[XCore] Target optimized library function __memcpy_4()

Summary:
If the src, dst and size of a memcpy are known to be 4 byte aligned we
can call __memcpy_4() instead of memcpy().

Reviewers: robertlytton

Reviewed By: robertlytton

CC: llvm-commits

Differential Revision: http://llvm-reviews.chandlerc.com/D2871

llvm-svn: 202395

lib/Target/XCore/XCoreSelectionDAGInfo.cpp

@@ -21,3 +21,36 @@ XCoreSelectionDAGInfo::XCoreSelectionDAGInfo(const XCoreTargetMachine &TM)
 
 XCoreSelectionDAGInfo::~XCoreSelectionDAGInfo() {
 }
+
+SDValue XCoreSelectionDAGInfo::
+EmitTargetCodeForMemcpy(SelectionDAG &DAG, SDLoc dl, SDValue Chain,
+                        SDValue Dst, SDValue Src, SDValue Size, unsigned Align,
+                        bool isVolatile, bool AlwaysInline,
+                        MachinePointerInfo DstPtrInfo,
+                        MachinePointerInfo SrcPtrInfo) const
+{
+  unsigned SizeBitWidth = Size.getValueType().getSizeInBits();
+  // Call __memcpy_4 if the src, dst and size are all 4 byte aligned.
+  if (!AlwaysInline && (Align & 3) == 0 &&
+      DAG.MaskedValueIsZero(Size, APInt(SizeBitWidth, 3))) {
+    const TargetLowering &TLI = *DAG.getTarget().getTargetLowering();
+    TargetLowering::ArgListTy Args;
+    TargetLowering::ArgListEntry Entry;
+    Entry.Ty = TLI.getDataLayout()->getIntPtrType(*DAG.getContext());
+    Entry.Node = Dst; Args.push_back(Entry);
+    Entry.Node = Src; Args.push_back(Entry);
+    Entry.Node = Size; Args.push_back(Entry);
+
+    TargetLowering::CallLoweringInfo
+    CLI(Chain, Type::getVoidTy(*DAG.getContext()), false, false, false, false,
+        0, TLI.getLibcallCallingConv(RTLIB::MEMCPY), /*isTailCall=*/false,
+        /*doesNotRet=*/false, /*isReturnValueUsed=*/false,
+        DAG.getExternalSymbol("__memcpy_4", TLI.getPointerTy()), Args, DAG, dl);
+    std::pair<SDValue,SDValue> CallResult =
+      TLI.LowerCallTo(CLI);
+    return CallResult.second;
+  }
+
+  // Otherwise have the target-independent code call memcpy.
+  return SDValue();
+}

lib/Target/XCore/XCoreSelectionDAGInfo.h

@@ -24,6 +24,15 @@ class XCoreSelectionDAGInfo : public TargetSelectionDAGInfo {
 public:
   explicit XCoreSelectionDAGInfo(const XCoreTargetMachine &TM);
   ~XCoreSelectionDAGInfo();
+
+  virtual SDValue
+  EmitTargetCodeForMemcpy(SelectionDAG &DAG, SDLoc dl,
+                          SDValue Chain,
+                          SDValue Op1, SDValue Op2,
+                          SDValue Op3, unsigned Align, bool isVolatile,
+                          bool AlwaysInline,
+                          MachinePointerInfo DstPtrInfo,
+                          MachinePointerInfo SrcPtrInfo) const;
 };
 
 }

test/CodeGen/XCore/byVal.ll

@@ -20,7 +20,7 @@ entry:
 ; CHECK: ldaw r5, sp[1]
 ; CHECK: ldc r2, 40
 ; CHECK: mov r0, r5
-; CHECK: bl memcpy
+; CHECK: bl __memcpy_4
 ; CHECK: mov r0, r5
 ; CHECK: bl f1
 ; CHECK: mov r0, r4

test/CodeGen/XCore/memcpy.ll

@@ -0,0 +1,32 @@
+; RUN: llc < %s -march=xcore | FileCheck %s
+
+; Optimize memcpy to __memcpy_4 if src, dst and size are all 4 byte aligned.
+define void @f1(i8* %dst, i8* %src, i32 %n) nounwind {
+; CHECK-LABEL: f1:
+; CHECK: bl __memcpy_4
+entry:
+  %0 = shl i32 %n, 2
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src, i32 %0, i32 4, i1 false)
+  ret void
+}
+
+; Can't optimize - size is not a multiple of 4.
+define void @f2(i8* %dst, i8* %src, i32 %n) nounwind {
+; CHECK-LABEL: f2:
+; CHECK: bl memcpy
+entry:
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src, i32 %n, i32 4, i1 false)
+  ret void
+}
+
+; Can't optimize - alignment is not a multiple of 4.
+define void @f3(i8* %dst, i8* %src, i32 %n) nounwind {
+; CHECK-LABEL: f3:
+; CHECK: bl memcpy
+entry:
+  %0 = shl i32 %n, 2
+  call void @llvm.memcpy.p0i8.p0i8.i32(i8* %dst, i8* %src, i32 %0, i32 2, i1 false)
+  ret void
+}
+
+declare void @llvm.memcpy.p0i8.p0i8.i32(i8* nocapture, i8* nocapture, i32, i32, i1) nounwind