[llvm] r193942 - Enable optimization of sin / cos pair into call to __sincos_stret for iOS7+.
Bob Wilson
bob.wilson at apple.com
Sat Nov 2 23:14:38 PDT 2013
Author: bwilson
Date: Sun Nov 3 01:14:38 2013
New Revision: 193942
URL: http://llvm.org/viewvc/llvm-project?rev=193942&view=rev
Log:
Enable optimization of sin / cos pair into call to __sincos_stret for iOS7+.
rdar://12856873
Patch by Evan Cheng, with a fix for rdar://13209539 by Tilmann Scheller
Added:
llvm/trunk/test/CodeGen/ARM/sincos.ll
Modified:
llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
llvm/trunk/lib/Target/ARM/ARMISelLowering.h
llvm/trunk/lib/Target/ARM/ARMSubtarget.cpp
llvm/trunk/lib/Target/ARM/ARMSubtarget.h
Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp?rev=193942&r1=193941&r2=193942&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp Sun Nov 3 01:14:38 2013
@@ -869,6 +869,18 @@ ARMTargetLowering::ARMTargetLowering(Tar
setOperationAction(ISD::FP32_TO_FP16, MVT::i32, Expand);
}
}
+
+ // 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->getTargetTriple().getOS() == Triple::IOS) {
+ // For iOS, we don't want to the normal expansion of a libcall to
+ // sincos. We want to issue a libcall to __sincos_stret.
+ setOperationAction(ISD::FSINCOS, MVT::f64, Custom);
+ setOperationAction(ISD::FSINCOS, MVT::f32, Custom);
+ }
+ }
// We have target-specific dag combine patterns for the following nodes:
// ARMISD::VMOVRRD - No need to call setTargetDAGCombine
@@ -5950,6 +5962,70 @@ static SDValue LowerADDC_ADDE_SUBC_SUBE(
Op.getOperand(1), Op.getOperand(2));
}
+SDValue ARMTargetLowering::LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const {
+ assert(Subtarget->isTargetDarwin());
+
+ // For iOS, we want to call an alternative entry point: __sincos_stret,
+ // return values are passed via sret.
+ SDLoc dl(Op);
+ SDValue Arg = Op.getOperand(0);
+ EVT ArgVT = Arg.getValueType();
+ Type *ArgTy = ArgVT.getTypeForEVT(*DAG.getContext());
+
+ MachineFrameInfo *FrameInfo = DAG.getMachineFunction().getFrameInfo();
+ const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+ // Pair of floats / doubles used to pass the result.
+ StructType *RetTy = StructType::get(ArgTy, ArgTy, NULL);
+
+ // Create stack object for sret.
+ const uint64_t ByteSize = TLI.getDataLayout()->getTypeAllocSize(RetTy);
+ const unsigned StackAlign = TLI.getDataLayout()->getPrefTypeAlignment(RetTy);
+ int FrameIdx = FrameInfo->CreateStackObject(ByteSize, StackAlign, false);
+ SDValue SRet = DAG.getFrameIndex(FrameIdx, TLI.getPointerTy());
+
+ ArgListTy Args;
+ ArgListEntry Entry;
+
+ Entry.Node = SRet;
+ Entry.Ty = RetTy->getPointerTo();
+ Entry.isSExt = false;
+ Entry.isZExt = false;
+ Entry.isSRet = true;
+ Args.push_back(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());
+
+ TargetLowering::
+ CallLoweringInfo CLI(DAG.getEntryNode(), Type::getVoidTy(*DAG.getContext()),
+ false, false, false, false, 0,
+ CallingConv::C, /*isTaillCall=*/false,
+ /*doesNotRet=*/false, /*isReturnValueUsed*/false,
+ Callee, Args, DAG, dl);
+ std::pair<SDValue, SDValue> CallResult = LowerCallTo(CLI);
+
+ SDValue LoadSin = DAG.getLoad(ArgVT, dl, CallResult.second, SRet,
+ MachinePointerInfo(), false, false, false, 0);
+
+ // Address of cos field.
+ SDValue Add = DAG.getNode(ISD::ADD, dl, getPointerTy(), SRet,
+ DAG.getIntPtrConstant(ArgVT.getStoreSize()));
+ SDValue LoadCos = DAG.getLoad(ArgVT, dl, LoadSin.getValue(1), Add,
+ MachinePointerInfo(), false, false, false, 0);
+
+ SDVTList Tys = DAG.getVTList(ArgVT, ArgVT);
+ return DAG.getNode(ISD::MERGE_VALUES, dl, Tys,
+ LoadSin.getValue(0), LoadCos.getValue(0));
+}
+
static SDValue LowerAtomicLoadStore(SDValue Op, SelectionDAG &DAG) {
// Monotonic load/store is legal for all targets
if (cast<AtomicSDNode>(Op)->getOrdering() <= Monotonic)
@@ -6081,6 +6157,7 @@ SDValue ARMTargetLowering::LowerOperatio
case ISD::SUBE: return LowerADDC_ADDE_SUBC_SUBE(Op, DAG);
case ISD::ATOMIC_LOAD:
case ISD::ATOMIC_STORE: return LowerAtomicLoadStore(Op, DAG);
+ case ISD::FSINCOS: return LowerFSINCOS(Op, DAG);
case ISD::SDIVREM:
case ISD::UDIVREM: return LowerDivRem(Op, DAG);
}
Modified: llvm/trunk/lib/Target/ARM/ARMISelLowering.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMISelLowering.h?rev=193942&r1=193941&r2=193942&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMISelLowering.h (original)
+++ llvm/trunk/lib/Target/ARM/ARMISelLowering.h Sun Nov 3 01:14:38 2013
@@ -448,6 +448,7 @@ namespace llvm {
const ARMSubtarget *ST) const;
SDValue LowerBUILD_VECTOR(SDValue Op, SelectionDAG &DAG,
const ARMSubtarget *ST) const;
+ SDValue LowerFSINCOS(SDValue Op, SelectionDAG &DAG) const;
SDValue LowerDivRem(SDValue Op, SelectionDAG &DAG) const;
/// isFMAFasterThanFMulAndFAdd - Return true if an FMA operation is faster
Modified: llvm/trunk/lib/Target/ARM/ARMSubtarget.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMSubtarget.cpp?rev=193942&r1=193941&r2=193942&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMSubtarget.cpp (original)
+++ llvm/trunk/lib/Target/ARM/ARMSubtarget.cpp Sun Nov 3 01:14:38 2013
@@ -281,6 +281,11 @@ unsigned ARMSubtarget::getMispredictionP
return SchedModel->MispredictPenalty;
}
+bool ARMSubtarget::hasSinCos() const {
+ return getTargetTriple().getOS() == Triple::IOS &&
+ !getTargetTriple().isOSVersionLT(7, 0);
+}
+
bool ARMSubtarget::enablePostRAScheduler(
CodeGenOpt::Level OptLevel,
TargetSubtargetInfo::AntiDepBreakMode& Mode,
Modified: llvm/trunk/lib/Target/ARM/ARMSubtarget.h
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/ARM/ARMSubtarget.h?rev=193942&r1=193941&r2=193942&view=diff
==============================================================================
--- llvm/trunk/lib/Target/ARM/ARMSubtarget.h (original)
+++ llvm/trunk/lib/Target/ARM/ARMSubtarget.h Sun Nov 3 01:14:38 2013
@@ -330,6 +330,10 @@ public:
const std::string & getCPUString() const { return CPUString; }
unsigned getMispredictionPenalty() const;
+
+ /// This function returns true if the target has sincos() routine in its
+ /// compiler runtime or math libraries.
+ bool hasSinCos() const;
/// enablePostRAScheduler - True at 'More' optimization.
bool enablePostRAScheduler(CodeGenOpt::Level OptLevel,
Added: llvm/trunk/test/CodeGen/ARM/sincos.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/ARM/sincos.ll?rev=193942&view=auto
==============================================================================
--- llvm/trunk/test/CodeGen/ARM/sincos.ll (added)
+++ llvm/trunk/test/CodeGen/ARM/sincos.ll Sun Nov 3 01:14:38 2013
@@ -0,0 +1,38 @@
+; RUN: llc < %s -mtriple=armv7-apple-ios6 -mcpu=cortex-a8 | FileCheck %s --check-prefix=NOOPT
+; RUN: llc < %s -mtriple=armv7-apple-ios7 -mcpu=cortex-a8 | FileCheck %s --check-prefix=SINCOS
+
+; Combine sin / cos into a single call.
+; rdar://12856873
+
+define float @test1(float %x) nounwind {
+entry:
+; SINCOS-LABEL: test1:
+; SINCOS: bl ___sincosf_stret
+
+; NOOPT-LABEL: test1:
+; NOOPT: bl _sinf
+; NOOPT: bl _cosf
+ %call = tail call float @sinf(float %x) nounwind readnone
+ %call1 = tail call float @cosf(float %x) nounwind readnone
+ %add = fadd float %call, %call1
+ ret float %add
+}
+
+define double @test2(double %x) nounwind {
+entry:
+; SINCOS-LABEL: test2:
+; SINCOS: bl ___sincos_stret
+
+; NOOPT-LABEL: test2:
+; NOOPT: bl _sin
+; NOOPT: bl _cos
+ %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
+}
+
+declare float @sinf(float) readonly
+declare double @sin(double) readonly
+declare float @cosf(float) readonly
+declare double @cos(double) readonly
More information about the llvm-commits
mailing list