Merge branch 'convolution_work' into 'master'

Convolution work

See merge request mtr/cxxbase!7

pyinterpol_ng/demo.py

@@ -42,11 +42,13 @@ def convolve(alm1, alm2, lmax):
 lmax=60
 kmax=13
 ncomp=1
-separate=True
-ncomp2 = ncomp if separate else 1
+separate=False
+nptg = 1000000
 epsilon = 1e-4
-ofactor = 2
-nthreads = 0
+ofactor = 1.5
+nthreads = 0  # use as many threads as available
+
+ncomp2 = ncomp if separate else 1
 
 # get random sky a_lm
 # the a_lm arrays follow the same conventions as those in healpy
@@ -93,13 +95,14 @@ plt.imshow(bar2-bar[:,:,0])
 plt.show()
 
 
-ptg=np.random.uniform(0.,1.,3*1000000).reshape(1000000,3)
+ptg=np.random.uniform(0.,1.,3*nptg).reshape(nptg,3)
 ptg[:,0]*=np.pi
 ptg[:,1]*=2*np.pi
 ptg[:,2]*=2*np.pi
 #foo = pyinterpol_ng.PyInterpolator(slm,blm,separate,lmax, kmax, epsilon=1e-6, nthreads=2)
 t0=time.time()
 bar=foo.interpol(ptg)
+del foo
 print("interpolation time: ", time.time()-t0)
 fake = np.random.uniform(0.,1., (ptg.shape[0],ncomp2))
 foo2 = pyinterpol_ng.PyInterpolator(lmax, kmax, ncomp2, epsilon=epsilon, ofactor=ofactor, nthreads=nthreads)

pyinterpol_ng/interpol_ng.h

@@ -22,6 +22,96 @@
 
 namespace mr {
 
+namespace detail_fft {
+
+using std::vector;
+
+template<typename T, typename T0> aligned_array<T> alloc_tmp_conv
+  (const fmav_info &info, size_t axis, size_t len)
+  {
+  auto othersize = info.size()/info.shape(axis);
+  constexpr auto vlen = native_simd<T0>::size();
+  auto tmpsize = len*((othersize>=vlen) ? vlen : 1);
+  return aligned_array<T>(tmpsize);
+  }
+
+template<typename Tplan, typename T, typename T0, typename Exec>
+MRUTIL_NOINLINE void general_convolve(const fmav<T> &in, fmav<T> &out,
+  const size_t axis, const vector<T0> &kernel, size_t nthreads,
+  const Exec &exec)
+  {
+  std::shared_ptr<Tplan> plan1, plan2;
+
+  size_t l_in=in.shape(axis), l_out=out.shape(axis);
+  size_t l_min=std::min(l_in, l_out), l_max=std::max(l_in, l_out);
+  MR_assert(kernel.size()==l_min/2+1, "bad kernel size");
+  plan1 = get_plan<Tplan>(l_in);
+  plan2 = get_plan<Tplan>(l_out);
+
+  execParallel(
+    util::thread_count(nthreads, in, axis, native_simd<T0>::size()),
+    [&](Scheduler &sched) {
+      constexpr auto vlen = native_simd<T0>::size();
+      auto storage = alloc_tmp_conv<T,T0>(in, axis, l_max);
+      multi_iter<vlen> it(in, out, axis, sched.num_threads(), sched.thread_num());
+#ifndef MRUTIL_NO_SIMD
+      if (vlen>1)
+        while (it.remaining()>=vlen)
+          {
+          it.advance(vlen);
+          auto tdatav = reinterpret_cast<add_vec_t<T> *>(storage.data());
+          exec(it, in, out, tdatav, *plan1, *plan2, kernel);
+          }
+#endif
+      while (it.remaining()>0)
+        {
+        it.advance(1);
+        auto buf = reinterpret_cast<T *>(storage.data());
+        exec(it, in, out, buf, *plan1, *plan2, kernel);
+        }
+    });  // end of parallel region
+  }
+
+struct ExecConvR1
+  {
+  template <typename T0, typename T, size_t vlen> void operator() (
+    const multi_iter<vlen> &it, const fmav<T0> &in, fmav<T0> &out,
+    T * buf, const pocketfft_r<T0> &plan1, const pocketfft_r<T0> &plan2,
+    const vector<T0> &kernel) const
+    {
+    size_t l_in = plan1.length(),
+           l_out = plan2.length(),
+           l_min = std::min(l_in, l_out);
+    copy_input(it, in, buf);
+    plan1.exec(buf, T0(1), true);
+    for (size_t i=0; i<l_min; ++i) buf[i]*=kernel[(i+1)/2];
+    for (size_t i=l_in; i<l_out; ++i) buf[i] = T(0);
+    plan2.exec(buf, T0(1), false);
+    copy_output(it, buf, out);
+    }
+  };
+
+template<typename T> void convolve_1d(const fmav<T> &in,
+  fmav<T> &out, size_t axis, const vector<T> &kernel, size_t nthreads=1)
+  {
+  MR_assert(axis<in.ndim(), "bad axis number");
+  MR_assert(in.ndim()==out.ndim(), "dimensionality mismatch");
+  if (in.data()==out.data())
+    MR_assert(in.stride()==out.stride(), "strides mismatch");
+  for (size_t i=0; i<in.ndim(); ++i)
+    if (i!=axis)
+      MR_assert(in.shape(i)==out.shape(i), "shape mismatch");
+  MR_assert(!((in.shape(axis)&1) || (out.shape(axis)&1)),
+    "input and output axis lengths must be even");
+  if (in.size()==0) return;
+  general_convolve<pocketfft_r<T>>(in, out, axis, kernel, nthreads,
+    ExecConvR1());
+  }
+
+}
+
+using detail_fft::convolve_1d;
+
 namespace detail_interpol_ng {
 
 using namespace std;
@@ -41,78 +131,54 @@ template<typename T> class Interpolator
     void correct(mav<T,2> &arr, int spin)
       {
       T sfct = (spin&1) ? -1 : 1;
-      mav<T,2> tmp({nphi,nphi});
-      tmp.apply([](T &v){v=0.;});
-      auto tmp0=tmp.template subarray<2>({0,0},{nphi0, nphi0});
-      fmav<T> ftmp0(tmp0);
-      for (size_t i=0; i<ntheta0; ++i)
+      mav<T,2> tmp({nphi,nphi0});
+      // copy and extend to second half
       for (size_t j=0; j<nphi0; ++j)
-          tmp0.v(i,j) = arr(i,j);
-      // extend to second half
+        tmp.v(0,j) = arr(0,j);
       for (size_t i=1, i2=nphi0-1; i+1<ntheta0; ++i,--i2)
         for (size_t j=0,j2=nphi0/2; j<nphi0; ++j,++j2)
           {
           if (j2>=nphi0) j2-=nphi0;
-          tmp0.v(i2,j) = sfct*tmp0(i,j2);
-          }
-      // FFT to frequency domain on minimal grid
-      r2r_fftpack(ftmp0,ftmp0,{0,1},true,true,T(1./(nphi0*nphi0)),nthreads);
-      // correct amplitude at Nyquist frequency
-      for (size_t i=0; i<nphi0; ++i)
-        {
-        tmp0.v(i,nphi0-1)*=0.5;
-        tmp0.v(nphi0-1,i)*=0.5;
+          tmp.v(i,j2) = arr(i,j2);
+          tmp.v(i2,j) = sfct*tmp(i,j2);
           }
-      auto fct = kernel.correction_factors(nphi, nphi0/2+1, nthreads);
-      for (size_t i=0; i<nphi0; ++i)
       for (size_t j=0; j<nphi0; ++j)
-          tmp0.v(i,j) *= fct[(i+1)/2] * fct[(j+1)/2];
-      auto tmp1=tmp.template subarray<2>({0,0},{nphi, nphi0});
-      fmav<T> ftmp1(tmp1);
-      // zero-padded FFT in theta direction
-      r2r_fftpack(ftmp1,ftmp1,{0},false,false,T(1),nthreads);
-      auto tmp2=tmp.template subarray<2>({0,0},{ntheta, nphi});
-      fmav<T> ftmp2(tmp2);
+        tmp.v(ntheta0-1,j) = arr(ntheta0-1,j);
+      auto fct = kernel.correction_factors(nphi, nphi0/2+1, nthreads);
+      for (auto &f:fct) f/=nphi0;
+      fmav<T> ftmp(tmp);
+      fmav<T> ftmp0(tmp.template subarray<2>({0,0},{nphi0, nphi0}));
+      convolve_1d(ftmp0, ftmp, 0, fct, nthreads);
+      fmav<T> ftmp2(tmp.template subarray<2>({0,0},{ntheta, nphi0}));
       fmav<T> farr(arr);
-      // zero-padded FFT in phi direction
-      r2r_fftpack(ftmp2,farr,{1},false,false,T(1),nthreads);
+      convolve_1d(ftmp2, farr, 1, fct, nthreads);
       }
     void decorrect(mav<T,2> &arr, int spin)
       {
       T sfct = (spin&1) ? -1 : 1;
-      mav<T,2> tmp({nphi,nphi});
-      fmav<T> ftmp(tmp);
-
-      for (size_t i=0; i<ntheta; ++i)
-        for (size_t j=0; j<nphi; ++j)
-          tmp.v(i,j) = arr(i,j);
+      mav<T,2> tmp({nphi,nphi0});
+      auto fct = kernel.correction_factors(nphi, nphi0/2+1, nthreads);
+      for (auto &f:fct) f/=nphi0;
+      fmav<T> farr(arr);
+      fmav<T> ftmp2(tmp.template subarray<2>({0,0},{ntheta, nphi0}));
+      convolve_1d(farr, ftmp2, 1, fct, nthreads);
       // extend to second half
       for (size_t i=1, i2=nphi-1; i+1<ntheta; ++i,--i2)
-        for (size_t j=0,j2=nphi/2; j<nphi; ++j,++j2)
+        for (size_t j=0,j2=nphi0/2; j<nphi0; ++j,++j2)
           {
-          if (j2>=nphi) j2-=nphi;
+          if (j2>=nphi0) j2-=nphi0;
           tmp.v(i2,j) = sfct*tmp(i,j2);
           }
-      r2r_fftpack(ftmp,ftmp,{1},true,true,T(1),nthreads);
-      auto tmp1=tmp.template subarray<2>({0,0},{nphi, nphi0});
-      fmav<T> ftmp1(tmp1);
-      r2r_fftpack(ftmp1,ftmp1,{0},true,true,T(1),nthreads);
-      auto fct = kernel.correction_factors(nphi, nphi0/2+1, nthreads);
-      auto tmp0=tmp.template subarray<2>({0,0},{nphi0, nphi0});
-      fmav<T> ftmp0(tmp0);
-      for (size_t i=0; i<nphi0; ++i)
+      fmav<T> ftmp(tmp);
+      fmav<T> ftmp0(tmp.template subarray<2>({0,0},{nphi0, nphi0}));
+      convolve_1d(ftmp, ftmp0, 0, fct, nthreads);
       for (size_t j=0; j<nphi0; ++j)
-          tmp0.v(i,j) *= fct[(i+1)/2] * fct[(j+1)/2];
-      // FFT to (theta, phi) domain on minimal grid
-      r2r_fftpack(ftmp0,ftmp0,{0,1},false, false,T(1./(nphi0*nphi0)),nthreads);
+        arr.v(0,j) = 0.5*tmp(0,j);
+      for (size_t i=1; i+1<ntheta0; ++i)
         for (size_t j=0; j<nphi0; ++j)
-        {
-        tmp0.v(0,j)*=0.5;
-        tmp0.v(ntheta0-1,j)*=0.5;
-        }
-      for (size_t i=0; i<ntheta0; ++i)
+          arr.v(i,j) = tmp(i,j);
       for (size_t j=0; j<nphi0; ++j)
-          arr.v(i,j) = tmp0(i,j);
+        arr.v(ntheta0-1,j) = 0.5*tmp(ntheta0-1,j);
       }
 
     vector<size_t> getIdx(const mav<T,2> &ptg) const

pyinterpol_ng/pyinterpol_ng.cc

@@ -242,16 +242,28 @@ PYBIND11_MODULE(pyinterpol_ng, m)
 
   m.doc() = pyinterpol_ng_DS;
 
-  py::class_<PyInterpolator<fptype>> (m, "PyInterpolator", pyinterpolator_DS)
+  using inter_d = PyInterpolator<double>;
+  py::class_<inter_d> (m, "PyInterpolator", pyinterpolator_DS)
     .def(py::init<const py::array &, const py::array &, bool, int64_t, int64_t, fptype, fptype, int>(),
       initnormal_DS, "sky"_a, "beam"_a, "separate"_a, "lmax"_a, "kmax"_a, "epsilon"_a, "ofactor"_a=fptype(1.5),
       "nthreads"_a=0)
     .def(py::init<int64_t, int64_t, int64_t, fptype, fptype, int>(), initadjoint_DS,
       "lmax"_a, "kmax"_a, "ncomp"_a, "epsilon"_a, "ofactor"_a=fptype(1.5), "nthreads"_a=0)
-    .def ("interpol", &PyInterpolator<fptype>::pyinterpol, interpol_DS, "ptg"_a)
-    .def ("deinterpol", &PyInterpolator<fptype>::pydeinterpol, deinterpol_DS, "ptg"_a, "data"_a)
-    .def ("getSlm", &PyInterpolator<fptype>::pygetSlm, getSlm_DS, "beam"_a)
-    .def ("support", &PyInterpolator<fptype>::support);
+    .def ("interpol", &inter_d::pyinterpol, interpol_DS, "ptg"_a)
+    .def ("deinterpol", &inter_d::pydeinterpol, deinterpol_DS, "ptg"_a, "data"_a)
+    .def ("getSlm", &inter_d::pygetSlm, getSlm_DS, "beam"_a)
+    .def ("support", &inter_d::support);
+//   using inter_f = PyInterpolator<float>;
+//   py::class_<inter_f> (m, "PyInterpolator_f", pyinterpolator_DS)
+//     .def(py::init<const py::array &, const py::array &, bool, int64_t, int64_t, fptype, fptype, int>(),
+//       initnormal_DS, "sky"_a, "beam"_a, "separate"_a, "lmax"_a, "kmax"_a, "epsilon"_a, "ofactor"_a=fptype(1.5),
+//       "nthreads"_a=0)
+//     .def(py::init<int64_t, int64_t, int64_t, fptype, fptype, int>(), initadjoint_DS,
+//       "lmax"_a, "kmax"_a, "ncomp"_a, "epsilon"_a, "ofactor"_a=fptype(1.5), "nthreads"_a=0)
+//     .def ("interpol", &inter_f::pyinterpol, interpol_DS, "ptg"_a)
+//     .def ("deinterpol", &inter_f::pydeinterpol, deinterpol_DS, "ptg"_a, "data"_a)
+//     .def ("getSlm", &inter_f::pygetSlm, getSlm_DS, "beam"_a)
+//     .def ("support", &inter_f::support);
 #if 1
   m.def("rotate_alm", &pyrotate_alm<fptype>, "alm"_a, "lmax"_a, "psi"_a, "theta"_a,
     "phi"_a);

src/mr_util/math/fft.h

@@ -604,21 +604,6 @@ template<typename T, typename T0> aligned_array<T> alloc_tmp
   auto tmpsize = axsize*((othersize>=vlen) ? vlen : 1);
   return aligned_array<T>(tmpsize);
   }
-template<typename T, typename T0> aligned_array<T> alloc_tmp
-  (const fmav_info &info, const shape_t &axes)
-  {
-  size_t fullsize=info.size();
-  size_t tmpsize=0;
-  for (size_t i=0; i<axes.size(); ++i)
-    {
-    auto axsize = info.shape(axes[i]);
-    auto othersize = fullsize/axsize;
-    constexpr auto vlen = native_simd<T0>::size();
-    auto sz = axsize*((othersize>=vlen) ? vlen : 1);
-    if (sz>tmpsize) tmpsize=sz;
-    }
-  return aligned_array<T>(tmpsize);
-  }
 
 template <typename T, size_t vlen> void copy_input(const multi_iter<vlen> &it,
   const fmav<Cmplx<T>> &src, Cmplx<native_simd<T>> *MRUTIL_RESTRICT dst)