Chapter 12 edits

[dankamongmen]

Very few edits here. You suddenly start using MiB in 12-5. I'm all for the MB/MiB distinction, but you should probably use it everywhere if you're going to use it here? I didn't convert MiB to MB (this is your call), but I did change several instances of Mib that clearly meant MiB to MiB.

• 12.1.2: i'd talk about GCC Function Multiversioning (https://gcc.gnu.org/wiki/FunctionMultiVersioning)
• 12.1.3 p208: avoiding FMA might improve performance, but it might hurt accuracy (FMA uses an internal buffer to hold the first result, avoiding the loss of intermediate precision)

[dendibakh]

Very few edits here. You suddenly start using MiB in 12-5. I'm all for the MB/MiB distinction, but you should probably use it everywhere if you're going to use it here? I didn't convert MiB to MB (this is your call), but I did change several instances of Mib that clearly meant MiB to MiB.

Thanks, Nick. I didn't pay too much attention when I was writing... I will clean up across the book.

• 12.1.2: i'd talk about GCC Function Multiversioning (https://gcc.gnu.org/wiki/FunctionMultiVersioning)

This is GCC- and x86-specific, and I haven't seen it used in any production software.

• 12.1.3 p208: avoiding FMA might improve performance, but it might hurt accuracy (FMA uses an internal buffer to hold the first result, avoiding the loss of intermediate precision)

I appreciate that you've mentioned it. It's surprising how many software vendors care about bit-exact results, so FMA for them is a no-go, even though technically, yes, it will give them more accurate FP calculations.

[dendibakh]

ARM likes lowercase names, I've seen it in many places. Even disassembled code is shown in lowercase. But now I checked their official website, at it all capitalized. And also it will look consistent with everything else. So I guess I'll leave capitalized. Fine.