[llvm-commits] [patch] Mostly mechanical removal of getPhysicalRegisterRegClass.patch

[Jakob Stoklund Olesen]

On Jun 28, 2010, at 8:15 PM, Rafael Espindola wrote:

>> I think I was confused.
>> 
>> I have taken a closer look at InstrEmitter.cpp now. In EmitCopyFromReg(), getPhysicalRegisterRegClass() is only used to provide an argument for copyRegToReg(), and TLI->getRegClassFor(VT) is already used to create the virtual register.
>> 
>> That is fine as is. It is safe to replace getPhysicalRegisterRegClass with getMinimalPhysRegClass there.
>> 
>> The same goes for the uses in ScheduleDAG*.cpp
> 
> Sorry it took me so long to reply. Was "lost" in other areas.
> 
> The attached patch adds a VT argument to getMinimalPhysRegClass and
> replaces the copy related uses of getPhysicalRegisterRegClass with it.
> The argument being that if we want to make a copy (or estimate its
> cost), it is better to use the smallest class as more efficient
> operations might be possible.

Looks good.

> The other changes that were needed
> *) Update the ARM copyRegToReg to handle SPR_8

Yup.

> *) Update an ARM test, since we now produce a neon move instead of a
> vfp move. This is because the optimization:
> 
> else if (DestRC == ARM::DPR_VFP2RegisterClass ||
>             SrcRC == ARM::DPR_VFP2RegisterClass)
>      // Always use neon reg-reg move if source or dest is NEON-only regclass.
>      Opc = ARM::VMOVDneon;
> 
> is now used.

This is a little worrying, but probably not your fault.

Maybe the NEONMoveFix pass doesn't work properly for such a small function?

/jakob

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 1929 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20100628/6fe9020e/attachment.bin>