am 1c4ad5ef: Merge branch \'upstream\' into merge-2012_09_10

* commit '1c4ad5ef4fab105f0c8af7edd026e00502fb6279': (446 commits)
  Revert r163556. Missed updates to tablegen files.
  Update function names to conform to guidelines.  No functional change intended.
  test/CodeGen/X86/ms-inline-asm.ll: Relax for non-darwin x86 targets. '##InlineAsm' could not be seen in other hosts.
  [ms-inline asm] Properly emit the asm directives when the AsmPrinterVariant and InlineAsmVariant don't match.
  Update test case for Release builds.
  Remove redundant semicolons which are null statements.
  Disable stack coloring because it makes dragonegg fail bootstrapping.
  [ms-inline asm] Pass the correct AsmVariant to the PrintAsmOperand() function and update the printOperand() function accordingly.
  [ms-inline asm] Add support for .att_syntax directive.
  Enable stack coloring.
  Don't attempt to use flags from predicated instructions.
  [Object] Extract Elf_Ehdr. Patch by Hemant Kulkarni!
  Stack Coloring: Handle the case where END markers come before BEGIN markers properly.
  Enhance PR11334 fix to support extload from v2f32/v4f32
  Add "blocked" heuristic to the Hexagon MI scheduler.
  Fold multiply by 0 or 1 when in UnsafeFPMath mode in SelectionDAG::getNode().
  whitespace
  Add boolean simplification support from CMOV
  Fix an assertion failure when optimising a shufflevector incorrectly into concat_vectors, and a followup bug with SelectionDAG::getNode() creating nodes with invalid types.
  Minor cleanup. No functional change.
  ...

1 files changed, 235 insertions, 30 deletions

diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
index 1c485a0..d7fa009 100644
--- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
+++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp
@@ -413,7 +413,7 @@ static char isNegatibleForFree(SDValue Op, bool LegalOperations,
         !TLI.isOperationLegalOrCustom(ISD::FSUB,  Op.getValueType()))
       return 0;
 
-    // fold (fsub (fadd A, B)) -> (fsub (fneg A), B)
+    // fold (fneg (fadd A, B)) -> (fsub (fneg A), B)
     if (char V = isNegatibleForFree(Op.getOperand(0), LegalOperations, TLI,
                                     Options, Depth + 1))
       return V;
@@ -2496,8 +2496,18 @@ SDValue DAGCombiner::visitAND(SDNode *N) {
         // lanes of the constant together.
         EVT VT = Vector->getValueType(0);
         unsigned BitWidth = VT.getVectorElementType().getSizeInBits();
+
+        // If the splat value has been compressed to a bitlength lower
+        // than the size of the vector lane, we need to re-expand it to
+        // the lane size.
+        if (BitWidth > SplatBitSize)
+          for (SplatValue = SplatValue.zextOrTrunc(BitWidth);
+               SplatBitSize < BitWidth;
+               SplatBitSize = SplatBitSize * 2)
+            SplatValue |= SplatValue.shl(SplatBitSize);
+
         Constant = APInt::getAllOnesValue(BitWidth);
-        for (unsigned i = 0, n = VT.getVectorNumElements(); i < n; ++i)
+        for (unsigned i = 0, n = SplatBitSize/BitWidth; i < n; ++i)
           Constant &= SplatValue.lshr(i*BitWidth).zextOrTrunc(BitWidth);
       }
     }
@@ -5681,6 +5691,127 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
                        DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
                                    N0.getOperand(1), N1));
 
+  // In unsafe math mode, we can fold chains of FADD's of the same value
+  // into multiplications.  This transform is not safe in general because
+  // we are reducing the number of rounding steps.
+  if (DAG.getTarget().Options.UnsafeFPMath &&
+      TLI.isOperationLegalOrCustom(ISD::FMUL, VT) &&
+      !N0CFP && !N1CFP) {
+    if (N0.getOpcode() == ISD::FMUL) {
+      ConstantFPSDNode *CFP00 = dyn_cast<ConstantFPSDNode>(N0.getOperand(0));
+      ConstantFPSDNode *CFP01 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1));
+
+      // (fadd (fmul c, x), x) -> (fmul c+1, x)
+      if (CFP00 && !CFP01 && N0.getOperand(1) == N1) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP00, 0),
+                                     DAG.getConstantFP(1.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N1, NewCFP);
+      }
+
+      // (fadd (fmul x, c), x) -> (fmul c+1, x)
+      if (CFP01 && !CFP00 && N0.getOperand(0) == N1) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP01, 0),
+                                     DAG.getConstantFP(1.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N1, NewCFP);
+      }
+
+      // (fadd (fadd x, x), x) -> (fmul 3.0, x)
+      if (!CFP00 && !CFP01 && N0.getOperand(0) == N0.getOperand(1) &&
+          N0.getOperand(0) == N1) {
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N1, DAG.getConstantFP(3.0, VT));
+      }
+
+      // (fadd (fmul c, x), (fadd x, x)) -> (fmul c+2, x)
+      if (CFP00 && !CFP01 && N1.getOpcode() == ISD::FADD &&
+          N1.getOperand(0) == N1.getOperand(1) &&
+          N0.getOperand(1) == N1.getOperand(0)) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP00, 0),
+                                     DAG.getConstantFP(2.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N0.getOperand(1), NewCFP);
+      }
+
+      // (fadd (fmul x, c), (fadd x, x)) -> (fmul c+2, x)
+      if (CFP01 && !CFP00 && N1.getOpcode() == ISD::FADD &&
+          N1.getOperand(0) == N1.getOperand(1) &&
+          N0.getOperand(0) == N1.getOperand(0)) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP01, 0),
+                                     DAG.getConstantFP(2.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N0.getOperand(0), NewCFP);
+      }
+    }
+
+    if (N1.getOpcode() == ISD::FMUL) {
+      ConstantFPSDNode *CFP10 = dyn_cast<ConstantFPSDNode>(N1.getOperand(0));
+      ConstantFPSDNode *CFP11 = dyn_cast<ConstantFPSDNode>(N1.getOperand(1));
+
+      // (fadd x, (fmul c, x)) -> (fmul c+1, x)
+      if (CFP10 && !CFP11 && N1.getOperand(1) == N0) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP10, 0),
+                                     DAG.getConstantFP(1.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N0, NewCFP);
+      }
+
+      // (fadd x, (fmul x, c)) -> (fmul c+1, x)
+      if (CFP11 && !CFP10 && N1.getOperand(0) == N0) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP11, 0),
+                                     DAG.getConstantFP(1.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N0, NewCFP);
+      }
+
+      // (fadd x, (fadd x, x)) -> (fmul 3.0, x)
+      if (!CFP10 && !CFP11 && N1.getOperand(0) == N1.getOperand(1) &&
+          N1.getOperand(0) == N0) {
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N0, DAG.getConstantFP(3.0, VT));
+      }
+
+      // (fadd (fadd x, x), (fmul c, x)) -> (fmul c+2, x)
+      if (CFP10 && !CFP11 && N1.getOpcode() == ISD::FADD &&
+          N1.getOperand(0) == N1.getOperand(1) &&
+          N0.getOperand(1) == N1.getOperand(0)) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP10, 0),
+                                     DAG.getConstantFP(2.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N0.getOperand(1), NewCFP);
+      }
+
+      // (fadd (fadd x, x), (fmul x, c)) -> (fmul c+2, x)
+      if (CFP11 && !CFP10 && N1.getOpcode() == ISD::FADD &&
+          N1.getOperand(0) == N1.getOperand(1) &&
+          N0.getOperand(0) == N1.getOperand(0)) {
+        SDValue NewCFP = DAG.getNode(ISD::FADD, N->getDebugLoc(), VT,
+                                     SDValue(CFP11, 0),
+                                     DAG.getConstantFP(2.0, VT));
+        return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                           N0.getOperand(0), NewCFP);
+      }
+    }
+
+    // (fadd (fadd x, x), (fadd x, x)) -> (fmul 4.0, x)
+    if (N0.getOpcode() == ISD::FADD && N1.getOpcode() == ISD::FADD &&
+        N0.getOperand(0) == N0.getOperand(1) &&
+        N1.getOperand(0) == N1.getOperand(1) &&
+        N0.getOperand(0) == N1.getOperand(0)) {
+      return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                         N0.getOperand(0),
+                         DAG.getConstantFP(4.0, VT));
+    }
+  }
+
   // FADD -> FMA combines:
   if ((DAG.getTarget().Options.AllowFPOpFusion == FPOpFusion::Fast ||
        DAG.getTarget().Options.UnsafeFPMath) &&
@@ -5692,8 +5823,8 @@ SDValue DAGCombiner::visitFADD(SDNode *N) {
       return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
                          N0.getOperand(0), N0.getOperand(1), N1);
     }
-  
-    // fold (fadd x, (fmul y, z)) -> (fma x, y, z)
+
+    // fold (fadd x, (fmul y, z)) -> (fma y, z, x)
     // Note: Commutes FADD operands.
     if (N1.getOpcode() == ISD::FMUL && N1->hasOneUse()) {
       return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT,
@@ -5867,6 +5998,7 @@ SDValue DAGCombiner::visitFMA(SDNode *N) {
   ConstantFPSDNode *N0CFP = dyn_cast<ConstantFPSDNode>(N0);
   ConstantFPSDNode *N1CFP = dyn_cast<ConstantFPSDNode>(N1);
   EVT VT = N->getValueType(0);
+  DebugLoc dl = N->getDebugLoc();
 
   if (N0CFP && N0CFP->isExactlyValue(1.0))
     return DAG.getNode(ISD::FADD, N->getDebugLoc(), VT, N1, N2);
@@ -5877,6 +6009,58 @@ SDValue DAGCombiner::visitFMA(SDNode *N) {
   if (N0CFP && !N1CFP)
     return DAG.getNode(ISD::FMA, N->getDebugLoc(), VT, N1, N0, N2);
 
+  // (fma x, c1, (fmul x, c2)) -> (fmul x, c1+c2)
+  if (DAG.getTarget().Options.UnsafeFPMath && N1CFP &&
+      N2.getOpcode() == ISD::FMUL &&
+      N0 == N2.getOperand(0) &&
+      N2.getOperand(1).getOpcode() == ISD::ConstantFP) {
+    return DAG.getNode(ISD::FMUL, dl, VT, N0,
+                       DAG.getNode(ISD::FADD, dl, VT, N1, N2.getOperand(1)));
+  }
+
+
+  // (fma (fmul x, c1), c2, y) -> (fma x, c1*c2, y)
+  if (DAG.getTarget().Options.UnsafeFPMath &&
+      N0.getOpcode() == ISD::FMUL && N1CFP &&
+      N0.getOperand(1).getOpcode() == ISD::ConstantFP) {
+    return DAG.getNode(ISD::FMA, dl, VT,
+                       N0.getOperand(0),
+                       DAG.getNode(ISD::FMUL, dl, VT, N1, N0.getOperand(1)),
+                       N2);
+  }
+
+  // (fma x, 1, y) -> (fadd x, y)
+  // (fma x, -1, y) -> (fadd (fneg x), y)
+  if (N1CFP) {
+    if (N1CFP->isExactlyValue(1.0))
+      return DAG.getNode(ISD::FADD, dl, VT, N0, N2);
+
+    if (N1CFP->isExactlyValue(-1.0) &&
+        (!LegalOperations || TLI.isOperationLegal(ISD::FNEG, VT))) {
+      SDValue RHSNeg = DAG.getNode(ISD::FNEG, dl, VT, N0);
+      AddToWorkList(RHSNeg.getNode());
+      return DAG.getNode(ISD::FADD, dl, VT, N2, RHSNeg);
+    }
+  }
+
+  // (fma x, c, x) -> (fmul x, (c+1))
+  if (DAG.getTarget().Options.UnsafeFPMath && N1CFP && N0 == N2) {
+    return DAG.getNode(ISD::FMUL, dl, VT,
+                       N0,
+                       DAG.getNode(ISD::FADD, dl, VT,
+                                   N1, DAG.getConstantFP(1.0, VT)));
+  }
+
+  // (fma x, c, (fneg x)) -> (fmul x, (c-1))
+  if (DAG.getTarget().Options.UnsafeFPMath && N1CFP &&
+      N2.getOpcode() == ISD::FNEG && N2.getOperand(0) == N0) {
+    return DAG.getNode(ISD::FMUL, dl, VT,
+                       N0,
+                       DAG.getNode(ISD::FADD, dl, VT,
+                                   N1, DAG.getConstantFP(-1.0, VT)));
+  }
+
+
   return SDValue();
 }
 
@@ -6225,6 +6409,30 @@ SDValue DAGCombiner::visitFNEG(SDNode *N) {
   SDValue N0 = N->getOperand(0);
   EVT VT = N->getValueType(0);
 
+  if (VT.isVector() && !LegalOperations) {
+    // If operand is a BUILD_VECTOR node, see if we can constant fold it.
+    if (N0.getOpcode() == ISD::BUILD_VECTOR) {
+      SmallVector<SDValue, 8> Ops;
+      for (unsigned i = 0, e = N0.getNumOperands(); i != e; ++i) {
+        SDValue Op = N0.getOperand(i);
+        if (Op.getOpcode() != ISD::UNDEF &&
+            Op.getOpcode() != ISD::ConstantFP)
+          break;
+        EVT EltVT = Op.getValueType();
+        SDValue FoldOp = DAG.getNode(ISD::FNEG, N0.getDebugLoc(), EltVT, Op);
+        if (FoldOp.getOpcode() != ISD::UNDEF &&
+            FoldOp.getOpcode() != ISD::ConstantFP)
+          break;
+        Ops.push_back(FoldOp);
+        AddToWorkList(FoldOp.getNode());
+      }
+
+      if (Ops.size() == N0.getNumOperands())
+        return DAG.getNode(ISD::BUILD_VECTOR, N->getDebugLoc(),
+                           VT, &Ops[0], Ops.size());
+    }
+  }
+
   if (isNegatibleForFree(N0, LegalOperations, DAG.getTargetLoweringInfo(),
                          &DAG.getTarget().Options))
     return GetNegatedExpression(N0, DAG, LegalOperations);
@@ -6246,6 +6454,17 @@ SDValue DAGCombiner::visitFNEG(SDNode *N) {
     }
   }
 
+  // (fneg (fmul c, x)) -> (fmul -c, x)
+  if (N0.getOpcode() == ISD::FMUL) {
+    ConstantFPSDNode *CFP1 = dyn_cast<ConstantFPSDNode>(N0.getOperand(1));
+    if (CFP1) {
+      return DAG.getNode(ISD::FMUL, N->getDebugLoc(), VT,
+                         N0.getOperand(0),
+                         DAG.getNode(ISD::FNEG, N->getDebugLoc(), VT,
+                                     N0.getOperand(1)));
+    }
+  }
+
   return SDValue();
 }
 
@@ -7876,29 +8095,15 @@ SDValue DAGCombiner::visitBUILD_VECTOR(SDNode *N) {
       if (VecIn1.getValueType().getSizeInBits()*2 != VT.getSizeInBits())
         return SDValue();
 
-      // If the element type of the input vector is not the same as
-      // the output element type, make concat_vectors based on input element
-      // type and then bitcast it to the output vector type.
-      //
-      // In another words avoid nodes like this:
-      //  <NODE> v16i8 = concat_vectors v4i16 v4i16
-      // Replace it with this one:
-      //  <NODE0> v8i16 = concat_vectors v4i16 v4i16
-      //  <NODE1> v16i8 = bitcast NODE0
-      EVT ItemType = VecIn1.getValueType().getVectorElementType();
-      if (ItemType != VT.getVectorElementType()) {
-        EVT ConcatVT = EVT::getVectorVT(*DAG.getContext(),
-                                ItemType,
-                                VecIn1.getValueType().getVectorNumElements()*2);
-        // Widen the input vector by adding undef values.
-        VecIn1 = DAG.getNode(ISD::CONCAT_VECTORS, dl, ConcatVT,
-                             VecIn1, DAG.getUNDEF(VecIn1.getValueType()));
-        VecIn1 = DAG.getNode(ISD::BITCAST, dl, VT, VecIn1);
-      } else
-        // Widen the input vector by adding undef values.
-        VecIn1 = DAG.getNode(ISD::CONCAT_VECTORS, dl, VT,
-                             VecIn1, DAG.getUNDEF(VecIn1.getValueType()));
+      // If the input vector type has a different base type to the output
+      // vector type, bail out.
+      if (VecIn1.getValueType().getVectorElementType() !=
+          VT.getVectorElementType())
+        return SDValue();
 
+      // Widen the input vector by adding undef values.
+      VecIn1 = DAG.getNode(ISD::CONCAT_VECTORS, N->getDebugLoc(), VT,
+                           VecIn1, DAG.getUNDEF(VecIn1.getValueType()));
     }
 
     // If VecIn2 is unused then change it to undef.
@@ -8749,7 +8954,7 @@ SDValue DAGCombiner::BuildUDIV(SDNode *N) {
 // to alias with anything but itself.  Provides base object and offset as
 // results.
 static bool FindBaseOffset(SDValue Ptr, SDValue &Base, int64_t &Offset,
-                           const GlobalValue *&GV, void *&CV) {
+                           const GlobalValue *&GV, const void *&CV) {
   // Assume it is a primitive operation.
   Base = Ptr; Offset = 0; GV = 0; CV = 0;
 
@@ -8774,8 +8979,8 @@ static bool FindBaseOffset(SDValue Ptr, SDValue &Base, int64_t &Offset,
   // for ConstantSDNodes since the same constant pool entry may be represented
   // by multiple nodes with different offsets.
   if (ConstantPoolSDNode *C = dyn_cast<ConstantPoolSDNode>(Base)) {
-    CV = C->isMachineConstantPoolEntry() ? (void *)C->getMachineCPVal()
-                                         : (void *)C->getConstVal();
+    CV = C->isMachineConstantPoolEntry() ? (const void *)C->getMachineCPVal()
+                                         : (const void *)C->getConstVal();
     Offset += C->getOffset();
     return false;
   }
@@ -8800,7 +9005,7 @@ bool DAGCombiner::isAlias(SDValue Ptr1, int64_t Size1,
   SDValue Base1, Base2;
   int64_t Offset1, Offset2;
   const GlobalValue *GV1, *GV2;
-  void *CV1, *CV2;
+  const void *CV1, *CV2;
   bool isFrameIndex1 = FindBaseOffset(Ptr1, Base1, Offset1, GV1, CV1);
   bool isFrameIndex2 = FindBaseOffset(Ptr2, Base2, Offset2, GV2, CV2);