FWIW, the assembly in the link is ancient. Treating FPU registers as a stack was considered a mistake 20 years ago. In this day and age it's been replaced with MMX and then SSE. It will technically still work because of backwards compatibility, but modern compilers will output newer, faster SSE instructions.
A bit of a nitpick here about MMX. Although the MMX registers were aliases of the floating point stack registers, the MMX instruction set consisted of strictly integer operations, not floating point.
Actually the FPU instruction set, even if stack-based, is quite complete and the implementations are effective, not often not slower than the more "modern" ones, except when there is a possibility for parallel processing. And SSE doesn't have 80-bits precision at all.
MMX is something that is probably not much useful today. But FPU and SSEx are both usable, especially when using assembly.
