[AIX] Implement caller arguments passed in stack memory

[cebowleratibm]

Summary

This patch implements the caller side of placing function call arguments in stack memory. This removes the current limitation where LLVM on AIX will report fatal error when arguments can't be contained in registers.

There is a particular oddity that a float argument that passes in a register and also in stack memory requires that the caller initialize both. From what AIX "ABI" documentation I have it's not clear that this needs to be done, however, it is necessary for compatibility with the AIX XL compiler so I think it's best to implement it the same way.

Note a later patch will follow to address the callee side.

[ZarkoCA]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

6873

mark this as const as well?

6886–6891

Can we add a test for this? Specifically checking if the full memory for the arg is initialized.

7091–7099

I would prefer to have the MemLoc code after the RegLoc. To me it's more intuitive that way since we use up registers first and then stack. It's purely a personal preference though and have no issue with keeping it as is.

7147–7151

Can we have a testcase for this?

[cebowleratibm]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

6886–6891

call_test_stackarg_float2 covers this scenario. There are STFS and STFD for the trailing float and double args.

7091–7099

I think this is a personal preference but I'll reorder the logic.

7147–7151

That's fair. test_stackarg_float2 covers has a double arg split across that last available GPR. I can add a case for the scenario where the double arg will fit in the last 2 available GPRs.

[cebowleratibm]

Added a test for a double that passes in fpr and 2 gpr.

Addressed non functional issues identified in the review.

[ZarkoCA]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

6886–6891

Yep, I missed it the first time, thanks for pointing it out.

7067

Does the value given by CCInfo.getNextStackOffset() account for the linkage save area? I've found that it doesn't and only gives the size of the PSA when working on the callee side.

[cebowleratibm]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

7067

LowerCall_AIX does:

CCInfo.AllocateStack(LinkageSize, PtrByteSize);
CCInfo.AnalyzeCallOperands(Outs, CC_AIX);

It needs to be done this way so that any stack arguments will get the correct stack offset. When AnalyzeCallOperands is done, the stack offset accounts for the linkage save area.

LowerFormalArguments_AIX should also AllocateStack the linkage save area as well. I think this can be done with the next patch for the callee work.

[ZarkoCA]

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

7067

Got it, missed the part CCInfo.AllocateStack(LinkageSize, PtrByteSize); of the code previously.

[ZarkoCA]

fyi@cebowleratibm To me, this patch looks good now. I'm holding on approving to give other reviewers a chance to have a look.

[sfertile]

Code changes look good. I 'll start on reviewing the tests now.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

6997

Moving the increment of the index into the loop makes this a lot more complicated even though its a small change to the code. If we have to peek the next memloc then its unavoidable, but did we consider marking the memloc we want to skip as custom and having the normal memloc handling skip it?

7146–7156

Really minor nit: there is a bunch of extra spaces on this blank line.

7150

Really minor nit: Same thing here, a bunch of spaces or a tab after the '{'.

[sfertile]

llvm/test/CodeGen/PowerPC/aix-cc-abi.ll

1098

This is unused in the test.

1117

Just to make sure I understand correctly:

This store is just for moving the double value into the gprs. We don't have to initialize the register image in the parameter save area (even though the call is variadic) because it maps into the gprs available for argument passing. If the gprs were exhausted but we still have available fprs then we have to initialize both the fpr and the register image in the PSA (this case being tested by the following test).

1265

Where do we initialize $r10? I don't think we are checking for the asm that should be generated from

; 64BIT-DAG:   STFD renamable $f1, 0, %stack.0 :: (store 8 into %stack.0)
; 64BIT-DAG:   renamable $x10 = LD 0, %stack.0 :: (load 8 from %stack.0)

[cebowleratibm]

Addressed Sean's comments. Notable change: when a float arg passes in FPR as well as PSA memory we will mark the memLoc custom. LowerFormalArguments_AIX can ignore custom memLoc. If a float arg passes in PSA memory and there was no FPR available, then the memLoc will be normal (non-custom.) I've updated LowerFormalArguments_AIX such that it will pass over custom memloc but continue to report_fatal_error on non-custom memloc. Zarko will have a patch coming to remove the report_fatal_error and implement memLoc for LowerFormalArguments_AIX.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

6997

I can use the customMem to make it cleaner to skip over these. I will use custom mem when the argument is also passed in FPR. If the argument is only passed in memory, it will remain a regular memLoc.

llvm/test/CodeGen/PowerPC/aix-cc-abi.ll

1098

carry over from call_test_stackarg_float. Good catch.

1117

Your understanding is correct. call_test_stackarg_float3, passes float args in FPR as well as PSA memory. call_test_stackarg_float2 tests that the required GPRs are initialized. The stack store is only an autolocal artifact required to initialize the GPRs. On Pwr8 up the operation can be done without going through memory but we can't test it yet because it requires vector instructions that we've blocked on AIX.

1265

I missed it. We're generating and it belongs in the test.

[cebowleratibm]

Used custom mem to tag memLocs for float args that also pass in register so the LowerFormalArguments_AIX can skip them easily.

Addressed minor cleanup and testcase updates from review.

[sfertile]

LGTM.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

6997

👍 Thank you. I really like this approach.

llvm/lib/Target/PowerPC/PPCISelLowering.cpp

static bool CC_AIX(unsigned ValNo, MVT ValVT, MVT LocVT,
   switch (ValVT.SimpleTy) {
   default:
     report_fatal_error("Unhandled value type for argument.");
   case MVT::i64:
     // i64 arguments should have been split to i32 for PPC32.
     assert(IsPPC64 && "PPC32 should have split i64 values.");
     LLVM_FALLTHROUGH;
   case MVT::i1:
-  case MVT::i32:
-    State.AllocateStack(PtrByteSize, PtrByteSize);
+  case MVT::i32: {
+    const unsigned Offset = State.AllocateStack(PtrByteSize, PtrByteSize);
+    const MVT RegVT = IsPPC64 ? MVT::i64 : MVT::i32;
     if (unsigned Reg = State.AllocateReg(IsPPC64 ? GPR_64 : GPR_32)) {
-      MVT RegVT = IsPPC64 ? MVT::i64 : MVT::i32;
       // Promote integers if needed.
       if (ValVT.getSizeInBits() < RegVT.getSizeInBits())
         LocInfo = ArgFlags.isSExt() ? CCValAssign::LocInfo::SExt
                                     : CCValAssign::LocInfo::ZExt;
       State.addLoc(CCValAssign::getReg(ValNo, ValVT, Reg, RegVT, LocInfo));
     }
     else
-      report_fatal_error("Handling of placing parameters on the stack is "
-                         "unimplemented!");
-    return false;
+      State.addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, RegVT, LocInfo));
 
+    return false;
+  }
   case MVT::f32:
   case MVT::f64: {
     // Parameter save area (PSA) is reserved even if the float passes in fpr.
     const unsigned StoreSize = LocVT.getStoreSize();
     // Floats are always 4-byte aligned in the PSA on AIX.
     // This includes f64 in 64-bit mode for ABI compatibility.
-    State.AllocateStack(IsPPC64 ? 8 : StoreSize, 4);
-    if (unsigned Reg = State.AllocateReg(FPR))
-      State.addLoc(CCValAssign::getReg(ValNo, ValVT, Reg, LocVT, LocInfo));
-    else
-      report_fatal_error("Handling of placing parameters on the stack is "
-                         "unimplemented!");
+    const unsigned Offset = State.AllocateStack(IsPPC64 ? 8 : StoreSize, 4);
+    unsigned FReg = State.AllocateReg(FPR);
+    if (FReg)
+      State.addLoc(CCValAssign::getReg(ValNo, ValVT, FReg, LocVT, LocInfo));
 
-    // AIX requires that GPRs are reserved for float arguments.
-    // Successfully reserved GPRs are only initialized for vararg calls.
+    // Reserve and initialize GPRs or initialize the PSA as required.
     MVT RegVT = IsPPC64 ? MVT::i64 : MVT::i32;
     for (unsigned I = 0; I < StoreSize; I += PtrByteSize) {
       if (unsigned Reg = State.AllocateReg(IsPPC64 ? GPR_64 : GPR_32)) {
+        assert(FReg && "An FPR should be available when a GPR is reserved.");
         if (State.isVarArg()) {
+          // Successfully reserved GPRs are only initialized for vararg calls.
           // Custom handling is required for:
           //   f64 in PPC32 needs to be split into 2 GPRs.
           //   f32 in PPC64 needs to occupy only lower 32 bits of 64-bit GPR.
           State.addLoc(
               CCValAssign::getCustomReg(ValNo, ValVT, Reg, RegVT, LocInfo));
         }
-      } else if (State.isVarArg()) {
-        report_fatal_error("Handling of placing parameters on the stack is "
-                           "unimplemented!");
+      } else {
+        // If there are insufficient GPRs, the PSA needs to be initialized.
+        // Initialization occurs even if an FPR was initialized for
+        // compatibility with the AIX XL compiler. The full memory for the
+        // argument will be initialized even if a prior word is saved in GPR.
+        State.addLoc(CCValAssign::getMem(ValNo, ValVT, Offset, LocVT, LocInfo));
+        break;
       }
     }
 
     return false;
   }
   }
   return true;
 }
SDValue PPCTargetLowering::LowerFormalArguments_AIX(
 
   // Reserve space for the linkage area on the stack.
   const unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
   // On AIX a minimum of 8 words is saved to the parameter save area.
   const unsigned MinParameterSaveArea = 8 * PtrByteSize;
   CCInfo.AllocateStack(LinkageSize + MinParameterSaveArea, PtrByteSize);
   CCInfo.AnalyzeFormalArguments(Ins, CC_AIX);
 
-  for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
-    CCValAssign &VA = ArgLocs[i];
-    SDValue ArgValue;
-    ISD::ArgFlagsTy Flags = Ins[i].Flags;
+  for (unsigned I = 0, E = ArgLocs.size(); I != E;) {
+    CCValAssign &VA = ArgLocs[I++];
+    ISD::ArgFlagsTy Flags = Ins[VA.getValNo()].Flags;
     if (VA.isRegLoc()) {
       EVT ValVT = VA.getValVT();
       MVT LocVT = VA.getLocVT();
       MVT::SimpleValueType SVT = ValVT.getSimpleVT().SimpleTy;
       unsigned VReg =
           MF.addLiveIn(VA.getLocReg(), getRegClassForSVT(SVT, IsPPC64));
-      ArgValue = DAG.getCopyFromReg(Chain, dl, VReg, LocVT);
+      SDValue ArgValue = DAG.getCopyFromReg(Chain, dl, VReg, LocVT);
       if (ValVT.isScalarInteger() &&
           (ValVT.getSizeInBits() < LocVT.getSizeInBits())) {
         ArgValue =
             truncateScalarIntegerArg(Flags, ValVT, DAG, ArgValue, LocVT, dl);
       }
       InVals.push_back(ArgValue);
+
+      // For compatibility with the AIX XL compiler, the float args in the
+      // parameter save area are initialized even if the argument is available
+      // in register.  The caller is required to initialize both the register
+      // and memory, however, the callee can choose to expect it in either.  The
+      // memloc is dismissed here because the argument is retrieved from the
+      // register.
+      if ((ValVT == MVT::f32 || ValVT == MVT::f64) && I != E &&
+          ArgLocs[I].isMemLoc() && ArgLocs[I].getValNo() == VA.getValNo())
+        ++I;
     } else {
       report_fatal_error("Handling of formal arguments on the stack is "
                          "unimplemented!");
     }
   }
 
   // Area that is at least reserved in the caller of this function.
   unsigned MinReservedArea = CCInfo.getNextStackOffset();
SDValue PPCTargetLowering::LowerCall_AIX(
   CCState CCInfo(CallConv, isVarArg, MF, ArgLocs, *DAG.getContext());
 
   // Reserve space for the linkage save area (LSA) on the stack.
   // In both PPC32 and PPC64 there are 6 reserved slots in the LSA:
   //   [SP][CR][LR][2 x reserved][TOC].
   // The LSA is 24 bytes (6x4) in PPC32 and 48 bytes (6x8) in PPC64.
   const unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
   const bool IsPPC64 = Subtarget.isPPC64();
+  const EVT PtrVT = getPointerTy(DAG.getDataLayout());
   const unsigned PtrByteSize = IsPPC64 ? 8 : 4;
   CCInfo.AllocateStack(LinkageSize, PtrByteSize);
   CCInfo.AnalyzeCallOperands(Outs, CC_AIX);
 
   // The prolog code of the callee may store up to 8 GPR argument registers to
   // the stack, allowing va_start to index over them in memory if the callee
   // is variadic.
   // Because we cannot tell if this is needed on the caller side, we have to
   // conservatively assume that it is needed.  As such, make sure we have at
   // least enough stack space for the caller to store the 8 GPRs.
   const unsigned MinParameterSaveAreaSize = 8 * PtrByteSize;
-  const unsigned NumBytes = LinkageSize + MinParameterSaveAreaSize;
+  const unsigned NumBytes = std::max(LinkageSize + MinParameterSaveAreaSize,
+                                     CCInfo.getNextStackOffset());
 
   // Adjust the stack pointer for the new arguments...
   // These operations are automatically eliminated by the prolog/epilog pass.
   Chain = DAG.getCALLSEQ_START(Chain, NumBytes, 0, dl);
   SDValue CallSeqStart = Chain;
 
   SmallVector<std::pair<unsigned, SDValue>, 8> RegsToPass;
+  SmallVector<SDValue, 8> MemOpChains;
+
+  // Set up a copy of the stack pointer for loading and storing any
+  // arguments that may not fit in the registers available for argument
+  // passing.
+  const SDValue StackPtr = IsPPC64 ? DAG.getRegister(PPC::X1, MVT::i64)
+                                   : DAG.getRegister(PPC::R1, MVT::i32);
 
   for (unsigned I = 0, E = ArgLocs.size(); I != E;) {
     CCValAssign &VA = ArgLocs[I++];
 
-    if (VA.isMemLoc())
-      report_fatal_error("Handling of placing parameters on the stack is "
-                         "unimplemented!");
-    if (!VA.isRegLoc())
-      report_fatal_error(
-          "Unexpected non-register location for function call argument.");
-
     SDValue Arg = OutVals[VA.getValNo()];
 
+    if (!VA.isRegLoc() && !VA.isMemLoc())
+      report_fatal_error("Unexpected location for function call argument.");
+
+    if (VA.isMemLoc()) {
+      SDValue PtrOff =
+          DAG.getConstant(VA.getLocMemOffset(), dl, StackPtr.getValueType());
+      PtrOff = DAG.getNode(ISD::ADD, dl, PtrVT, StackPtr, PtrOff);
+      MemOpChains.push_back(
+          DAG.getStore(Chain, dl, Arg, PtrOff, MachinePointerInfo()));
+
+      continue;
+    }
+
     if (!VA.needsCustom()) {
       switch (VA.getLocInfo()) {
       default:
         report_fatal_error("Unexpected argument extension type.");
       case CCValAssign::Full:
         break;
       case CCValAssign::ZExt:
         Arg = DAG.getNode(ISD::ZERO_EXTEND, dl, VA.getLocVT(), Arg);
else {
       // The 2 GPRs are marked custom and expected to be adjacent in ArgLocs.
       assert(Arg.getValueType() == MVT::f64 && isVarArg && !IsPPC64 &&
              "Unexpected custom register for argument!");
       CCValAssign &GPR1 = VA;
       SDValue MSWAsI64 = DAG.getNode(ISD::SRL, dl, MVT::i64, ArgAsInt,
                                      DAG.getConstant(32, dl, MVT::i8));
       RegsToPass.push_back(std::make_pair(
           GPR1.getLocReg(), DAG.getZExtOrTrunc(MSWAsI64, dl, MVT::i32)));
-      assert(I != E && "A second custom GPR is expected!");
+      
+      if (I != E) {
+        // If only 1 GPR was available, there will only be one custom GPR and the argument will also pass in memory.
+        CCValAssign &PeekArg = ArgLocs[I];
+        if (PeekArg.isRegLoc() && PeekArg.getValNo() == PeekArg.getValNo()) {          
+          assert(PeekArg.needsCustom() && "A second custom GPR is expected.");
           CCValAssign &GPR2 = ArgLocs[I++];
-      assert(GPR2.isRegLoc() && GPR2.getValNo() == GPR1.getValNo() &&
-             GPR2.needsCustom() && "A second custom GPR is expected!");
           RegsToPass.push_back(std::make_pair(
               GPR2.getLocReg(), DAG.getZExtOrTrunc(ArgAsInt, dl, MVT::i32)));
         }
       }
+    }
+  }
+
+  if (!MemOpChains.empty())
+    Chain = DAG.getNode(ISD::TokenFactor, dl, MVT::Other, MemOpChains);
 
   // For indirect calls, we need to save the TOC base to the stack for
   // restoration after the call.
   if (!isTailCall && !isPatchPoint &&
       !isFunctionGlobalAddress(Callee) && !isa<ExternalSymbolSDNode>(Callee)) {
     const MCRegister TOCBaseReg = Subtarget.getTOCPointerRegister();
     const MCRegister StackPtrReg = Subtarget.getStackPointerRegister();
     const MVT PtrVT = Subtarget.isPPC64() ? MVT::i64 : MVT::i32;

llvm/test/CodeGen/PowerPC/aix-cc-abi.ll

 entry:
   %0 = load double, double* @d1, align 8
   call double @test_fpr_max(double %0, double %0, double %0, double %0, double %0, double %0, double %0, double %0, double %0, double %0, double %0, double %0, double %0)
   ret void
 }
 
 ; CHECK-LABEL: name: call_test_fpr_max{{.*}}
 
-; 32BIT:      renamable $r3 = LWZtoc @d1, $r2 :: (load 4 from got)
-; 32BIT-NEXT: renamable $f1 = LFD 0, killed renamable $r3 :: (dereferenceable load 8 from @d1)
-; 32BIT-NEXT: ADJCALLSTACKDOWN 56, 0, implicit-def dead $r1, implicit $r1
-; 32BIT-NEXT: $f2 = COPY renamable $f1
-; 32BIT-NEXT: $f3 = COPY renamable $f1
-; 32BIT-NEXT: $f4 = COPY renamable $f1
-; 32BIT-NEXT: $f5 = COPY renamable $f1
-; 32BIT-NEXT: $f6 = COPY renamable $f1
-; 32BIT-NEXT: $f7 = COPY renamable $f1
-; 32BIT-NEXT: $f8 = COPY renamable $f1
-; 32BIT-NEXT: $f9 = COPY renamable $f1
-; 32BIT-NEXT: $f10 = COPY renamable $f1
-; 32BIT-NEXT: $f11 = COPY renamable $f1
-; 32BIT-NEXT: $f12 = COPY renamable $f1
-; 32BIT-NEXT: $f13 = COPY renamable $f1
-; 32BIT-NEXT: BL_NOP <mcsymbol .test_fpr_max>, csr_aix32, implicit-def dead $lr, implicit $rm, implicit $f1, implicit killed $f2, implicit killed $f3, implicit killed $f4, implicit killed $f5, implicit killed $f6, implicit killed $f7, implicit killed $f8, implicit killed $f9, implicit killed $f10, implicit killed $f11, implicit killed $f12, implicit killed $f13, implicit $r2, implicit-def $r1
-; 32BIT-NEXT: ADJCALLSTACKUP 56, 0, implicit-def dead $r1, implicit $r1
+; 32BIT:      renamable $r[[REG:[0-9]+]] = LWZtoc @d1, $r2 :: (load 4 from got)
+; 32BIT-NEXT: renamable $f1 = LFD 0, killed renamable $r[[REG]] :: (dereferenceable load 8 from @d1)
+; 32BIT-NEXT: ADJCALLSTACKDOWN 128, 0, implicit-def dead $r1, implicit $r1
+; 32BIT-DAG:  STFD renamable $f1, 56, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 64, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 72, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 80, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 88, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 96, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 104, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 112, $r1 :: (store 8)
+; 32BIT-DAG:  STFD renamable $f1, 120, $r1 :: (store 8)
+; 32BIT-DAG:  $f2 = COPY renamable $f1
+; 32BIT-DAG:  $f3 = COPY renamable $f1
+; 32BIT-DAG:  $f4 = COPY renamable $f1
+; 32BIT-DAG:  $f5 = COPY renamable $f1
+; 32BIT-DAG:  $f6 = COPY renamable $f1
+; 32BIT-DAG:  $f7 = COPY renamable $f1
+; 32BIT-DAG:  $f8 = COPY renamable $f1
+; 32BIT-DAG:  $f9 = COPY renamable $f1
+; 32BIT-DAG:  $f10 = COPY renamable $f1
+; 32BIT-DAG:  $f11 = COPY renamable $f1
+; 32BIT-DAG:  $f12 = COPY renamable $f1
+; 32BIT-DAG:  $f13 = COPY renamable $f1
+; 32BIT-NEXT: BL_NOP <mcsymbol .test_fpr_max>, csr_aix32, implicit-def dead $lr, implicit $rm, implicit $f1, implicit killed $f2, implicit killed $f3, implicit killed $f4, implicit killed $f5, implicit killed $f6, implicit killed $f7, implicit killed $f8, implicit killed $f9, implicit killed $f10, implicit killed $f11, implicit killed $f12, implicit killed $f13, implicit $r2, implicit-def $r1, implicit-def dead $f1
+; 32BIT-NEXT: ADJCALLSTACKUP 128, 0, implicit-def dead $r1, implicit $r1
 
-; 64BIT:      renamable $x3 = LDtoc @d1, $x2 :: (load 8 from got)
-; 64BIT-NEXT: renamable $f1 = LFD 0, killed renamable $x3 :: (dereferenceable load 8 from @d1)
-; 64BIT-NEXT: ADJCALLSTACKDOWN 112, 0, implicit-def dead $r1, implicit $r1
-; 64BIT-NEXT: $f2 = COPY renamable $f1
-; 64BIT-NEXT: $f3 = COPY renamable $f1
-; 64BIT-NEXT: $f4 = COPY renamable $f1
-; 64BIT-NEXT: $f5 = COPY renamable $f1
-; 64BIT-NEXT: $f6 = COPY renamable $f1
-; 64BIT-NEXT: $f7 = COPY renamable $f1
-; 64BIT-NEXT: $f8 = COPY renamable $f1
-; 64BIT-NEXT: $f9 = COPY renamable $f1
-; 64BIT-NEXT: $f10 = COPY renamable $f1
-; 64BIT-NEXT: $f11 = COPY renamable $f1
-; 64BIT-NEXT: $f12 = COPY renamable $f1
-; 64BIT-NEXT: $f13 = COPY renamable $f1
+; CHECKASM-LABEL: .call_test_fpr_max:
+
+; ASM32PWR4:       stwu 1, -128(1)
+; ASM32PWR4-NEXT:  lwz [[REG:[0-9]+]], LC2(2)
+; ASM32PWR4-NEXT:  lfd 1, 0([[REG]])
+; ASM32PWR4-DAG:   stfd 1, 56(1)
+; ASM32PWR4-DAG:   stfd 1, 64(1)
+; ASM32PWR4-DAG:   stfd 1, 72(1)
+; ASM32PWR4-DAG:   stfd 1, 80(1)
+; ASM32PWR4-DAG:   stfd 1, 88(1)
+; ASM32PWR4-DAG:   stfd 1, 96(1)
+; ASM32PWR4-DAG:   stfd 1, 104(1)
+; ASM32PWR4-DAG:   stfd 1, 112(1)
+; ASM32PWR4-DAG:   stfd 1, 120(1)
+; ASM32PWR4-DAG:   fmr 2, 1
+; ASM32PWR4-DAG:   fmr 3, 1
+; ASM32PWR4-DAG:   fmr 4, 1
+; ASM32PWR4-DAG:   fmr 5, 1
+; ASM32PWR4-DAG:   fmr 6, 1
+; ASM32PWR4-DAG:   fmr 7, 1
+; ASM32PWR4-DAG:   fmr 8, 1
+; ASM32PWR4-DAG:   fmr 9, 1
+; ASM32PWR4-DAG:   fmr 10, 1
+; ASM32PWR4-DAG:   fmr 11, 1
+; ASM32PWR4-DAG:   fmr 12, 1
+; ASM32PWR4-DAG:   fmr 13, 1
+; ASM32PWR4-NEXT:  bl .test_fpr_max
+; ASM32PWR4-NEXT:  nop
+; ASM32PWR4-NEXT:  addi 1, 1, 128
+
+; 64BIT:      renamable $x[[REGD1ADDR:[0-9]+]] = LDtoc @d1, $x2 :: (load 8 from got)
+; 64BIT-NEXT: renamable $f1 = LFD 0, killed renamable $x[[REGD1ADDR:[0-9]+]] :: (dereferenceable load 8 from @d1)
+; 64BIT-NEXT: ADJCALLSTACKDOWN 152, 0, implicit-def dead $r1, implicit $r1
+; 64BIT-DAG:  STFD renamable $f1, 112, $x1 :: (store 8)
+; 64BIT-DAG:  STFD renamable $f1, 120, $x1 :: (store 8)
+; 64BIT-DAG:  STFD renamable $f1, 128, $x1 :: (store 8)
+; 64BIT-DAG:  STFD renamable $f1, 136, $x1 :: (store 8)
+; 64BIT-DAG:  STFD renamable $f1, 144, $x1 :: (store 8)
+; 64BIT-DAG:  $f2 = COPY renamable $f1
+; 64BIT-DAG:  $f3 = COPY renamable $f1
+; 64BIT-DAG:  $f4 = COPY renamable $f1
+; 64BIT-DAG:  $f5 = COPY renamable $f1
+; 64BIT-DAG:  $f6 = COPY renamable $f1
+; 64BIT-DAG:  $f7 = COPY renamable $f1
+; 64BIT-DAG:  $f8 = COPY renamable $f1
+; 64BIT-DAG:  $f9 = COPY renamable $f1
+; 64BIT-DAG:  $f10 = COPY renamable $f1
+; 64BIT-DAG:  $f11 = COPY renamable $f1
+; 64BIT-DAG:  $f12 = COPY renamable $f1
+; 64BIT-DAG:  $f13 = COPY renamable $f1
 ; 64BIT-NEXT: BL8_NOP <mcsymbol .test_fpr_max>, csr_aix64, implicit-def dead $lr8, implicit $rm, implicit $f1, implicit killed $f2, implicit killed $f3, implicit killed $f4, implicit killed $f5, implicit killed $f6, implicit killed $f7, implicit killed $f8, implicit killed $f9, implicit killed $f10, implicit killed $f11, implicit killed $f12, implicit killed $f13, implicit $x2, implicit-def $r1
-; 64BIT-NEXT: ADJCALLSTACKUP 112, 0, implicit-def dead $r1, implicit $r1
+; 64BIT-NEXT: ADJCALLSTACKUP 152, 0, implicit-def dead $r1, implicit $r1
+
+; ASM64PWR4:       stdu 1, -160(1)
+; ASM64PWR4-NEXT:  ld [[REG:[0-9]+]], LC2(2)
+; ASM64PWR4-NEXT:  lfd 1, 0([[REG]])
+; ASM64PWR4-DAG:   stfd 1, 112(1)
+; ASM64PWR4-DAG:   stfd 1, 120(1)
+; ASM64PWR4-DAG:   stfd 1, 128(1)
+; ASM64PWR4-DAG:   stfd 1, 136(1)
+; ASM64PWR4-DAG:   stfd 1, 144(1)
+; ASM64PWR4-DAG:   fmr 2, 1
+; ASM64PWR4-DAG:   fmr 3, 1
+; ASM64PWR4-DAG:   fmr 4, 1
+; ASM64PWR4-DAG:   fmr 5, 1
+; ASM64PWR4-DAG:   fmr 6, 1
+; ASM64PWR4-DAG:   fmr 7, 1
+; ASM64PWR4-DAG:   fmr 8, 1
+; ASM64PWR4-DAG:   fmr 9, 1
+; ASM64PWR4-DAG:   fmr 10, 1
+; ASM64PWR4-DAG:   fmr 11, 1
+; ASM64PWR4-DAG:   fmr 12, 1
+; ASM64PWR4-DAG:   fmr 13, 1
+; ASM64PWR4-NEXT:  bl .test_fpr_max
+; ASM64PWR4-NEXT:  nop
+; ASM64PWR4-NEXT:  addi 1, 1, 160
 
 define double @test_fpr_max(double %d1, double %d2, double %d3, double %d4, double %d5, double %d6, double %d7, double %d8, double %d9, double %d10, double %d11, double %d12, double %d13) {
 entry:
   %add = fadd double %d1, %d2
   %add1 = fadd double %add, %d3
   %add2 = fadd double %add1, %d4
   %add3 = fadd double %add2, %d5
   %add4 = fadd double %add3, %d6

 ; ASM64PWR4-NEXT: ld [[REG:[0-9]+]], LC1(2)
 ; ASM64PWR4-NEXT: li 5, 42
 ; ASM64PWR4-NEXT: lfs 1, 0([[REG]])
 ; ASM64PWR4-NEXT: ld [[REG:[0-9]+]], LC2(2)
 ; ASM64PWR4-NEXT: stfd 1, 112(1)
 ; ASM64PWR4-NEXT: lfd 2, 0([[REG]])
 ; ASM64PWR4-NEXT: li 3, 42
 ; ASM64PWR4-NEXT: stfd 2, 120(1)
-; ASM64PWR4-DAG: ld 4, 112(1)
-; ASM64PWR4-DAG: ld 6, 120(1)
+; ASM64PWR4-DAG:  ld 4, 112(1)
+; ASM64PWR4-DAG:  ld 6, 120(1)
 ; ASM64PWR4-NEXT: bl .test_vararg
 ; ASM64PWR4-NEXT: nop
 
 define void @call_test_vararg4() {
 entry:
   %0 = load float, float* @f1, align 4
   call void (i32, ...) @test_vararg(i32 42, float %0)
   ret void

 ; ASM64PWR4:      stdu 1, -128(1)
 ; ASM64PWR4-NEXT: ld [[REG:[0-9]+]], LC1(2)
 ; ASM64PWR4-NEXT: lfs 1, 0([[REG]])
 ; ASM64PWR4-NEXT: li 3, 42
 ; ASM64PWR4-NEXT: stfs 1, 124(1)
 ; ASM64PWR4-NEXT: lwz 4, 124(1)
 ; ASM64PWR4-NEXT: bl .test_vararg
 ; ASM64PWR4-NEXT: nop
+
+@c = common global i8 0, align 1
+@si = common global i16 0, align 2
+@i = common global i32 0, align 4
+@lli = common global i64 0, align 8
+@f = common global float 0.000000e+00, align 4
+@d = common global double 0.000000e+00, align 8
+
+; Basic saving of integral type arguments to the parameter save area.
+define void @call_test_stackarg_int() {
+entry:
+  %0 = load i8, i8* @c, align 1
+  %1 = load i16, i16* @si, align 2
+  %2 = load i32, i32* @i, align 4
+  %3 = load i64, i64* @lli, align 8
+  %4 = load i32, i32* @i, align 4
+  call void @test_stackarg_int(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i8 zeroext %0, i16 signext %1, i32 %2, i64 %3, i32 %4)
+  ret void
+}
+
+declare void @test_stackarg_int(i32, i32, i32, i32, i32, i32, i32, i32, i8 zeroext, i16 signext, i32, i64, i32)
+
+; CHECK-LABEL:     name: call_test_stackarg_int{{.*}}
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; 32BIT-DAG:  renamable $r[[REGCADDR:[0-9]+]] = LWZtoc @c, $r2 :: (load 4 from got)
+; 32BIT-DAG:  renamable $r[[REGC:[0-9]+]] = LBZ 0, killed renamable $r[[REGCADDR]] :: (dereferenceable load 1 from @c)
+; 32BIT-DAG:  renamable $r[[REGSIADDR:[0-9]+]] = LWZtoc @si, $r2 :: (load 4 from got)
+; 32BIT-DAG:  renamable $r[[REGSI:[0-9]+]] = LHA 0, killed renamable $r[[REGSIADDR]] :: (dereferenceable load 2 from @si)
+; 32BIT-DAG:  renamable $r[[REGIADDR:[0-9]+]] = LWZtoc @i, $r2 :: (load 4 from got)
+; 32BIT-DAG:  renamable $r[[REGI:[0-9]+]] = LWZ 0, killed renamable $r[[REGIADDR]] :: (dereferenceable load 4 from @i)
+; 32BIT-DAG:  renamable $r[[REGLLIADDR:[0-9]+]] = LWZtoc @lli, $r2 :: (load 4 from got)
+; 32BIT-DAG:  renamable $r[[REGLLI1:[0-9]+]] = LWZ 0, renamable $r[[REGLLIADDR]] :: (dereferenceable load 4 from @lli, align 8)
+; 32BIT-DAG:  renamable $r[[REGLLI2:[0-9]+]] = LWZ 4, killed renamable $r[[REGLLIADDR]] :: (dereferenceable load 4 from @lli + 4)
+; 32BIT-NEXT: ADJCALLSTACKDOWN 80, 0, implicit-def dead $r1, implicit $r1
+; 32BIT-DAG:  STW killed renamable $r[[REGC]], 56, $r1 :: (store 4)
+; 32BIT-DAG:  STW killed renamable $r[[REGSI]], 60, $r1 :: (store 4)
+; 32BIT-DAG:  STW killed renamable $r[[REGI]], 64, $r1 :: (store 4)
+; 32BIT-DAG:  STW killed renamable $r[[REGLLI1]], 68, $r1 :: (store 4)
+; 32BIT-DAG:  STW killed renamable $r[[REGLLI2]], 72, $r1 :: (store 4)
+; 32BIT-DAG:  STW renamable $r[[REGI]], 76, $r1 :: (store 4)
+; 32BIT-DAG:  $r3 = LI 1
+; 32BIT-DAG:  $r4 = LI 2
+; 32BIT-DAG:  $r5 = LI 3
+; 32BIT-DAG:  $r6 = LI 4
+; 32BIT-DAG:  $r7 = LI 5
+; 32BIT-DAG:  $r8 = LI 6
+; 32BIT-DAG:  $r9 = LI 7
+; 32BIT-DAG:  $r10 = LI 8
+; 32BIT-NEXT: BL_NOP <mcsymbol .test_stackarg_int>, csr_aix32, implicit-def dead $lr, implicit $rm, implicit $r3, implicit $r4, implicit $r5, implicit $r6, implicit $r7, implicit $r8, implicit $r9, implicit $r10, implicit $r2, implicit-def $r1
+; 32BIT-NEXT: ADJCALLSTACKUP 80, 0, implicit-def dead $r1, implicit $r1
+
+; Basic saving of floating point type arguments to the parameter save area.
+; The float and double arguments will pass in both fpr as well as parameter save area.
+define void @call_test_stackarg_float() {
+entry:
+  %0 = load float, float* @f, align 4
+  %1 = load double, double* @d, align 8
+  call void @test_stackarg_float(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, float %0, double %1)
+  ret void
+}
+
+declare void @test_stackarg_float(i32, i32, i32, i32, i32, i32, i32, i32, float, double)
+
+; CHECK-LABEL:     name:            call_test_stackarg_float
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; 32BIT-DAG:   renamable $r[[REGF:[0-9]+]] = LWZtoc @f, $r2 :: (load 4 from got)
+; 32BIT-DAG:   renamable $f1 = LFS 0, killed renamable $r[[REGF]] :: (dereferenceable load 4 from @f)
+; 32BIT-DAG:   renamable $r[[REGD:[0-9]+]] = LWZtoc @d, $r2 :: (load 4 from got)
+; 32BIT-DAG:   renamable $f2 = LFD 0, killed renamable $r[[REGD]] :: (dereferenceable load 8 from @d)
+; 32BIT-NEXT:  ADJCALLSTACKDOWN 68, 0, implicit-def dead $r1, implicit $r1
+; 32BIT-DAG:   STFS renamable $f1, 56, $r1 :: (store 4)
+; 32BIT-DAG:   STFD renamable $f2, 60, $r1 :: (store 8)
+; 32BIT-DAG:   $r3 = LI 1
+; 32BIT-DAG:   $r4 = LI 2
+; 32BIT-DAG:   $r5 = LI 3
+; 32BIT-DAG:   $r6 = LI 4
+; 32BIT-DAG:   $r7 = LI 5
+; 32BIT-DAG:   $r8 = LI 6
+; 32BIT-DAG:   $r9 = LI 7
+; 32BIT-DAG:   $r10 = LI 8
+; 32BIT-NEXT:  BL_NOP <mcsymbol .test_stackarg_float>, csr_aix32, implicit-def dead $lr, implicit $rm, implicit $r3, implicit $r4, implicit killed $r5, implicit killed $r6, implicit killed $r7, implicit killed $r8, implicit killed $r9, implicit killed $r10, implicit $f1, implicit $f2, implicit $r2, implicit-def $r1
+; 32BIT-NEXT:  ADJCALLSTACKUP 68, 0, implicit-def dead $r1, implicit $r1
+
+; CHECKASM-LABEL: .call_test_stackarg_float:
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; ASM32PWR4:      stwu 1, -80(1)
+; ASM32PWR4-DAG:  lwz [[REGF:[0-9]+]], LC8(2)
+; ASM32PWR4-DAG:  lfs 1, 0([[REGF]])
+; ASM32PWR4-DAG:  lwz [[REGD:[0-9]+]], LC9(2)
+; ASM32PWR4-DAG:  lfd 2, 0([[REGD:[0-9]+]])
+; ASM32PWR4-DAG:  stfs 1, 56(1)
+; ASM32PWR4-DAG:  stfd 2, 60(1)
+; ASM32PWR4-DAG:  li 3, 1
+; ASM32PWR4-DAG:  li 4, 2
+; ASM32PWR4-DAG:  li 5, 3
+; ASM32PWR4-DAG:  li 6, 4
+; ASM32PWR4-DAG:  li 7, 5
+; ASM32PWR4-DAG:  li 8, 6
+; ASM32PWR4-DAG:  li 9, 7
+; ASM32PWR4-DAG:  li 10, 8
+; ASM32PWR4-NEXT: bl .test_stackarg_float
+; ASM32PWR4-NEXT: nop
+; ASM32PWR4-NEXT: addi 1, 1, 80
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; 64BIT-DAG:   renamable $x[[REGF:[0-9]+]] = LDtoc @f, $x2 :: (load 8 from got)
+; 64BIT-DAG:   renamable $f1 = LFS 0, killed renamable $x[[REGF]] :: (dereferenceable load 4 from @f)
+; 64BIT-DAG:   renamable $x[[REGD:[0-9]+]] = LDtoc @d, $x2 :: (load 8 from got)
+; 64BIT-DAG:   renamable $f2 = LFD 0, killed renamable $x[[REGD]] :: (dereferenceable load 8 from @d)
+; 64BIT-NEXT:  ADJCALLSTACKDOWN 128, 0, implicit-def dead $r1, implicit $r1
+; 64BIT-DAG:   STFS renamable $f1, 112, $x1 :: (store 4)
+; 64BIT-DAG:   STFD renamable $f2, 120, $x1 :: (store 8)
+; 64BIT-DAG:   $x3 = LI8 1
+; 64BIT-DAG:   $x4 = LI8 2
+; 64BIT-DAG:   $x5 = LI8 3
+; 64BIT-DAG:   $x6 = LI8 4
+; 64BIT-DAG:   $x7 = LI8 5
+; 64BIT-DAG:   $x8 = LI8 6
+; 64BIT-DAG:   $x9 = LI8 7
+; 64BIT-DAG:   $x10 = LI8 8
+; 64BIT-NEXT:  BL8_NOP <mcsymbol .test_stackarg_float>, csr_aix64, implicit-def dead $lr8, implicit $rm, implicit $x3, implicit $x4, implicit killed $x5, implicit killed $x6, implicit killed $x7, implicit killed $x8, implicit killed $x9, implicit killed $x10, implicit $f1, implicit $f2, implicit $x2, implicit-def $r1
+; 64BIT-NEXT:  ADJCALLSTACKUP 128, 0, implicit-def dead $r1, implicit $r1
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; ASM64PWR4:      stdu 1, -128(1)
+; ASM64PWR4-DAG:  ld [[REGF:[0-9]+]], LC7(2)
+; ASM64PWR4-DAG:  lfs 1, 0([[REGF]])
+; ASM64PWR4-DAG:  ld [[REGD:[0-9]+]], LC8(2)
+; ASM64PWR4-DAG:  lfd 2, 0([[REGD]])
+; ASM64PWR4-DAG:  stfs 1, 112(1)
+; ASM64PWR4-DAG:  stfd 2, 120(1)
+; ASM64PWR4-DAG:  li 3, 1
+; ASM64PWR4-DAG:  li 4, 2
+; ASM64PWR4-DAG:  li 5, 3
+; ASM64PWR4-DAG:  li 6, 4
+; ASM64PWR4-DAG:  li 7, 5
+; ASM64PWR4-DAG:  li 8, 6
+; ASM64PWR4-DAG:  li 9, 7
+; ASM64PWR4-DAG:  li 10, 8
+; ASM64PWR4-NEXT: bl .test_stackarg_float
+; ASM64PWR4-NEXT: nop
+; ASM64PWR4-NEXT: addi 1, 1, 128
+
+; A double arg will pass on the stack in PPC32 if there is only one available GPR.
+define void @call_test_stackarg_float2() {
+entry:
+  %0 = load double, double* @d, align 8
+  %1 = load float, float* @f, align 4
+  call void (i32, i32, i32, i32, i32, i32, i32, ...) @test_stackarg_float2(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, double %0, float %1)
+  ret void
+}
+
+declare void @test_stackarg_float2(i32, i32, i32, i32, i32, i32, i32, ...)
+
+; CHECK-LABEL:     name: call_test_stackarg_float2{{.*}}
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; In 32-bit the double arg is written to memory because it cannot be fully stored in the last 32-bit GPR.
+; 32BIT-DAG:   renamable $r[[REGD:[0-9]+]] = LWZtoc @d, $r2 :: (load 4 from got)
+; 32BIT-DAG:   renamable $f1 = LFD 0, killed renamable $r[[REGD]] :: (dereferenceable load 8 from @d)
+; 32BIT-DAG:   renamable $r[[REGF:[0-9]+]] = LWZtoc @f, $r2 :: (load 4 from got)
+; 32BIT-DAG:   renamable $f2 = LFS 0, killed renamable $r[[REGF]] :: (dereferenceable load 4 from @f)
+; 32BIT-DAG:   ADJCALLSTACKDOWN 64, 0, implicit-def dead $r1, implicit $r1
+; 32BIT-DAG:   STFD renamable $f1, 52, $r1 :: (store 8)
+; 32BIT-DAG:   STFS renamable $f2, 60, $r1 :: (store 4)
+; 32BIT-DAG:   $r3 = LI 1
+; 32BIT-DAG:   $r4 = LI 2
+; 32BIT-DAG:   $r5 = LI 3
+; 32BIT-DAG:   $r6 = LI 4
+; 32BIT-DAG:   $r7 = LI 5
+; 32BIT-DAG:   $r8 = LI 6
+; 32BIT-DAG:   $r9 = LI 7
+; 32BIT-DAG:   STFD renamable $f1, 0, %stack.0 :: (store 8 into %stack.0)
+; 32BIT-DAG:   renamable $r10 = LWZ 0, %stack.0 :: (load 4 from %stack.0, align 8)
+; 32BIT-NEXT:  BL_NOP <mcsymbol .test_stackarg_float2>, csr_aix32, implicit-def dead $lr, implicit $rm, implicit $r3, implicit killed $r4, implicit killed $r5, implicit killed $r6, implicit killed $r7, implicit killed $r8, implicit killed $r9, implicit $f1, implicit $r10, implicit $f2, implicit $r2, implicit-def $r1
+; 32BIT-NEXT:  ADJCALLSTACKUP 64, 0, implicit-def dead $r1, implicit $r1
+
+; CHECKASM-LABEL: .call_test_stackarg_float2:
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; ASM32PWR4:       stwu 1, -80(1)
+; ASM32PWR4-DAG:   lwz [[REGD:[0-9]+]], LC9(2)
+; ASM32PWR4-DAG:   lfd 1, 0([[REGD]])
+; ASM32PWR4-DAG:   lwz [[REGF:[0-9]+]], LC8(2)
+; ASM32PWR4-DAG:   lfs 2, 0([[REGF]])
+; ASM32PWR4-DAG:   stfd 1, 52(1)
+; ASM32PWR4-DAG:   stfs 2, 60(1)
+; ASM32PWR4-DAG:   li 3, 1
+; ASM32PWR4-DAG:   li 4, 2
+; ASM32PWR4-DAG:   li 5, 3
+; ASM32PWR4-DAG:   li 6, 4
+; ASM32PWR4-DAG:   li 7, 5
+; ASM32PWR4-DAG:   li 8, 6
+; ASM32PWR4-DAG:   li 9, 7
+; ASM32PWR4-DAG:   stfd 1, 72(1)
+; ASM32PWR4-DAG:   lwz 10, 72(1)
+; ASM32PWR4-NEXT:  bl .test_stackarg_float2
+; ASM32PWR4-NEXT:  nop
+; ASM32PWR4-NEXT:  addi 1, 1, 80
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; In 64-bit the double arg is not written to memory because it is fully stored in the last 64-bit GPR.
+; 64BIT-DAG:   renamable $x[[REGD:[0-9]+]] = LDtoc @d, $x2 :: (load 8 from got)
+; 64BIT-DAG:   renamable $f1 = LFD 0, killed renamable $x[[REGD]] :: (dereferenceable load 8 from @d)
+; 64BIT-DAG:   renamable $x[[REGF:[0-9]+]] = LDtoc @f, $x2 :: (load 8 from got)
+; 64BIT-DAG:   renamable $f2 = LFS 0, killed renamable $x[[REGF]] :: (dereferenceable load 4 from @f)
+; 64BIT-DAG:   ADJCALLSTACKDOWN 120, 0, implicit-def dead $r1, implicit $r1
+; 64BIT-DAG:   STFS renamable $f2, 112, $x1 :: (store 4)
+; 64BIT-DAG:   $x3 = LI8 1
+; 64BIT-DAG:   $x4 = LI8 2
+; 64BIT-DAG:   $x5 = LI8 3
+; 64BIT-DAG:   $x6 = LI8 4
+; 64BIT-DAG:   $x7 = LI8 5
+; 64BIT-DAG:   $x8 = LI8 6
+; 64BIT-DAG:   $x9 = LI8 7
+; 64BIT-DAG:   STFD renamable $f1, 0, %stack.0 :: (store 8 into %stack.0)
+; 64BIT-DAG:   renamable $x10 = LD 0, %stack.0 :: (load 8 from %stack.0)
+; 64BIT-NEXT:  BL8_NOP <mcsymbol .test_stackarg_float2>, csr_aix64, implicit-def dead $lr8, implicit $rm, implicit $x3, implicit killed $x4, implicit killed $x5, implicit killed $x6, implicit killed $x7, implicit killed $x8, implicit killed $x9, implicit $f1, implicit $x10, implicit $f2, implicit $x2, implicit-def $r1
+
+; 64BIT-NEXT: ADJCALLSTACKUP 120, 0, implicit-def dead $r1, implicit $r1
+
+; The DAG block permits some invalid inputs for the benefit of allowing more valid orderings.
+; ASM64PWR4:       stdu 1, -128(1)
+; ASM64PWR4-DAG:   ld [[REGD:[0-9]+]], LC8(2)
+; ASM64PWR4-DAG:   lfd 1, 0([[REGD]])
+; ASM64PWR4-DAG:   ld [[REGF:[0-9]+]], LC7(2)
+; ASM64PWR4-DAG:   lfs 2, 0([[REGF]])
+; ASM64PWR4-DAG:   stfs 2, 112(1)
+; ASM64PWR4-DAG:   li 3, 1
+; ASM64PWR4-DAG:   li 4, 2
+; ASM64PWR4-DAG:   li 5, 3
+; ASM64PWR4-DAG:   li 6, 4
+; ASM64PWR4-DAG:   li 7, 5
+; ASM64PWR4-DAG:   li 8, 6
+; ASM64PWR4-DAG:   li 9, 7
+; ASM64PWR4-NEXT:  bl .test_stackarg_float2
+; ASM64PWR4-NEXT:  nop
+; ASM64PWR4-NEXT:  addi 1, 1, 128

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

-; RUN: not llc -mtriple powerpc-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
-; RUN: not llc -mtriple powerpc64-ibm-aix-xcoff < %s 2>&1 | FileCheck %s
-
-define void @bar() {
-entry:
-  call void @foo(i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9)
-  ret void
-}
-
-declare void @foo(i32, i32, i32, i32, i32, i32, i32, i32, i32)
-
-; CHECK: LLVM ERROR: Handling of placing parameters on the stack is unimplemented!