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]
Other format: [Raw text]

Re: Improve unrolled size estimates


>> +struct loop_size
>> +{
>> +  /* Number of instructions in the loop.  */
>> +  int overall;
>> +  /* Number of instructions that will be likely optimized out in
>> +     peeled iterations of loop  (i.e. computation based on induction
>> +     variable where induction variable starts at known constant.)  */
>> +  int eliminated_by_peeling;

Space after "int foo;" lines.

> Is it worth handling this very special case?  Note that you are 
> recursively asking for stuff again, so this may become quadratic, 
> especially as simple_iv isn't exactly very cheap.

Why instead not using two bitmaps for known-yes and known-no results of
constant_after_peeling for SSA_NAMEs?  Or alternatively a cache of
simple_iv results; I would like to use simple_iv in VRP too, to get good
bounds for simple instance variables instead of punting to
[something,+INF].  I'm not negative at all about Honza's approach with
one of these changes.

Paolo


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