
  int jump = hlp.lineJump();
  const T * ku = hlp.kernel.data();
  const T * kv = hlp.kernel.data()+supp;

  // Loop over sampling points
#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
    UVW<T> coord = baselines.effectiveCoord(idx(ipart));
    hlp.prep(coord.u, coord.v);
    auto * ptr = hlp.p0w;
    auto v(vis(ipart)*emb);
    if (have_wgt)
      v*=wgt(ipart);
    for (size_t cu=0; cu<supp; ++cu)
      {
      complex<T> tmp(v*ku[cu]);
      for (size_t cv=0; cv<supp; ++cv)
        ptr[cv] += tmp*kv[cv];
      ptr+=jump;
      }
    }
} // end of parallel region
  }

template<typename T> void ms2grid_c
  (const Baselines<T> &baselines, const GridderConfig<T> &gconf,
  const const_mav<uint32_t,1> &idx, const const_mav<complex<T>,2> &ms,
  mav<complex<T>,2> &grid, const const_mav<T,2> &wgt)
  {
  auto nrows = baselines.Nrows();
  auto nchan = baselines.Nchannels();
  size_t nvis = idx.shape(0);
  checkShape(ms.shape(), {nrows, nchan});
  bool have_wgt = wgt.size()!=0;
  if (have_wgt) checkShape(wgt.shape(), {nrows, nchan});

  size_t nu=gconf.Nu(), nv=gconf.Nv();
  checkShape(grid.shape(), {nu, nv});
  myassert(grid.contiguous(), "grid is not contiguous");
  T beta = gconf.Beta();
  size_t supp = gconf.Supp();

#pragma omp parallel num_threads(nthreads)
{
  Helper<T> hlp(gconf, nullptr, grid.data());
  T emb = exp(-2*beta);
  int jump = hlp.lineJump();
  const T * ku = hlp.kernel.data();
  const T * kv = hlp.kernel.data()+supp;

  // 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 * ptr = hlp.p0w;
    auto v(ms(row,chan)*emb);
    if (have_wgt)
      v*=wgt(row,chan);
    for (size_t cu=0; cu<supp; ++cu)
      {
      complex<T> tmp(v*ku[cu]);
      for (size_t cv=0; cv<supp; ++cv)
        ptr[cv] += tmp*kv[cv];
      ptr+=jump;
      }
    }
} // end of parallel region
  }

template<typename T> void grid2vis_c
  (const Baselines<T> &baselines, const GridderConfig<T> &gconf,
  const const_mav<uint32_t,1> &idx, const const_mav<complex<T>,2> &grid,
  mav<complex<T>,1> &vis, const const_mav<T,1> &wgt)
  {
  size_t nvis = vis.shape(0);
  checkShape(idx.shape(), {nvis});
  bool have_wgt = wgt.size()!=0;
  if (have_wgt) checkShape(wgt.shape(), {nvis});

  size_t nu=gconf.Nu(), nv=gconf.Nv();
  checkShape(grid.shape(), {nu, nv});
  myassert(grid.contiguous(), "grid is not contiguous");
  T beta = gconf.Beta();
  size_t supp = gconf.Supp();

  // Loop over sampling points
#pragma omp parallel num_threads(nthreads)
{
  Helper<T> hlp(gconf, grid.data(), nullptr);
  T emb = exp(-2*beta);
  int jump = hlp.lineJump();
  const T * ku = hlp.kernel.data();
  const T * kv = hlp.kernel.data()+supp;

#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
    UVW<T> coord = baselines.effectiveCoord(idx(ipart));
    hlp.prep(coord.u, coord.v);
    complex<T> r = 0;
    const auto * ptr = hlp.p0r;
    for (size_t cu=0; cu<supp; ++cu)
      {
      complex<T> tmp(0);
      for (size_t cv=0; cv<supp; ++cv)
        tmp += ptr[cv] * kv[cv];
      r += tmp*ku[cu];
      ptr += jump;
      }
    if (have_wgt) r*=wgt[ipart];
    vis[ipart] = r*emb;
    }
}
  }

template<typename T> void grid2ms_c
  (const Baselines<T> &baselines, const GridderConfig<T> &gconf,
  const const_mav<uint32_t,1> &idx, const const_mav<complex<T>,2> &grid,
  mav<complex<T>,2> &ms, const const_mav<T,2> &wgt)
  {
  auto nrows = baselines.Nrows();
  auto nchan = baselines.Nchannels();
  size_t nvis = idx.shape(0);
  checkShape(ms.shape(), {nrows, nchan});
  bool have_wgt = wgt.size()!=0;
  if (have_wgt) checkShape(wgt.shape(), {nrows, nchan});

  size_t nu=gconf.Nu(), nv=gconf.Nv();
  checkShape(grid.shape(), {nu, nv});
  myassert(grid.contiguous(), "grid is not contiguous");
  T beta = gconf.Beta();
  size_t supp = gconf.Supp();

  // Loop over sampling points
#pragma omp parallel num_threads(nthreads)
{
  Helper<T> hlp(gconf, grid.data(), nullptr);
  T emb = exp(-2*beta);
  int jump = hlp.lineJump();
  const T * ku = hlp.kernel.data();
  const T * kv = hlp.kernel.data()+supp;

#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);
    complex<T> r = 0;
    const auto * ptr = hlp.p0r;
    for (size_t cu=0; cu<supp; ++cu)
      {
      complex<T> tmp(0);
      for (size_t cv=0; cv<supp; ++cv)
        tmp += ptr[cv] * kv[cv];
      r += tmp*ku[cu];
      ptr += jump;
      }
    if (have_wgt) r*=wgt(row,chan);
    ms(row,chan) = r*emb;
    }
}
  }

template<typename T> void vis2dirty_wstack(const Baselines<T> &baselines,
  const GridderConfig<T> &gconf, const const_mav<uint32_t,1> &idx,
  const const_mav<complex<T>,1> &vis, mav<complex<T>,2> &dirty)
  {
  auto nx_dirty=gconf.Nxdirty();
  auto ny_dirty=gconf.Nydirty();
  auto nu=gconf.Nu();
  auto nv=gconf.Nv();
  auto psx=gconf.Pixsize_x();
  auto psy=gconf.Pixsize_y();
  size_t nvis = vis.shape(0);
  checkShape(idx.shape(),vis.shape());
  checkShape(dirty.shape(),{nx_dirty,ny_dirty});

  // determine w values for every visibility, and min/max w;
  T wmin=T(1e38), wmax=T(-1e38);
  vector<T> wval(nvis);
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
    wval[ipart] = baselines.effectiveCoord(idx(ipart)).w;
    wmin = min(wmin,wval[ipart]);
    wmax = max(wmax,wval[ipart]);
    }
cout << "data w range: " << wmin << " to " << wmax << endl;

  double x0 = -0.5*nx_dirty*psx,
         y0 = -0.5*ny_dirty*psy;
  double nmin = sqrt(max(1.-x0*x0-y0*y0,0.))-1.;
  double dw = 0.25/abs(nmin);

  double w_eps=1e-7; // FIXME
  auto w_supp = get_supp(w_eps);
  EC_Kernel_with_correction<T> kernel(w_supp);
  wmin -= 0.5*w_supp*dw;
  wmax += 0.5*w_supp*dw;
  size_t nplanes = size_t((wmax-wmin)/dw)+2;
cout << "nplanes: " << nplanes << endl;
  vector<size_t> nvis_plane(nplanes,0);
  vector<int> minplane(nvis);
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
    int plane0 = int((wval[ipart]-wmin)/dw+0.5*w_supp)-int(w_supp-1);
    minplane[ipart]=plane0;
    for (int i=max<int>(0,plane0); i<min<int>(nplanes,plane0+w_supp); ++i)
      ++nvis_plane[i];
    }

  for (ptrdiff_t i=0; i<nx_dirty; ++i)
    for (ptrdiff_t j=0; j<ny_dirty; ++j)
      dirty(i,j) = 0.;
  for (size_t iw=0; iw<nplanes; ++iw)
    {
cout << "working on w plane #" << iw << " containing " << nvis_plane[iw]
     << " visibilities" << endl;
    if (nvis_plane[iw]==0) continue;
    double wcur = wmin+iw*dw;
    size_t cnt=0;
    vector<complex<T>> vis_loc_ (nvis_plane[iw]);
    auto vis_loc = mav<complex<T>,1>(vis_loc_.data(), {nvis_plane[iw]});
    vector<uint32_t> idx_loc_ (nvis_plane[iw]);
    auto idx_loc = mav<uint32_t,1>(idx_loc_.data(), {nvis_plane[iw]});
    for (size_t ipart=0; ipart<nvis; ++ipart)
      if ((iw>=minplane[ipart]) && (iw<minplane[ipart]+w_supp))
        {
myassert(cnt<nvis_plane[iw],"must not happen");
        double x=min(1.,2./(w_supp*dw)*abs(wcur-wval[ipart]));
        vis_loc[cnt] = vis[ipart]*kernel(x);
        idx_loc[cnt] = idx[ipart];
        ++cnt;
        }
myassert(cnt==nvis_plane[iw],"must not happen 2");
    auto grid_=vector<complex<T>>(nu*nv);
    auto grid=mav<complex<T>,2>(grid_.data(),{nu, nv});
    vis2grid_c(baselines, gconf, const_mav<uint32_t,1>(idx_loc), const_mav<complex<T>,1>(vis_loc), grid, const_mav<T,1>(nullptr,{0}));
    auto tdirty_=vector<complex<T>>(nx_dirty*ny_dirty);
    auto tdirty=mav<complex<T>,2>(tdirty_.data(),{nx_dirty, ny_dirty});
    gconf.grid2dirty_c(grid, tdirty);
    gconf.apply_wscreen(tdirty, tdirty, wcur, false);
    for (ptrdiff_t i=0; i<nx_dirty; ++i)
      for (ptrdiff_t j=0; j<ny_dirty; ++j)
        dirty(i,j) += tdirty(i,j);
    }
  // correct for w gridding
cout << "applying correction for gridding in w direction" << endl;

#pragma omp parallel num_threads(nthreads)
{
#pragma omp for schedule(static)
  for (size_t i=0; i<=nx_dirty/2; ++i)
    {
    double fx = x0+i*psx;
    fx *= fx;
    for (size_t j=0; j<=ny_dirty/2; ++j)
      {
      double fy = y0+j*psy;
      fy*=fy;
#if 1
      auto nm1=cos(sqrt(fx+fy)) -1; // cosine profile
#else
      auto nm1 = (-fx-fy)/(sqrt(1.-fx-fy)+1.);
#endif
      T fct = kernel.corfac(nm1*dw);
      size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
      dirty(i,j)*=fct;
      if ((i>0)&&(i<i2))
        {
        dirty(i2,j)*=fct;
        if ((j>0)&&(j<j2))
          dirty(i2,j2)*=fct;
        }
      if ((j>0)&&(j<j2))
        dirty(i,j2)*=fct;
      }
    }
}
  }

Martin Reinecke's avatar
Martin Reinecke committed
947
948
} // namespace gridder

Martin Reinecke's avatar
Martin Reinecke committed
949
#endif