Bug 114811 – LLInt ARM backend should not use the d8 register as scratch register

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Gabor Rapcsanyi]

LLInt ARM backend is using d8 register as scratch register but the ARM ABI says that d8-d15 registers must be preserved across subroutine calls.

[Gabor Rapcsanyi]

Created attachment 198731 [details]
proposed fix

[Benjamin Poulain]

I don't know much about LLInt, but why does the ABI matters at all here?
I would think the only thing that matters is follow the JIT conventions(?).

[Filip Pizlo]

(In reply to comment #2)
> I don't know much about LLInt, but why does the ABI matters at all here?
> I would think the only thing that matters is follow the JIT conventions(?).

I think that the LLInt uses more scratch registers, and it doesn't have to use the same scratch registers as the JIT.  So this change looks valid to me.

[WebKit Commit Bot]

Comment on attachment 198731 [details]
proposed fix

Clearing flags on attachment: 198731

Committed r148705: <http://trac.webkit.org/changeset/148705>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[SangGyu Lee]

Is there any reason to duplicate my preceding bug report and fix (https://bugs.webkit.org/show_bug.cgi?id=114495) as new one?

[Benjamin Poulain]

(In reply to comment #6)
> Is there any reason to duplicate my preceding bug report and fix (https://bugs.webkit.org/show_bug.cgi?id=114495) as new one?

In WebKit, we care about:
-bug report with the best patch.
-bug report with the best info.

The order in which bugs are reported does not matter. We dupe backward or forward to the best bug report.

[Gabor Rapcsanyi]

(In reply to comment #6)
> Is there any reason to duplicate my preceding bug report and fix (https://bugs.webkit.org/show_bug.cgi?id=114495) as new one?

Sorry, I forgot to check before I reported and fixed it. I found this bug after debugging a bug for a week.

[SangGyu Lee]

Hello, Benjamin Poulain 

> In WebKit, we care about:
> -bug report with the best patch.
> -bug report with the best info.

> The order in which bugs are reported does not matter. We dupe backward or 
> forward to the best bug report.

What do you mean by "best patch" or "best info"?

Do you mean followings?

1) I didn't make my suggested patch as attached patch.
2) I didn't request to reviewer more aggressively (by using IRC or by sending mail again and again)

I cannot find the difference between mine and this one:

title(114495-mine): LLInt should not use d8 register as scratch register
title(114811-this): LLInt ARM backend should not use the d8 register as scratch register 

and here is my bug report:

Currently, LLInt uses d8 register as scratch register as followings:

ARM_SCRATCH_FPR = SpecialRegister.new("d8")    in arm.rb
C_LOOP_SCRATCH_FPR = SpecialRegister.new("d8") in cloop.rb

However, AAPCS §5.1.2.1 says

Registers s16-s31 (d8-d15, q4-q7) must be preserved across subroutine calls; registers s0-s15 (d0-d7, q0-q3) do not need to be preserved (and can be used for passing arguments or returning results in standard procedure-call variants). Registers d16-d31 (q8-q15), if present, do not need to be preserved.

Therefore it should not use d8 for that purpose.

I think it would be safe to use d6.
(Register d7 is already in use for other purpose. )

ARM_SCRATCH_FPR = SpecialRegister.new("d6")    in arm.rb
C_LOOP_SCRATCH_FPR = SpecialRegister.new("d6") in cloop.rb

[Csaba Osztrogonác]

Gábor ran into the same bug and then reported and fixed it quickly.
And then apologized to you because he didn't check/find if there was 
already a bug report about this bug. Additionally it isn't so easy to 
find it, because there are 15847 unconfirmed/new bugs now. If you file
a new bug report without cc-ing/poke-ing folks maybe nobody will notice
that you filed a new bug report.

[Mark Lam]

*** Bug 114495 has been marked as a duplicate of this bug. ***