[llvm-commits] [llvm] r74429 - /llvm/trunk/lib/Target/X86/X86.td

[David Greene]

On Monday 29 June 2009 12:56, Nate Begeman wrote:
> On Jun 29, 2009, at 9:54 AM, David Greene wrote:
> > Author: greened
> > Date: Mon Jun 29 11:54:06 2009
> > New Revision: 74429
> >
> > URL: http://llvm.org/viewvc/llvm-project?rev=74429&view=rev
> > Log:
> >
> > Add processor descriptions for Istanbul and Shanghai.
> >
> > Modified:
> >    llvm/trunk/lib/Target/X86/X86.td
> >
> > Modified: llvm/trunk/lib/Target/X86/X86.td
> > URL:
> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86.td?rev=
> >74429&r1=74428&r2=74429&view=diff
> >
> > =
> > =
> > =
> > =
> > =
> > =
> > =
> > =
> > ======================================================================
> > --- llvm/trunk/lib/Target/X86/X86.td (original)
> > +++ llvm/trunk/lib/Target/X86/X86.td Mon Jun 29 11:54:06 2009
> > @@ -119,6 +119,10 @@
> >                                Feature3DNowA, Feature64Bit,
> > FeatureSlowBTMem]>;
> > def : Proc<"barcelona",       [FeatureSSE3,   FeatureSSE4A,
> >                                Feature3DNowA, Feature64Bit,
> > FeatureSlowBTMem]>;
> > +def : Proc<"istanbul",        [Feature3DNowA, Feature64Bit,
> > FeatureSSE4A,
> > +                               Feature3DNowA]>;
> > +def : Proc<"shanghai",        [Feature3DNowA, Feature64Bit,
> > FeatureSSE4A,
> > +                               Feature3DNowA]>;
>
> Is there some reason to have both, since they have the exact same
> feature strings?  Typically we've avoided added every Intel variant
> that has the same features as a previous one.

They have different cache characteristics and core counts.  That's actually 
one area where LLVM could use some work.  Right now we're missing a lot of 
architectural detail.

                                    -Dave