|
|
Ziemowit Laski wrote:
On 17 Jan 2005, at 5.07, Paolo Bonzini wrote:
Ziemowit Laski wrote:
This patch (finally!) fixes the mysterious loss of volatile-ness
experienced by PARM_DECLs
in ObjC methods. I'm not terribly proud of the objc.dg/volatile-1.m
test case, as it only really works for PowerPC Darwin targets; I'm
hoping people can add assembly patterns to search for on their own
platforms. Or perhaps there is a better, target-independent way of
testing this; suggestions are welcome and encouraged.
You can scan the ".optimized" dump.
I'm not a DejaGuru -- could you show me how that is done? :-) Also,
would that approach be any more portable than what I cooked up?
Yes, because you test what you want: that the front-end left the
assignment until it hands the function to the tree->RTL expander.
That's enough, I think, unless the test done this way passes before your
patch (which I doubt).
It would be something like
/* { dg-final { scan-tree-dump-times "a = 1" 1 "optimized"} } */
Paolo
|
|