Bug 194656 – lowerStackArgs should lower Lea32/64 on ARM64 to Add

• Status: RESOLVED
• Resolution: FIXED
• Component: JavaScriptCore

[Saam Barati]

Because it's not really an address. It's just an add on arm. So we end up using the wrong computation since we think we have a scaled address.

[Saam Barati]

Created attachment 362031 [details]
WIP

I want to add some more tests. But the lowering itself is done

[Saam Barati]

Created attachment 362070 [details]
WIP

[Saam Barati]

Created attachment 362086 [details]
patch

[Yusuke Suzuki]

Comment on attachment 362086 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=362086&action=review

r=me, I understand the issue, interesting.

> Source/JavaScriptCore/b3/air/AirLowerStackArgs.cpp:77
> +            if (isARM64() && (inst.kind.opcode == Lea32 || inst.kind.opcode == Lea64)) {
> +                // On ARM64, Lea is just an add. We can't handle this below because
> +                // taking into account the Width to see if we can compute the immediate
> +                // is wrong.

Should we add some notes in AirOpcode.opcodes? Basically, UA (UseAddr) typed address would cause a similar problem. (Currently, only LEA uses this type).

[Saam Barati]

Comment on attachment 362086 [details]
patch

View in context: https://bugs.webkit.org/attachment.cgi?id=362086&action=review

>> Source/JavaScriptCore/b3/air/AirLowerStackArgs.cpp:77
>> +                // is wrong.
> 
> Should we add some notes in AirOpcode.opcodes? Basically, UA (UseAddr) typed address would cause a similar problem. (Currently, only LEA uses this type).

I don’t fully follow this comment. Can you expand on it. I feel like I’m missing something. What do you mean by UseAdder? What do you mean by “typed adresses”?

[Saam Barati]

Oh I see what you mean. I’ll add a comment

[Saam Barati]

landed in:
https://trac.webkit.org/changeset/241577/webkit

[Radar WebKit Bug Importer]

<rdar://problem/48101457>