842167 - Baseline: Fix sunspider date tests

Status: RESOLVED WORKSFORME

(Core :: JavaScript Engine, defect)

Description

[Brian Hackett [Laid off!]]

On my machine, I get these scores for SS and its date tests, tofte and xparb.  Builds are threadsafe, x86/darwin, two cores:

trunk w/parallel:
  overall: 271.6
  date: 40.1

baseline w/parallel and 1000 use count:
  overall: 280.9
  date: 53.5

recent d8:
  overall: 246.9
  d8: 21.4

The difference on the date tests is about the same as the difference on the entire benchmark.  If baseline were to run as fast on the date tests as trunk, its score would be 267.5, and if it were to run as fast on the date tests as d8, its score would be 248.8.

In other words, there isn't much value in looking at other tests in sunspider.  If these two are fixed then our performance will be roughly in line with v8 and we will never ever have to look at sunspider again (right!?).

I haven't looked into the reasons for the difference, but suspect that eval() related stuff is the bulk of it and that fixing this will be less about Baseline than about the VM.

Comment 1

[Guilherme Lima]

Results

Nightly
overall: 256.2ms +/- 8.7%
date: 30.3ms +/- 5.3%
format-tofte: 14.5ms +/- 3.5%
format-xparb: 15.8ms +/- 7.6%

Chrome 32
overall: 271.7ms +/- 1.9%
date: 27.4ms +/- 5.7%
format-tofte: 15.1ms +/- 7.9%
format-xparb: 12.3ms +/- 7.3%

Comment 2

[André Bargull [:anba]]

Resolving as WFM given the results in comment #1.