The OpenCL and CUDA backend functions in ArrayFire are asynchronous whereas the CPU backend is not. We need to unify the behavior of all of the backends so that they are consistent.

Function coverage w.r.t each file in <REPOSITORY_ROOT>/src/backend/cpu/*.cpp:

• approx.cpp
• Array.cpp
• assign.cpp
• bilateral.cpp
• blas.cpp
• cholesky.cpp
• convolve.cpp
• copy.cpp
• diagonal.cpp
• diff.cpp
• fast.cpp
• fftconvolve.cpp
• fft.cpp
• gradient.cpp
• harris.cpp
• hist_graphics.cpp
• histogram.cpp
• homography.cpp
• hsv_rgb.cpp
• identity.cpp
• iir.cpp
• image.cpp
• index.cpp
• inverse.cpp
• iota.cpp
• ireduce.cpp
• join.cpp
• lookup.cpp
• lu.cpp
• match_template.cpp
• math.cpp
• meanshift.cpp
• medfilt.cpp
• morph.cpp
• nearest_neighbour.cpp
• orb.cpp
• plot3.cpp
• plot.cpp
• qr.cpp
• random.cpp
• range.cpp
• reduce.cpp
• regions.cpp
• reorder.cpp
• resize.cpp
• rotate.cpp
• scan.cpp
• select.cpp
• set.cpp
• shift.cpp
• sift.cpp
• sobel.cpp
• solve.cpp
• sort_by_key.cpp
• sort.cpp
• sort_index.cpp
• surface.cpp
• susan.cpp
• svd.cpp
• tile.cpp
• transform.cpp
• transpose.cpp
• triangle.cpp
• unwrap.cpp
• where.cpp
• wrap.cpp