AST nodes have at most one bailout/typefeedback ID now, saving lots of memory.

AST nodes have at most one bailout/typefeedback ID now, saving lots of memory.

This is basically https://codereview.chromium.org/569573002/ done right:

During construction, each node type tells its parent how many IDs it
needs in addition to the parent's ones. This is done all the way up in
the class hierarchy until a node's parent doesn't need any ID. At that
point we know how many IDs in summary are needed, and we reserve the
whole range at once, saving only the base ID of that range. All IDs
are now calculated via simple offsets to that base ID. To all
performaniacs: The C++ compiler simplifies the constant calculation to
a simple load and the addition of a single constant.

Note that the actual code is much simpler than all that prose above. :-)
It's basically how compilers for OO languages figure out vtable entries.

We still have lots of holes due to padding in the AST nodes, but this
will be addressed in a separate CL.

BUG=chromium:417697
LOG=y
R=mvstanton@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=24524

[Sven Panne, 2014-10-10 09:46:18]

svenpanne@chromium.org changed reviewers:
+ bmeurer@chromium.org

[Sven Panne, 2014-10-10 09:46:18]

One ID to rule them all, One ID for the TypeFeedbackOracle to find them,
One ID to bring them all and in the DeoptimizationOutputData bind them
In the Land of the Deoptimizer where the Shadows lie.

:-D

[Sven Panne, 2014-10-10 10:31:27]

svenpanne@chromium.org changed reviewers:
+ mvstanton@chromium.org
- bmeurer@chromium.org

[mvstanton, 2014-10-10 10:42:03]

Nice space savings!
LGTM.

[Sven Panne, 2014-10-10 10:52:41]

Committed patchset #4 (id:100001) manually as 24524 (presubmit successful).

[wingo, 2014-10-10 12:00:19]

On 2014/10/10 10:52:41, Sven Panne wrote:
> Committed patchset #4 (id:100001) manually as 24524 (presubmit successful).

Nice!  FWIW d8 --no-lazy http://wingolog.org/pub/jquery-parser-test.js on this
goes down from 480.576M instructions to 478.964M instructions.  So it saves time
and space :)

[Sven Panne, 2014-10-10 12:44:51]

On 2014/10/10 12:00:19, wingo wrote:
> On 2014/10/10 10:52:41, Sven Panne wrote:
> > Committed patchset #4 (id:100001) manually as 24524 (presubmit successful).
> 
> Nice!  FWIW d8 --no-lazy http://wingolog.org/pub/jquery-parser-test.js on this
> goes down from 480.576M instructions to 478.964M instructions.  So it saves
time
> and space :)

As expected, the savings with this CL alone are not impressive due to padding.
The *real* savings come here: https://codereview.chromium.org/646803004/ (at
least for asm.js-relevant node types).