GIL acquisition deadlock related to frame dimensions

[dzhoshkun]

Related PR: #33

The GIL acquisition deadlock (or so it seems from the core dump) seems to be reproducible when using somewhat unusual frame dimensions. In particular I was able to reproduce it by setting a crop region of x, y = 223, 321, width, height = 506, 607 with 5185540 configured using: cmake -D CMAKE_BUILD_TYPE=Debug -D BUILD_PYTHON=ON -D USE_NUMPY=ON -D ENABLE_NONFREE=ON -D USE_EPIPHAN_DVI2PCIE_DUO=ON -D USE_EPIPHANSDK=ON -D BUILD_PYTHON=ON -D BUILD_TESTS=ON ../../GIFT-Grab/src. And using this with BGRA captures from Epiphan DVI2PCIe Duo with the #46 branch.

Not sure why any particular frame dimensions should lead to a GIL acquisition deadlock. Could this be related to the increased processing time leading a thread to keep the GIL longer than intended?

[dzhoshkun]

Also the following crop region seems to reproduce the same problem: x, y = 150, 78 and width, height = 420, 420

[dzhoshkun]

But not x, y = 0, 0 and width, height = 420, 420

[dzhoshkun]

Looks like this is related to unaligned memory. For instance x=1 leads to this problem, while x=32 does not.

[dzhoshkun]

Odd x values seem to reproduce the problem. x=2 reproduces it as well.