On Wed, Dec 18, 2013 at 4:46 PM, Vincent Belliard <
[email protected]> wrote:

> This change will trigger OSR faster and most of all, it will trigger OSR
> for big functions:
> on ARM, with the LUA benchmark, a function is very big and, when ticks
> equals 255, the function is still not optimized (and on IA32, it is
> optimized just before ticks equals 255). [...]
>

Just for the record: The Lua VM benchmark exposes at least 3 problems in v8:

   * The mysterious ad-hoc OSR limit we're talking about in this thread.

   * A totally wrong representation inference leading to a deopt loop.

   * If the previous deopt loop is fixed, it will lead to another deopt
loop for Math.round. :-P

I hope that I'll have some time to fix the last problem, but what's the
status of the 2nd problem? Toon? Ben?

-- 
-- 
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.

Reply via email to