[Sbcl-devel] Improve "x86-64 MOVE-FROM-SIGNED overflow detection using IMUL"

[<lut...@fr...>]

Hi Juho,

in the CVS log message for version 0.9.14.17 you wrote:

> Implement x86-64 MOVE-FROM-SIGNED overflow detection using IMUL
> instead of using three separate shifts / checks, which is somewhat
> faster and smaller.

This is a nice trick! I think it can be improved further. You had:

  (inst mov y x)
  (inst imul y 8)
  (inst jmp :o bignum)

As the IMUL instruction can be used with three operands it can subsume
the MOV, saving three bytes and one clock of latency:

  (inst imul y x 8)
  (inst jmp :o bignum)

Attached is a patch to this effect.

Yours

Lutz Euler