Open Bug 806638 Opened 8 years ago Updated 7 years ago
Monkey: Fix V8 Benchmark Performance
IonMonkey is 4,000 points slower than V8, on the V8-v7 benchmark suite. That leaves us with a 40% gap to cover. This is a tracking bug for the steps necessary to cover that gap.
Depends on: 768739
Depends on: 768745
Depends on: 765980
Depends on: 806646
Depends on: 768572
Okay with the most interesting bugs filed, here is a rough approximation of what we should be able to achieve on our overall v8 score, by fixing the worst performing benchmarks: regexp: +13% earley-boyer: +9% raytrace: +9% delta-blue: +5% crypto: +3% ----------------- total: +39%
(In reply to David Anderson [:dvander] from comment #1) > Okay with the most interesting bugs filed, here is a rough approximation of > what we should be able to achieve on our overall v8 score, by fixing the > worst performing benchmarks: > > regexp: +13% > earley-boyer: +9% > raytrace: +9% > delta-blue: +5% > crypto: +3% > ----------------- > total: +39% Remember that the total isn't additive in this way: making the above improvements to the score increases v8-bench's final score from 7814 -> 8136 locally, still far away from v8's 11127. In other words, if we make the changes above, we only see a 4% total improvement to the benchmark.
Script used to run the geometricMean() algorithm over proposed v8 improvements, so we can see the effect on the total more easily.
Sorry, I should have been clearer in comment #2: The number is the improvement we'd see to the overall v8 score if we brought that individual benchmark to parity with Crankshaft.
Depends on: 807188
You need to log in before you can comment on or make changes to this bug.