Map::Hash() calculation made deterministic in predictable mode.

Map::Hash() calculation made deterministic in predictable mode.

BUG=v8:3563
LOG=N
R=jkummerow@chromium.org

Committed: https://code.google.com/p/v8/source/detail?r=24281

[Igor Sheludko, 2014-09-29 08:18:27]

Patchset #1 (id:1) has been deleted

[Igor Sheludko, 2014-09-29 08:21:52]

ishell@chromium.org changed reviewers:
+ jkummerow@chromium.org

[Igor Sheludko, 2014-09-29 08:21:53]

PTAL

[Jakob Kummerow, 2014-09-29 09:25:30]

LGTM.

I wonder if we should do this in general (i.e. without FLAG_predictable too).

[Sven Panne, 2014-09-29 09:48:52]

On 2014/09/29 09:25:30, Jakob wrote:
> LGTM.
> 
> I wonder if we should do this in general (i.e. without FLAG_predictable too).

This is just what I thought, too: My gut feeling is that 18 + 18 + 8 bits input
for the 32bit hash function are enough.

[Igor Sheludko, 2014-09-29 10:18:32]

On 2014/09/29 09:48:52, Sven Panne wrote:
> On 2014/09/29 09:25:30, Jakob wrote:
> > LGTM.
> > 
> > I wonder if we should do this in general (i.e. without FLAG_predictable
too).
> 
> This is just what I thought, too: My gut feeling is that 18 + 18 + 8 bits
input
> for the 32bit hash function are enough.

Good point, thanks!

[Jakob Kummerow, 2014-09-29 10:38:21]

https://codereview.chromium.org/610363002/diff/40001/src/objects.cc
File src/objects.cc (right):

https://codereview.chromium.org/610363002/diff/40001/src/objects.cc#newcode9018
src/objects.cc:9018: hash ^= ObjectAddressForHashing(prototype()) << 2;
To document our offline discussion for posterity: I think it would make sense to
shift this value further left, so all 32 hash bits are (potentially) filled with
input bits.

[Igor Sheludko, 2014-09-29 10:48:49]

Addressed all the comments.

[Jakob Kummerow, 2014-09-29 10:51:05]

Thanks, still LGTM.

[Igor Sheludko, 2014-09-29 11:29:55]

Committed patchset #3 (id:60001) manually as 24281 (presubmit successful).