989641 - Move FlattenedMResumePointIter into LRecover.

Status: RESOLVED FIXED

(Core :: JavaScript Engine: JIT, defect)

Description

[Nicolas B. Pierron [:nbp]]

LRecover goals is to provide a linear way to iterate over the list of dependencies of recover instructions (MResumePoint / MInstructions).

FlattenedMResumePointIter is doing this operation 3 times and save the recorded sequence into a vector of MResumePoint.  We should cache the result of the vector and keep the result of the flatten vector instead of computing it 3 times.

Comment 1

[Nicolas B. Pierron [:nbp]]

Attachment: Move FlattenedMResumePointIter into LRecover

This patch is still using LRecover, I will rename the class to LRecoverInfo locally, as mentioned on Bug 989344.

Comment 2

[Jan de Mooij [:jandem]]

Comment on attachment 8398977 [details] [diff] [review]
Move FlattenedMResumePointIter into LRecover

Review of attachment 8398977 [details] [diff] [review]:
-----------------------------------------------------------------

Nice cleanup, r=me with comments addressed.

::: js/src/jit/LIR.cpp
@@ +144,5 @@
> +    MResumePoint *it = mir_;
> +
> +    // Sort operations in the order in which we need to restore the stack.  This
> +    // implies that outer frames as well as operations needed to recover the
> +    // inner frame content are located before. The top-level resume point should

Nit: "This implies that outer frames, as well as operations needed to recover the current frame, are located before the current frame. The inner-most resume point should be the last element in the list."

::: js/src/jit/LIR.h
@@ +878,5 @@
>  
>  class LRecover : public TempObject
>  {
> +  public:
> +    typedef Vector<MResumePoint *, 0, IonAllocPolicy> Instructions;

s/0/2, some inline storage to avoid the alloc-more-memory dance in the common case. There will always be at least 1 element in the vector.

@@ +881,5 @@
> +  public:
> +    typedef Vector<MResumePoint *, 0, IonAllocPolicy> Instructions;
> +
> +  private:
> +    // List of instructions needed to recover the stack frames.

Add: "Outer frames are stored before inner frames."

@@ +889,1 @@
>      MResumePoint *mir_;

Do we still need to store this one separately if we also store it in the Vector?

Comment 3

[Nicolas B. Pierron [:nbp]]

(In reply to Jan de Mooij [:jandem] from comment #2)
> @@ +889,1 @@
> >      MResumePoint *mir_;
> 
> Do we still need to store this one separately if we also store it in the
> Vector?

No, I will assert about it in the init function and modify the mir accessor to pick the last element of the vector.

Comment 4

[Nicolas B. Pierron [:nbp]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/10b769a8f5f2

Comment 5

[Ryan VanderMeulen [:RyanVM]]

Backed out for B2G bustage in the push.
https://hg.mozilla.org/integration/mozilla-inbound/rev/c4d1adee6057

https://tbpl.mozilla.org/php/getParsedLog.php?id=37290188&tree=Mozilla-Inbound

Comment 6

[Nicolas B. Pierron [:nbp]]

https://hg.mozilla.org/integration/mozilla-inbound/rev/704fd93fb58d

Comment 7

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/704fd93fb58d