Advanced Micro Devices Inc (AMD)

[wbmw]

Sgolds, Re: perhaps I am old-fashioned in my terminology, but I always considered 'double precision' to mean twice the size of the standard register word. In the case of AMD64, that would mean 128 bits.

No. The floating point guidelines comply with the IEEE 754 standard. "Double-precision" always means 64-bit. Extended-precision is 80-bit, while single-precision is 32-bit. You are not using an old-fashion terminology, you are using a non-existant terminology.

Re: Chapter 1 goes into a lot more detail. Looks like they did go to 128-bit registers to come up with a more powerful SSE, SSE2 set.

Wrong again. Intel's SSE specification also uses 128-bit registers, which pack either two double-precision or four single-precision data variables into a single register. AMD's only contribution is to offer 16 registers, which is double the number that Intel uses. Furthermore, those extra eight are only accessible via AMD's x86-64 instruction extensions, not as an enhanced version of SSE-1/2.