Custom lower unaligned 32 bit stores and loads into libcalls. This is

a big code size win since before they were expanding to upto 16
instructions.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@75901 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/XCore/XCoreISelLowering.cpp b/lib/Target/XCore/XCoreISelLowering.cpp
index d65495076e4d2e6b1bc7aad91c1e20425431df9f..88502549a59985a841891aaf99723f5261f49bbd 100644 (file)
--- a/lib/Target/XCore/XCoreISelLowering.cpp
+++ b/lib/Target/XCore/XCoreISelLowering.cpp
@@ -131,7 +131,11 @@ XCoreTargetLowering::XCoreTargetLowering(XCoreTargetMachine &XTM)
 
   setLoadExtAction(ISD::SEXTLOAD, MVT::i8, Expand);
   setLoadExtAction(ISD::ZEXTLOAD, MVT::i16, Expand);
-  
+
+  // Custom expand misaligned loads / stores.
+  setOperationAction(ISD::LOAD, MVT::i32, Custom);
+  setOperationAction(ISD::STORE, MVT::i32, Custom);
+
   // Varargs
   setOperationAction(ISD::VAEND, MVT::Other, Expand);
   setOperationAction(ISD::VACOPY, MVT::Other, Expand);
@@ -159,6 +163,8 @@ LowerOperation(SDValue Op, SelectionDAG &DAG) {
   case ISD::GlobalTLSAddress: return LowerGlobalTLSAddress(Op, DAG);
   case ISD::ConstantPool:     return LowerConstantPool(Op, DAG);
   case ISD::JumpTable:        return LowerJumpTable(Op, DAG);
+  case ISD::LOAD:             return LowerLOAD(Op, DAG);
+  case ISD::STORE:            return LowerSTORE(Op, DAG);
   case ISD::SELECT_CC:        return LowerSELECT_CC(Op, DAG);
   case ISD::VAARG:            return LowerVAARG(Op, DAG);
   case ISD::VASTART:          return LowerVASTART(Op, DAG);
@@ -319,6 +325,87 @@ LowerJumpTable(SDValue Op, SelectionDAG &DAG)
   return DAG.getNode(XCoreISD::DPRelativeWrapper, dl, MVT::i32, JTI);
 }
 
+SDValue XCoreTargetLowering::
+LowerLOAD(SDValue Op, SelectionDAG &DAG)
+{
+  LoadSDNode *LD = cast<LoadSDNode>(Op);
+  assert(LD->getExtensionType() == ISD::NON_EXTLOAD && "Unexpected extension type");
+  assert(LD->getMemoryVT() == MVT::i32 && "Unexpected load MVT");
+  if (allowsUnalignedMemoryAccesses()) {
+    return SDValue();
+  }
+  unsigned ABIAlignment = getTargetData()->
+    getABITypeAlignment(LD->getMemoryVT().getTypeForMVT(*DAG.getContext()));
+  // Leave aligned load alone.
+  if (LD->getAlignment() >= ABIAlignment) {
+    return SDValue();
+  }
+  SDValue Chain = LD->getChain();
+  SDValue BasePtr = LD->getBasePtr();
+  DebugLoc dl = Op.getDebugLoc();
+  
+  // Lower to a call to __misaligned_load(BasePtr).
+  const Type *IntPtrTy = getTargetData()->getIntPtrType();
+  TargetLowering::ArgListTy Args;
+  TargetLowering::ArgListEntry Entry;
+  
+  Entry.Ty = IntPtrTy;
+  Entry.Node = BasePtr;
+  Args.push_back(Entry);
+  
+  std::pair<SDValue, SDValue> CallResult =
+        LowerCallTo(Chain, IntPtrTy, false, false,
+                    false, false, 0, CallingConv::C, false,
+                    DAG.getExternalSymbol("__misaligned_load", getPointerTy()),
+                    Args, DAG, dl);
+
+  SDValue Ops[] =
+    { CallResult.first, CallResult.second };
+
+  return DAG.getMergeValues(Ops, 2, dl);
+}
+
+SDValue XCoreTargetLowering::
+LowerSTORE(SDValue Op, SelectionDAG &DAG)
+{
+  StoreSDNode *ST = cast<StoreSDNode>(Op);
+  assert(!ST->isTruncatingStore() && "Unexpected store type");
+  assert(ST->getMemoryVT() == MVT::i32 && "Unexpected store MVT");
+  if (allowsUnalignedMemoryAccesses()) {
+    return SDValue();
+  }
+  unsigned ABIAlignment = getTargetData()->
+    getABITypeAlignment(ST->getMemoryVT().getTypeForMVT(*DAG.getContext()));
+  // Leave aligned store alone.
+  if (ST->getAlignment() >= ABIAlignment) {
+    return SDValue();
+  }
+  SDValue Chain = ST->getChain();
+  SDValue BasePtr = ST->getBasePtr();
+  SDValue Value = ST->getValue();
+  DebugLoc dl = Op.getDebugLoc();
+  
+  // Lower to a call to __misaligned_store(BasePtr, Value).
+  const Type *IntPtrTy = getTargetData()->getIntPtrType();
+  TargetLowering::ArgListTy Args;
+  TargetLowering::ArgListEntry Entry;
+  
+  Entry.Ty = IntPtrTy;
+  Entry.Node = BasePtr;
+  Args.push_back(Entry);
+  
+  Entry.Node = Value;
+  Args.push_back(Entry);
+  
+  std::pair<SDValue, SDValue> CallResult =
+        LowerCallTo(Chain, Type::VoidTy, false, false,
+                    false, false, 0, CallingConv::C, false,
+                    DAG.getExternalSymbol("__misaligned_store", getPointerTy()),
+                    Args, DAG, dl);
+
+  return CallResult.second;
+}
+
 SDValue XCoreTargetLowering::
 ExpandADDSUB(SDNode *N, SelectionDAG &DAG)
 {

diff --git a/lib/Target/XCore/XCoreISelLowering.h b/lib/Target/XCore/XCoreISelLowering.h
index 753ea819c2bdbeb2cce6595edf081f736c76f61b..e7a48a97265c589dcec3fd93e4d6523a67869872 100644 (file)
--- a/lib/Target/XCore/XCoreISelLowering.h
+++ b/lib/Target/XCore/XCoreISelLowering.h
@@ -101,6 +101,8 @@ namespace llvm {
                                     SelectionDAG &DAG);
 
     // Lower Operand specifics
+    SDValue LowerLOAD(SDValue Op, SelectionDAG &DAG);
+    SDValue LowerSTORE(SDValue Op, SelectionDAG &DAG);
     SDValue LowerRET(SDValue Op, SelectionDAG &DAG);
     SDValue LowerCALL(SDValue Op, SelectionDAG &DAG);
     SDValue LowerFORMAL_ARGUMENTS(SDValue Op, SelectionDAG &DAG);

diff --git a/test/CodeGen/XCore/unaligned_load.ll b/test/CodeGen/XCore/unaligned_load.ll
new file mode 100644 (file)
index 0000000..a6a5089
--- /dev/null
+++ b/test/CodeGen/XCore/unaligned_load.ll
@@ -0,0 +1,9 @@
+; RUN: llvm-as < %s | llc -march=xcore > %t1.s
+; RUN: grep "bl __misaligned_load" %t1.s | count 1
+
+; Byte aligned load. Expands to call to __misaligned_load.
+define i32 @align1(i32* %p) nounwind {
+entry:
+       %0 = load i32* %p, align 1              ; <i32> [#uses=1]
+       ret i32 %0
+}

diff --git a/test/CodeGen/XCore/unaligned_store.ll b/test/CodeGen/XCore/unaligned_store.ll
new file mode 100644 (file)
index 0000000..b7a5192
--- /dev/null
+++ b/test/CodeGen/XCore/unaligned_store.ll
@@ -0,0 +1,9 @@
+; RUN: llvm-as < %s | llc -march=xcore > %t1.s
+; RUN: grep "bl __misaligned_store" %t1.s | count 1
+
+; Byte aligned store. Expands to call to __misaligned_store.
+define void @align1(i32* %p, i32 %val) nounwind {
+entry:
+       store i32 %val, i32* %p, align 1
+       ret void
+}