Teach SDISel to combine fsin / fcos into a fsincos node if the following

conditions are met:
1. They share the same operand and are in the same BB.
2. Both outputs are used.
3. The target has a native instruction that maps to ISD::FSINCOS node or
   the target provides a sincos library call.

Implemented the generic optimization in sdisel and enabled it for
Mac OSX. Also added an additional optimization for x86_64 Mac OSX by
using an alternative entry point __sincos_stret which returns the two
results in xmm0 / xmm1.

rdar://13087969
PR13204


git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@173755 91177308-0d34-0410-b5e6-96231b3b80d8

10 files changed, 102 insertions, 15 deletions

diff --git a/lib/Target/ARM/ARMISelLowering.cpp b/lib/Target/ARM/ARMISelLowering.cpp
index bdbf45ca04..3d283a5ac9 100644
--- a/lib/Target/ARM/ARMISelLowering.cpp
+++ b/lib/Target/ARM/ARMISelLowering.cpp
@@ -781,6 +781,8 @@ ARMTargetLowering::ARMTargetLowering(TargetMachine &TM)
   setOperationAction(ISD::FSIN,      MVT::f32, Expand);
   setOperationAction(ISD::FCOS,      MVT::f32, Expand);
   setOperationAction(ISD::FCOS,      MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS,   MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS,   MVT::f32, Expand);
   setOperationAction(ISD::FREM,      MVT::f64, Expand);
   setOperationAction(ISD::FREM,      MVT::f32, Expand);
   if (!TM.Options.UseSoftFloat && Subtarget->hasVFP2() &&
diff --git a/lib/Target/Hexagon/HexagonISelLowering.cpp b/lib/Target/Hexagon/HexagonISelLowering.cpp
index 1a0e581051..9c7243bdce 100644
--- a/lib/Target/Hexagon/HexagonISelLowering.cpp
+++ b/lib/Target/Hexagon/HexagonISelLowering.cpp
@@ -1364,6 +1364,8 @@ HexagonTargetLowering::HexagonTargetLowering(HexagonTargetMachine
     setOperationAction(ISD::FSIN , MVT::f32, Expand);
     setOperationAction(ISD::FCOS , MVT::f32, Expand);
     setOperationAction(ISD::FREM , MVT::f32, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
     setOperationAction(ISD::CTPOP, MVT::i32, Expand);
     setOperationAction(ISD::CTTZ , MVT::i32, Expand);
     setOperationAction(ISD::CTTZ_ZERO_UNDEF, MVT::i32, Expand);
diff --git a/lib/Target/MBlaze/MBlazeISelLowering.cpp b/lib/Target/MBlaze/MBlazeISelLowering.cpp
index 8a9f0922b1..5b3c6fede3 100644
--- a/lib/Target/MBlaze/MBlazeISelLowering.cpp
+++ b/lib/Target/MBlaze/MBlazeISelLowering.cpp
@@ -81,6 +81,7 @@ MBlazeTargetLowering::MBlazeTargetLowering(MBlazeTargetMachine &TM)
   setOperationAction(ISD::FCOPYSIGN,  MVT::f64, Expand);
   setOperationAction(ISD::FSIN,       MVT::f32, Expand);
   setOperationAction(ISD::FCOS,       MVT::f32, Expand);
+  setOperationAction(ISD::FSINCOS,    MVT::f32, Expand);
   setOperationAction(ISD::FPOWI,      MVT::f32, Expand);
   setOperationAction(ISD::FPOW,       MVT::f32, Expand);
   setOperationAction(ISD::FLOG,       MVT::f32, Expand);
diff --git a/lib/Target/Mips/MipsISelLowering.cpp b/lib/Target/Mips/MipsISelLowering.cpp
index 8925ddeaf3..7b8557dcd0 100644
--- a/lib/Target/Mips/MipsISelLowering.cpp
+++ b/lib/Target/Mips/MipsISelLowering.cpp
@@ -421,6 +421,8 @@ MipsTargetLowering(MipsTargetMachine &TM)
   setOperationAction(ISD::FSIN,              MVT::f64,   Expand);
   setOperationAction(ISD::FCOS,              MVT::f32,   Expand);
   setOperationAction(ISD::FCOS,              MVT::f64,   Expand);
+  setOperationAction(ISD::FSINCOS,           MVT::f32,   Expand);
+  setOperationAction(ISD::FSINCOS,           MVT::f64,   Expand);
   setOperationAction(ISD::FPOWI,             MVT::f32,   Expand);
   setOperationAction(ISD::FPOW,              MVT::f32,   Expand);
   setOperationAction(ISD::FPOW,              MVT::f64,   Expand);
diff --git a/lib/Target/PowerPC/PPCISelLowering.cpp b/lib/Target/PowerPC/PPCISelLowering.cpp
index 9966b2cc90..9a68927103 100644
--- a/lib/Target/PowerPC/PPCISelLowering.cpp
+++ b/lib/Target/PowerPC/PPCISelLowering.cpp
@@ -132,11 +132,13 @@ PPCTargetLowering::PPCTargetLowering(PPCTargetMachine &TM)
   // We don't support sin/cos/sqrt/fmod/pow
   setOperationAction(ISD::FSIN , MVT::f64, Expand);
   setOperationAction(ISD::FCOS , MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
   setOperationAction(ISD::FREM , MVT::f64, Expand);
   setOperationAction(ISD::FPOW , MVT::f64, Expand);
   setOperationAction(ISD::FMA  , MVT::f64, Legal);
   setOperationAction(ISD::FSIN , MVT::f32, Expand);
   setOperationAction(ISD::FCOS , MVT::f32, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
   setOperationAction(ISD::FREM , MVT::f32, Expand);
   setOperationAction(ISD::FPOW , MVT::f32, Expand);
   setOperationAction(ISD::FMA  , MVT::f32, Legal);
diff --git a/lib/Target/Sparc/SparcISelLowering.cpp b/lib/Target/Sparc/SparcISelLowering.cpp
index 168640f457..7664abde51 100644
--- a/lib/Target/Sparc/SparcISelLowering.cpp
+++ b/lib/Target/Sparc/SparcISelLowering.cpp
@@ -759,10 +759,12 @@ SparcTargetLowering::SparcTargetLowering(TargetMachine &TM)
 
   setOperationAction(ISD::FSIN , MVT::f64, Expand);
   setOperationAction(ISD::FCOS , MVT::f64, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
   setOperationAction(ISD::FREM , MVT::f64, Expand);
   setOperationAction(ISD::FMA  , MVT::f64, Expand);
   setOperationAction(ISD::FSIN , MVT::f32, Expand);
   setOperationAction(ISD::FCOS , MVT::f32, Expand);
+  setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
   setOperationAction(ISD::FREM , MVT::f32, Expand);
   setOperationAction(ISD::FMA  , MVT::f32, Expand);
   setOperationAction(ISD::CTPOP, MVT::i32, Expand);
diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index a4eae0a8f8..ca606daaba 100644
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -605,10 +605,12 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FGETSIGN, MVT::i32, Custom);
 
     // We don't support sin/cos/fmod
-    setOperationAction(ISD::FSIN , MVT::f64, Expand);
-    setOperationAction(ISD::FCOS , MVT::f64, Expand);
-    setOperationAction(ISD::FSIN , MVT::f32, Expand);
-    setOperationAction(ISD::FCOS , MVT::f32, Expand);
+    setOperationAction(ISD::FSIN   , MVT::f64, Expand);
+    setOperationAction(ISD::FCOS   , MVT::f64, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
+    setOperationAction(ISD::FSIN   , MVT::f32, Expand);
+    setOperationAction(ISD::FCOS   , MVT::f32, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
 
     // Expand FP immediates into loads from the stack, except for the special
     // cases we handle.
@@ -633,8 +635,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FCOPYSIGN, MVT::f32, Custom);
 
     // We don't support sin/cos/fmod
-    setOperationAction(ISD::FSIN , MVT::f32, Expand);
-    setOperationAction(ISD::FCOS , MVT::f32, Expand);
+    setOperationAction(ISD::FSIN   , MVT::f32, Expand);
+    setOperationAction(ISD::FCOS   , MVT::f32, Expand);
+    setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
 
     // Special cases we handle for FP constants.
     addLegalFPImmediate(APFloat(+0.0f)); // xorps
@@ -644,8 +647,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     addLegalFPImmediate(APFloat(-1.0)); // FLD1/FCHS
 
     if (!TM.Options.UnsafeFPMath) {
-      setOperationAction(ISD::FSIN           , MVT::f64  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f64  , Expand);
+      setOperationAction(ISD::FSIN   , MVT::f64, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f64, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
     }
   } else if (!TM.Options.UseSoftFloat) {
     // f32 and f64 in x87.
@@ -659,10 +663,12 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::FCOPYSIGN, MVT::f32, Expand);
 
     if (!TM.Options.UnsafeFPMath) {
-      setOperationAction(ISD::FSIN           , MVT::f32  , Expand);
-      setOperationAction(ISD::FSIN           , MVT::f64  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f32  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f64  , Expand);
+      setOperationAction(ISD::FSIN   , MVT::f64, Expand);
+      setOperationAction(ISD::FSIN   , MVT::f32, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f64, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f32, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f64, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f32, Expand);
     }
     addLegalFPImmediate(APFloat(+0.0)); // FLD0
     addLegalFPImmediate(APFloat(+1.0)); // FLD1
@@ -699,8 +705,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     }
 
     if (!TM.Options.UnsafeFPMath) {
-      setOperationAction(ISD::FSIN           , MVT::f80  , Expand);
-      setOperationAction(ISD::FCOS           , MVT::f80  , Expand);
+      setOperationAction(ISD::FSIN   , MVT::f80, Expand);
+      setOperationAction(ISD::FCOS   , MVT::f80, Expand);
+      setOperationAction(ISD::FSINCOS, MVT::f80, Expand);
     }
 
     setOperationAction(ISD::FFLOOR, MVT::f80, Expand);
@@ -748,7 +755,9 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setOperationAction(ISD::INSERT_SUBVECTOR, VT,Expand);
     setOperationAction(ISD::FABS, VT, Expand);
     setOperationAction(ISD::FSIN, VT, Expand);
+    setOperationAction(ISD::FSINCOS, VT, Expand);
     setOperationAction(ISD::FCOS, VT, Expand);
+    setOperationAction(ISD::FSINCOS, VT, Expand);
     setOperationAction(ISD::FREM, VT, Expand);
     setOperationAction(ISD::FMA,  VT, Expand);
     setOperationAction(ISD::FPOWI, VT, Expand);
@@ -1281,6 +1290,19 @@ X86TargetLowering::X86TargetLowering(X86TargetMachine &TM)
     setLibcallName(RTLIB::SRA_I128, 0);
   }
 
+  // Combine sin / cos into one node or libcall if possible.
+  if (Subtarget->hasSinCos()) {
+    setLibcallName(RTLIB::SINCOS_F32, "sincosf");
+    setLibcallName(RTLIB::SINCOS_F64, "sincos");
+    if (Subtarget->isTargetDarwin() && Subtarget->is64Bit()) {
+      // For MacOSX, we don't want to the normal expansion of a libcall to
+      // sincos. We want to issue a libcall to __sincos_stret to avoid memory
+      // traffic.
+      setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
+      setOperationAction(ISD::FSINCOS, MVT::f32, Custom);
+    }
+  }
+
   // We have target-specific dag combine patterns for the following nodes:
   setTargetDAGCombine(ISD::VECTOR_SHUFFLE);
   setTargetDAGCombine(ISD::EXTRACT_VECTOR_ELT);
@@ -12014,6 +12036,50 @@ static SDValue LowerADDC_ADDE_SUBC_SUBE(SDValue Op, SelectionDAG &DAG) {
                      Op.getOperand(1), Op.getOperand(2));
 }
 
+SDValue X86TargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const {
+  assert(Subtarget->isTargetDarwin());
+ 
+  // For MacOSX, we want to call an alternative entry point: __sincos_stret,
+  // which returns the values in two XMM registers.
+  DebugLoc dl = Op.getDebugLoc();
+  SDValue Arg = Op.getOperand(0);
+  EVT ArgVT = Arg.getValueType();
+  Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
+  
+  ArgListTy Args;
+  ArgListEntry Entry;
+  
+  Entry.Node = Arg;
+  Entry.Ty = ArgTy;
+  Entry.isSExt = false;
+  Entry.isZExt = false;
+  Args.push_back(Entry);
+  
+  const char *LibcallName = (ArgVT == MVT::f64)
+    ? "__sincos_stret" : "__sincosf_stret";
+  SDValue Callee = DAG.getExternalSymbol(LibcallName, getPointerTy());
+  
+  StructType *RetTy = StructType::get(ArgTy, ArgTy, NULL);
+  TargetLowering::
+  CallLoweringInfo CLI(DAG.getEntryNode(), RetTy,
+                       false, false, false, false, 0,
+                       CallingConv::C, /*isTaillCall=*/false,
+                       /*doesNotRet=*/false, /*isReturnValueUsed*/true,
+                       Callee, Args, DAG, dl);
+  std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
+#if 1
+  return CallResult.first;
+#else
+  SDValue RetSin = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, ArgVT,
+                               CallResult.first, DAG.getIntPtrConstant(0));
+  SDValue RetCos = DAG.getNode(ISD::EXTRACT_ELEMENT, dl, ArgVT,
+                               CallResult.first, DAG.getIntPtrConstant(1));
+  
+  SDVTList Tys = DAG.getVTList(ArgVT, ArgVT);
+  return DAG.getNode(ISD::MERGE_VALUES, dl, Tys, RetSin, RetCos);
+#endif
+}
+
 /// LowerOperation - Provide custom lowering hooks for some operations.
 ///
 SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
@@ -12096,6 +12162,7 @@ SDValue X86TargetLowering::LowerOperation(SDValue Op, SelectionDAG &DAG) const {
   case ISD::ADD:                return LowerADD(Op, DAG);
   case ISD::SUB:                return LowerSUB(Op, DAG);
   case ISD::SDIV:               return LowerSDIV(Op, DAG);
+  case ISD::FSINCOS:            return LowerFSINCOS(Op, DAG);
   }
 }
 
diff --git a/lib/Target/X86/X86ISelLowering.h b/lib/Target/X86/X86ISelLowering.h
index 5e84e277c0..6758ed1404 100644
--- a/lib/Target/X86/X86ISelLowering.h
+++ b/lib/Target/X86/X86ISelLowering.h
@@ -838,8 +838,8 @@ namespace llvm {
     SDValue LowerFLT_ROUNDS_(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerShift(SDValue Op, SelectionDAG &DAG) const;
     SDValue LowerSDIV(SDValue Op, SelectionDAG &DAG) const;
-
     SDValue LowerSIGN_EXTEND_INREG(SDValue Op, SelectionDAG &DAG) const;
+    SDValue LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const;
 
     // Utility functions to help LowerVECTOR_SHUFFLE & LowerBUILD_VECTOR
     SDValue LowerVectorBroadcast(SDValue Op, SelectionDAG &DAG) const;
diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
index 53c28f4fce..dad95c6800 100644
--- a/lib/Target/X86/X86Subtarget.cpp
+++ b/lib/Target/X86/X86Subtarget.cpp
@@ -155,6 +155,11 @@ const char *X86Subtarget::getBZeroEntry() const {
   return 0;
 }
 
+bool X86Subtarget::hasSinCos() const {
+  return getTargetTriple().isMacOSX() &&
+    !getTargetTriple().isMacOSXVersionLT(10, 9);
+}
+
 /// IsLegalToCallImmediateAddr - Return true if the subtarget allows calls
 /// to immediate address.
 bool X86Subtarget::IsLegalToCallImmediateAddr(const TargetMachine &TM) const {
diff --git a/lib/Target/X86/X86Subtarget.h b/lib/Target/X86/X86Subtarget.h
index b325f62126..eb587a5761 100644
--- a/lib/Target/X86/X86Subtarget.h
+++ b/lib/Target/X86/X86Subtarget.h
@@ -328,6 +328,10 @@ public:
   /// memset with zero passed as the second argument. Otherwise it
   /// returns null.
   const char *getBZeroEntry() const;
+  
+  /// This function returns true if the target has sincos() routine in its
+  /// compiler runtime or math libraries.
+  bool hasSinCos() const;
 
   /// enablePostRAScheduler - run for Atom optimization.
   bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,