We should get all of the V8 benchmarks running in Jaeger mode so we can start seeing what perf effects we're getting there, especially since those benchmarks are harder to trace, and should thus benefit more in practice from JM. The first two JSOps that we abort on are deflocalfun_fc and initmethod, but there are others lurking.
ETA is Friday 5 Mar 2010. I don't even know how many ops need to be added, so this is a very rough estimate.
Created attachment 430125 [details] [diff] [review] WIP (almost done) Almost done. I just need to update the throwpoline for x64.
Created attachment 430217 [details] [diff] [review] Patch
Attachment #430125 - Attachment is obsolete: true
Landed. I currently see a 6% speedup on the v8 benchmarks on my machine. It's a start.
Status: NEW → RESOLVED
Last Resolved: 9 years ago
Resolution: --- → FIXED
ETA was 5 Mar 2010. Actual was 3 Mar 2010. Main reason for the estimation error was that I didn't know how many ops I needed to do so I gave a very conservative estimate.
You need to log in before you can comment on or make changes to this bug.