Greetings! I'm cc'ing this to the Debian parisc experts as they are best qualified to answer your 'is it worth it' question. My guess is probably no, unless it is simply a matter of defining ATL_prefetch as the correct asm in some file, adding the 2.0 abi to the CFLAGS, and retiming. If it is any more work than this I suggest we wait until there is someone actually requesting it :-).
Take care, [EMAIL PROTECTED] (R Clint Whaley) writes: > Camm, > > OK, I was about to reply that the only PA-RISC machine I have access to > is a decade old, but turns out HP has kept compaq's testdrive, and so > I have managed to get access to some more modern machines. > > However, the question I have is, does anyone care? I cannot > find a single addressable PA-RISC tech document on HP's website. They > are switching to the Itanium for the future, and I am unaware of legions > of people using PA-RISC machines (though I've had a couple write in for > help). So, does it make sense to spend a lot of time on this arch, given > that it appears to be a dead end? > > Obviously, ATLAS can be installed on the machines, so the question is not one > of whether it's possible, but whether it is worth time to make it easy, > create defaults, look for speedups, etc . . . > > >Atlas has a well developed facility for using C macros to call various > >flavors of prefetch via __asm__ statements. This is at least used in > >this manner on i386 and powerpc, on which gcc also does not emit such > >instructions. Not that this warrants any immediate work, but my > >understanding is that some speedup of l1 and l2 in particular could be > >had with minimal labor. > > Maybe someone wants to point me at the techdoc that addresses these > extensions, > and while you are at it, docs that describe everything else, since I can't > find it? > > Cheers, > Clint > > > -- Camm Maguire [EMAIL PROTECTED] ========================================================================== "The earth is but one country, and mankind its citizens." -- Baha'u'llah
