This is the mail archive of the 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: [PATCH] Fix inaccurate estimation of probabilities in thescheduler

On Mon, 2005-07-04 at 07:42, Andrey Belevantsev wrote:
> The patch also introduces two parameters instead of hard-coded scheduler 
> values. These are maximal latency of the instruction, which is to be 
> considered for interblock code motion, and minimal relative basic block 
> probability. The default values are 3 and 40%, respectively.

Vlad said he is OK with the sched part of the patch.  I am OK with the
IA-64 and other parts.

I am a little concerned about the docs though.  There are currently no
target ports that modify params.  So IA-64 would be the first, and you
are making it dependent on optimization options.  This probably needs to
be documented somehow.  Otherwise, users may get confused that the
compiler does not behave as documented when they try to use the --param
option.  Maybe just a note that some targets may use different default
values, and that it may depend on optimization options?  If we try to
document the actual machine dependent values, we might have trouble with
the docs getting out of date, because people won't remember to change
the docs when they change the code.

Maybe an option to dump --param values would be useful, so that users
can see the actual machine dependent values.  That way we don't have to
document them.  This could be printed to the .s file for instance when
an extra -fdump-params (or whatever) option is used.  The --param docs
could then refer users to this output.
Jim Wilson, GNU Tools Support,

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