Index: lib/Target/PowerPC/CMakeLists.txt
===================================================================
--- lib/Target/PowerPC/CMakeLists.txt
+++ lib/Target/PowerPC/CMakeLists.txt
@@ -16,6 +16,7 @@
   PPCBoolRetToInt.cpp
   PPCAsmPrinter.cpp
   PPCBranchSelector.cpp
+  PPCCCState.cpp
   PPCCTRLoops.cpp
   PPCHazardRecognizers.cpp
   PPCInstrInfo.cpp
Index: lib/Target/PowerPC/PPCCCState.h
===================================================================
--- lib/Target/PowerPC/PPCCCState.h
+++ lib/Target/PowerPC/PPCCCState.h
@@ -0,0 +1,42 @@
+//===---- PPCCCState.h - CCState with PowerPC specific extensions -----------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef PPCCCSTATE_H
+#define PPCCCSTATE_H
+
+#include "PPCISelLowering.h"
+#include "llvm/ADT/SmallVector.h"
+#include "llvm/CodeGen/CallingConvLower.h"
+
+namespace llvm {
+
+class PPCCCState : public CCState {
+public:
+
+  void
+  PreAnalyzeCallOperands(const SmallVectorImpl<ISD::OutputArg> &Outs);
+  void
+  PreAnalyzeFormalArguments(const SmallVectorImpl<ISD::InputArg> &Ins);
+
+private:
+
+  // Records whether the value has been lowered from an ppcf128.
+  SmallVector<bool, 4> OriginalArgWasPPCF128;
+
+public:
+  PPCCCState(CallingConv::ID CC, bool isVarArg, MachineFunction &MF,
+             SmallVectorImpl<CCValAssign> &locs, LLVMContext &C)
+        : CCState(CC, isVarArg, MF, locs, C) {}
+
+  bool WasOriginalArgPPCF128(unsigned ValNo) { return OriginalArgWasPPCF128[ValNo]; }
+  void clearWasPPCF128() { OriginalArgWasPPCF128.clear(); }
+};
+}
+
+#endif
Index: lib/Target/PowerPC/PPCCCState.cpp
===================================================================
--- lib/Target/PowerPC/PPCCCState.cpp
+++ lib/Target/PowerPC/PPCCCState.cpp
@@ -0,0 +1,36 @@
+//===---- PPCCCState.cpp - CCState with PowerPC specific extensions ---------===//
+//
+//                     The LLVM Compiler Infrastructure
+//
+// This file is distributed under the University of Illinois Open Source
+// License. See LICENSE.TXT for details.
+//
+//===----------------------------------------------------------------------===//
+
+#include "PPCCCState.h"
+#include "PPCSubtarget.h"
+#include "llvm/IR/Module.h"
+using namespace llvm;
+
+// Identify lowered values that originated from ppcf128 arguments and record
+// this.
+void PPCCCState::PreAnalyzeCallOperands(
+    const SmallVectorImpl<ISD::OutputArg> &Outs) {
+  for (const auto &I : Outs) {
+    if (I.ArgVT == llvm::MVT::ppcf128)
+      OriginalArgWasPPCF128.push_back(true);
+    else
+      OriginalArgWasPPCF128.push_back(false);
+  }
+}
+
+void PPCCCState::PreAnalyzeFormalArguments(
+    const SmallVectorImpl<ISD::InputArg> &Ins) {
+  for (const auto &I : Ins) {
+    if (I.ArgVT == llvm::MVT::ppcf128) {
+      OriginalArgWasPPCF128.push_back(true);
+    } else {
+      OriginalArgWasPPCF128.push_back(false);
+    }
+  }
+}
\ No newline at end of file
Index: lib/Target/PowerPC/PPCCallingConv.td
===================================================================
--- lib/Target/PowerPC/PPCCallingConv.td
+++ lib/Target/PowerPC/PPCCallingConv.td
@@ -23,6 +23,9 @@
                        "(State.getMachineFunction().getSubtarget()).",
                      F),
           A>;
+class CCIfOrigArgWasNotPPCF128<CCAction A>
+    : CCIf<"!static_cast<PPCCCState *>(&State)->WasOriginalArgPPCF128(ValNo)",
+           A>;
 
 //===----------------------------------------------------------------------===//
 // Return Value Calling Convention
@@ -131,7 +134,14 @@
 
   // The ABI requires i64 to be passed in two adjacent registers with the first
   // register having an odd register number.
-  CCIfType<[i32], CCIfSplit<CCCustom<"CC_PPC32_SVR4_Custom_AlignArgRegs">>>,
+  CCIfType<[i32],
+  CCIfSplit<CCIfSubtarget<"useSoftFloat()", 
+            CCIfOrigArgWasNotPPCF128<
+            CCCustom<"CC_PPC32_SVR4_Custom_AlignArgRegs">>>>>,
+  
+  CCIfType<[i32],
+  CCIfSplit<CCIfNotSubtarget<"useSoftFloat()", 
+                            CCCustom<"CC_PPC32_SVR4_Custom_AlignArgRegs">>>>,
 
   // The 'nest' parameter, if any, is passed in R11.
   CCIfNest<CCAssignToReg<[R11]>>,
Index: lib/Target/PowerPC/PPCFastISel.cpp
===================================================================
--- lib/Target/PowerPC/PPCFastISel.cpp
+++ lib/Target/PowerPC/PPCFastISel.cpp
@@ -16,6 +16,7 @@
 #include "PPC.h"
 #include "MCTargetDesc/PPCPredicates.h"
 #include "PPCCallingConv.h"
+#include "PPCCCState.h"
 #include "PPCISelLowering.h"
 #include "PPCMachineFunctionInfo.h"
 #include "PPCSubtarget.h"
Index: lib/Target/PowerPC/PPCISelLowering.cpp
===================================================================
--- lib/Target/PowerPC/PPCISelLowering.cpp
+++ lib/Target/PowerPC/PPCISelLowering.cpp
@@ -14,6 +14,7 @@
 #include "PPCISelLowering.h"
 #include "MCTargetDesc/PPCPredicates.h"
 #include "PPCCallingConv.h"
+#include "PPCCCState.h"
 #include "PPCMachineFunctionInfo.h"
 #include "PPCPerfectShuffle.h"
 #include "PPCTargetMachine.h"
@@ -2847,14 +2848,17 @@
 
   // Assign locations to all of the incoming arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
+  PPCCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                  *DAG.getContext());
 
   // Reserve space for the linkage area on the stack.
   unsigned LinkageSize = Subtarget.getFrameLowering()->getLinkageSize();
   CCInfo.AllocateStack(LinkageSize, PtrByteSize);
+  if (Subtarget.useSoftFloat())
+    CCInfo.PreAnalyzeFormalArguments(Ins);
 
   CCInfo.AnalyzeFormalArguments(Ins, CC_PPC32_SVR4);
+  CCInfo.clearWasPPCF128();
 
   for (unsigned i = 0, e = ArgLocs.size(); i != e; ++i) {
     CCValAssign &VA = ArgLocs[i];
@@ -4741,12 +4745,14 @@
 
   // Assign locations to all of the outgoing arguments.
   SmallVector<CCValAssign, 16> ArgLocs;
-  CCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
+  PPCCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                  *DAG.getContext());
 
   // Reserve space for the linkage area on the stack.
   CCInfo.AllocateStack(Subtarget.getFrameLowering()->getLinkageSize(),
                        PtrByteSize);
+  if (Subtarget.useSoftFloat())
+    CCInfo.PreAnalyzeCallOperands(Outs);
 
   if (isVarArg) {
     // Handle fixed and variable vector arguments differently.
@@ -4779,7 +4785,8 @@
     // All arguments are treated the same.
     CCInfo.AnalyzeCallOperands(Outs, CC_PPC32_SVR4);
   }
-
+  CCInfo.clearWasPPCF128();
+  
   // Assign locations to all of the outgoing aggregate by value arguments.
   SmallVector<CCValAssign, 16> ByValArgLocs;
   CCState CCByValInfo(CallConv, isVarArg, DAG.getMachineFunction(),
Index: test/CodeGen/PowerPC/ppc32-align-long-double-sf.ll
===================================================================
--- test/CodeGen/PowerPC/ppc32-align-long-double-sf.ll
+++ test/CodeGen/PowerPC/ppc32-align-long-double-sf.ll
@@ -0,0 +1,21 @@
+; RUN: llc -O2 -mtriple=powerpc-unknown-linux-gnu < %s | FileCheck %s
+
+@x = global ppc_fp128 0xM405EDA5E353F7CEE0000000000000000, align 16
+@.str = private unnamed_addr constant [5 x i8] c"%Lf\0A\00", align 1
+
+
+define void @foo() #0 {
+entry:
+  %0 = load ppc_fp128, ppc_fp128* @x, align 16
+  %call = tail call i32 (i8*, ...) @printf(i8* getelementptr inbounds ([5 x i8], [5 x i8]* @.str, i32 0, i32 0), ppc_fp128 %0)
+  ret void
+}
+; Do not skip register r4 because of register alignment in soft float mode. Instead skipping 
+; put in r4 part of first argument for printf function (long double).
+; CHECK: lwzu 4, x@l({{[0-9]+}})
+
+declare i32 @printf(i8* nocapture readonly, ...) #0
+
+attributes #0 = { "use-soft-float"="true" }
+
+                        
\ No newline at end of file