This is the mail archive of the
gcc-patches@gcc.gnu.org
mailing list for the GCC project.
Re: [itanium-sched-branch] Insert nops for MM-insns and tune placing stop bits [patch]
Richard Henderson wrote:
>
> On Mon, Oct 07, 2002 at 06:32:33PM -0400, Vladimir Makarov wrote:
> > + { "tune-stop-bits", MASK_TUNE_STOP_BITS, \
> > + N_("Enable tuning stop bits for better scheduling")}, \
>
> Might I suggest "early-stop-bits"? "Tune" makes it sound like
> it should be a universal improvement, whereas this is apparently
> not the case.
>
Yes, that is better. I'll rename it.
Actually, I expected that it will improve any benchmark (so it would
be universal). Unfortunately, it did not happen (at least for
SPECint2000).
> > + Permit to place stop bits not right before insn requiring it. It might
> > + improve insn scheduling.
>
> Allow stop bits to be placed earlier than immediately
> preceeding the instruction that triggered the stop bit.
> This can improve instruction scheduling, but does not
> always do so.
>
> It may also improve code density, correct?
Yes, that is right. In the most cases, it results in generating more
compact code. But with my point of view we should not pay much
attention to code density for Itanium. Intel compiler generates 1.5-2
times bigger but faster code. Intel compiler is also much faster than
gcc. Improving the compilation speed was the major reason for the
project.
Richard, thanks for the feedback (especially because it is not the
main line). I'll commit your variant of the option name and description
today.
Vlad