3 files changed, 40 insertions, 19 deletions

diff --git a/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCISelLowering.cpp b/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
index 459525ed4ee9..b6db0abcc86d 100644
--- a/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -7296,9 +7296,17 @@ SDValue PPCTargetLowering::LowerFormalArguments_AIX(
       if (!ArgVT.isVector() && !ValVT.isVector() && ArgVT.isInteger() &&
           ValVT.isInteger() &&
           ArgVT.getScalarSizeInBits() < ValVT.getScalarSizeInBits()) {
-        SDValue ArgValueTrunc = DAG.getNode(
-            ISD::TRUNCATE, dl, ArgVT.getSimpleVT() == MVT::i1 ? MVT::i8 : ArgVT,
-            ArgValue);
+        // It is possible to have either real integer values
+        // or integers that were not originally integers.
+        // In the latter case, these could have came from structs,
+        // and these integers would not have an extend on the parameter.
+        // Since these types of integers do not have an extend specified
+        // in the first place, the type of extend that we do should not matter.
+        EVT TruncatedArgVT = ArgVT.isSimple() && ArgVT.getSimpleVT() == MVT::i1
+                                 ? MVT::i8
+                                 : ArgVT;
+        SDValue ArgValueTrunc =
+            DAG.getNode(ISD::TRUNCATE, dl, TruncatedArgVT, ArgValue);
         SDValue ArgValueExt =
             ArgSignExt ? DAG.getSExtOrTrunc(ArgValueTrunc, dl, ValVT)
                        : DAG.getZExtOrTrunc(ArgValueTrunc, dl, ValVT);
@@ -9586,12 +9594,14 @@ static bool isValidSplatLoad(const PPCSubtarget &Subtarget, const SDValue &Op,
   return false;
 }
 
-bool isValidMtVsrBmi(APInt &BitMask, BuildVectorSDNode &BVN) {
+bool isValidMtVsrBmi(APInt &BitMask, BuildVectorSDNode &BVN,
+                     bool IsLittleEndian) {
   assert(BVN.getNumOperands() > 0 && "Unexpected 0-size build vector");
 
   BitMask.clearAllBits();
   EVT VT = BVN.getValueType(0);
-  APInt ConstValue(VT.getSizeInBits(), 0);
+  unsigned VTSize = VT.getSizeInBits();
+  APInt ConstValue(VTSize, 0);
 
   unsigned EltWidth = VT.getScalarSizeInBits();
 
@@ -9601,8 +9611,10 @@ bool isValidMtVsrBmi(APInt &BitMask, BuildVectorSDNode &BVN) {
 
     if (!CN)
       return false;
-
-    ConstValue.insertBits(CN->getAPIntValue().zextOrTrunc(EltWidth), BitPos);
+    // The elements in a vector register are ordered in reverse byte order
+    // between little-endian and big-endian modes.
+    ConstValue.insertBits(CN->getAPIntValue().zextOrTrunc(EltWidth),
+                          IsLittleEndian ? BitPos : VTSize - EltWidth - BitPos);
     BitPos += EltWidth;
   }
 
@@ -9633,7 +9645,8 @@ SDValue PPCTargetLowering::LowerBUILD_VECTOR(SDValue Op,
     // we do not convert it to MTVSRBMI.
     // The xxleqv instruction sets a vector with all ones.
     // The xxlxor instruction sets a vector with all zeros.
-    if (isValidMtVsrBmi(BitMask, *BVN) && BitMask != 0 && BitMask != 0xffff) {
+    if (isValidMtVsrBmi(BitMask, *BVN, Subtarget.isLittleEndian()) &&
+        BitMask != 0 && BitMask != 0xffff) {
       SDValue SDConstant = DAG.getTargetConstant(BitMask, dl, MVT::i32);
       MachineSDNode *MSDNode =
           DAG.getMachineNode(PPC::MTVSRBMI, dl, MVT::v16i8, SDConstant);
@@ -15314,6 +15327,12 @@ SDValue PPCTargetLowering::DAGCombineExtBoolTrunc(SDNode *N,
     }
   }
 
+  // Convert PromOps to handles before doing any RAUW operations, as these
+  // may CSE with existing nodes, deleting the originals.
+  std::list<HandleSDNode> PromOpHandles;
+  for (auto &PromOp : PromOps)
+    PromOpHandles.emplace_back(PromOp);
+
   // Replace all inputs, either with the truncation operand, or a
   // truncation or extension to the final output type.
   for (unsigned i = 0, ie = Inputs.size(); i != ie; ++i) {
@@ -15337,10 +15356,6 @@ SDValue PPCTargetLowering::DAGCombineExtBoolTrunc(SDNode *N,
         DAG.getAnyExtOrTrunc(InSrc, dl, N->getValueType(0)));
   }
 
-  std::list<HandleSDNode> PromOpHandles;
-  for (auto &PromOp : PromOps)
-    PromOpHandles.emplace_back(PromOp);
-
   // Replace all operations (these are all the same, but have a different
   // (promoted) return type). DAG.getNode will validate that the types of
   // a binary operator match, so go through the list in reverse so that
diff --git a/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCInstrInfo.td b/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCInstrInfo.td
index 99ef89a7fdc0..dbbf06f47153 100644
--- a/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCInstrInfo.td
+++ b/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCInstrInfo.td
@@ -3252,7 +3252,8 @@ def PPC32GOT: PPCEmitTimePseudo<(outs gprc:$rD), (ins), "#PPC32GOT",
 
 // Get the _GLOBAL_OFFSET_TABLE_ in PIC mode.
 // This uses two output registers, the first as the real output, the second as a
-// temporary register, used internally in code generation.
+// temporary register, used internally in code generation. A "bl" also clobbers LR.
+let Defs = [LR] in
 def PPC32PICGOT: PPCEmitTimePseudo<(outs gprc:$rD, gprc:$rT), (ins), "#PPC32PICGOT",
                 []>, NoEncode<"$rT">;
 
diff --git a/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp b/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp
index 76dca4794e05..f1230407b164 100644
--- a/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp
+++ b/contrib/llvm-project/llvm/lib/Target/PowerPC/PPCRegisterInfo.cpp
@@ -1102,13 +1102,20 @@ void PPCRegisterInfo::lowerCRBitSpilling(MachineBasicBlock::iterator II,
     SpillsKnownBit = true;
     break;
   default:
+    // When spilling a CR bit, the super register may not be explicitly defined
+    // (i.e. it can be defined by a CR-logical that only defines the subreg) so
+    // we state that the CR field is undef. Also, in order to preserve the kill
+    // flag on the CR bit, we add it as an implicit use.
+
     // On Power10, we can use SETNBC to spill all CR bits. SETNBC will set all
     // bits (specifically, it produces a -1 if the CR bit is set). Ultimately,
     // the bit that is of importance to us is bit 32 (bit 0 of a 32-bit
     // register), and SETNBC will set this.
     if (Subtarget.isISA3_1()) {
       BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::SETNBC8 : PPC::SETNBC), Reg)
-          .addReg(SrcReg, RegState::Undef);
+          .addReg(SrcReg, RegState::Undef)
+          .addReg(SrcReg, RegState::Implicit |
+                              getKillRegState(MI.getOperand(0).isKill()));
       break;
     }
 
@@ -1122,16 +1129,14 @@ void PPCRegisterInfo::lowerCRBitSpilling(MachineBasicBlock::iterator II,
           SrcReg == PPC::CR4LT || SrcReg == PPC::CR5LT ||
           SrcReg == PPC::CR6LT || SrcReg == PPC::CR7LT) {
         BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::SETB8 : PPC::SETB), Reg)
-          .addReg(getCRFromCRBit(SrcReg), RegState::Undef);
+            .addReg(getCRFromCRBit(SrcReg), RegState::Undef)
+            .addReg(SrcReg, RegState::Implicit |
+                                getKillRegState(MI.getOperand(0).isKill()));
         break;
       }
     }
 
     // We need to move the CR field that contains the CR bit we are spilling.
-    // The super register may not be explicitly defined (i.e. it can be defined
-    // by a CR-logical that only defines the subreg) so we state that the CR
-    // field is undef. Also, in order to preserve the kill flag on the CR bit,
-    // we add it as an implicit use.
     BuildMI(MBB, II, dl, TII.get(LP64 ? PPC::MFOCRF8 : PPC::MFOCRF), Reg)
       .addReg(getCRFromCRBit(SrcReg), RegState::Undef)
       .addReg(SrcReg,