This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: new cpu's for rs6000 target



On Friday, September 21, 2001, at 12:13 PM, Daniel Egger wrote:

> On Fre, 2001-09-21 at 21:03, Dale Johannesen wrote:
>
>> That's true.  Well, how about doing it this way then:
>>
>> (define_function_unit "iu" 1 0
>>    (and (eq_attr "type" "compare,delayed_compare")
>>         (eq_attr "cpu" "ppc750,ppc7400"))
>>    1 1  [(eq_attr "type" "!branch")])
>>
>> (define_function_unit "iu" 1 0
>>    (and (eq_attr "type" "compare,delayed_compare")
>>         (eq_attr "cpu" "ppc750,ppc7400"))
>>    3 1  [(eq_attr "type" "branch")])
>
>> I'll try that and see how it works.
>
> I think this matter is too new for me to grok all the effects
> of this change in one go. What are you trying to achieve with
> the additional restrictions?

This means:  if the compare/delayed_compare is followed by a dependent 
branch, assume a 3-cycle issue delay; if followed by anything else, assume 
a 1-cycle issue delay.
It seems to be a slight improvement (over 3 1 throughout) in practice, 
although the differences are small.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]