Implement the trampoline intrinsics and nest parameter for AIX. (#149388)

We can expand the init intrinsic to create a descriptor for the nested
procedure by combining the entry point and TOC pointer from the global
descriptor with the nest argument. The normal indirect call sequence
then calls the nested procedure through the descriptor like all other
calls. Patch also implements support for a nest parameter by mapping it
to gpr 11.

Author: mandlebug

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

@@ -3925,9 +3925,6 @@ SDValue PPCTargetLowering::LowerVACOPY(SDValue Op, SelectionDAG &DAG) const {
 
 SDValue PPCTargetLowering::LowerADJUST_TRAMPOLINE(SDValue Op,
                                                   SelectionDAG &DAG) const {
-  if (Subtarget.isAIXABI())
-    report_fatal_error("ADJUST_TRAMPOLINE operation is not supported on AIX.");
-
   return Op.getOperand(0);
 }
 
@@ -3984,16 +3981,72 @@ SDValue PPCTargetLowering::LowerINLINEASM(SDValue Op, SelectionDAG &DAG) const {
 
 SDValue PPCTargetLowering::LowerINIT_TRAMPOLINE(SDValue Op,
                                                 SelectionDAG &DAG) const {
-  if (Subtarget.isAIXABI())
-    report_fatal_error("INIT_TRAMPOLINE operation is not supported on AIX.");
-
   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
   SDLoc dl(Op);
 
   EVT PtrVT = getPointerTy(DAG.getDataLayout());
+
+  if (Subtarget.isAIXABI()) {
+    // On AIX we create a trampoline descriptor by combining the
+    // entry point and TOC from the global descriptor (FPtr) with the
+    // nest argument as the environment pointer.
+    uint64_t PointerSize = Subtarget.isPPC64() ? 8 : 4;
+    MaybeAlign PointerAlign(PointerSize);
+    auto MMOFlags = Subtarget.hasInvariantFunctionDescriptors()
+                        ? (MachineMemOperand::MODereferenceable |
+                           MachineMemOperand::MOInvariant)
+                        : MachineMemOperand::MONone;
+
+    uint64_t TOCPointerOffset = 1 * PointerSize;
+    uint64_t EnvPointerOffset = 2 * PointerSize;
+    SDValue SDTOCPtrOffset = DAG.getConstant(TOCPointerOffset, dl, PtrVT);
+    SDValue SDEnvPtrOffset = DAG.getConstant(EnvPointerOffset, dl, PtrVT);
+
+    const Value *TrampolineAddr =
+        cast<SrcValueSDNode>(Op.getOperand(4))->getValue();
+    const Function *Func =
+        cast<Function>(cast<SrcValueSDNode>(Op.getOperand(5))->getValue());
+
+    SDValue OutChains[3];
+
+    // Copy the entry point address from the global descriptor to the
+    // trampoline buffer.
+    SDValue LoadEntryPoint =
+        DAG.getLoad(PtrVT, dl, Chain, FPtr, MachinePointerInfo(Func, 0),
+                    PointerAlign, MMOFlags);
+    SDValue EPLoadChain = LoadEntryPoint.getValue(1);
+    OutChains[0] = DAG.getStore(EPLoadChain, dl, LoadEntryPoint, Trmp,
+                                MachinePointerInfo(TrampolineAddr, 0));
+
+    // Copy the TOC pointer from the global descriptor to the trampoline
+    // buffer.
+    SDValue TOCFromDescriptorPtr =
+        DAG.getNode(ISD::ADD, dl, PtrVT, FPtr, SDTOCPtrOffset);
+    SDValue TOCReg = DAG.getLoad(PtrVT, dl, Chain, TOCFromDescriptorPtr,
+                                 MachinePointerInfo(Func, TOCPointerOffset),
+                                 PointerAlign, MMOFlags);
+    SDValue TrampolineTOCPointer =
+        DAG.getNode(ISD::ADD, dl, PtrVT, Trmp, SDTOCPtrOffset);
+    SDValue TOCLoadChain = TOCReg.getValue(1);
+    OutChains[1] =
+        DAG.getStore(TOCLoadChain, dl, TOCReg, TrampolineTOCPointer,
+                     MachinePointerInfo(TrampolineAddr, TOCPointerOffset));
+
+    // Store the nest argument into the environment pointer in the trampoline
+    // buffer.
+    SDValue EnvPointer = DAG.getNode(ISD::ADD, dl, PtrVT, Trmp, SDEnvPtrOffset);
+    OutChains[2] =
+        DAG.getStore(Chain, dl, Nest, EnvPointer,
+                     MachinePointerInfo(TrampolineAddr, EnvPointerOffset));
+
+    SDValue TokenFactor =
+        DAG.getNode(ISD::TokenFactor, dl, MVT::Other, OutChains);
+    return TokenFactor;
+  }
+
   bool isPPC64 = (PtrVT == MVT::i64);
   Type *IntPtrTy = DAG.getDataLayout().getIntPtrType(*DAG.getContext());
 
@@ -6865,9 +6918,6 @@ static bool CC_AIX(unsigned ValNo, MVT ValVT, MVT LocVT,
   if (ValVT == MVT::f128)
     report_fatal_error("f128 is unimplemented on AIX.");
 
-  if (ArgFlags.isNest())
-    report_fatal_error("Nest arguments are unimplemented.");
-
   static const MCPhysReg GPR_32[] = {// 32-bit registers.
                                      PPC::R3, PPC::R4, PPC::R5, PPC::R6,
                                      PPC::R7, PPC::R8, PPC::R9, PPC::R10};
@@ -6882,6 +6932,14 @@ static bool CC_AIX(unsigned ValNo, MVT ValVT, MVT LocVT,
 
   const ArrayRef<MCPhysReg> GPRs = IsPPC64 ? GPR_64 : GPR_32;
 
+  if (ArgFlags.isNest()) {
+    MCRegister EnvReg = State.AllocateReg(IsPPC64 ? PPC::X11 : PPC::R11);
+    if (!EnvReg)
+      report_fatal_error("More then one nest argument.");
+    State.addLoc(CCValAssign::getReg(ValNo, ValVT, EnvReg, RegVT, LocInfo));
+    return false;
+  }
+
   if (ArgFlags.isByVal()) {
     const Align ByValAlign(ArgFlags.getNonZeroByValAlign());
     if (ByValAlign > StackAlign)

llvm/test/CodeGen/PowerPC/aix-nest-param.ll

@@ -1,5 +1,5 @@
-; RUN: not --crash llc -mtriple powerpc-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
-; RUN: not --crash llc -mtriple powerpc64-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
+; RUN: llc -mtriple powerpc-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
+; RUN: llc -mtriple powerpc64-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
 
 define ptr @nest_receiver(ptr nest %arg) nounwind {
   ret ptr %arg
@@ -9,5 +9,10 @@ define ptr @nest_caller(ptr %arg) nounwind {
   %result = call ptr @nest_receiver(ptr nest %arg)
   ret ptr %result
 }
+; CHECK-LABEL: .nest_receiver:
+; CHECK:         mr      3, 11
+; CHECK:         blr
 
-; CHECK: LLVM ERROR: Nest arguments are unimplemented.
+; CHECK-LABEL: .nest_caller:
+; CHECK:         mr      11, 3
+; CHECK:         bl .nest_receiver

llvm/test/CodeGen/PowerPC/aix-trampoline.ll

@@ -1,7 +1,7 @@
-; RUN: not --crash llc -mtriple powerpc-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
-; RUN: not --crash llc -mtriple powerpc64-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
-
-; CHECK: LLVM ERROR: INIT_TRAMPOLINE operation is not supported on AIX.
+; RUN: llc -mtriple powerpc-ibm-aix-xcoff < %s 2>&1 | \
+; RUN: FileCheck %s --check-prefix=32BIT
+; RUN: llc -mtriple powerpc64-ibm-aix-xcoff < %s 2>&1 -mattr=-altivec | \
+; RUN: FileCheck %s --check-prefix=64BIT
 
 define void @create_trampoline(ptr %buffer, ptr %nval) nounwind {
 entry:
@@ -12,3 +12,17 @@ entry:
 declare i32 @nested(i32);
 
 declare void @llvm.init.trampoline(ptr, ptr, ptr) nounwind
+
+; 32BIT:     stw 4, 8(3)
+; 32BIT:     lwz [[FuncDesc:[0-9]+]], L..C0(2)
+; 32BIT-DAG: lwz [[SCRATCH1:[0-9]+]], 0([[FuncDesc]])
+; 32BIT-DAG: lwz [[SCRATCH2:[0-9]+]], 4([[FuncDesc]])
+; 32BIT-DAG: stw [[SCRATCH1]], 0(3)
+; 32BIT-DAG: stw [[SCRATCH2]], 4(3)
+
+; 64BIT:     std 4, 16(3)
+; 64BIT-DAG: ld [[FuncDesc:[0-9]+]], L..C0(2)
+; 64BIT-DAG: ld [[SCRATCH1:[0-9]+]], 0([[FuncDesc]])
+; 64BIT-DAG: ld [[SCRATCH2:[0-9]+]], 8([[FuncDesc]])
+; 64BIT-DAG: std [[SCRATCH1]], 0(3)
+; 64BIT-DAG: std [[SCRATCH2]], 8(3)