x86 performance

[Stephen Weeks]

> Which raises an interesting point -- in that non-allocating loop, there is
> a check for 24 additional bytes at each entry; probably for the
> continuation where I print out the result, but it looks like it got pushed
> too far back into the loop.)

This feels like a bug.  I'll look into it.

> Notice that the x86 leal is 4 bytes and the c-codegen leal is 3 bytes. 

This is still better than the mov and add, right?  I wouldn't worry about it.