Author: Eugene Nalimov
Date: 19:50:59 06/26/03
Go up one level in this thread
I didn't look at GCC sources, but I looked at sources of some other compilers, and understand x86 and PPC architecture well enough, so I think I know that x86 and PPC backends should be vastly different, and each should contain lot of platform-specific and unique code. Thanks, Eugene On June 26, 2003 at 17:40:39, Robert Hyatt wrote: >On June 26, 2003 at 16:47:06, Eugene Nalimov wrote: > >>On June 26, 2003 at 16:11:42, Robert Hyatt wrote: >> >>>On June 25, 2003 at 13:20:46, Tom Kerrigan wrote: >>> >>>>On June 25, 2003 at 04:52:12, Dann Corbit wrote: >>>> >>>>>On June 25, 2003 at 03:55:03, Andreas Guettinger wrote: >>>>> >>>>>>Apple Hardware VP Defends Benchmarks: >>>>>> >>>>>>http://apple.slashdot.org/apple/03/06/24/2154256.shtml?tid=126&tid=181 >>>>> >>>>>I'll be darned. An oinking weasel. >>>> >>>>It obviously doesn't pass the smell test when Apple's scores disagree with the >>>>officially submitted SPEC scores so dramatically, even if the VP does try to >>>>justify their testing methodology. >>>> >>>>The guy mentions that the PPC scores could have been higher if they had used a >>>>different compiler? Uhhh, why didn't they do that and avoid this whole mess? >>>> >>>>-Tom >>> >>> >>>His testing methodology was not _that_ bad. He _did_ use the same compiler for >>>both processors, which is certainly reasonable. >>> >>>Whether he used that specific compiler because it made the g5 look better is >>>another issue, although it is doubtful that the gcc guys have got any great >>>g5 customizations built in yet. >>> >>>One _could_ make a case for testing either way. (a) using the same compiler; >>>(b) using the _best_ compiler for each respective machine. >> >>By "the same compiler" you mean "the same front end?" >> >>Thanks, >>Eugene > >Actually I was thinking more about the "back end". IE emitting code and >optimizing same. The parsing and converting to some intermediate form is >not that interesting from this perspective. It's what happens after that >that becomes "interesting". > >In the case of GCC, there is plenty of evidence that the optimizer guys >do a lot of communicating, so that if one discovers a new trick, all the >backenders use it if possible and applicable. Vendors seem to be more >protective of their "tricks" for reasons that must relate to software >marketing rather than hardware marketing/performance. > > > > > >> >>>The classic problem with (b) is that humans are influencing the outcome in a >>>big way, because you not only measure raw hardware performance, you measure how >>>good the optimizing gurus are at their craft. Either way is open to lots of >>>criticism, unfortunately. >>> >>>SPEC is still going to be the best comparison since each vendor is free to >>>use the fastest compiler and settings he can find, so long as the result >>>produces correct and validated answers. The "gurus" still count, of course, >>>but absolute is absolute.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.