Increased PTT performances

[gabknight]

• cythonized find_closest(.) from dipy.core.shpere to dipy.direction.pmf

• optimized pmf.get_pmf_value(.): only interpolated the direction of interest instead of the whole pmf.

• improved cython code in dipy.direction.ptt_direction_getter.pyx

On my tests on the disco dataset, the performances improved by 5X.

[codecov]

Codecov Report

Merging #2875 (357a0bd) into master (50f4325) will increase coverage by 0.00%.
The diff coverage is n/a.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #2875   +/-   ##
=======================================
  Coverage   81.48%   81.48%           
=======================================
  Files         144      144           
  Lines       20062    20062           
  Branches     3194     3194           
=======================================
+ Hits        16347    16348    +1     
  Misses       2907     2907           
+ Partials      808      807    -1     

see 1 file with indirect coverage changes

[gabknight]

@skoudoro @Garyfallidis I had to modify the cpdef to cdef in the fast_numpy functions to change the inputs from double[:] to double* . As such, I also had to remove the tests. Do we move forward like this? Another option would be to add _c to the cdef function names, and create cpdef functions with the same names that calls their respective _c functions. This option would be fine, but solely used for the testing. What do you think?

[skoudoro]

I am ok, with that only for fast_numpy module.

In parallel, we need to figure out a way to add unit test for cdef functions

[skoudoro]

Can you comment then instead or deleting them for now ?

[gabknight]

This is ready for review. Further improvements will come in another PR.

[skoudoro]

Thank you @gabknight,

I am going ahead and merge this PR. Looks good to me.