Re: Performance of Integer Multiplication on PIII

[pete at ltoi dot iap dot physik dot tu-darmstadt dot de]

>> Thus, if (and you could see this in the .s files) gcc uses to much
>> equivalent replacement instructions for imul, then, even on the P4, with his
>> (not fully pipelined) imul, your handcoded version runs faster ...
>> thus leading us again, to the question, in what respect, the new gcc-3.0.x
>> x86 backend is improved?
>Yes, even when there are sequential dependencies, large expansion may
>lose due to overflowing trace cache.  Besides, it's useful to have
>options which work well on a variety of processors.

You mean, on i[3456] and P4 (we may call it i786 for convienience)?

Then "work well on a variety of processors" probably needs a proper
definition which, then, contains "values" (timings, or ... don't know)
for all archs & consider the fine point here, that you, by pure change
(... no, unknown wisdom, right ;-) code (nearly) perfect for the P6 and now
compare this P6 peek performance code with it's poorer performing
predeccessors/successors, which needs a different approach, in order to
achive their peek performance.

gcc's approach for i586 seems to work well, whereas in the i686 case the
compiler seems not to know whenever to use imul or not. Or, as Honza
wrote it:

>Gcc do have set of costs for each supported x86 variant
that means i[3456] & P4, with all P6 variants "aliased" to PPro.

> and for size
>optimization separately.  The model is an estimate (it does not handle leas
>and
>2 address machines perfectly), as the algorithm to construct sequences is not
>trying all possibilities, so it is possible that the produced sequences are
>suboptimal, but in case it makes truly big differences, we probably should
>investigate it.
>
>Honza

>>
>> - Supports: -march=athlon
>> - ?
>I know that the SuSE people have been working hard on this.
>I only wish it were not so difficult to change the OS on these AthlonMP
>SCSI drive boxes which come from Taiwan with Win98 installed.  Sorry to
>be OT, but does anything work? SuSE?
Though, the new K7 variant is supported by the new 2.4 Kernels, SMP too.
 ... am i not up to date?

Peter