Hough indexing of lazy EBSD patterns errors with PyEBSDIndex 0.3

[hakonanes]

Hough indexing of lazy EBSD patterns, where the patterns are in a Dask array, work fine with PyEBSDIndex 0.2. This was because every EBSD pattern was converted to a NumPy array by PyEBSDIndex (as a side-effect!) before being passed to Numba for further processing.

Numba requires NumPy arrays. In PyEBSDIndex 0.3, our Dask array is not converted before being passed to Numba, resulting in an error.

@drowenhorst-nrl: Do you have any suggestions for how we can solve this? It was very nice to be able to pass the entire Dask array to PyEBSDIndex so that you could operate on it chunk by chunk, without using having to write it into memory.

A minimal working example:

>>> import kikuchipy as kp
>>> s = kp.data.nickel_ebsd_large(lazy=True)
>>> pl = s.xmap.phases
>>> indexer = s.detector.get_indexer(pl)
>>> xmap = s.hough_indexing(pl, indexer)
{
    "name": "TypingError",
    "message": "Failed in nopython mode pipeline (step: nopython frontend)
non-precise type pyobject
During: typing of argument at /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/radon_fast.py (272)

File \"../../../../opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/radon_fast.py\", line 272:
  def radon_faster(self,imageIn,padding = np.array([0,0]), fixArtifacts = False, background = None, normalization=True):
      <source elided>

  @staticmethod
  ^ 

This error may have been caused by the following argument(s):
- argument 0: Cannot determine Numba type of <class 'dask.array.core.Array'>
",
    "stack": "---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/opencl/band_detect_cl.py:104, in BandDetect.find_bands(self, patternsIn, verbose, clparams, chunksize, useCPU, **kwargs)
    103 #rdnNorm, clparams, rdnNorm_gpu = self.calc_rdn(patterns[chnk[0]:chnk[1],:,:], clparams, use_gpu=self.CLOps[0])
--> 104 rdnNorm, clparams = self.radon_fasterCL(patterns[chnk[0]:chnk[1],:,:], padding=self.padding,
    105                                                                fixArtifacts=False, background=self.backgroundsub,
    106                                                                returnBuff=True, clparams=clparams)
    108 #rdnNorm, clparams = self.rdn_mask(rdnNorm, clparams=clparams, returnBuff=False)
    109 
    110 #if (self.EDAXIQ == True): # I think EDAX actually uses the convolved radon for IQ
   (...)
    114   #rdnNorm_nocov = np.zeros((nRp,nTp,nImCL),dtype=np.float32)
    115   #cl.enqueue_copy(clparams.queue,rdnNorm_nocov,rdnNorm,is_blocking=True)

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/opencl/band_detect_cl.py:270, in BandDetect.radon_fasterCL(self, image, padding, fixArtifacts, background, returnBuff, clparams)
    268 #radon_gpu = cl.Buffer(ctx,mf.READ_WRITE,size=radon.nbytes)
    269 #radon_gpu = cl.Buffer(ctx,mf.READ_WRITE | mf.COPY_HOST_PTR,hostbuf=radon)
--> 270 image_gpu = cl.Buffer(ctx,mf.READ_ONLY | mf.COPY_HOST_PTR,hostbuf=image)
    271 imstep = np.uint64(np.product(shapeIm[-2:]))

TypeError: a bytes-like object is required, not 'Array'

During handling of the above exception, another exception occurred:

TypingError                               Traceback (most recent call last)
File /Users/hakon/kode/scratches/kp_scratch.py:3
      1 #%%
      2 indexer = s.detector.get_indexer(pl)
----> 3 xmap = s.hough_indexing(pl, indexer)

File ~/kode/personal/kikuchipy/kikuchipy/signals/ebsd.py:1687, in EBSD.hough_indexing(self, phase_list, indexer, chunksize, verbose, return_index_data, return_band_data)
   1684 if self._lazy:  # pragma: no cover
   1685     patterns = patterns.rechunk({0: chunksize, 1: -1, 2: -1})
-> 1687 xmap, index_data, band_data = _hough_indexing(
   1688     patterns=patterns,
   1689     phase_list=phase_list,
   1690     nav_shape=nav_shape,
   1691     step_sizes=step_sizes,
   1692     indexer=indexer,
   1693     chunksize=chunksize,
   1694     verbose=verbose,
   1695 )
   1697 xmap.scan_unit = _get_navigation_axes_unit(am)
   1699 if return_index_data and return_band_data:

File ~/kode/personal/kikuchipy/kikuchipy/indexing/_hough_indexing.py:235, in _hough_indexing(patterns, phase_list, nav_shape, step_sizes, indexer, chunksize, verbose)
    232 print(info_message)
    234 tic = time()
--> 235 index_data, band_data, _, _ = indexer.index_pats(
    236     patsin=patterns, verbose=verbose, chunksize=chunksize
    237 )
    238 toc = time()
    239 patterns_per_second = n_patterns / (toc - tic)

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/_ebsd_index_single.py:524, in EBSDIndexer.index_pats(self, patsin, patstart, npats, xyloc, clparams, PC, verbose, chunksize)
    521 if npats == -1:
    522     npats = npoints
--> 524 banddata, bandnorm = self._detectbands(pats, PC, xyloc=xyloc, clparams=clparams, verbose=verbose,
    525                                        chunksize=chunksize)
    526 tic = timer()
    528 indxData, banddata = self._indexbandsphase(banddata, bandnorm, verbose=verbose)

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/_ebsd_index_single.py:618, in EBSDIndexer._detectbands(self, pats, PC, xyloc, clparams, verbose, chunksize)
    617 def _detectbands(self, pats, PC, xyloc=None, clparams=None, verbose=0, chunksize=528):
--> 618     banddata = self.bandDetectPlan.find_bands(
    619         pats, clparams=clparams, verbose=verbose, chunksize=chunksize
    620     )
    621     #  shpBandDat = banddata.shape
    622     if PC is None:

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/opencl/band_detect_cl.py:220, in BandDetect.find_bands(self, patternsIn, verbose, clparams, chunksize, useCPU, **kwargs)
    218 except Exception as e: # something went wrong - try the CPU
    219   print(e)
--> 220   bandData = band_detect.BandDetect.find_bands(self, patternsIn, verbose=verbose, chunksize=-1, **kwargs)
    222 return bandData

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/band_detect.py:411, in BandDetect.find_bands(self, patternsIn, verbose, chunksize, **kwargs)
    409 for chnk in chunk_start_end:
    410   tic1 = timer()
--> 411   rdnNorm = self.radonPlan.radon_faster(patterns[chnk[0]:chnk[1],:,:], self.padding, fixArtifacts=False, background=self.backgroundsub)
    412   rdntime += timer() - tic1
    413   tic1 = timer()

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/radon_fast.py:259, in Radon.radon_faster(self, imageIn, padding, fixArtifacts, background, normalization)
    256 radon = np.full([self.nRho + 2 * padding[0],self.nTheta + 2 * padding[1], nIm],self.missingval,dtype=np.float32)
    257 shp = radon.shape
--> 259 counter = self.rdn_loops(image,self.indexPlan,nIm,nPx,indxDim,radon,
    260                          np.asarray(padding), np.float32(normalization > 0))
    262 if (fixArtifacts == True):
    263   radon[:,padding[1],:] = radon[:,padding[1]+1,:]

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/numba/core/dispatcher.py:468, in _DispatcherBase._compile_for_args(self, *args, **kws)
    464         msg = (f\"{str(e).rstrip()} \
\
This error may have been caused \"
    465                f\"by the following argument(s):\
{args_str}\
\")
    466         e.patch_message(msg)
--> 468     error_rewrite(e, 'typing')
    469 except errors.UnsupportedError as e:
    470     # Something unsupported is present in the user code, add help info
    471     error_rewrite(e, 'unsupported_error')

File /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/numba/core/dispatcher.py:409, in _DispatcherBase._compile_for_args.<locals>.error_rewrite(e, issue_type)
    407     raise e
    408 else:
--> 409     raise e.with_traceback(None)

TypingError: Failed in nopython mode pipeline (step: nopython frontend)
non-precise type pyobject
During: typing of argument at /opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/radon_fast.py (272)

File \"../../../../opt/homebrew/Caskroom/miniforge/base/envs/kp/lib/python3.10/site-packages/pyebsdindex/radon_fast.py\", line 272:
  def radon_faster(self,imageIn,padding = np.array([0,0]), fixArtifacts = False, background = None, normalization=True):
      <source elided>

  @staticmethod
  ^ 

This error may have been caused by the following argument(s):
- argument 0: Cannot determine Numba type of <class 'dask.array.core.Array'>
"
}

[hakonanes]

I believe they are related to these changes in band_detect_cl.py: https://github.com/USNavalResearchLaboratory/PyEBSDIndex/commit/88136e60b8bc6a6c53c507bfc4dda1a39163e864#diff-b1a7457abbba36e579c00423396bc962373e6fcb044b4c43a3c2ad0384cca8be

[drowenhorst-nrl]

What is the data type of your dask array?

[hakonanes]

In this example it is uint8.

[drowenhorst-nrl]

Trying to figure out exactly what is up ... I see Numba error, but seems like it is failing on making an opencl buffer?

[drowenhorst-nrl]

Ah - I see what is up ... I think. When it fails to run the OpenCL version, it attempts the numba version. Both are failing for similar reasons. Let me look at what I can do here.

[hakonanes]

Thank you for having a look, really appreciate it! Any potentially useful information you have about the problem is useful.

[drowenhorst-nrl]

If only all the fixes where this easy (at least I hope it is this easy). Maybe not the most elegant solution, but it looks like throwing an image = np.asarray(image) at the top of the radon functions should work. This will make a copy of the chunk of patterns read in, but I think I was doing that before anyway. While I have made many improvements on making PyEBSDIndex faster, I have not worried much about the memory efficiency.

Some insight:

• Through some staring some Windows system monitoring tools, it looked like I was spending just about as much time transferring patterns to the GPU as calculating (as is the way with GPU compute). I figured it would not be that bad, that if the patterns are uint8, or uint16, I could just send them as-is to the GPU, and convert to float on the GPU. In fact - I made the decision that if the patterns were float -- I actually convert them to uint16, and then scale the results (anyone who wants 32-bit precision on pattern quality really needs to justify it to me I guess).
• Then I figured, well, I might as well do my data transpose on the GPU as well.

Thus, I stoped converting to float on the CPU (which would have made a copy in this case), and tried sending a version as is to the GPU. OpenCL and Numba will work with numpy, but not Dask arrays ... thus the issue on both pipelines.

I will send in a PR for you to test.

[drowenhorst-nrl]

BTW - old version on my machine: 780 pats/s, new 1180 pats/s

[hakonanes]

Your changes in https://github.com/USNavalResearchLaboratory/PyEBSDIndex/pull/60 fixes my issue here, thanks!

[drowenhorst-nrl]

PyEBSDIndex 0.3.2 is now out on PyPi and conda-forge, and should be the fix for this issue.

[hakonanes]

Thank you so much for the quick handling of my issue here. Have a good weekend!