motmot.FastImage – SIMD image processing

Description

FastImage implements low-level image processing operations designed to operate very quickly by using SIMD instructions. This is achieved by calling the Framewave 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 image processing 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.