[llvm-commits] [llvm] r104653 - /llvm/trunk/lib/Target/ARM/ARMCodeEmitter.cpp

[Shih-wei Liao]

I hope to address the valid concern. I spent time today in retrying.
Unfortunately I think it's hard to find a way to generalize these
instructions. BFC and BFI are really "special" ones. So I think they deserve
these 5 lines of encoding in the patch.
Compared with 1600 lines in ARMCodeEmitter.cpp, maybe these lines are not
adding too much. My apologies for not being able to cut down to 0.

On Wed, May 26, 2010 at 2:15 AM, Shih-wei Liao <sliao at google.com> wrote:

> I agree that it's probably too easy to add one more to the many cases in
> ARMCodeEmitter.cpp. My bad. If changing ARM*.td can fix the bugs completely,
> that will be the best. Anyone knows of a better patch that's doable now?
> That will be great. I tried and stopped after the irregularities like
> widthm1 encoding and so on.
>
> Probably we should overhaul the ARM*.td some time soon. Currently these 2
> if-conditions can make the ARM JIT usable first, and I agree that I should
> strive for a better solution. Stay tuned.
>
> On Tue, May 25, 2010 at 10:07 PM, Evan Cheng <evan.cheng at apple.com> wrote:
> >
> > Sorry I have been behind on patch reviews. I have some concerns about
> these patches. We want to prevent special casing for individual
> instructions. Is it not possible to add generic support for these?
> >
> > Evan
> >
> > On May 25, 2010, at 5:25 PM, Shih-wei Liao <sliao at google.com> wrote:
> >
> > > Author: sliao
> > > Date: Tue May 25 19:25:05 2010
> > > New Revision: 104653
> > >
> > > URL: http://llvm.org/viewvc/llvm-project?rev=104653&view=rev
> > > Log:
> > > Adding the missing implementation of Bitfield's "clear" and "insert".
> > > Fixing http://llvm.org/bugs/show_bug.cgi?id=7222.
> > >
> > > Modified:
> > >    llvm/trunk/lib/Target/ARM/ARMCodeEmitter.cpp
> > >
> > > Modified: llvm/trunk/lib/Target/ARM/ARMCodeEmitter.cpp
> > > URL:
> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMCodeEmitter.cpp?rev=104653&r1=104652&r2=104653&view=diff
> > >
> ==============================================================================
> > > --- llvm/trunk/lib/Target/ARM/ARMCodeEmitter.cpp (original)
> > > +++ llvm/trunk/lib/Target/ARM/ARMCodeEmitter.cpp Tue May 25 19:25:05
> 2010
> > > @@ -781,10 +781,6 @@
> > >                                                    unsigned ImplicitRn)
> {
> > >   const TargetInstrDesc &TID = MI.getDesc();
> > >
> > > -  if (TID.Opcode == ARM::BFC) {
> > > -    report_fatal_error("ARMv6t2 JIT is not yet supported.");
> > > -  }
> > > -
> > >   // Part of binary is determined by TableGn.
> > >   unsigned Binary = getBinaryCodeForInstr(MI);
> > >
> > > @@ -820,6 +816,15 @@
> > >       Binary |= ((Hi16 >> 12) & 0xF) << 16;
> > >       emitWordLE(Binary);
> > >       return;
> > > +  } else if((TID.Opcode == ARM::BFC) || (TID.Opcode == ARM::BFI)) {
> > > +      uint32_t v = ~MI.getOperand(2).getImm();
> > > +      int32_t lsb = CountTrailingZeros_32(v);
> > > +      int32_t msb = (32 - CountLeadingZeros_32(v)) - 1;
> > > +      // Insts[20-16] = msb, Insts[11-7] = lsb
> > > +      Binary |= (msb & 0x1F) << 16;
> > > +      Binary |= (lsb & 0x1F) << 7;
> > > +      emitWordLE(Binary);
> > > +      return;
> > >   }
> > >
> > >   // If this is a two-address operand, skip it. e.g. MOVCCr operand 1.
> > >
> > >
> > > _______________________________________________
> > > llvm-commits mailing list
> > > llvm-commits at cs.uiuc.edu
> > > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
>
> --
> Thanks,
>   Shih-wei
>



-- 
Thanks,
  Shih-wei
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20100526/ee136429/attachment.html>