Fixing newDistArray such that it takes axis of PFFT pencil into account

Author: mikaem

azure-pipelines.yml

@@ -1,8 +1,8 @@
 jobs:
 - job: macOS
-  displayName: macOS-10.13
+  displayName: macOS-10.14
   pool:
-    vmImage: 'macOS-10.13'
+    vmImage: 'macOS-10.14'
   strategy:
     matrix:
       Python27:
@@ -11,6 +11,8 @@ jobs:
         python.version: '3.6'
       Python37:
         python.version: '3.7'
+      Python38:
+        python.version: '3.8'
 
   steps:
   - bash: echo "##vso[task.prependpath]$CONDA/bin"

mpi4py_fft/distarray.py

@@ -481,7 +481,7 @@ def newDistArray(pfft, forward_output=True, val=0, rank=0, view=False):
         dtype = pfft.forward.input_array.dtype
     global_shape = (len(global_shape),)*rank + global_shape
     z = DistArray(global_shape, subcomm=p0.subcomm, val=val, dtype=dtype,
-                  rank=rank)
+                  alignment=p0.axis, rank=rank)
     return z.v if view else z
 
 def Function(*args, **kwargs): #pragma: no cover

mpi4py_fft/io/generate_xdmf.py

@@ -114,7 +114,7 @@ def generate_xdmf(h5filename, periodic=True, order='paraview'):
         order here.
     """
     import h5py
-    f = h5py.File(h5filename)
+    f = h5py.File(h5filename, 'a')
     keys = []
     f.visit(keys.append)
     assert order.lower() in ('paraview', 'visit')

tests/test_fftw.py

@@ -103,7 +103,7 @@ def test_fftw():
                                 idct = fftw.idctn(input_array, None, axes, type, threads, iflags, output_array=oa)
                                 A2 = idct(B, implicit=True, normalize=True)
                                 assert allclose(A, A2), np.linalg.norm(A-A2)
-                                if typecode is not 'g' and not type is 4:
+                                if typecode != 'g' and type != 4:
                                     B2 = scipy_dctn(A, axes=axes, type=type)
                                     assert allclose(B, B2), np.linalg.norm(B-B2)
 
@@ -112,7 +112,7 @@ def test_fftw():
                                 idst = fftw.idstn(input_array, None, axes, type, threads, iflags, output_array=oa)
                                 A2 = idst(B, implicit=True, normalize=True)
                                 assert allclose(A, A2), np.linalg.norm(A-A2)
-                                if typecode is not 'g' and not type is 4:
+                                if typecode != 'g' and type != 4:
                                     B2 = scipy_dstn(A, axes=axes, type=type)
                                     assert allclose(B, B2), np.linalg.norm(B-B2)
 
@@ -132,7 +132,7 @@ def test_fftw():
                             M = fftw.get_normalization(kds, input_array.shape, axes)
                             assert allclose(C2*M, A)
                             # Test vs scipy for transforms available in scipy
-                            if typecode is not 'g' and not any(f in kds for f in (fftw.FFTW_RODFT11, fftw.FFTW_REDFT11)):
+                            if typecode != 'g' and not any(f in kds for f in (fftw.FFTW_RODFT11, fftw.FFTW_REDFT11)):
                                 for m, ts in enumerate(tsf):
                                     A = eval('scipy.fftpack.'+ts[:-1])(A, axis=axes[m], type=int(ts[-1]))
                                 assert allclose(C, A), np.linalg.norm(C-A)

tests/test_speed.py

@@ -12,7 +12,7 @@
 N = (64, 64, 64)
 loops = 50
 axis = 1
-threads = 1
+threads = 4
 implicit = True
 flags = (fftw.FFTW_PATIENT, fftw.FFTW_DESTROY_INPUT)
 
@@ -73,7 +73,7 @@
     ifft = fftw.ifftn(output_array, None, axes, threads, flags)
     t0 = time()
     for i in range(loops):
-        B2 = ifft(C, normalize_idft=True, implicit=implicit)
+        B2 = ifft(C, normalize=True, implicit=implicit)
     ftime[1].append(time()-t0)
     assert np.allclose(B, B2), np.linalg.norm(B-B2)
 
@@ -140,7 +140,7 @@
     t0 = time()
     for i in range(loops):
         C2[:] = C
-        D2 = irfft(C2, normalize_idft=True, implicit=implicit)
+        D2 = irfft(C2, normalize=True, implicit=implicit)
     ftime[1].append(time()-t0)
     assert np.allclose(D, D2), np.linalg.norm(D-D2)