Updates:
Status: Accepted
Comment #1 on issue 2451 by [email protected]: Math.round with
negative numbers causes a method to de/re-optimize repeatedly and finally
disabled
http://code.google.com/p/v8/issues/detail?id=2451
FYI: I've analyzed the deopts and the problem is that we infer an unboxed
integer as the representation for the result. This is not the right thing
to do, the spec explicitly says that values in the interval [-0.5, 0) are
mapped to -0, so we deopt in these cases.
Let's see if I can find a fix which doesn't regress our performance in the
usual benchmarks...
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
