Pythonic API for a subset of the Intel Integrated Performance Primitives (Intel IPP) library
# motmot.FastImage – SIMD image processing
FastImage implements low-level image processing operations designed to operate very quickly by using SIMD instructions. This is achieved by calling the Intel IPP library. A bridge to numpy is made through the array interface.
Allocation of aligned memory
For the SIMD instructions to perform at maximal speed, images must be aligned on 32-byte boundaries. FastImage relies on the underlying IPP library to allocate the memory, trusting that it knows best:
import motmot.FastImage.FastImage as FastImage import numpy as np # Allocate the image fi_im1 = FastImage.FastImage8u( FastImage.Size(4,5) ) # width, height # Get a numpy view im1 = np.asarray( fi_im1 ) assert im1.shape == (5,4) # height, width
In the above example, im1.strides will be (32,1), indicating that each row is aligned on a 32 byte boundary.