enable SDISel sincos optimization for GNU environments

- add sincos to runtime library if target triple environment is GNU
- added canCombineSinCosLibcall() which checks that sincos is in the RTL and
  if the environment is GNU then unsafe fpmath is enabled (required to
  preserve errno)
- extended sincos-opt lit test

Reviewed by: Hal Finkel

llvm-svn: 175283

3 files changed, 73 insertions, 23 deletions

diff --git a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
index 6a3e03b3936..4a0176bc031 100644
--- a/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/LegalizeDAG.cpp
@@ -15,6 +15,7 @@
 #include "llvm/ADT/DenseMap.h"
 #include "llvm/ADT/SmallPtrSet.h"
 #include "llvm/ADT/SmallVector.h"
+#include "llvm/ADT/Triple.h"
 #include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/MachineFunction.h"
 #include "llvm/CodeGen/MachineJumpTableInfo.h"
@@ -2111,6 +2112,20 @@ static bool isSinCosLibcallAvailable(SDNode *Node, const TargetLowering &TLI) {
   return TLI.getLibcallName(LC) != 0;
 }
 
+/// canCombineSinCosLibcall - Return true if sincos libcall is available and
+/// can be used to combine sin and cos.
+static bool canCombineSinCosLibcall(SDNode *Node, const TargetLowering &TLI,
+                                    const TargetMachine &TM) {
+  if (!isSinCosLibcallAvailable(Node, TLI))
+    return false;
+  // GNU sin/cos functions set errno while sincos does not. Therefore
+  // combining sin and cos is only safe if unsafe-fpmath is enabled.
+  bool isGNU = Triple(TM.getTargetTriple()).getEnvironment() == Triple::GNU;
+  if (isGNU && !TM.Options.UnsafeFPMath)
+    return false;
+  return true;
+}
+
 /// useSinCos - Only issue sincos libcall if both sin and cos are
 /// needed.
 static bool useSinCos(SDNode *Node) {
@@ -3149,7 +3164,7 @@ void SelectionDAGLegalize::ExpandNode(SDNode *Node) {
     // Turn fsin / fcos into ISD::FSINCOS node if there are a pair of fsin /
     // fcos which share the same operand and both are used.
     if ((TLI.isOperationLegalOrCustom(ISD::FSINCOS, VT) ||
-         isSinCosLibcallAvailable(Node, TLI))
+         canCombineSinCosLibcall(Node, TLI, TM))
         && useSinCos(Node)) {
       SDVTList VTs = DAG.getVTList(VT, VT);
       Tmp1 = DAG.getNode(ISD::FSINCOS, dl, VTs, Node->getOperand(0));
diff --git a/llvm/lib/CodeGen/TargetLoweringBase.cpp b/llvm/lib/CodeGen/TargetLoweringBase.cpp
index 10aa9d684ba..f2329dc720c 100644
--- a/llvm/lib/CodeGen/TargetLoweringBase.cpp
+++ b/llvm/lib/CodeGen/TargetLoweringBase.cpp
@@ -14,6 +14,7 @@
 #include "llvm/Target/TargetLowering.h"
 #include "llvm/ADT/BitVector.h"
 #include "llvm/ADT/STLExtras.h"
+#include "llvm/ADT/Triple.h"
 #include "llvm/CodeGen/Analysis.h"
 #include "llvm/CodeGen/MachineFrameInfo.h"
 #include "llvm/CodeGen/MachineFunction.h"
@@ -34,7 +35,7 @@ using namespace llvm;
 
 /// InitLibcallNames - Set default libcall names.
 ///
-static void InitLibcallNames(const char **Names) {
+static void InitLibcallNames(const char **Names, const TargetMachine &TM) {
   Names[RTLIB::SHL_I16] = "__ashlhi3";
   Names[RTLIB::SHL_I32] = "__ashlsi3";
   Names[RTLIB::SHL_I64] = "__ashldi3";
@@ -341,12 +342,20 @@ static void InitLibcallNames(const char **Names) {
   Names[RTLIB::SYNC_FETCH_AND_NAND_4] = "__sync_fetch_and_nand_4";
   Names[RTLIB::SYNC_FETCH_AND_NAND_8] = "__sync_fetch_and_nand_8";
   
-  // These are generally not available.
-  Names[RTLIB::SINCOS_F32] = 0;
-  Names[RTLIB::SINCOS_F64] = 0;
-  Names[RTLIB::SINCOS_F80] = 0;
-  Names[RTLIB::SINCOS_F128] = 0;
-  Names[RTLIB::SINCOS_PPCF128] = 0;
+  if (Triple(TM.getTargetTriple()).getEnvironment() == Triple::GNU) {
+    Names[RTLIB::SINCOS_F32] = "sincosf";
+    Names[RTLIB::SINCOS_F64] = "sincos";
+    Names[RTLIB::SINCOS_F80] = "sincosl";
+    Names[RTLIB::SINCOS_F128] = "sincosl";
+    Names[RTLIB::SINCOS_PPCF128] = "sincosl";
+  } else {
+    // These are generally not available.
+    Names[RTLIB::SINCOS_F32] = 0;
+    Names[RTLIB::SINCOS_F64] = 0;
+    Names[RTLIB::SINCOS_F80] = 0;
+    Names[RTLIB::SINCOS_F128] = 0;
+    Names[RTLIB::SINCOS_PPCF128] = 0;
+  }
 }
 
 /// InitLibcallCallingConvs - Set default libcall CallingConvs.
@@ -726,7 +735,7 @@ TargetLoweringBase::TargetLoweringBase(const TargetMachine &tm,
   SupportJumpTables = true;
   MinimumJumpTableEntries = 4;
 
-  InitLibcallNames(LibcallRoutineNames);
+  InitLibcallNames(LibcallRoutineNames, TM);
   InitCmpLibcallCCs(CmpLibcallCCs);
   InitLibcallCallingConvs(LibcallCallingConvs);
 }
diff --git a/llvm/test/CodeGen/X86/sincos-opt.ll b/llvm/test/CodeGen/X86/sincos-opt.ll
index 65c84177454..f364d1fc2dc 100644
--- a/llvm/test/CodeGen/X86/sincos-opt.ll
+++ b/llvm/test/CodeGen/X86/sincos-opt.ll
@@ -1,18 +1,24 @@
-; RUN: llc < %s -mtriple=x86_64-apple-macosx10.9.0 -mcpu=core2 | FileCheck %s --check-prefix=SINCOS
-; RUN: llc < %s -mtriple=x86_64-apple-macosx10.8.0 -mcpu=core2 | FileCheck %s --check-prefix=NOOPT
+; RUN: llc < %s -mtriple=x86_64-apple-macosx10.9.0 -mcpu=core2 | FileCheck %s --check-prefix=OSX_SINCOS
+; RUN: llc < %s -mtriple=x86_64-apple-macosx10.8.0 -mcpu=core2 | FileCheck %s --check-prefix=OSX_NOOPT
+; RUN: llc < %s -mtriple=x86_64-pc-linux-gnu -mcpu=core2 -enable-unsafe-fp-math | FileCheck %s --check-prefix=GNU_SINCOS
 
 ; Combine sin / cos into a single call.
 ; rdar://13087969
 
 define float @test1(float %x) nounwind {
 entry:
-; SINCOS: test1:
-; SINCOS: callq ___sincosf_stret
-; SINCOS: addss %xmm1, %xmm0
+; GNU_SINCOS: test1:
+; GNU_SINCOS: callq sincosf
+; GNU_SINCOS: movss 4(%rsp), %xmm0
+; GNU_SINCOS: addss (%rsp), %xmm0
 
-; NOOPT: test1
-; NOOPT: callq _cosf
-; NOOPT: callq _sinf
+; OSX_SINCOS: test1:
+; OSX_SINCOS: callq ___sincosf_stret
+; OSX_SINCOS: addss %xmm1, %xmm0
+
+; OSX_NOOPT: test1
+; OSX_NOOPT: callq _cosf
+; OSX_NOOPT: callq _sinf
   %call = tail call float @sinf(float %x) nounwind readnone
   %call1 = tail call float @cosf(float %x) nounwind readnone
   %add = fadd float %call, %call1
@@ -21,20 +27,40 @@ entry:
 
 define double @test2(double %x) nounwind {
 entry:
-; SINCOS: test2:
-; SINCOS: callq ___sincos_stret
-; SINCOS: addsd %xmm1, %xmm0
+; GNU_SINCOS: test2:
+; GNU_SINCOS: callq sincos
+; GNU_SINCOS: movsd 16(%rsp), %xmm0
+; GNU_SINCOS: addsd 8(%rsp), %xmm0
+
+; OSX_SINCOS: test2:
+; OSX_SINCOS: callq ___sincos_stret
+; OSX_SINCOS: addsd %xmm1, %xmm0
 
-; NOOPT: test2
-; NOOPT: callq _cos
-; NOOPT: callq _sin
+; OSX_NOOPT: test2
+; OSX_NOOPT: callq _cos
+; OSX_NOOPT: callq _sin
   %call = tail call double @sin(double %x) nounwind readnone
   %call1 = tail call double @cos(double %x) nounwind readnone
   %add = fadd double %call, %call1
   ret double %add
 }
 
+define x86_fp80 @test3(x86_fp80 %x) nounwind {
+entry:
+; GNU_SINCOS: test3:
+; GNU_SINCOS: callq sinl
+; GNU_SINCOS: callq cosl
+; GNU_SINCOS: ret
+  %call = tail call x86_fp80 @sinl(x86_fp80 %x) nounwind
+  %call1 = tail call x86_fp80 @cosl(x86_fp80 %x) nounwind
+  %add = fadd x86_fp80 %call, %call1
+  ret x86_fp80 %add
+}
+
 declare float  @sinf(float) readonly
 declare double @sin(double) readonly
 declare float @cosf(float) readonly
 declare double @cos(double) readonly
+
+declare x86_fp80 @sinl(x86_fp80)
+declare x86_fp80 @cosl(x86_fp80)