Comment #12 on issue 2513 by [email protected]: life benchmark slowness
http://code.google.com/p/v8/issues/detail?id=2513
Ok, I added an option called EXPLICIT_ZEXT to emscripten, that does LLVM
zext explicitly using ?1:0. This is still preffed off by default, still
investigating the performance implications. So far I see that on the plus
side, this improves performance on v8 on this specific benchmark as well as
IE10, as expected. But I see no change in the rest of the benchmark suite
with this option on, so this issue (of booleans sent into bitwise
operators) might be quite limited in scope. On the minus side, performance
is slightly slower in firefox and safari, but again, limited to this
benchmark only.
It's possible this benchmark is 'special' somehow, but that seems unlikely
- it's very basic, straightforward C code that runs a game of life
simulation. So it's possible the ?1:0 pattern is causing something to
change in other emscripten optimizations, offsetting direct benefits
of ?1:0 in other benchmarks, I will need to look more into that.
--
--
v8-dev mailing list
[email protected]
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
For more options, visit https://groups.google.com/groups/opt_out.
