RFR: Fix Windows aarch64 builds

[Andrew Haley]

On Mon, 27 Sep 2021 17:41:45 GMT, Aleksey Shipilev <shade at openjdk.org> wrote:

> Current GHA point the problem with Windows aarch64 builds:
>   https://github.com/shipilev/loom/runs/3721167733
>  
> It passes now:
>   https://github.com/shipilev/loom/runs/3722819426
> 
> Additional testing:
>  - [x] Windows aarch64 GHA builds fine
>  - [x] Linux aarch64 `tier1_loom` still passes

This is an obviously-correct fix, but if I were doing it I'd be tempted to try a size_t overload to `MacroAssembler::sub()` instead. Mind you, that would probably have side effects elsewhere.

-------------

Marked as reviewed by aph (Committer).

PR: https://git.openjdk.java.net/loom/pull/70