1 files changed, 45 insertions, 23 deletions

diff --git a/lib/Target/PowerPC/PPCVSXFMAMutate.cpp b/lib/Target/PowerPC/PPCVSXFMAMutate.cpp
index 6b19a2f7118b2..7c22cb22bfa5f 100644
--- a/lib/Target/PowerPC/PPCVSXFMAMutate.cpp
+++ b/lib/Target/PowerPC/PPCVSXFMAMutate.cpp
@@ -38,8 +38,14 @@
 
 using namespace llvm;
 
-static cl::opt<bool> DisableVSXFMAMutate("disable-ppc-vsx-fma-mutation",
-cl::desc("Disable VSX FMA instruction mutation"), cl::Hidden);
+// Temporarily disable FMA mutation by default, since it doesn't handle
+// cross-basic-block intervals well.
+// See: http://lists.llvm.org/pipermail/llvm-dev/2016-February/095669.html
+//      http://reviews.llvm.org/D17087
+static cl::opt<bool> DisableVSXFMAMutate(
+    "disable-ppc-vsx-fma-mutation",
+    cl::desc("Disable VSX FMA instruction mutation"), cl::init(true),
+    cl::Hidden);
 
 #define DEBUG_TYPE "ppc-vsx-fma-mutate"
 
@@ -99,7 +105,7 @@ protected:
         //                         %RM<imp-use>; VSLRC:%vreg16,%vreg18,%vreg9
         // and we remove: %vreg5<def> = COPY %vreg9; VSLRC:%vreg5,%vreg9
 
-        SlotIndex FMAIdx = LIS->getInstructionIndex(MI);
+        SlotIndex FMAIdx = LIS->getInstructionIndex(*MI);
 
         VNInfo *AddendValNo =
           LIS->getInterval(MI->getOperand(1).getReg()).Query(FMAIdx).valueIn();
@@ -168,21 +174,32 @@ protected:
         if (OtherUsers || KillsAddendSrc)
           continue;
 
-        // Find one of the product operands that is killed by this instruction.
 
+        // The transformation doesn't work well with things like:
+        //    %vreg5 = A-form-op %vreg5, %vreg11, %vreg5;
+        // unless vreg11 is also a kill, so skip when it is not,
+        // and check operand 3 to see it is also a kill to handle the case:
+        //   %vreg5 = A-form-op %vreg5, %vreg5, %vreg11;
+        // where vreg5 and vreg11 are both kills. This case would be skipped
+        // otherwise.
+        unsigned OldFMAReg = MI->getOperand(0).getReg();
+
+        // Find one of the product operands that is killed by this instruction.
         unsigned KilledProdOp = 0, OtherProdOp = 0;
-        if (LIS->getInterval(MI->getOperand(2).getReg())
-                     .Query(FMAIdx).isKill()) {
+        unsigned Reg2 = MI->getOperand(2).getReg();
+        unsigned Reg3 = MI->getOperand(3).getReg();
+        if (LIS->getInterval(Reg2).Query(FMAIdx).isKill()
+            && Reg2 != OldFMAReg) {
           KilledProdOp = 2;
           OtherProdOp  = 3;
-        } else if (LIS->getInterval(MI->getOperand(3).getReg())
-                     .Query(FMAIdx).isKill()) {
+        } else if (LIS->getInterval(Reg3).Query(FMAIdx).isKill()
+            && Reg3 != OldFMAReg) {
           KilledProdOp = 3;
           OtherProdOp  = 2;
         }
 
-        // If there are no killed product operands, then this transformation is
-        // likely not profitable.
+        // If there are no usable killed product operands, then this
+        // transformation is likely not profitable.
         if (!KilledProdOp)
           continue;
 
@@ -212,14 +229,6 @@ protected:
         bool KilledProdRegUndef = MI->getOperand(KilledProdOp).isUndef();
         bool OtherProdRegUndef  = MI->getOperand(OtherProdOp).isUndef();
 
-        unsigned OldFMAReg = MI->getOperand(0).getReg();
-
-        // The transformation doesn't work well with things like:
-        //    %vreg5 = A-form-op %vreg5, %vreg11, %vreg5;
-        // so leave such things alone.
-        if (OldFMAReg == KilledProdReg)
-          continue;
-
         // If there isn't a class that fits, we can't perform the transform.
         // This is needed for correctness with a mixture of VSX and Altivec
         // instructions to make sure that a low VSX register is not assigned to
@@ -236,23 +245,33 @@ protected:
         MI->getOperand(0).setReg(KilledProdReg);
         MI->getOperand(1).setReg(KilledProdReg);
         MI->getOperand(3).setReg(AddendSrcReg);
-        MI->getOperand(2).setReg(OtherProdReg);
 
         MI->getOperand(0).setSubReg(KilledProdSubReg);
         MI->getOperand(1).setSubReg(KilledProdSubReg);
         MI->getOperand(3).setSubReg(AddSubReg);
-        MI->getOperand(2).setSubReg(OtherProdSubReg);
 
         MI->getOperand(1).setIsKill(KilledProdRegKill);
         MI->getOperand(3).setIsKill(AddRegKill);
-        MI->getOperand(2).setIsKill(OtherProdRegKill);
 
         MI->getOperand(1).setIsUndef(KilledProdRegUndef);
         MI->getOperand(3).setIsUndef(AddRegUndef);
-        MI->getOperand(2).setIsUndef(OtherProdRegUndef);
 
         MI->setDesc(TII->get(AltOpc));
 
+        // If the addend is also a multiplicand, replace it with the addend
+        // source in both places.
+        if (OtherProdReg == AddendMI->getOperand(0).getReg()) {
+          MI->getOperand(2).setReg(AddendSrcReg);
+          MI->getOperand(2).setSubReg(AddSubReg);
+          MI->getOperand(2).setIsKill(AddRegKill);
+          MI->getOperand(2).setIsUndef(AddRegUndef);
+        } else {
+          MI->getOperand(2).setReg(OtherProdReg);
+          MI->getOperand(2).setSubReg(OtherProdSubReg);
+          MI->getOperand(2).setIsKill(OtherProdRegKill);
+          MI->getOperand(2).setIsUndef(OtherProdRegUndef);
+        }
+
         DEBUG(dbgs() << " -> " << *MI);
 
         // The killed product operand was killed here, so we can reuse it now
@@ -312,7 +331,7 @@ protected:
         // Remove the (now unused) copy.
 
         DEBUG(dbgs() << "  removing: " << *AddendMI << '\n');
-        LIS->RemoveMachineInstrFromMaps(AddendMI);
+        LIS->RemoveMachineInstrFromMaps(*AddendMI);
         AddendMI->eraseFromParent();
 
         Changed = true;
@@ -323,6 +342,9 @@ protected:
 
 public:
     bool runOnMachineFunction(MachineFunction &MF) override {
+      if (skipFunction(*MF.getFunction()))
+        return false;
+
       // If we don't have VSX then go ahead and return without doing
       // anything.
       const PPCSubtarget &STI = MF.getSubtarget<PPCSubtarget>();