PocketMatrix

The division per triangle is not the bottleneck of your polygon rasterizer. It's more important to get the scanline drawn as fast as possible.
Cache coherency plays a great role in texture mapping also.

Check out this document also (I'm not sure if you've seen it before)
http://www.multi.fi/~mbc/sources/fatmap2.txt

Cheers

To avoid jittering of your texture you'll probably have to make all coordinates fixed point (even the ones on screen) and use the subpixel portion of the screen coordinates to correct the ones of the texture while you're stepping through through it.

Read the link I've provided, it explains how to write a subpixel/subtexel accurate texture mapper.

Thanks,
'twas not easy, but I dug up the thread. I'm assuming you're referring to this one:
http://www.pocketmatrix.com/forums/view ... s&start=30

Let me put the technique into words to make sure that I understand how it works. By intertwining the uv coordinates as follows:
A = [u7|v7|u6|v6|u5|v5|u4|v4|u3|v3|u2|v2|u1|v1|u0|v0]

A = 0 means uv = (0,0)
A = 1 means uv = (0,1)
A = 2 means uv = (1,0) etc.

So this improves cache coherency because sequential data is preloaded into the cache. So depending on the texture screen orientation we get more cache hits even if we traverse in the v direction. This also means that the texture has to be rearanged such that the intertwined addressing accesses the corresponding pixels of the original texture.

Did I get the gist of it?