[Mesa-dev] [PATCH 1/3] R600: Expand v2i32 load/store instead of custom lowering

Tom Stellard tom at stellard.net
Thu Jun 20 14:30:25 PDT 2013


On Wed, Jun 19, 2013 at 08:51:01AM -0700, Tom Stellard wrote:
> On Mon, Jun 17, 2013 at 04:11:38PM -0500, Aaron Watry wrote:
> > The custom lowering causes llc to crash with a segfault.
> > 
> > Ideally, the custom lowering can be fixed, but this allows
> > programs which load/store v2i32 to work without crashing.
> > 
> > Signed-off-by: Aaron Watry<awatry at gmail.com>
> > ---
> >  lib/Target/R600/R600ISelLowering.cpp | 4 ++--
> >  test/CodeGen/R600/load.vec.ll        | 6 ++++++
> >  2 files changed, 8 insertions(+), 2 deletions(-)
> > 
> > diff --git a/lib/Target/R600/R600ISelLowering.cpp b/lib/Target/R600/R600ISelLowering.cpp
> > index 9cedadb..812df83 100644
> > --- a/lib/Target/R600/R600ISelLowering.cpp
> > +++ b/lib/Target/R600/R600ISelLowering.cpp
> > @@ -86,7 +86,7 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
> >  
> >    // Legalize loads and stores to the private address space.
> >    setOperationAction(ISD::LOAD, MVT::i32, Custom);
> > -  setOperationAction(ISD::LOAD, MVT::v2i32, Custom);
> > +  setOperationAction(ISD::LOAD, MVT::v2i32, Expand);
> >    setOperationAction(ISD::LOAD, MVT::v4i32, Custom);
> >    setLoadExtAction(ISD::EXTLOAD, MVT::v4i8, Custom);
> >    setLoadExtAction(ISD::EXTLOAD, MVT::i8, Custom);
> > @@ -94,7 +94,7 @@ R600TargetLowering::R600TargetLowering(TargetMachine &TM) :
> >    setLoadExtAction(ISD::ZEXTLOAD, MVT::v4i8, Custom);
> >    setOperationAction(ISD::STORE, MVT::i8, Custom);
> >    setOperationAction(ISD::STORE, MVT::i32, Custom);
> > -  setOperationAction(ISD::STORE, MVT::v2i32, Custom);
> > +  setOperationAction(ISD::STORE, MVT::v2i32, Expand);
> 
> We have to use custom lowering in order for things like constant loads
> and global stores to work, so I think we need to fix whatever crash is
> happening with custom lowering.
>

As you pointed out on IRC, since v2[if]32 loads are expanded to [if]32
loads and we custom lower those, this patch should work fine.  I will
push this series.

Thanks,
Tom
> >    setOperationAction(ISD::STORE, MVT::v4i32, Custom);
> >  
> >    setOperationAction(ISD::LOAD, MVT::i32, Custom);
> > diff --git a/test/CodeGen/R600/load.vec.ll b/test/CodeGen/R600/load.vec.ll
> > index 08e034e..da1149a 100644
> > --- a/test/CodeGen/R600/load.vec.ll
> > +++ b/test/CodeGen/R600/load.vec.ll
> > @@ -1,6 +1,10 @@
> > +; RUN: llc < %s -march=r600 -mcpu=redwood | FileCheck --check-prefix=EG-CHECK  %s
> >  ; RUN: llc < %s -march=r600 -mcpu=SI | FileCheck --check-prefix=SI-CHECK  %s
> >  
> >  ; load a v2i32 value from the global address space.
> > +; EG-CHECK: @load_v2i32
> > +; EG-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 4
> > +; EG-CHECK: VTX_READ_32 T{{[0-9]+}}.X, T{{[0-9]+}}.X, 0
> >  ; SI-CHECK: @load_v2i32
> >  ; SI-CHECK: BUFFER_LOAD_DWORDX2 VGPR{{[0-9]+}}
> >  define void @load_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %in) {
> > @@ -10,6 +14,8 @@ define void @load_v2i32(<2 x i32> addrspace(1)* %out, <2 x i32> addrspace(1)* %i
> >  }
> >  
> >  ; load a v4i32 value from the global address space.
> > +; EG-CHECK: @load_v4i32
> > +; EG-CHECK: VTX_READ_128 T{{[0-9]+}}.XYZW, T{{[0-9]+}}.X, 0
> >  ; SI-CHECK: @load_v4i32
> >  ; SI-CHECK: BUFFER_LOAD_DWORDX4 VGPR{{[0-9]+}}
> >  define void @load_v4i32(<4 x i32> addrspace(1)* %out, <4 x i32> addrspace(1)* %in) {
> > -- 
> > 1.8.1.2
> > 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits


More information about the mesa-dev mailing list