[webkit-reviews] review granted: [Bug 215232] REGRESSION (r257667): 1.9x more CPU time in IPC::SharedBufferDataReference decoding during Netflix playback : [Attachment 406872] Patch

[bugzilla-daemon at webkit.org]

Geoffrey Garen <ggaren at apple.com> has granted Andy Estes <aestes at apple.com>'s
request for review:
Bug 215232: REGRESSION (r257667): 1.9x more CPU time in
IPC::SharedBufferDataReference decoding during Netflix playback
https://bugs.webkit.org/show_bug.cgi?id=215232

Attachment 406872: Patch

https://bugs.webkit.org/attachment.cgi?id=406872&action=review




--- Comment #9 from Geoffrey Garen <ggaren at apple.com> ---
Comment on attachment 406872
  --> https://bugs.webkit.org/attachment.cgi?id=406872
Patch

View in context: https://bugs.webkit.org/attachment.cgi?id=406872&action=review

r=me

> Source/WebKit/ChangeLog:43
> +	   LegacySharedBufferDataReference type, which reverts to the old
behaior of producing an

behavior

> Source/WebKit/ChangeLog:44
> +	   Resolve this slowdown by reverting the argument encoding changes
made to
> +	   IPC::SharedBufferDataReference in r257667. For compatibility with
code added after r257667,
> +	   IPC::SharedBufferDataReference retains its new behavior of copying
into a SharedBuffer when
> +	   decoded, but this type is now only used in cases where the
SharedBuffer can be ref'd instead
> +	   of copied. Pre-r257667 instances of SharedBufferDataReference were
changed to a new
> +	   LegacySharedBufferDataReference type, which reverts to the old
behaior of producing an
> +	   IPC::DataReference when decoded, avoiding many copies.

So LegacySharedBufferDataReference is the fast IPC data type and
SharedBufferDataReference is the slow IPC data type? That is... surprising.

Are we in danger of new adoptions of SharedBufferDataReference causing perf
regressions?