Bug 195595 – blocksInPreOrder and blocksInPostOrder should reserve the right capacity for their result vector

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

[Robin Morisset]

Since we know that the result vector will be exactly the size of m_blocks (it is just a permutation).
We should also remove the inline capacity of that vector: even most small functions will have more than 5 BB.

[Robin Morisset]

Created attachment 364332 [details]
Patch

[Saam Barati]

Comment on attachment 364332 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=364332&action=review

> Source/JavaScriptCore/dfg/DFGBasicBlock.h:260
> +typedef Vector<BasicBlock*> BlockList;

What’s the motivation here?

[Saam Barati]

(In reply to Robin Morisset from comment #0)
> We should also remove the inline capacity of that vector: even most small
> functions will have more than 5 BB.

Based on what data? This is actually a bit surprising to me

[Robin Morisset]

(In reply to Saam Barati from comment #3)
> (In reply to Robin Morisset from comment #0)
> > We should also remove the inline capacity of that vector: even most small
> > functions will have more than 5 BB.
> 
> Based on what data? This is actually a bit surprising to me

I had just observed that while debugging some small test cases.

I just tested it on JetStream2. Out of 39158 calls to blocksIn{Pre/Post}Order, only 13676 had 5 or less BasicBlocks. So the current inline capacity is only used in 1/3 of cases, and it otherwise wastes space for no reason.
Here are some more numbers in case we really want an inline capacity here for some reason:
- 1 BB: 7806
- <= 4 BB: 12129
- <= 8 BB: 17974
- <= 16 BB: 25612
- <= 32 BB: 32558
- <= 34 BB: 37144

[Robin Morisset]

> - <= 34 BB: 37144

This one should have been <= 64 BB of course.

[WebKit Commit Bot]

Comment on attachment 364332 [details]
Patch

Clearing flags on attachment: 364332

Committed r242802: <https://trac.webkit.org/changeset/242802>

[WebKit Commit Bot]

All reviewed patches have been landed.  Closing bug.

[Radar WebKit Bug Importer]

<rdar://problem/48813529>