[Webkit-unassigned] [Bug 213374] Bring back llint on x86_32

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=213374

--- Comment #9 from Keith Miller <keith_miller at apple.com> ---
(In reply to Angelos Oikonomopoulos from comment #6)
> I should note that the purpose of this patch is not better performance on
> x86_32. Rather, it is to enable faster testing of llint-related changes on
> 32-bit platforms.

Can you clarify how your patch achieves this goal? It's not immediately obvious. So it's hard for me to provide actionable feedback.

(In reply to Angelos Oikonomopoulos from comment #8)
> So, the main thing I'm looking for feedback on is the magical treatment of
> PB on x86 (32 bits).
> 
> Instead of abstracting away all accesses to PB behind offlineasm macros, we
> leave PB as a register and, on X86,
> - - rewrite the actual PB-using instruction to reference a temporary reg
> (which we get by spilling a value)
> - prepend instructions to load the value of PB from the stack in the
> temporary reg (if it's being read from)
> - append instructions to write it back to the stack from the temporary reg
> 
> The alternative would be to mediate all accesses to PB via offlineasm macros
> on all platforms. This way, we don't complicate the llint .asm code for the
> benefit of a single platform.

Generally, I'd rather have a special .asm file for X86, if it's particularly special. We don't look at the offline assembler code too often, so keeping that code as clean as possible has a lot of value.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.webkit.org/pipermail/webkit-unassigned/attachments/20200720/f5f2b147/attachment.htm>