[llvm-commits] [RFC/Patch 1.5/2 V2] ppc64-elf-linux-abi mark IsCodeGenOnly for specific operands for tblgen

[Hal Finkel]

----- Original Message -----
> From: "Will Schmidt" <will_schmidt at vnet.ibm.com>
> To: "Hal Finkel" <hfinkel at anl.gov>
> Cc: llvm-commits at cs.uiuc.edu
> Sent: Thursday, October 4, 2012 8:18:37 AM
> Subject: Re: [llvm-commits] [RFC/Patch 1.5/2 V2] ppc64-elf-linux-abi mark IsCodeGenOnly for specific operands for
> tblgen
> 
> On Wed, 2012-10-03 at 18:05 -0500, Hal Finkel wrote:
> > On Wed, 03 Oct 2012 17:09:13 -0500
> > Will Schmidt <will_schmidt at vnet.ibm.com> wrote:
> > 
> > > subj: [RFC/Patch 1.5/2 V2 ] ppc64-elf-linux-abi Add instruction
> > > tokens
> > > for tblgen
> > > 
> > > Slightly reworked per request. This content is broken out from
> > > the
> > > Sep14 patch "Add Instruction tokens for tblgen".
> > > 
> > > - Mark the BCC and BLR defs as isCodeGenOnly per error output
> > > from
> > > llvm-tblgen -gen-asm-matcher.
> > 
> > Do we want to mark BCC as isCodeGenOnly, or do we want to mark the
> > more-specialized mnemonics as isCodeGenOnly? If BCC is the most
> > generic
> > form, are there aliases for all other forms, and if there are, do
> > we
> > want separate instruction entries for all of them?
> 
> I've looked a bit, and I don't see anywhere that BCC is used as a
> basis
> for other mnemonics. At least, not in the .td files.
> 
> The tblgen output does call out the "'${cond:cc}'" operand modifier
> as
> not supported, which seems to be unique to the BCC def.
> 
> tblgen specifically reports
> Included from PowerPC/PPC.td:68:
> PowerPC//PPCInstrInfo.td:444:5: error: matchable with operand
> modifier '${cond:cc}' not supported by asm matcher.  Mark
> isCodeGenOnly!
>     def BCC : BForm<16, 0, 0, (outs), (ins pred:$cond,
>     condbrtarget:$dst),
>     ^
> 

Fair enough. In that case, LGTM.

 -Hal

> 
> Thanks for feedback,
> -Will
> 
> > 
> > Thanks again,
> > Hal
> > 
> > > 
> > > This change should not affect the current behavior of the build,
> > > and
> > > should be safe to commit after part 1/2.
> > > 
> > > Thanks,
> > > -Will
> > > 
> > > 
> > 
> > 
> > 
> 
> 
>