there's a lot of floating point operations in the zspan rendering code without any concurrent execution between them, and a working fixed-point version of the span8 rendering code (albeit in C, not asm)
if the fixedspan and zspan code could be interwoven and run concurrently, that could be a massive speedup over the current code, since most of the fixedspan instructions should be able to run "for free" between the zspan ones.
there's a lot of floating point operations in the zspan rendering code without any concurrent execution between them, and a working fixed-point version of the span8 rendering code (albeit in C, not asm)
if the fixedspan and zspan code could be interwoven and run concurrently, that could be a massive speedup over the current code, since most of the fixedspan instructions should be able to run "for free" between the zspan ones.