gridder_cxx.h 51.9 KB
Newer Older
Martin Reinecke's avatar
Martin Reinecke committed
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
#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>
Martin Reinecke's avatar
Martin Reinecke committed
30
#include <array>
Martin Reinecke's avatar
Martin Reinecke committed
31
32

#include "pocketfft_hdronly.h"
Martin Reinecke's avatar
Martin Reinecke committed
33
34

#if defined(__GNUC__)
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
35
#define NOINLINE __attribute__((noinline))
Martin Reinecke's avatar
Martin Reinecke committed
36
#define ALIGNED(align) __attribute__ ((aligned(align)))
Martin Reinecke's avatar
Martin Reinecke committed
37
#define RESTRICT __restrict__
Martin Reinecke's avatar
Martin Reinecke committed
38
39
40
#else
#define NOINLINE
#define ALIGNED(align)
Martin Reinecke's avatar
Martin Reinecke committed
41
#define RESTRICT
Martin Reinecke's avatar
Martin Reinecke committed
42
#endif
Martin Reinecke's avatar
Martin Reinecke committed
43

Martin Reinecke's avatar
Martin Reinecke committed
44
45
namespace gridder {

46
47
namespace detail {

Martin Reinecke's avatar
Martin Reinecke committed
48
49
using namespace std;

50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
template<typename T> struct VLEN { static constexpr size_t val=1; };

#if (defined(__AVX512F__))
template<> struct VLEN<float> { static constexpr size_t val=16; };
template<> struct VLEN<double> { static constexpr size_t val=8; };
#elif (defined(__AVX__))
template<> struct VLEN<float> { static constexpr size_t val=8; };
template<> struct VLEN<double> { static constexpr size_t val=4; };
#elif (defined(__SSE2__))
template<> struct VLEN<float> { static constexpr size_t val=4; };
template<> struct VLEN<double> { static constexpr size_t val=2; };
#elif (defined(__VSX__))
template<> struct VLEN<float> { static constexpr size_t val=4; };
template<> struct VLEN<double> { static constexpr size_t val=2; };
#endif

Martin Reinecke's avatar
Martin Reinecke committed
66
template<typename T, size_t ndim> class mav
Martin Reinecke's avatar
Martin Reinecke committed
67
68
69
70
71
  {
  static_assert((ndim>0) && (ndim<3), "only supports 1D and 2D arrays");

  private:
    T *d;
Martin Reinecke's avatar
Martin Reinecke committed
72
73
    array<size_t, ndim> shp;
    array<ptrdiff_t, ndim> str;
Martin Reinecke's avatar
Martin Reinecke committed
74
75

  public:
Martin Reinecke's avatar
Martin Reinecke committed
76
77
78
79
    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
80
81
82
83
84
85
      : 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
86
87
    operator mav<const T, ndim>() const
      { return mav<const T, ndim>(d,shp,str); }
88
    T &operator[](size_t i) const
Martin Reinecke's avatar
Martin Reinecke committed
89
      { return operator()(i); }
90
    T &operator()(size_t i) const
Martin Reinecke's avatar
Martin Reinecke committed
91
92
      {
      static_assert(ndim==1, "ndim must be 1");
Martin Reinecke's avatar
Martin Reinecke committed
93
      return d[str[0]*i];
Martin Reinecke's avatar
Martin Reinecke committed
94
      }
95
    T &operator()(size_t i, size_t j) const
Martin Reinecke's avatar
Martin Reinecke committed
96
97
      {
      static_assert(ndim==2, "ndim must be 2");
Martin Reinecke's avatar
Martin Reinecke committed
98
      return d[str[0]*i + str[1]*j];
Martin Reinecke's avatar
Martin Reinecke committed
99
100
      }
    size_t shape(size_t i) const { return shp[i]; }
Martin Reinecke's avatar
Martin Reinecke committed
101
    const array<size_t,ndim> &shape() const { return shp; }
Martin Reinecke's avatar
Martin Reinecke committed
102
103
104
105
106
107
    size_t size() const
      {
      size_t res=1;
      for (auto v: shp) res*=v;
      return res;
      }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
108
    ptrdiff_t stride(size_t i) const { return str[i]; }
109
    T *data() const
Martin Reinecke's avatar
Martin Reinecke committed
110
      { return d; }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
111
112
113
114
115
116
117
118
119
120
    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
121
122
123
124
125
126
127
128
129
130
    void fill(const T &val) const
      {
      if (ndim==1)
        for (size_t i=0; i<shp[0]; ++i)
          d[str[0]*i]=val;
      else if (ndim==2)
        for (size_t i=0; i<shp[0]; ++i)
          for (size_t j=0; j<shp[1]; ++j)
            d[str[0]*i + str[1]*j] = val;
      }
Martin Reinecke's avatar
Martin Reinecke committed
131
132
  };

Martin Reinecke's avatar
updates    
Martin Reinecke committed
133
template<typename T, size_t ndim> using const_mav = mav<const T, ndim>;
Martin Reinecke's avatar
Martin Reinecke committed
134
135
136
137
138
139
140
141
template<typename T, size_t ndim> const_mav<T, ndim> cmav (const mav<T, ndim> &mav)
  { return const_mav<T, ndim>(mav.data(), mav.shape()); }
template<typename T, size_t ndim> const_mav<T, ndim> nullmav()
  {
  array<size_t,ndim> shp;
  shp.fill(0);
  return const_mav<T, ndim>(nullptr, shp);
  }
Martin Reinecke's avatar
Martin Reinecke committed
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
template<typename T, size_t ndim> class tmpStorage
  {
  private:
    vector<T> d;
    mav<T,ndim> mav_;

    static size_t prod(const array<size_t,ndim> &shp)
      {
      size_t res=1;
      for (auto v: shp) res*=v;
      return res;
      }

  public:
    tmpStorage(const array<size_t,ndim> &shp)
      : d(prod(shp)), mav_(d.data(), shp) {}
Martin Reinecke's avatar
Martin Reinecke committed
158
159
    mav<T,ndim> &getMav() { return mav_; }
    const_mav<T,ndim> getCmav() { return cmav(mav_); }
Martin Reinecke's avatar
Martin Reinecke committed
160
161
162
    void fill(const T & val)
      { std::fill(d.begin(), d.end(), val); }
  };
Martin Reinecke's avatar
Martin Reinecke committed
163
164
165
166
167
168
169
//
// basic utilities
//

void myassert(bool cond, const char *msg)
  {
  if (cond) return;
Martin Reinecke's avatar
Martin Reinecke committed
170
  throw runtime_error(msg);
Martin Reinecke's avatar
Martin Reinecke committed
171
172
173
  }

template<size_t ndim> void checkShape
Martin Reinecke's avatar
Martin Reinecke committed
174
  (const array<size_t, ndim> &shp1, const array<size_t, ndim> &shp2)
Martin Reinecke's avatar
Martin Reinecke committed
175
176
177
178
179
180
181
182
183
184
185
186
  {
  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);
Martin Reinecke's avatar
Martin Reinecke committed
187
188
189
  T tmp = v1+v2;
  if (tmp<0)
    tmp=fmod(v1,v2)+v2;
Martin Reinecke's avatar
Martin Reinecke committed
190
191
192
193
194
195
196
197
198
199
  return (tmp==v2) ? T(0) : tmp;
  }

//
// 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
200
void legendre_prep(int n, vector<double> &x, vector<double> &w, size_t nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
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
  {
  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
260
  static const vector<double> maxmaperr { 1e8, 0.32, 0.021, 6.2e-4,
Martin Reinecke's avatar
Martin Reinecke committed
261
262
263
264
265
266
267
    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
268
  throw runtime_error("requested epsilon too small - minimum is 2e-13");
Martin Reinecke's avatar
Martin Reinecke committed
269
270
  }

Martin Reinecke's avatar
Martin Reinecke committed
271
272
273
274
275
struct RowChan
  {
  size_t row, chan;
  };

276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
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
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
template<typename T> void complex2hartley
  (const const_mav<complex<T>, 2> &grid, const mav<T,2> &grid2, size_t nthreads)
  {
  myassert(grid.shape()==grid2.shape(), "shape mismatch");
  size_t nu=grid.shape(0), nv=grid.shape(1);

#pragma omp parallel for num_threads(nthreads)
  for (size_t u=0; u<nu; ++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;
      grid2(u,v) += T(0.5)*(grid( u, v).real()+grid( u, v).imag()+
                            grid(xu,xv).real()-grid(xu,xv).imag());
      }
    }
  }

template<typename T> void hartley2complex
  (const const_mav<T,2> &grid, const mav<complex<T>,2> &grid2, size_t nthreads)
  {
  myassert(grid.shape()==grid2.shape(), "shape mismatch");
  size_t nu=grid.shape(0), nv=grid.shape(1);

#pragma omp parallel for num_threads(nthreads)
  for (size_t u=0; u<nu; ++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);
      grid2(u,v) = std::complex<T>(v1+v2, v1-v2);
      }
    }
  }

template<typename T> void hartley2_2D(const const_mav<T,2> &in, const mav<T,2> &out, size_t nthreads)
  {
  myassert(in.shape()==out.shape(), "shape mismatch");
  size_t nu=in.shape(0), nv=in.shape(1);
  ptrdiff_t sz=ptrdiff_t(sizeof(T));
  pocketfft::stride_t stri{sz*in.stride(0), sz*in.stride(1)};
  pocketfft::stride_t stro{sz*out.stride(0), sz*out.stride(1)};
  auto d_i = in.data();
  auto ptmp = out.data();
  pocketfft::r2r_separable_hartley({nu, nv}, stri, stro, {0,1}, d_i, ptmp, T(1),
    nthreads);
#pragma omp parallel for num_threads(nthreads)
  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
342
343
344
template<typename T> class EC_Kernel
  {
  protected:
Martin Reinecke's avatar
Martin Reinecke committed
345
    T beta;
Martin Reinecke's avatar
Martin Reinecke committed
346
347

  public:
Martin Reinecke's avatar
Martin Reinecke committed
348
    EC_Kernel(size_t supp) : beta(2.3*supp) {}
Martin Reinecke's avatar
Martin Reinecke committed
349
350
351
352
353
354
355
356
    T operator()(T v) const { return exp(beta*(sqrt(T(1)-v*v)-T(1))); }
  };

template<typename T> class EC_Kernel_with_correction: public EC_Kernel<T>
  {
  protected:
    static constexpr T pi = T(3.141592653589793238462643383279502884197L);
    int p;
357
    vector<double> x, wgt, psi;
Martin Reinecke's avatar
Martin Reinecke committed
358
    size_t supp;
Martin Reinecke's avatar
Martin Reinecke committed
359
360
361

  public:
    using EC_Kernel<T>::operator();
Martin Reinecke's avatar
Martin Reinecke committed
362
    EC_Kernel_with_correction(size_t supp_, size_t nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
363
      : EC_Kernel<T>(supp_), p(int(1.5*supp_+2)), supp(supp_)
Martin Reinecke's avatar
Martin Reinecke committed
364
      {
Martin Reinecke's avatar
Martin Reinecke committed
365
      legendre_prep(2*p,x,wgt,nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
366
367
368
369
370
371
372
373
374
375
376
      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]);
Martin Reinecke's avatar
tweaks    
Martin Reinecke committed
377
      return T(1./(supp*tmp));
Martin Reinecke's avatar
Martin Reinecke committed
378
379
380
381
      }
  };
/* 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
382
383
vector<double> correction_factors (size_t n, size_t nval, size_t supp,
  size_t nthreads)
Martin Reinecke's avatar
Martin Reinecke committed
384
  {
Martin Reinecke's avatar
Martin Reinecke committed
385
  EC_Kernel_with_correction<double> kernel(supp, nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
386
  vector<double> res(nval);
Martin Reinecke's avatar
Martin Reinecke committed
387
388
389
390
391
392
393
394
395
396
397
398
399
400
  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); }
401
402
403
404
405
406
407
  void Flip() { u=-u; v=-v; w=-w; }
  bool FixW()
    {
    bool flip = w<0;
    if (flip) Flip();
    return flip;
    }
Martin Reinecke's avatar
Martin Reinecke committed
408
409
410
411
412
  };

template<typename T> class Baselines
  {
  protected:
Martin Reinecke's avatar
Martin Reinecke committed
413
414
    vector<UVW<T>> coord;
    vector<T> f_over_c;
Martin Reinecke's avatar
Martin Reinecke committed
415
    size_t nrows, nchan;
416
    size_t shift, mask;
Martin Reinecke's avatar
Martin Reinecke committed
417
418

  public:
Martin Reinecke's avatar
updates    
Martin Reinecke committed
419
    Baselines(const const_mav<T,2> &coord_, const const_mav<T,1> &freq)
Martin Reinecke's avatar
Martin Reinecke committed
420
421
422
423
424
      {
      constexpr double speedOfLight = 299792458.;
      myassert(coord_.shape(1)==3, "dimension mismatch");
      nrows = coord_.shape(0);
      nchan = freq.shape(0);
425
426
427
      shift=0;
      while((size_t(1)<<shift)<nchan) ++shift;
      mask=(size_t(1)<<shift)-1;
Martin Reinecke's avatar
Martin Reinecke committed
428
429
430
      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)
431
432
        {
        myassert(freq[i]>0, "negative channel frequency encountered");
Martin Reinecke's avatar
Martin Reinecke committed
433
        f_over_c[i] = freq(i)/speedOfLight;
434
        }
Martin Reinecke's avatar
Martin Reinecke committed
435
436
437
438
439
      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));
      }

Martin Reinecke's avatar
Martin Reinecke committed
440
    RowChan getRowChan(uint32_t index) const
441
      { return RowChan{index>>shift, index&mask}; }
Martin Reinecke's avatar
Martin Reinecke committed
442
443
444
445
446

    UVW<T> effectiveCoord(const RowChan &rc) const
      { return coord[rc.row]*f_over_c[rc.chan]; }
    UVW<T> effectiveCoord(uint32_t index) const
      { return effectiveCoord(getRowChan(index)); }
Martin Reinecke's avatar
Martin Reinecke committed
447
448
    size_t Nrows() const { return nrows; }
    size_t Nchannels() const { return nchan; }
449
450
    uint32_t getIdx(size_t irow, size_t ichan) const
      { return ichan+(irow<<shift); }
Martin Reinecke's avatar
Martin Reinecke committed
451

Martin Reinecke's avatar
Martin Reinecke committed
452
    void effectiveUVW(const mav<const uint32_t,1> &idx, mav<T,2> &res) const
Martin Reinecke's avatar
Martin Reinecke committed
453
454
455
456
457
458
459
460
461
462
463
464
465
      {
      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
466
    template<typename T2> void ms2vis(const mav<const T2,2> &ms,
Martin Reinecke's avatar
Martin Reinecke committed
467
      const mav<const uint32_t,1> &idx, mav<T2,1> &vis, size_t nthreads) const
Martin Reinecke's avatar
Martin Reinecke committed
468
469
470
471
472
473
474
475
      {
      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)
        {
Martin Reinecke's avatar
Martin Reinecke committed
476
477
        auto rc = getRowChan(idx(i));
        vis[i] = ms(rc.row, rc.chan);
Martin Reinecke's avatar
Martin Reinecke committed
478
479
480
        }
      }

Martin Reinecke's avatar
Martin Reinecke committed
481
    template<typename T2> void vis2ms(const mav<const T2,1> &vis,
Martin Reinecke's avatar
Martin Reinecke committed
482
      const mav<const uint32_t,1> &idx, mav<T2,2> &ms, size_t nthreads) const
Martin Reinecke's avatar
Martin Reinecke committed
483
484
485
486
487
488
489
490
      {
      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)
        {
Martin Reinecke's avatar
Martin Reinecke committed
491
492
        auto rc = getRowChan(idx(i));
        ms(rc.row, rc.chan) += vis(i);
Martin Reinecke's avatar
Martin Reinecke committed
493
494
495
496
497
498
499
500
        }
      }
  };

template<typename T> class GridderConfig
  {
  protected:
    size_t nx_dirty, ny_dirty;
501
    T eps, psx, psy;
Martin Reinecke's avatar
Martin Reinecke committed
502
503
    size_t supp, nsafe, nu, nv;
    T beta;
Martin Reinecke's avatar
Martin Reinecke committed
504
    vector<T> cfu, cfv;
Martin Reinecke's avatar
Martin Reinecke committed
505
    size_t nthreads;
Martin Reinecke's avatar
Martin Reinecke committed
506

Martin Reinecke's avatar
Martin Reinecke committed
507
    complex<T> wscreen(double x, double y, double w, bool adjoint) const
Martin Reinecke's avatar
Martin Reinecke committed
508
509
      {
      constexpr double pi = 3.141592653589793238462643383279502884197;
510
511
512
      double tmp = 1-x-y;
      if (tmp<0) return 0.;
      double nm1 = (-x-y)/(sqrt(tmp)+1); // more accurate form of sqrt(1-x-y)
Martin Reinecke's avatar
Martin Reinecke committed
513
514
515
516
517
518
519
520
      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,
Martin Reinecke's avatar
Martin Reinecke committed
521
      double pixsize_x, double pixsize_y, size_t nthreads_)
Martin Reinecke's avatar
Martin Reinecke committed
522
523
524
      : 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
525
        nu(max(2*nsafe,2*nx_dirty)), nv(max(2*nsafe,2*ny_dirty)),
Martin Reinecke's avatar
Martin Reinecke committed
526
        beta(2.3*supp),
Martin Reinecke's avatar
Martin Reinecke committed
527
        cfu(nx_dirty), cfv(ny_dirty), nthreads(nthreads_)
Martin Reinecke's avatar
Martin Reinecke committed
528
529
530
531
532
533
534
      {
      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");

Martin Reinecke's avatar
Martin Reinecke committed
535
      auto tmp = correction_factors(nu, nx_dirty/2+1, supp, nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
536
537
538
539
      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];
Martin Reinecke's avatar
Martin Reinecke committed
540
      tmp = correction_factors(nv, ny_dirty/2+1, supp, nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
      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
556
    size_t Nthreads() const { return nthreads; }
Martin Reinecke's avatar
Martin Reinecke committed
557

Martin Reinecke's avatar
Martin Reinecke committed
558
    void apply_taper(const mav<const T,2> &img, mav<T,2> &img2, bool divide) const
Martin Reinecke's avatar
Martin Reinecke committed
559
      {
Martin Reinecke's avatar
Martin Reinecke committed
560
561
      checkShape(img.shape(), {nx_dirty, ny_dirty});
      checkShape(img2.shape(), {nx_dirty, ny_dirty});
Martin Reinecke's avatar
Martin Reinecke committed
562
563
564
      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
565
            img2(i,j) = img(i,j)/(cfu[i]*cfv[j]);
Martin Reinecke's avatar
Martin Reinecke committed
566
567
568
      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
569
            img2(i,j) = img(i,j)*cfu[i]*cfv[j];
Martin Reinecke's avatar
Martin Reinecke committed
570
571
      }

572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
    void grid2dirty(const const_mav<T,2> &grid, const mav<T,2> &dirty) const
      {
      checkShape(grid.shape(), {nu,nv});
      checkShape(dirty.shape(), {nx_dirty,ny_dirty});
      tmpStorage<T,2> tmpdat({nu,nv});
      auto tmav = tmpdat.getMav();
      hartley2_2D<T>(grid, tmav, nthreads);
      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;
          dirty(i,j) = tmav(i2,j2)*cfu[i]*cfv[j];
          }
      }

Martin Reinecke's avatar
Martin Reinecke committed
590
    void grid2dirty_c(const mav<const complex<T>,2> &grid, mav<complex<T>,2> &dirty) const
Martin Reinecke's avatar
Martin Reinecke committed
591
      {
Martin Reinecke's avatar
Martin Reinecke committed
592
593
      checkShape(grid.shape(), {nu,nv});
      checkShape(dirty.shape(), {nx_dirty,ny_dirty});
594
595
      tmpStorage<complex<T>,2> tmpdat({nu,nv});
      auto tmp = tmpdat.getMav();
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
596
      constexpr auto sc = ptrdiff_t(sizeof(complex<T>));
Martin Reinecke's avatar
Martin Reinecke committed
597
598
599
      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
600
601
602
603
604
605
606
      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
607
          dirty(i,j) = tmp(i2,j2)*cfu[i]*cfv[j];
Martin Reinecke's avatar
Martin Reinecke committed
608
609
          }
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
610

611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
    void dirty2grid(const const_mav<T,2> &dirty, mav<T,2> &grid) const
      {
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(grid.shape(), {nu, nv});
      grid.fill(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;
          grid(i2,j2) = dirty(i,j)*cfu[i]*cfv[j];
          }
      hartley2_2D<T>(grid, grid, nthreads);
      }

Martin Reinecke's avatar
updates    
Martin Reinecke committed
628
629
    void dirty2grid_c(const const_mav<complex<T>,2> &dirty,
      mav<complex<T>,2> &grid) const
Martin Reinecke's avatar
Martin Reinecke committed
630
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
631
632
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(grid.shape(), {nu, nv});
633
      grid.fill(0);
Martin Reinecke's avatar
Martin Reinecke committed
634
635
636
637
638
639
640
      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
641
          grid(i2,j2) = dirty(i,j)*cfu[i]*cfv[j];
Martin Reinecke's avatar
Martin Reinecke committed
642
          }
Martin Reinecke's avatar
cleanup    
Martin Reinecke committed
643
      constexpr auto sc = ptrdiff_t(sizeof(complex<T>));
Martin Reinecke's avatar
updates    
Martin Reinecke committed
644
      pocketfft::stride_t strides{grid.stride(0)*sc,grid.stride(1)*sc};
Martin Reinecke's avatar
Martin Reinecke committed
645
      pocketfft::c2c({nu,nv}, strides, strides, {0,1}, pocketfft::FORWARD,
Martin Reinecke's avatar
updates    
Martin Reinecke committed
646
        grid.data(), grid.data(), T(1), nthreads);
Martin Reinecke's avatar
Martin Reinecke committed
647
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
648

Martin Reinecke's avatar
Martin Reinecke committed
649
    void getpix(T u_in, T v_in, T &u, T &v, int &iu0, int &iv0) const
Martin Reinecke's avatar
Martin Reinecke committed
650
651
652
      {
      u=fmodulo(u_in*psx, T(1))*nu,
      iu0 = int(u-supp*0.5 + 1 + nu) - nu;
Martin Reinecke's avatar
Martin Reinecke committed
653
      iu0 = min<int>(iu0, (nu+nsafe)-supp);
Martin Reinecke's avatar
Martin Reinecke committed
654
655
      v=fmodulo(v_in*psy, T(1))*nv;
      iv0 = int(v-supp*0.5 + 1 + nv) - nv;
Martin Reinecke's avatar
Martin Reinecke committed
656
      iv0 = min<int>(iv0, (nv+nsafe)-supp);
Martin Reinecke's avatar
Martin Reinecke committed
657
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
658
659
660

    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
661
      {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
662
663
      checkShape(dirty.shape(), {nx_dirty, ny_dirty});
      checkShape(dirty2.shape(), {nx_dirty, ny_dirty});
Martin Reinecke's avatar
Martin Reinecke committed
664
665
      double x0 = -0.5*nx_dirty*psx,
             y0 = -0.5*ny_dirty*psy;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
666
#pragma omp parallel for num_threads(nthreads) schedule(static)
Martin Reinecke's avatar
Martin Reinecke committed
667
668
669
670
671
672
673
674
      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
675
          dirty2(i,j) = dirty(i,j)*ws; // lower left
Martin Reinecke's avatar
Martin Reinecke committed
676
677
678
          size_t i2 = nx_dirty-i, j2 = ny_dirty-j;
          if ((i>0)&&(i<i2))
            {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
679
            dirty2(i2,j) = dirty(i2,j)*ws; // lower right
Martin Reinecke's avatar
Martin Reinecke committed
680
            if ((j>0)&&(j<j2))
Martin Reinecke's avatar
updates    
Martin Reinecke committed
681
              dirty2(i2,j2) = dirty(i2,j2)*ws; // upper right
Martin Reinecke's avatar
Martin Reinecke committed
682
683
            }
          if ((j>0)&&(j<j2))
Martin Reinecke's avatar
updates    
Martin Reinecke committed
684
            dirty2(i,j2) = dirty(i,j2)*ws; // upper left
Martin Reinecke's avatar
Martin Reinecke committed
685
686
687
          }
        }
      }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
688

Martin Reinecke's avatar
Martin Reinecke committed
689
690
691
692
  };

constexpr int logsquare=4;

Martin Reinecke's avatar
Martin Reinecke committed
693
template<typename T, typename T2=complex<T>> class Helper
Martin Reinecke's avatar
Martin Reinecke committed
694
695
696
697
698
699
700
701
702
703
704
  {
  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
705
    vector<T2> rbuf, wbuf;
Martin Reinecke's avatar
Martin Reinecke committed
706
707
708
    bool do_w_gridding;
    T w0, xdw;
    size_t nexp;
709
    size_t nvecs;
Martin Reinecke's avatar
Martin Reinecke committed
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

    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
751
    T kernel[64] ALIGNED(64);
Martin Reinecke's avatar
Martin Reinecke committed
752

Martin Reinecke's avatar
Martin Reinecke committed
753
    Helper(const GridderConfig<T> &gconf_, const T2 *grid_r_, T2 *grid_w_, T w0_=-1, T dw_=-1)
Martin Reinecke's avatar
Martin Reinecke committed
754
755
756
757
758
759
      : 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)),
Martin Reinecke's avatar
Martin Reinecke committed
760
761
762
763
764
        do_w_gridding(dw_>0),
        w0(w0_),
        xdw(T(1)/dw_),
        nexp(2*supp + do_w_gridding),
        nvecs(VLEN<T>::val*((nexp+VLEN<T>::val-1)/VLEN<T>::val))
Martin Reinecke's avatar
Martin Reinecke committed
765
766
767
768
      {}
    ~Helper() { if (grid_w) dump(); }

    int lineJump() const { return sv; }
Martin Reinecke's avatar
Martin Reinecke committed
769
    T Wfac() const { return kernel[2*supp]; }
Martin Reinecke's avatar
Martin Reinecke committed
770
    void prep(const UVW<T> &in)
Martin Reinecke's avatar
Martin Reinecke committed
771
772
      {
      T u, v;
Martin Reinecke's avatar
Martin Reinecke committed
773
      gconf.getpix(in.u, in.v, u, v, iu0, iv0);
Martin Reinecke's avatar
Martin Reinecke committed
774
775
776
777
778
      T xsupp=T(2)/supp;
      auto x0 = xsupp*(iu0-u);
      auto y0 = xsupp*(iv0-v);
      for (int i=0; i<supp; ++i)
        {
Martin Reinecke's avatar
Martin Reinecke committed
779
780
        kernel[i  ] = x0+i*xsupp;
        kernel[i+supp] = y0+i*xsupp;
Martin Reinecke's avatar
Martin Reinecke committed
781
        }
Martin Reinecke's avatar
Martin Reinecke committed
782
      if (do_w_gridding)
Martin Reinecke's avatar
Martin Reinecke committed
783
        kernel[2*supp] = min(T(1), xdw*xsupp*abs(w0-in.w));
Martin Reinecke's avatar
Martin Reinecke committed
784
      for (size_t i=nexp; i<nvecs; ++i)
785
786
        kernel[i]=0;
      for (size_t i=0; i<nvecs; ++i)
Martin Reinecke's avatar
Martin Reinecke committed
787
        kernel[i] = exp(beta*(sqrt(T(1)-kernel[i]*kernel[i])-T(1)));
Martin Reinecke's avatar
Martin Reinecke committed
788
789
790
791
792
793
794
795
796
797
798
799
      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;
      }
  };

800
template<class T, class T2, class T3> class VisServ
Martin Reinecke's avatar
updates    
Martin Reinecke committed
801
  {
802
803
804
805
806
807
808
  private:
    const Baselines<T> &baselines;
    const const_mav<uint32_t,1> &idx;
    T2 vis;
    const T3 &wgt;
    size_t nvis;
    bool have_wgt;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
809

810
811
812
813
814
815
816
817
818
819
820
  public:
    VisServ(const Baselines<T> &baselines_,
      const const_mav<uint32_t,1> &idx_, T2 vis_, const T3 &wgt_)
      : baselines(baselines_), idx(idx_), vis(vis_), wgt(wgt_)
      {
      nvis = vis.shape(0);
      checkShape(idx.shape(), {nvis});
      have_wgt = wgt.size()!=0;
      if (have_wgt) checkShape(wgt.shape(), {nvis});
      }
    size_t Nvis() const { return nvis; }
821
    const Baselines<T> &getBaselines() const { return baselines; }
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
    UVW<T> getCoord(size_t i) const
      { return baselines.effectiveCoord(idx[i]); }
    complex<T> getVis(size_t i) const
      { return have_wgt ? vis(i)*wgt(i) : vis(i); }
    uint32_t getIdx(size_t i) const { return idx[i]; }
    void setVis (size_t i, const complex<T> &v) const
      { vis(i) = have_wgt ? v*wgt(i) : v; }
    void addVis (size_t i, const complex<T> &v) const
      { vis(i) += have_wgt ? v*wgt(i) : v; }
  };
template<class T, class T2, class T3> class MsServ
  {
  private:
    const Baselines<T> &baselines;
    const const_mav<uint32_t,1> &idx;
    T2 ms;
    const T3 &wgt;
    size_t nvis;
    bool have_wgt;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
841

842
843
844
845
  public:
    MsServ(const Baselines<T> &baselines_,
    const const_mav<uint32_t,1> &idx_, T2 ms_, const T3 &wgt_)
      : baselines(baselines_), idx(idx_), ms(ms_), wgt(wgt_)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
846
      {
847
848
849
850
851
852
      auto nrows = baselines.Nrows();
      auto nchan = baselines.Nchannels();
      nvis = idx.shape(0);
      checkShape(ms.shape(), {nrows, nchan});
      have_wgt = wgt.size()!=0;
      if (have_wgt) checkShape(wgt.shape(), {nrows, nchan});
Martin Reinecke's avatar
updates    
Martin Reinecke committed
853
      }
Martin Reinecke's avatar
Martin Reinecke committed
854
855
    MsServ(const MsServ &orig, const const_mav<uint32_t,1> &newidx)
      : MsServ(orig.baselines, newidx, orig.ms, orig.wgt) {}
856
    size_t Nvis() const { return nvis; }
857
    const Baselines<T> &getBaselines() const { return baselines; }
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
    UVW<T> 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);
      }
    uint32_t getIdx(size_t i) const { return idx[i]; }
    void setVis (size_t i, const complex<T> &v) const
      {
      auto rc = baselines.getRowChan(idx(i));
      ms(rc.row, rc.chan) = have_wgt ? v*wgt(rc.row, rc.chan) : v;
      }
    void addVis (size_t i, const complex<T> &v) const
      {
      auto rc = baselines.getRowChan(idx(i));
      ms(rc.row, rc.chan) += have_wgt ? v*wgt(rc.row, rc.chan) : v;
      }
  };
Martin Reinecke's avatar
updates    
Martin Reinecke committed
878

879
template<typename T, typename Serv> void x2grid_c
Martin Reinecke's avatar
Martin Reinecke committed
880
  (const GridderConfig<T> &gconf, const Serv &srv, mav<complex<T>,2> &grid, T w0=-1, T dw=-1)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
881
882
883
884
885
  {
  size_t nu=gconf.Nu(), nv=gconf.Nv();
  checkShape(grid.shape(), {nu, nv});
  myassert(grid.contiguous(), "grid is not contiguous");
  size_t supp = gconf.Supp();
Martin Reinecke's avatar
Martin Reinecke committed
886
  size_t nthreads = gconf.Nthreads();
Martin Reinecke's avatar
Martin Reinecke committed
887
  bool do_w_gridding=dw>0;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
888
889
890

#pragma omp parallel num_threads(nthreads)
{
Martin Reinecke's avatar
Martin Reinecke committed
891
  Helper<T> hlp(gconf, nullptr, grid.data(), w0, dw);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
892
  int jump = hlp.lineJump();
Martin Reinecke's avatar
Martin Reinecke committed
893
894
  const T * RESTRICT ku = hlp.kernel;
  const T * RESTRICT kv = hlp.kernel+supp;
895
  size_t np = srv.Nvis();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
896
897
898

  // Loop over sampling points
#pragma omp for schedule(guided,100)
899
  for (size_t ipart=0; ipart<np; ++ipart)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
900
    {
901
    UVW<T> coord = srv.getCoord(ipart);
902
    auto flip = coord.FixW();
Martin Reinecke's avatar
Martin Reinecke committed
903
    hlp.prep(coord);
Martin Reinecke's avatar
Martin Reinecke committed
904
    auto * RESTRICT ptr = hlp.p0w;
Martin Reinecke's avatar
Martin Reinecke committed
905
906
    auto v(srv.getVis(ipart));
    if (do_w_gridding) v*=hlp.Wfac();
907
    if (flip) v=conj(v);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
908
909
910
    for (size_t cu=0; cu<supp; ++cu)
      {
      complex<T> tmp(v*ku[cu]);
Martin Reinecke's avatar
Martin Reinecke committed
911
912
913
914
915
916
917
918
919
      size_t cv=0;
      for (; cv<supp-3; 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)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
920
921
922
923
924
925
926
        ptr[cv] += tmp*kv[cv];
      ptr+=jump;
      }
    }
} // end of parallel region
  }

927
template<typename T> void vis2grid_c
Martin Reinecke's avatar
updates    
Martin Reinecke committed
928
  (const Baselines<T> &baselines, const GridderConfig<T> &gconf,
929
  const const_mav<uint32_t,1> &idx, const const_mav<complex<T>,1> &vis,
Martin Reinecke's avatar
Martin Reinecke committed
930
  mav<complex<T>,2> &grid, const const_mav<T,1> &wgt, T w0=-1, T dw=-1)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
931
  {
Martin Reinecke's avatar
Martin Reinecke committed
932
  x2grid_c(gconf, VisServ<T,const_mav<complex<T>,1>,const_mav<T,1>> (baselines, idx, vis, wgt), grid, w0, dw);
933
934
935
936
937
938
939
940
941
  }

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)
  {
  x2grid_c(gconf, MsServ<T,const_mav<complex<T>,2>,const_mav<T,2>> (baselines, idx, ms, wgt), grid);
  }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
942

943
template<typename T, typename Serv> void grid2x_c
Martin Reinecke's avatar
Martin Reinecke committed
944
945
  (const GridderConfig<T> &gconf, const const_mav<complex<T>,2> &grid, const Serv &srv,
  T w0=-1, T dw=-1)
946
  {
Martin Reinecke's avatar
updates    
Martin Reinecke committed
947
948
949
950
  size_t nu=gconf.Nu(), nv=gconf.Nv();
  checkShape(grid.shape(), {nu, nv});
  myassert(grid.contiguous(), "grid is not contiguous");
  size_t supp = gconf.Supp();
Martin Reinecke's avatar
Martin Reinecke committed
951
  size_t nthreads = gconf.Nthreads();
Martin Reinecke's avatar
Martin Reinecke committed
952
  bool do_w_gridding=dw>0;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
953
954
955
956

  // Loop over sampling points
#pragma omp parallel num_threads(nthreads)
{
Martin Reinecke's avatar
Martin Reinecke committed
957
  Helper<T> hlp(gconf, grid.data(), nullptr, w0, dw);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
958
  int jump = hlp.lineJump();
Martin Reinecke's avatar
Martin Reinecke committed
959
960
  const T * RESTRICT ku = hlp.kernel;
  const T * RESTRICT kv = hlp.kernel+supp;
961
  size_t np = srv.Nvis();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
962
963

#pragma omp for schedule(guided,100)
964
  for (size_t ipart=0; ipart<np; ++ipart)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
965
    {
966
    UVW<T> coord = srv.getCoord(ipart);
967
    auto flip = coord.FixW();
Martin Reinecke's avatar
Martin Reinecke committed
968
    hlp.prep(coord);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
969
    complex<T> r = 0;
Martin Reinecke's avatar
Martin Reinecke committed
970
    const auto * RESTRICT ptr = hlp.p0r;
Martin Reinecke's avatar
updates    
Martin Reinecke committed
971
972
973
    for (size_t cu=0; cu<supp; ++cu)
      {
      complex<T> tmp(0);
Martin Reinecke's avatar
Martin Reinecke committed
974
975
976
977
978
979
980
      size_t cv=0;
      for (; cv<supp-3; 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)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
981
982
983
984
        tmp += ptr[cv] * kv[cv];
      r += tmp*ku[cu];
      ptr += jump;
      }
985
    if (flip) r=conj(r);
Martin Reinecke's avatar
Martin Reinecke committed
986
987
    if (do_w_gridding) r*=hlp.Wfac();
    srv.setVis(ipart, r);
Martin Reinecke's avatar
updates    
Martin Reinecke committed
988
989
990
    }
}
  }
991
992
993
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,
Martin Reinecke's avatar
Martin Reinecke committed
994
  mav<complex<T>,1> &vis, const const_mav<T,1> &wgt, T w0=-1, T dw=-1)
995
  {
Martin Reinecke's avatar
Martin Reinecke committed
996
  grid2x_c(gconf, grid, VisServ<T,mav<complex<T>,1>,const_mav<T,1>> (baselines, idx, vis, wgt), w0, dw);
997
  }
Martin Reinecke's avatar
updates    
Martin Reinecke committed
998
999
1000
1001
1002
1003

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)
  {
1004
  grid2x_c(gconf, grid, MsServ<T,mav<complex<T>,2>,const_mav<T,2>> (baselines, idx, ms, wgt));
Martin Reinecke's avatar
updates    
Martin Reinecke committed
1005
1006
  }

Martin Reinecke's avatar
Martin Reinecke committed
1007
1008
1009
1010
1011
1012
1013
1014
template<typename T> void apply_holo
  (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> &ogrid, const const_mav<T,1> &wgt)
  {
  size_t nu=gconf.Nu(), nv=gconf.Nv();
  checkShape(grid.shape(), {nu, nv});
  size_t nvis = idx.shape(0);
Martin Reinecke's avatar
Martin Reinecke committed
1015
  size_t nthreads = gconf.Nthreads();
Martin Reinecke's avatar
Martin Reinecke committed
1016
1017
1018
1019
  bool have_wgt = wgt.size()!=0;
  if (have_wgt) checkShape(wgt.shape(), {nvis});
  checkShape(ogrid.shape(), {nu, nv});

Martin Reinecke's avatar
Martin Reinecke committed
1020
  ogrid.fill(0);
Martin Reinecke's avatar
Martin Reinecke committed
1021
1022
1023
1024
1025
1026
1027
1028

  size_t supp = gconf.Supp();

  // Loop over sampling points
#pragma omp parallel num_threads(nthreads)
{
  Helper<T> hlp(gconf, grid.data(), ogrid.data());
  int jump = hlp.lineJump();
1029
1030
  const T * ku = hlp.kernel;
  const T * kv = hlp.kernel+supp;
Martin Reinecke's avatar
Martin Reinecke committed
1031
1032
1033
1034

#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
1035
    UVW<T> coord = baselines.effectiveCoord(idx(ipart));
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
1036
    coord.FixW();
Martin Reinecke's avatar
Martin Reinecke committed
1037
    hlp.prep(coord);
Martin Reinecke's avatar
Martin Reinecke committed
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
    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)
      {
      auto twgt = wgt(ipart);
      r*=twgt*twgt;
      }
    auto * wptr = hlp.p0w;
    for (size_t cu=0; cu<supp; ++cu)
      {
      complex<T> tmp(r*ku[cu]);
      for (size_t cv=0; cv<supp; ++cv)
        wptr[cv] += tmp*kv[cv];
      wptr += jump;
      }
    }
}
  }

template<typename T> void get_correlations
  (const Baselines<T> &baselines, const GridderConfig<T> &gconf,
  const const_mav<uint32_t,1> &idx, int du, int dv,
  mav<T,2> &ogrid, const const_mav<T,1> &wgt)
  {
  size_t nu=gconf.Nu(), nv=gconf.Nv();
  size_t nvis = idx.shape(0);
  bool have_wgt = wgt.size()!=0;
  if (have_wgt) checkShape(wgt.shape(), {nvis});
  checkShape(ogrid.shape(), {nu, nv});
  size_t supp = gconf.Supp();
  myassert(size_t(abs(du))<supp, "|du| must be smaller than Supp");
  myassert(size_t(abs(dv))<supp, "|dv| must be smaller than Supp");
Martin Reinecke's avatar
Martin Reinecke committed
1078
  size_t nthreads = gconf.Nthreads();
Martin Reinecke's avatar
Martin Reinecke committed
1079

Martin Reinecke's avatar
Martin Reinecke committed
1080
  ogrid.fill(0);
Martin Reinecke's avatar
Martin Reinecke committed
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096

  size_t u0, u1, v0, v1;
  if (du>=0)
    { u0=0; u1=supp-du; }
  else
    { u0=-du; u1=supp; }
  if (dv>=0)
    { v0=0; v1=supp-dv; }
  else
    { v0=-dv; v1=supp; }

  // Loop over sampling points
#pragma omp parallel num_threads(nthreads)
{
  Helper<T,T> hlp(gconf, nullptr, ogrid.data());
  int jump = hlp.lineJump();
1097
1098
  const T * ku = hlp.kernel;
  const T * kv = hlp.kernel+supp;
Martin Reinecke's avatar
Martin Reinecke committed
1099
1100
1101
1102
1103

#pragma omp for schedule(guided,100)
  for (size_t ipart=0; ipart<nvis; ++ipart)
    {
    UVW<T> coord = baselines.effectiveCoord(idx(ipart));
Martin Reinecke's avatar
fixes    
Martin Reinecke committed
1104
    coord.FixW();
Martin Reinecke's avatar
Martin Reinecke committed
1105
    hlp.prep(coord);
Martin Reinecke's avatar
Martin Reinecke committed
1106
    auto * wptr = hlp.p0w + u0*jump;
Martin Reinecke's avatar
Martin Reinecke committed
1107
    auto f0 = T(1);
Martin Reinecke's avatar
Martin Reinecke committed
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
    if (have_wgt)
      {
      auto twgt = wgt(ipart);
      f0*=twgt*twgt;
      }
    for (size_t cu=u0; cu<u1; ++cu)
      {
      auto f1=ku[cu]*ku[cu+du]*f0;
      for (size_t cv=v0; cv<v1; ++cv)
        wptr[cv] += f1*kv[cv]*kv[cv+dv];
      wptr += jump;
      }
    }
}
  }


1125
template<typename T, typename T2> void apply_wcorr(const GridderConfig<T> &gconf,
Martin Reinecke's avatar
Martin Reinecke committed
1126
  const mav<T2,2> &dirty, const EC_Kernel_with_correction<T> &kernel, double dw)
Martin Reinecke's avatar
Martin Reinecke committed
1127
1128
1129
  {
  auto nx_dirty=gconf.Nxdirty();
  auto ny_dirty=gconf.Nydirty();
Martin Reinecke's avatar
Martin Reinecke committed
1130
  size_t nthreads = gconf.Nthreads();
Martin Reinecke's avatar
Martin Reinecke committed
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
  auto psx=gconf.Pixsize_x();
  auto psy=gconf.Pixsize_y();
  double x0 = -0.5*nx_dirty*psx,
         y0 = -0.5*ny_dirty*psy;
#pragma omp parallel for schedule(static) num_threads(nthreads)
  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;
1144
1145
1146
1147
1148
1149
1150
      T fct = 0;
      double tmp = 1.-fx-fy;
      if (tmp>=0)
        {
        auto nm1 = (-fx-fy)/(sqrt(1.-fx-fy)+1.); // accurate form of sqrt(1-x-y)
        fct = (nm1<=-1) ? 0. : kernel.corfac(nm1*dw);
        }
Martin Reinecke's avatar
Martin Reinecke committed
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
      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;
      }
    }
  }

1165
1166
1167
1168
1169
1170
1171
1172
1173
template<typename T, typename Serv> void x2dirty(
  const GridderConfig<T> &gconf, const Serv &srv, const mav<T,2> &dirty, size_t verbosity=0)
  {
  auto nu=gconf.Nu();
  auto nv=gconf.Nv();

  if (verbosity>0)
    cout << "Gridding without w-stacking: " << srv.Nvis()
         << " visibilities" << endl;
Martin Reinecke's avatar
Martin Reinecke committed
1174
  if (verbosity>0) cout << "Using " << gconf.Nthreads() << " threads" << endl;
1175
1176
1177
1178
1179

  tmpStorage<complex<T>,2> grid_({nu,nv});
  auto grid=grid_.getMav();
  grid_.fill(0.);
  x2grid_c(gconf, srv, grid);
Martin Reinecke's avatar
Martin Reinecke committed
1180
1181
1182
1183
  tmpStorage<T,2> rgrid_({nu, nv});
  auto rgrid=rgrid_.getMav();
  complex2hartley(cmav(grid), rgrid, gconf.Nthreads());
  gconf.grid2dirty(rgrid, dirty);
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
  }

template<typename T, typename Serv> void dirty2x(
  const GridderConfig<T> &gconf,  const const_mav<T,2> &dirty,
  const Serv &srv, size_t verbosity=0)
  {
  auto nu=gconf.Nu();
  auto nv=gconf.Nv();
  size_t nvis = srv.Nvis();

Martin Reinecke's avatar
Martin Reinecke committed
1194
1195
1196
1197
1198
  if (verbosity>0)
    cout << "Degridding without w-stacking: " << srv.Nvis()
         << " visibilities" << endl;
  if (verbosity>0) cout << "Using " << gconf.Nthreads() << " threads" << endl;

1199
1200
1201
  for (size_t i=0; i<nvis; ++i)
    srv.setVis(i, 0.);

Martin Reinecke's avatar
Martin Reinecke committed
1202
  tmpStorage<T,2> grid_({nu,nv});
1203
  auto grid=grid_.getMav();
Martin Reinecke's avatar
Martin Reinecke committed
1204
1205
1206
1207
1208
  gconf.dirty2grid(dirty, grid);
  tmpStorage<complex<T>,2> grid2_({nu,nv});
  auto grid2=grid2_.getMav();
  hartley2complex(cmav(grid), grid2, gconf.Nthreads());
  grid2x_c(gconf, cmav(grid2), srv);
1209
1210
  }

Martin Reinecke's avatar
Martin Reinecke committed
1211
template<typename T, typename Serv> void wstack_common(
Martin Reinecke's avatar
Martin Reinecke committed
1212
  const GridderConfig<T> &gconf, const Serv &srv, T &wmin,
Martin Reinecke's avatar
Martin Reinecke committed
1213
  double &dw, size_t &nplanes, vector<size_t> &nvis_plane, vector<int> &minplane, size_t verbosity)
Martin Reinecke's avatar
updates    
Martin Reinecke committed
1214
1215
1216
1217
1218
  {
  auto nx_dirty=gconf.Nxdirty();
  auto ny_dirty=gconf.Nydirty();
  auto psx=gconf.Pixsize_x();
  auto psy=gconf.Pixsize_y();
1219
  size_t nvis = srv.Nvis();
Martin Reinecke's avatar
Martin Reinecke committed
1220
  size_t nthreads = gconf.Nthreads();
Martin Reinecke's avatar
updates    
Martin Reinecke committed
1221
1222

  // determine w values for every visibility, and min/max w;