[Webkit-unassigned] [Bug 69500] DFG should not enforce speculations on a variable if those speculations are never used

[bugzilla-daemon at webkit.org]

https://bugs.webkit.org/show_bug.cgi?id=69500





--- Comment #2 from Filip Pizlo <fpizlo at apple.com>  2011-10-05 21:01:46 PST ---
Created an attachment (id=109911)
 --> (https://bugs.webkit.org/attachment.cgi?id=109911&action=review)
the patch

It's not clear to me if this change is actually a progression.  Empirically, it isn't - it's neutral at best.

Analytically, it's no longer clear to me that this was a good idea.  There are two reasons why the current ToT strategy is best:

- Currently if we store an int even if we're never using it as an int, we get the benefit of only doing a 32-bit store rather than a 64-bit store.  The difference between doing a 32-bit store and a 64-bit store may be greater than the difference between doing a branch and not doing a branch (where the branch is the int check).  On 32_64, this will be a win.

- It's often the case that an-int-store-not-used-as-int comes from a value source that is statically an int, like a constant, or a computation.  In those cases, the 32-bit store is a pure win, since there is no need for the branch.

I'll leave this up here.  Maybe I'll change my mind, but right now I'm thinking this was a nifty, but failed, experiment.

-- 
Configure bugmail: https://bugs.webkit.org/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are the assignee for the bug.