Bug 113662 – Regions should be allocated from the same contiguous segment of virtual memory

• Component: JavaScriptCore
• Status: RESOLVED FIXED

[Mark Hahnenberg]

Instead of letting the OS spread our Regions all over the place, we should allocate them all within some range of each other. This change will open the door to some other optimizations, e.g. doing simple range checks for our write barriers and compressing JSCell pointers to 32-bits.

We can do this by having a SuperRegion for each BlockAllocator (i.e. per-VM) which subclasses MetaAllocator along the lines of the FixedVMPoolExecutableAllocator we use for the JIT.

[Mark Hahnenberg]

Created attachment 195921 [details]
Patch

[Mark Hahnenberg]

(In reply to comment #1)
> Created an attachment (id=195921) [details]
> Patch

This patch is actually a slight speedup on some benchmarks.

[Mark Hahnenberg]

Created attachment 195922 [details]
Patch

[Mark Hahnenberg]

(In reply to comment #3)
> Created an attachment (id=195922) [details]
> Patch

Had to add more ifdefs.

[Filip Pizlo]

Comment on attachment 195922 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=195922&action=review

Other than the one comment, looks good!

> Source/WTF/wtf/MetaAllocator.h:67
> +    WTF_EXPORT_PRIVATE MetaAllocator(size_t allocationGranule, size_t pageSize = 0);

Why not say pageSize = WTF::pageSize() and then get rid of the pageSize initialization conditional inside MetaAllocator::MetaAllocator?

[Mark Hahnenberg]

(In reply to comment #5)
> (From update of attachment 195922 [details])
> View in context: https://bugs.webkit.org/attachment.cgi?id=195922&action=review
> 
> Other than the one comment, looks good!
> 
> > Source/WTF/wtf/MetaAllocator.h:67
> > +    WTF_EXPORT_PRIVATE MetaAllocator(size_t allocationGranule, size_t pageSize = 0);
> 
> Why not say pageSize = WTF::pageSize() and then get rid of the pageSize initialization conditional inside MetaAllocator::MetaAllocator?

Good call.

[EFL EWS Bot]

Comment on attachment 195922 [details]
Patch

Attachment 195922 [details] did not pass efl-ews (efl):
Output: http://webkit-commit-queue.appspot.com/results/17392062

[Mark Hahnenberg]

Committed r147324: <http://trac.webkit.org/changeset/147324>

[Mark Hahnenberg]

So much build carnage...

http://trac.webkit.org/changeset/147328
http://trac.webkit.org/changeset/147329
http://trac.webkit.org/changeset/147330
http://trac.webkit.org/changeset/147332
http://trac.webkit.org/changeset/147334
http://trac.webkit.org/changeset/147335

[Geoffrey Garen]

Comment on attachment 195922 [details]
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=195922&action=review

> Source/JavaScriptCore/heap/SuperRegion.cpp:34
> +const size_t SuperRegion::s_fixedHeapMemoryPoolSize = 4 * 1024 * MB;

This seems a bit excessive, compared to most heaps. If the only cost of going over is a bit of performance, maybe we should make this smaller, like 1GB or 2GB.