R600/SI: Custom lower CONCAT_VECTORS

This will lower them using register copies rather than loads and stores
to the stack.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@215270 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/R600/SIISelLowering.cpp b/lib/Target/R600/SIISelLowering.cpp
index f7717da..f9be144 100644 (file)
--- a/lib/Target/R600/SIISelLowering.cpp
+++ b/lib/Target/R600/SIISelLowering.cpp
@@ -196,10 +196,12 @@ SITargetLowering::SITargetLowering(TargetMachine &TM) :
       case ISD::BITCAST:
       case ISD::EXTRACT_VECTOR_ELT:
       case ISD::INSERT_VECTOR_ELT:
-      case ISD::CONCAT_VECTORS:
       case ISD::INSERT_SUBVECTOR:
       case ISD::EXTRACT_SUBVECTOR:
         break;
+      case ISD::CONCAT_VECTORS:
+        setOperationAction(Op, VT, Custom);
+        break;
       default:
         setOperationAction(Op, VT, Expand);
         break;

diff --git a/test/CodeGen/R600/concat_vectors.ll b/test/CodeGen/R600/concat_vectors.ll
index c38f3ee..753282b 100644 (file)
--- a/test/CodeGen/R600/concat_vectors.ll
+++ b/test/CodeGen/R600/concat_vectors.ll
@@ -77,8 +77,7 @@ define void @test_concat_v4f32(<8 x float> addrspace(1)* %out, <4 x float> %a, <
 }
 
 ; FUNC-LABEL: @test_concat_v8f32
-; FIXME: This is currently being expanded with loads / stores to the stack.
-; FIXME-SI-NOT: S_MOV_B32 s{{[0-9]}}, 0x80f000
+; SI-NOT: S_MOV_B32 s{{[0-9]}}, 0x80f000
 ; SI-NOT: MOVREL
 define void @test_concat_v8f32(<16 x float> addrspace(1)* %out, <8 x float> %a, <8 x float> %b) nounwind {
   %concat = shufflevector <8 x float> %a, <8 x float> %b, <16 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14, i32 15>