Stop resetting DisableTailCalls in TargetMachine::resetTargetOptions

[ahatanak]

Summary

This patch makes changes in TargetMachine::resetTargetOptions to stop setting DisableTallCalls based on function attribute "disable-tail-calls". DisableTailCalls is the last non-fast-math option that is set in resetTargetOptions, so we should be able to remove this function entirely after the work to propagate IR-level fast-math flags to DAG nodes is completed.

Also, this patch enables cl:opt option "disable-tail-calls" to override function attribute "disable-tail-calls".

[echristo]

Hi Akira,

Can you add a "no attribute" function to the testcase as well? Also probably test with x86 since you're changing that backend as well that matches the ARM one?

I assume there's a front end patch that I'm missing that goes along with this?

Thanks!

-eric

[echristo]

Hi Akira,

Just saw the frontend patch, so go ahead and ignore that question :)

-eric

[ahatanak]

Added a function that doesn't have any attributes and a test case for x86_64.

[echristo]

LGTM, thanks!

-eric

llvm/trunk/include/llvm/CodeGen/CommandFlags.h

 // command line tools. The tools "llc" and "opt" both use this file to prevent
 // flag duplication.
 //
 //===----------------------------------------------------------------------===//
 
 #ifndef LLVM_CODEGEN_COMMANDFLAGS_H
 #define LLVM_CODEGEN_COMMANDFLAGS_H
 
+#include "llvm/ADT/StringExtras.h"
 #include "llvm/IR/Module.h"
 #include "llvm/MC/MCTargetOptionsCommandFlags.h"
 #include "llvm//MC/SubtargetFeature.h"
 #include "llvm/Support/CodeGen.h"
 #include "llvm/Support/CommandLine.h"
 #include "llvm/Support/Host.h"
 #include "llvm/Target/TargetMachine.h"
 #include "llvm/Target/TargetOptions.h"
static inline TargetOptions InitTargetOptionsFromCodeGenFlags() {
   Options.NoInfsFPMath = EnableNoInfsFPMath;
   Options.NoNaNsFPMath = EnableNoNaNsFPMath;
   Options.HonorSignDependentRoundingFPMathOption =
       EnableHonorSignDependentRoundingFPMath;
   if (FloatABIForCalls != FloatABI::Default)
     Options.FloatABIType = FloatABIForCalls;
   Options.NoZerosInBSS = DontPlaceZerosInBSS;
   Options.GuaranteedTailCallOpt = EnableGuaranteedTailCallOpt;
-  Options.DisableTailCalls = DisableTailCalls;
   Options.StackAlignmentOverride = OverrideStackAlignment;
   Options.TrapFuncName = TrapFuncName;
   Options.PositionIndependentExecutable = EnablePIE;
   Options.UseInitArray = !UseCtors;
   Options.DataSections = DataSections;
   Options.FunctionSections = FunctionSections;
   Options.UniqueSectionNames = UniqueSectionNames;
 
if (!Features.empty())
       NewAttrs = NewAttrs.addAttribute(Ctx, AttributeSet::FunctionIndex,
                                        "target-features", Features);
 
     if (DisableFPElim.getNumOccurrences() > 0)
       NewAttrs = NewAttrs.addAttribute(Ctx, AttributeSet::FunctionIndex,
                                        "no-frame-pointer-elim",
                                        DisableFPElim ? "true" : "false");
 
+    if (DisableTailCalls.getNumOccurrences() > 0)
+      NewAttrs = NewAttrs.addAttribute(Ctx, AttributeSet::FunctionIndex,
+                                       "disable-tail-calls",
+                                       toStringRef(DisableTailCalls));
+
     // Let NewAttrs override Attrs.
     NewAttrs = Attrs.addAttributes(Ctx, AttributeSet::FunctionIndex, NewAttrs);
     F.setAttributes(NewAttrs);
   }
 }
 
 #endif

llvm/trunk/include/llvm/Target/TargetOptions.h

namespace llvm {
   public:
     TargetOptions()
         : PrintMachineCode(false),
           LessPreciseFPMADOption(false), UnsafeFPMath(false),
           NoInfsFPMath(false), NoNaNsFPMath(false),
           HonorSignDependentRoundingFPMathOption(false),
           NoZerosInBSS(false),
           GuaranteedTailCallOpt(false),
-          DisableTailCalls(false), StackAlignmentOverride(0),
+          StackAlignmentOverride(0),
           EnableFastISel(false), PositionIndependentExecutable(false),
           UseInitArray(false), DisableIntegratedAS(false),
           CompressDebugSections(false), FunctionSections(false),
           DataSections(false), UniqueSectionNames(true), TrapUnreachable(false),
           TrapFuncName(), FloatABIType(FloatABI::Default),
           AllowFPOpFusion(FPOpFusion::Standard), Reciprocals(TargetRecip()),
           JTType(JumpTable::Single),
           ThreadModel(ThreadModel::POSIX) {}
public:
     /// GuaranteedTailCallOpt - This flag is enabled when -tailcallopt is
     /// specified on the commandline. When the flag is on, participating targets
     /// will perform tail call optimization on all calls which use the fastcc
     /// calling convention and which satisfy certain target-independent
     /// criteria (being at the end of a function, having the same return type
     /// as their parent function, etc.), using an alternate ABI if necessary.
     unsigned GuaranteedTailCallOpt : 1;
 
-    /// DisableTailCalls - This flag controls whether we will use tail calls.
-    /// Disabling them may be useful to maintain a correct call stack.
-    unsigned DisableTailCalls : 1;
-
     /// StackAlignmentOverride - Override default stack alignment for target.
     unsigned StackAlignmentOverride;
 
     /// EnableFastISel - This flag enables fast-path instruction selection
     /// which trades away generated code quality in favor of reducing
     /// compile time.
     unsigned EnableFastISel : 1;
 

 #define ARE_EQUAL(X) LHS.X == RHS.X
   return
     ARE_EQUAL(UnsafeFPMath) &&
     ARE_EQUAL(NoInfsFPMath) &&
     ARE_EQUAL(NoNaNsFPMath) &&
     ARE_EQUAL(HonorSignDependentRoundingFPMathOption) &&
     ARE_EQUAL(NoZerosInBSS) &&
     ARE_EQUAL(GuaranteedTailCallOpt) &&
-    ARE_EQUAL(DisableTailCalls) &&
     ARE_EQUAL(StackAlignmentOverride) &&
     ARE_EQUAL(EnableFastISel) &&
     ARE_EQUAL(PositionIndependentExecutable) &&
     ARE_EQUAL(UseInitArray) &&
     ARE_EQUAL(TrapUnreachable) &&
     ARE_EQUAL(TrapFuncName) &&
     ARE_EQUAL(FloatABIType) &&
     ARE_EQUAL(AllowFPOpFusion) &&

llvm/trunk/lib/Target/ARM/ARMISelLowering.cpp

ARMTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   CallingConv::ID CallConv              = CLI.CallConv;
   bool doesNotRet                       = CLI.DoesNotReturn;
   bool isVarArg                         = CLI.IsVarArg;
 
   MachineFunction &MF = DAG.getMachineFunction();
   bool isStructRet    = (Outs.empty()) ? false : Outs[0].Flags.isSRet();
   bool isThisReturn   = false;
   bool isSibCall      = false;
+  auto Attr = MF.getFunction()->getFnAttribute("disable-tail-calls");
 
   // Disable tail calls if they're not supported.
-  if (!Subtarget->supportsTailCall() || MF.getTarget().Options.DisableTailCalls)
+  if (!Subtarget->supportsTailCall() || Attr.getValueAsString() == "true")
     isTailCall = false;
 
   if (isTailCall) {
     // Check if it's really possible to do a tail call.
     isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv,
                     isVarArg, isStructRet, MF.getFunction()->hasStructRetAttr(),
                                                    Outs, OutVals, Ins, DAG);
     if (!isTailCall && CLI.CS && CLI.CS->isMustTailCall())
bool ARMTargetLowering::isUsedByReturnOnly(SDNode *N, SDValue &Chain) const {
   Chain = TCChain;
   return true;
 }
 
 bool ARMTargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
   if (!Subtarget->supportsTailCall())
     return false;
 
-  if (!CI->isTailCall() || getTargetMachine().Options.DisableTailCalls)
+  auto Attr =
+      CI->getParent()->getParent()->getFnAttribute("disable-tail-calls");
+  if (!CI->isTailCall() || Attr.getValueAsString() == "true")
     return false;
 
   return !Subtarget->isThumb1Only();
 }
 
 // Trying to write a 64 bit value so need to split into two 32 bit values first,
 // and pass the lower and high parts through.
 static SDValue LowerWRITE_REGISTER(SDValue Op, SelectionDAG &DAG) {

llvm/trunk/lib/Target/Hexagon/HexagonISelLowering.cpp

HexagonTargetLowering::LowerReturn(SDValue Chain,
   if (Flag.getNode())
     RetOps.push_back(Flag);
 
   return DAG.getNode(HexagonISD::RET_FLAG, dl, MVT::Other, RetOps);
 }
 
 bool HexagonTargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
   // If either no tail call or told not to tail call at all, don't.
-  if (!CI->isTailCall() || HTM.Options.DisableTailCalls)
+  auto Attr =
+      CI->getParent()->getParent()->getFnAttribute("disable-tail-calls");
+  if (!CI->isTailCall() || Attr.getValueAsString() == "true")
     return false;
 
   return true;
 }
 
 /// LowerCallResult - Lower the result values of an ISD::CALL into the
 /// appropriate copies out of appropriate physical registers.  This assumes that
 /// Chain/InFlag are the input chain/flag to use, and that TheCall is the call
HexagonTargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   HexagonCCState CCInfo(CallConv, isVarArg, DAG.getMachineFunction(), ArgLocs,
                         *DAG.getContext(), NumNamedVarArgParams);
 
   if (isVarArg)
     CCInfo.AnalyzeCallOperands(Outs, CC_Hexagon_VarArg);
   else
     CCInfo.AnalyzeCallOperands(Outs, CC_Hexagon);
 
-  if (DAG.getTarget().Options.DisableTailCalls)
+  auto Attr = MF.getFunction()->getFnAttribute("disable-tail-calls");
+  if (Attr.getValueAsString() == "true")
     isTailCall = false;
 
   if (isTailCall) {
     bool StructAttrFlag = MF.getFunction()->hasStructRetAttr();
     isTailCall = IsEligibleForTailCallOptimization(Callee, CallConv,
                                                    isVarArg, IsStructRet,
                                                    StructAttrFlag,
                                                    Outs, OutVals, Ins, DAG);

llvm/trunk/lib/Target/TargetMachine.cpp

do {                                                                         \
     if (F.hasFnAttribute(Y))                                                   \
       Options.X = (F.getFnAttribute(Y).getValueAsString() == "true");          \
   } while (0)
 
   RESET_OPTION(LessPreciseFPMADOption, "less-precise-fpmad");
   RESET_OPTION(UnsafeFPMath, "unsafe-fp-math");
   RESET_OPTION(NoInfsFPMath, "no-infs-fp-math");
   RESET_OPTION(NoNaNsFPMath, "no-nans-fp-math");
-  RESET_OPTION(DisableTailCalls, "disable-tail-calls");
 }
 
 /// getRelocationModel - Returns the code generation relocation model. The
 /// choices are static, PIC, and dynamic-no-pic, and target default.
 Reloc::Model TargetMachine::getRelocationModel() const {
   if (!CodeGenInfo)
     return Reloc::Default;
   return CodeGenInfo->getRelocationModel();

llvm/trunk/lib/Target/X86/X86ISelLowering.cpp

 
 /// \brief Return true if the calling convention is a C calling convention.
 static bool IsCCallConvention(CallingConv::ID CC) {
   return (CC == CallingConv::C || CC == CallingConv::X86_64_Win64 ||
           CC == CallingConv::X86_64_SysV);
 }
 
 bool X86TargetLowering::mayBeEmittedAsTailCall(CallInst *CI) const {
-  if (!CI->isTailCall() || getTargetMachine().Options.DisableTailCalls)
+  auto Attr =
+      CI->getParent()->getParent()->getFnAttribute("disable-tail-calls");
+  if (!CI->isTailCall() || Attr.getValueAsString() == "true")
     return false;
 
   CallSite CS(CI);
   CallingConv::ID CalleeCC = CS.getCallingConv();
   if (!IsTailCallConvention(CalleeCC) && !IsCCallConvention(CalleeCC))
     return false;
 
   return true;
X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
   bool isVarArg                         = CLI.IsVarArg;
 
   MachineFunction &MF = DAG.getMachineFunction();
   bool Is64Bit        = Subtarget->is64Bit();
   bool IsWin64        = Subtarget->isCallingConvWin64(CallConv);
   StructReturnType SR = callIsStructReturn(Outs);
   bool IsSibcall      = false;
   X86MachineFunctionInfo *X86Info = MF.getInfo<X86MachineFunctionInfo>();
+  auto Attr = MF.getFunction()->getFnAttribute("disable-tail-calls");
 
-  if (MF.getTarget().Options.DisableTailCalls)
+  if (Attr.getValueAsString() == "true")
     isTailCall = false;
 
   if (Subtarget->isPICStyleGOT() &&
       !MF.getTarget().Options.GuaranteedTailCallOpt) {
     // If we are using a GOT, disable tail calls to external symbols with
     // default visibility. Tail calling such a symbol requires using a GOT
     // relocation, which forces early binding of the symbol. This breaks code
     // that require lazy function symbol resolution. Using musttail or

llvm/trunk/lib/Transforms/IPO/PassManagerBuilder.cpp

static cl::opt<bool> EnableLoopDistribute(
     "enable-loop-distribute", cl::init(false), cl::Hidden,
     cl::desc("Enable the new, experimental LoopDistribution Pass"));
 
 PassManagerBuilder::PassManagerBuilder() {
     OptLevel = 2;
     SizeLevel = 0;
     LibraryInfo = nullptr;
     Inliner = nullptr;
-    DisableTailCalls = false;
     DisableUnitAtATime = false;
     DisableUnrollLoops = false;
     BBVectorize = RunBBVectorization;
     SLPVectorize = RunSLPVectorization;
     LoopVectorize = RunLoopVectorization;
     RerollLoops = RunLoopRerolling;
     LoadCombine = RunLoadCombine;
     DisableGVNLoadPRE = false;
else
     MPM.add(createScalarReplAggregatesPass(-1, false));
   MPM.add(createEarlyCSEPass());              // Catch trivial redundancies
   MPM.add(createJumpThreadingPass());         // Thread jumps.
   MPM.add(createCorrelatedValuePropagationPass()); // Propagate conditionals
   MPM.add(createCFGSimplificationPass());     // Merge & remove BBs
   MPM.add(createInstructionCombiningPass());  // Combine silly seq's
   addExtensionsToPM(EP_Peephole, MPM);
 
-  if (!DisableTailCalls)
-    MPM.add(createTailCallEliminationPass()); // Eliminate tail calls
   MPM.add(createCFGSimplificationPass());     // Merge & remove BBs
   MPM.add(createReassociatePass());           // Reassociate expressions
   // Rotate Loop - disable header duplication at -Oz
   MPM.add(createLoopRotatePass(SizeLevel == 2 ? 0 : -1));
   MPM.add(createLICMPass());                  // Hoist loop invariants
   MPM.add(createLoopUnswitchPass(SizeLevel || OptLevel < 3));
   MPM.add(createInstructionCombiningPass());
   MPM.add(createIndVarSimplifyPass());        // Canonicalize indvars

llvm/trunk/lib/Transforms/Scalar/TailRecursionElimination.cpp

static bool CanTRE(Function &F) {
 
   return true;
 }
 
 bool TailCallElim::runOnFunction(Function &F) {
   if (skipOptnoneFunction(F))
     return false;
 
+  if (F.getFnAttribute("disable-tail-calls").getValueAsString() == "true")
+    return false;
+
   bool AllCallsAreTailCalls = false;
   bool Modified = markTails(F, AllCallsAreTailCalls);
   if (AllCallsAreTailCalls)
     Modified |= runTRE(F);
   return Modified;
 }
 
 namespace {

llvm/trunk/test/CodeGen/ARM/disable-tail-calls.ll

+; RUN: llc < %s -march arm | FileCheck %s --check-prefix=NO-OPTION
+; RUN: llc < %s -march arm -disable-tail-calls | FileCheck %s --check-prefix=DISABLE-TRUE
+; RUN: llc < %s -march arm -disable-tail-calls=false | FileCheck %s --check-prefix=DISABLE-FALSE
+
+; Check that command line option "-disable-tail-calls" overrides function
+; attribute "disable-tail-calls".
+
+; NO-OPTION-LABEL: {{\_?}}func_attr
+; NO-OPTION: bl {{\_?}}callee
+
+; DISABLE-FALSE-LABEL: {{\_?}}func_attr
+; DISABLE-FALSE: b {{\_?}}callee
+
+; DISABLE-TRUE-LABEL: {{\_?}}func_attr
+; DISABLE-TRUE: bl {{\_?}}callee
+
+define i32 @func_attr(i32 %a) #0 {
+entry:
+  %call = tail call i32 @callee(i32 %a)
+  ret i32 %call
+}
+
+; NO-OPTION-LABEL: {{\_?}}func_noattr
+; NO-OPTION: b {{\_?}}callee
+
+; DISABLE-FALSE-LABEL: {{\_?}}func_noattr
+; DISABLE-FALSE: b {{\_?}}callee
+
+; DISABLE-TRUE-LABEL: {{\_?}}func_noattr
+; DISABLE-TRUE: bl {{\_?}}callee
+
+define i32 @func_noattr(i32 %a) {
+entry:
+  %call = tail call i32 @callee(i32 %a)
+  ret i32 %call
+}
+
+declare i32 @callee(i32)
+
+attributes #0 = { "disable-tail-calls"="true" }

llvm/trunk/test/CodeGen/X86/disable-tail-calls.ll

+; RUN: llc < %s -march x86-64 | FileCheck %s --check-prefix=NO-OPTION
+; RUN: llc < %s -march x86-64 -disable-tail-calls | FileCheck %s --check-prefix=DISABLE-TRUE
+; RUN: llc < %s -march x86-64 -disable-tail-calls=false | FileCheck %s --check-prefix=DISABLE-FALSE
+
+; Check that command line option "-disable-tail-calls" overrides function
+; attribute "disable-tail-calls".
+
+; NO-OPTION-LABEL: {{\_?}}func_attr
+; NO-OPTION: callq {{\_?}}callee
+
+; DISABLE-FALSE-LABEL: {{\_?}}func_attr
+; DISABLE-FALSE: jmp {{\_?}}callee
+
+; DISABLE-TRUE-LABEL: {{\_?}}func_attr
+; DISABLE-TRUE: callq {{\_?}}callee
+
+define i32 @func_attr(i32 %a) #0 {
+entry:
+  %call = tail call i32 @callee(i32 %a)
+  ret i32 %call
+}
+
+; NO-OPTION-LABEL: {{\_?}}func_noattr
+; NO-OPTION: jmp {{\_?}}callee
+
+; DISABLE-FALSE-LABEL: {{\_?}}func_noattr
+; DISABLE-FALSE: jmp {{\_?}}callee
+
+; DISABLE-TRUE-LABEL: {{\_?}}func_noattr
+; DISABLE-TRUE: callq {{\_?}}callee
+
+define i32 @func_noattr(i32 %a) {
+entry:
+  %call = tail call i32 @callee(i32 %a)
+  ret i32 %call
+}
+
+declare i32 @callee(i32)
+
+attributes #0 = { "disable-tail-calls"="true" }