llvm dev - Feb 2012 - [LLVMdev] Eliminating copies between overlapping register classes

Hi,

I have two register classes A and B, where A contains a subset of the
registers in B:

A = [R0, R1, R2, ... R128]

B = [RO, R1, R2, ... R128,
     T0, T1, T2, ... T128]

I am using the Greedy Register Allocator, and I would expect the register
allocator to eliminate this copy:

%vreg0<def> = COPY %vreg1; B:%vreg0 A:%vreg1

but instead I end up with

%R0<def> = COPY %R1

Is there any way I can get the register allocator to eliminate these
kinds of copies?

Thanks,
Tom

On Feb 22, 2012, at 12:01 PM, Tom Stellard wrote:
> Hi,
> 
> I have two register classes A and B, where A contains a subset of the
> registers in B:
> 
> A = [R0, R1, R2, ... R128]
> 
> B = [RO, R1, R2, ... R128,
>     T0, T1, T2, ... T128]
> 
> I am using the Greedy Register Allocator, and I would expect the register
> allocator to eliminate this copy:
> 
> %vreg0<def> = COPY %vreg1; B:%vreg0 A:%vreg1
> 
> but instead I end up with
> 
> %R0<def> = COPY %R1
> 
> Is there any way I can get the register allocator to eliminate these
> kinds of copies?
Such copies should be eliminated as aggressively as normal copies. Perhaps the
debug output from RegisterCoalescer can tell you why they aren't.

/jakob

On Wed, Feb 22, 2012 at 07:00:55PM -0800, Jakob Stoklund Olesen
wrote:
> 
> On Feb 22, 2012, at 12:01 PM, Tom Stellard wrote:
> 
> > Hi,
> > 
> > I have two register classes A and B, where A contains a subset of the
> > registers in B:
> > 
> > A = [R0, R1, R2, ... R128]
> > 
> > B = [RO, R1, R2, ... R128,
> >     T0, T1, T2, ... T128]
> > 
> > I am using the Greedy Register Allocator, and I would expect the
register
> > allocator to eliminate this copy:
> > 
> > %vreg0<def> = COPY %vreg1; B:%vreg0 A:%vreg1
> > 
> > but instead I end up with
> > 
> > %R0<def> = COPY %R1
> > 
> > Is there any way I can get the register allocator to eliminate these
> > kinds of copies?
> 
> Such copies should be eliminated as aggressively as normal copies. Perhaps
the debug output from RegisterCoalescer can tell you why they aren't.
> 
> /jakob
Thanks for the debugging type, I caught my mistake.  Class A wasn't
actually a subset of B. The definitions were actually like this:

A = [R0, R1, R2, ... R128, CC]

B = [RO, R1, R2, ... R128,
     T0, T1, T2, ... T128

Thanks,
Tom