Patch [tree-ssa]: prevent edge probability overflow

Dale Johannesen dalej@apple.com
Sat Feb 28 00:51:00 GMT 2004


On Feb 27, 2004, at 2:01 PM, Richard Henderson wrote:

> On Fri, Feb 27, 2004 at 01:54:52PM -0800, Richard Henderson wrote:
>> It is *not* obvious.  There are existing places that probabilities
>> are manipulated.  You should be frobbing those, not random new places.
>
> Revise that.  No idea why we're doing computation here at all.
> We've found a known-taken edge.  There is now *one* outgoing edge.
> We can therefore set the probability at 100%.

Logically, yes (duh).  Unfortunately that's a bit less obvious, as it 
changes
the result in the (erroneous) case where the input edges don't add up...
I suppose a test cycle is in order.



More information about the Gcc-patches mailing list