[dataflow]: patch committed to improve -O0 performance.

Paolo Bonzini paolo.bonzini@lu.unisi.ch
Wed Mar 28 22:23:00 GMT 2007


Kenneth Zadeck wrote:
> Paolo Bonzini wrote:
>> I'm wondering, many passes that use DF_LIVE could actually use
>> DF_LR.  If they sum up substantially, DF_UR/DF_LIVE could become
>> optional problems.
>>
>> A nit:
>>
>>   
> They can, but the information is not as precise. 

I mean, maybe some can without changes in code generation.  Possibly
they look at liveness information only for registers that they know
to have been initialized.

Paolo



More information about the Gcc-patches mailing list