Add support for trampolines on the XCore.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@124722 91177308-0d34-0410-b5e6-96231b3b80d8

5 files changed, 109 insertions, 1 deletions

diff --git a/lib/Target/XCore/XCoreCallingConv.td b/lib/Target/XCore/XCoreCallingConv.td
index 8107e329bd..b20d71f49c 100644
--- a/lib/Target/XCore/XCoreCallingConv.td
+++ b/lib/Target/XCore/XCoreCallingConv.td
@@ -24,6 +24,9 @@ def CC_XCore : CallingConv<[
   // Promote i8/i16 arguments to i32.
   CCIfType<[i8, i16], CCPromoteToType<i32>>,
 
+  // The 'nest' parameter, if any, is passed in R11.
+  CCIfNest<CCAssignToReg<[R11]>>,
+
   // The first 4 integer arguments are passed in integer registers.
   CCIfType<[i32], CCAssignToReg<[R0, R1, R2, R3]>>,
 
diff --git a/lib/Target/XCore/XCoreFrameLowering.cpp b/lib/Target/XCore/XCoreFrameLowering.cpp
index 0645c1654a..057822074e 100644
--- a/lib/Target/XCore/XCoreFrameLowering.cpp
+++ b/lib/Target/XCore/XCoreFrameLowering.cpp
@@ -100,6 +100,11 @@ void XCoreFrameLowering::emitPrologue(MachineFunction &MF) const {
   DebugLoc dl = MBBI != MBB.end() ? MBBI->getDebugLoc() : DebugLoc();
 
   bool FP = hasFP(MF);
+  bool Nested = MF.getFunction()->getAttributes().hasAttrSomewhere(Attribute::Nest);
+
+  if (Nested) {
+    loadFromStack(MBB, MBBI, XCore::R11, 0, dl, TII);
+  }
 
   // Work out frame sizes.
   int FrameSize = MFI->getStackSize();
diff --git a/lib/Target/XCore/XCoreISelLowering.cpp b/lib/Target/XCore/XCoreISelLowering.cpp
index 29fadf69a3..6ec5843cff 100644
--- a/lib/Target/XCore/XCoreISelLowering.cpp
+++ b/lib/Target/XCore/XCoreISelLowering.cpp
@@ -148,7 +148,10 @@ XCoreTargetLowering::XCoreTargetLowering(XCoreTargetMachine &XTM)
   setOperationAction(ISD::STACKSAVE, MVT::Other, Expand);
   setOperationAction(ISD::STACKRESTORE, MVT::Other, Expand);
   setOperationAction(ISD::DYNAMIC_STACKALLOC, MVT::i32, Expand);
-  
+
+  // TRAMPOLINE is custom lowered.
+  setOperationAction(ISD::TRAMPOLINE, MVT::Other, Custom);
+
   maxStoresPerMemset = maxStoresPerMemsetOptSize = 4;
   maxStoresPerMemmove = maxStoresPerMemmoveOptSize
     = maxStoresPerMemcpy = maxStoresPerMemcpyOptSize = 2;
@@ -178,6 +181,7 @@ LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::ADD:
   case ISD::SUB:              return ExpandADDSUB(Op.getNode(), DAG);
   case ISD::FRAMEADDR:        return LowerFRAMEADDR(Op, DAG);
+  case ISD::TRAMPOLINE:       return LowerTRAMPOLINE(Op, DAG);
   default:
     llvm_unreachable("unimplemented operand");
     return SDValue();
@@ -794,6 +798,64 @@ SDValue XCoreTargetLowering::LowerFRAMEADDR(SDValue Op,
                             RegInfo->getFrameRegister(MF), MVT::i32);
 }
 
+SDValue XCoreTargetLowering::
+LowerTRAMPOLINE(SDValue Op, SelectionDAG &DAG) const {
+  SDValue Chain = Op.getOperand(0);
+  SDValue Trmp = Op.getOperand(1); // trampoline
+  SDValue FPtr = Op.getOperand(2); // nested function
+  SDValue Nest = Op.getOperand(3); // 'nest' parameter value
+
+  const Value *TrmpAddr = cast<SrcValueSDNode>(Op.getOperand(4))->getValue();
+
+  // .align 4
+  // LDAPF_u10 r11, nest
+  // LDW_2rus r11, r11[0]
+  // STWSP_ru6 r11, sp[0]
+  // LDAPF_u10 r11, fptr
+  // LDW_2rus r11, r11[0]
+  // BAU_1r r11
+  // nest:
+  // .word nest
+  // fptr:
+  // .word fptr
+  SDValue OutChains[5];
+
+  SDValue Addr = Trmp;
+
+  DebugLoc dl = Op.getDebugLoc();
+  OutChains[0] = DAG.getStore(Chain, dl, DAG.getConstant(0x0a3cd805, MVT::i32),
+                              Addr, MachinePointerInfo(TrmpAddr), false, false,
+                              0);
+
+  Addr = DAG.getNode(ISD::ADD, dl, MVT::i32, Trmp,
+                     DAG.getConstant(4, MVT::i32));
+  OutChains[1] = DAG.getStore(Chain, dl, DAG.getConstant(0xd80456c0, MVT::i32),
+                              Addr, MachinePointerInfo(TrmpAddr, 4), false,
+                              false, 0);
+
+  Addr = DAG.getNode(ISD::ADD, dl, MVT::i32, Trmp,
+                     DAG.getConstant(8, MVT::i32));
+  OutChains[2] = DAG.getStore(Chain, dl, DAG.getConstant(0x27fb0a3c, MVT::i32),
+                              Addr, MachinePointerInfo(TrmpAddr, 8), false,
+                              false, 0);
+
+  Addr = DAG.getNode(ISD::ADD, dl, MVT::i32, Trmp,
+                     DAG.getConstant(12, MVT::i32));
+  OutChains[3] = DAG.getStore(Chain, dl, Nest, Addr,
+                              MachinePointerInfo(TrmpAddr, 12), false, false,
+                              0);
+
+  Addr = DAG.getNode(ISD::ADD, dl, MVT::i32, Trmp,
+                     DAG.getConstant(16, MVT::i32));
+  OutChains[4] = DAG.getStore(Chain, dl, FPtr, Addr,
+                              MachinePointerInfo(TrmpAddr, 16), false, false,
+                              0);
+
+  SDValue Ops[] =
+    { Trmp, DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains, 5) };
+  return DAG.getMergeValues(Ops, 2, dl);
+}
+
 //===----------------------------------------------------------------------===//
 //                      Calling Convention Implementation
 //===----------------------------------------------------------------------===//
diff --git a/lib/Target/XCore/XCoreISelLowering.h b/lib/Target/XCore/XCoreISelLowering.h
index febc198f4f..7e5dd2e8e5 100644
--- a/lib/Target/XCore/XCoreISelLowering.h
+++ b/lib/Target/XCore/XCoreISelLowering.h
@@ -147,6 +147,7 @@ namespace llvm {
     SDValue LowerUMUL_LOHI(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSMUL_LOHI(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerFRAMEADDR(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerTRAMPOLINE(SDValue Op, SelectionDAG &DAG) const;
   
     // Inline asm support
     std::vector<unsigned>
diff --git a/test/CodeGen/XCore/trampoline.ll b/test/CodeGen/XCore/trampoline.ll
new file mode 100644
index 0000000000..18cc45edbf
--- /dev/null
+++ b/test/CodeGen/XCore/trampoline.ll
@@ -0,0 +1,37 @@
+; RUN: llc < %s -march=xcore | FileCheck %s
+
+%struct.FRAME.f = type { i32, i32 ()* }
+
+define void @f() nounwind {
+entry:
+; CHECK: f:
+; CHECK ldap r11, g.1101
+; CHECK stw r11, sp[7]
+  %TRAMP.23 = alloca [20 x i8], align 2
+  %FRAME.0 = alloca %struct.FRAME.f, align 4
+  %TRAMP.23.sub = getelementptr inbounds [20 x i8]* %TRAMP.23, i32 0, i32 0
+  %FRAME.02 = bitcast %struct.FRAME.f* %FRAME.0 to i8*
+  %tramp = call i8* @llvm.init.trampoline(i8* %TRAMP.23.sub, i8* bitcast (i32 (%struct.FRAME.f*)* @g.1101 to i8*), i8* %FRAME.02)
+  %0 = getelementptr inbounds %struct.FRAME.f* %FRAME.0, i32 0, i32 1
+  %1 = bitcast i8* %tramp to i32 ()*
+  store i32 ()* %1, i32 ()** %0, align 4
+  %2 = getelementptr inbounds %struct.FRAME.f* %FRAME.0, i32 0, i32 0
+  store i32 1, i32* %2, align 4
+  call void @h(i32 ()* %1) nounwind
+  ret void
+}
+
+define internal i32 @g.1101(%struct.FRAME.f* nocapture nest %CHAIN.1) nounwind readonly {
+entry:
+; CHECK: g.1101:
+; CHECK: ldw r11, sp[0]
+; CHECK-NEXT: ldw r0, r11[0]
+; CHECK-NEXT: retsp 0
+  %0 = getelementptr inbounds %struct.FRAME.f* %CHAIN.1, i32 0, i32 0
+  %1 = load i32* %0, align 4
+  ret i32 %1
+}
+
+declare i8* @llvm.init.trampoline(i8*, i8*, i8*) nounwind
+
+declare void @h(i32 ()*)