Add support for scalarizing/splitting vector bswap.

Summary:
  SLP Vectorization of intrinsics (r203707) has exposed cases where the
  expansion of vector bswap is failing (PR19151).

Reviewers: hfinkel

CC: chandlerc
Differential Revision: http://llvm-reviews.chandlerc.com/D3104

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@204163 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
index 4f57bc320111a226fb96580c9b485ca447dab2a6..940a9c90593d561144342b23d60440c62490a9d2 100644 (file)
--- a/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
+++ b/lib/CodeGen/SelectionDAG/LegalizeVectorTypes.cpp
@@ -65,6 +65,7 @@ void DAGTypeLegalizer::ScalarizeVectorResult(SDNode *N, unsigned ResNo) {
   case ISD::UNDEF:             R = ScalarizeVecRes_UNDEF(N); break;
   case ISD::VECTOR_SHUFFLE:    R = ScalarizeVecRes_VECTOR_SHUFFLE(N); break;
   case ISD::ANY_EXTEND:
+  case ISD::BSWAP:
   case ISD::CTLZ:
   case ISD::CTPOP:
   case ISD::CTTZ:
@@ -533,6 +534,7 @@ void DAGTypeLegalizer::SplitVectorResult(SDNode *N, unsigned ResNo) {
     SplitVecRes_VECTOR_SHUFFLE(cast<ShuffleVectorSDNode>(N), Lo, Hi);
     break;
 
+  case ISD::BSWAP:
   case ISD::CONVERT_RNDSAT:
   case ISD::CTLZ:
   case ISD::CTTZ:

diff --git a/test/CodeGen/Mips/bswap.ll b/test/CodeGen/Mips/bswap.ll
index 4824388da1092342cdfc3e4068345c9f58751841..22b3db41a635050fb944a22ae8536e8203630f6c 100644 (file)
--- a/test/CodeGen/Mips/bswap.ll
+++ b/test/CodeGen/Mips/bswap.ll
@@ -22,7 +22,26 @@ entry:
   ret i64 %or.7
 }
 
+define <4 x i32> @bswapv4i32(<4 x i32> %x) nounwind readnone {
+entry:
+; MIPS32-LABEL: bswapv4i32:
+; MIPS32: wsbh $[[R0:[0-9]+]]
+; MIPS32: rotr ${{[0-9]+}}, $[[R0]], 16
+; MIPS32: wsbh $[[R0:[0-9]+]]
+; MIPS32: rotr ${{[0-9]+}}, $[[R0]], 16
+; MIPS32: wsbh $[[R0:[0-9]+]]
+; MIPS32: rotr ${{[0-9]+}}, $[[R0]], 16
+; MIPS32: wsbh $[[R0:[0-9]+]]
+; MIPS32: rotr ${{[0-9]+}}, $[[R0]], 16
+; mips16: .ent bswapv4i32
+  %ret = call <4 x i32> @llvm.bswap.v4i32(<4 x i32> %x)
+  ret <4 x i32> %ret
+}
+
+
+
 declare i32 @llvm.bswap.i32(i32) nounwind readnone
 
 declare i64 @llvm.bswap.i64(i64) nounwind readnone
 
+declare <4 x i32> @llvm.bswap.v4i32(<4 x i32>) nounwind readnone