Name Description Size
filter_sse2_intrinsics.c Functions in this file look at most 3 pixels (a,b,c) to predict the 4th (d). They're positioned like this: prev: c b row: a d The Sub filter predicts d=a, Avg d=(a+b)/2, and Paeth predicts d to be whichever of a, b, or c is closest to p=a+b-c. 11323
intel_init.c The techniques used to implement each of these filters in SSE operate on one pixel at a time. So they generally speed up 3bpp images about 3x, 4bpp images about 4x. They can scale up to 6 and 8 bpp images and down to 2 bpp images, but they'd not likely have any benefit for 1bpp images. Most of these can be implemented using only MMX and 64-bit registers, but they end up a bit slower than using the equally-ubiquitous SSE2. 1793