The default bug view has changed. See this FAQ.

Clean up JSScript::jitArityCheck{Normal,Ctor}

RESOLVED FIXED in mozilla14

Status

()

Core
JavaScript Engine
RESOLVED FIXED
5 years ago
5 years ago

People

(Reporter: njn, Assigned: njn)

Tracking

unspecified
mozilla14
Points:
---

Firefox Tracking Flags

(Not tracked)

Details

(Whiteboard: [MemShrink])

Attachments

(1 attachment)

(Assignee)

Description

5 years ago
Created attachment 612090 [details] [diff] [review]
patch

Currently, JSScript has jitCtor, jitNormal, jitArityCheckNormal,
jitArityCheckCtor.  And JITScript has arityCheckEntry.

arityCheckEntry is unused.  Well, it's used very briefly in finishThisUp(),
but it's written and then immediately read and never read again elsewhere,
so a local variable would suffice.

jitArityCheck{Ctor,Normal} serve two purposes.  First, they hold the arity
check entries for the corresponding JITScripts (which is odd;  presumably
JITScript::arityCheckEntry held that at some point in the past).  Second,
they indicate something extra about the JSScript -- if equal to NULL, the
script hasn't been compiled, and if equal to JS_UNJITTABLE_SCRIPT, the
script failed compilation.

This is ugly.  The attached patch cleans it up.  Specifically:

- JSScript::jit{Ctor,Normal} take on the tri-valued
  NULL/UNJITTABLE_SCRIPT/JITScript role.  This means that a script marked as
  unjittable cannot have a JITScript, whereas it could before.
  
  To avoid accidental misuse of the tri-value (e.g. assuming that a non-NULL
  value is a valid JITScript) these values are now encapsulated within the
  new JITScriptHandle class.

- script->jitArityCheck{Ctor,Normal} are moved into
  script->jit{Ctor,Normal}->arityCheckEntry.  This shrinks JSScript by two
  words.

  In finishThisUp() and destroyChunk() we were doing an odd little
  normal-or-ctor? dance w.r.t. arityCheckEntry that we were weren't doing
  with fastEntry and argsCheckEntry.  AFAICT this had exactly the same
  effect as just accessing arityCheckEntry directly, so I've made that change
  (and all the tests pass).

  The code generated by CallCompiler::generateFullCallStub is slightly more
  complex because it has to look in JITScript for the arityCheckEntry, but
  that's rare and so doesn't matter.
Attachment #612090 - Flags: review?(dvander)
Attachment #612090 - Flags: review?(dvander) → review+
(Assignee)

Comment 1

5 years ago
https://hg.mozilla.org/integration/mozilla-inbound/rev/55ba87dd42e2
http://hg.mozilla.org/mozilla-central/rev/55ba87dd42e2
Status: ASSIGNED → RESOLVED
Last Resolved: 5 years ago
Resolution: --- → FIXED
Target Milestone: --- → mozilla14
(In reply to Nicholas Nethercote [:njn] from comment #1)
> https://hg.mozilla.org/integration/mozilla-inbound/rev/55ba87dd42e2

This broke nomethodjit builds.
(Assignee)

Comment 4

5 years ago
> This broke nomethodjit builds.

Thanks for the info.  Easter Monday is a holiday here, I'll fix it on Tuesday.
(Assignee)

Comment 5

5 years ago
Unbreak --disable-methodjit:

https://hg.mozilla.org/integration/mozilla-inbound/rev/8e79f8755d65
I'd rather you didn't land patches with r=Ms2ger if I haven't seen them.
https://hg.mozilla.org/mozilla-central/rev/8e79f8755d65
You need to log in before you can comment on or make changes to this bug.