[llvm] r213892 - [X86] Optimize stackmap shadows on X86.
Eric Christopher
echristo at gmail.com
Thu Jul 24 16:56:47 PDT 2014
On Thu, Jul 24, 2014 at 4:37 PM, Lang Hames <lhames at gmail.com> wrote:
> Hi Eric,
>
>> So you've create an EmitAndCountInstruction that you use for
>> everything except nop...
>
> Yep. I didn't want the shadow tracker counting the NOP bytes that it itself
> was outputting. Doing so wouldn't cause any harm (the padding size is
> calculated before any NOPs are emitted, so wouldn't be affected), but it
> would lead to double-encoding NOPs unnecessarily. The downside of this
> conservatism is that you don't get to take advantage of any NOPs already
> present in the code following your stackmap, however I expect this to be
> rare.
>
Except you make sure and turn it off before you emit it seems?
>> + bool Count
>> Ahem. A boolean called Count? Especially when you have a function called
>> count.
>
> Yeah, that's probably not very helpful is it? :)
> This variable is a latch: It ensures that we only double encoding
> instructions when we're actually in the shadow of a stackmap. Prior to
> hitting a stackmap, or after we're out of its shadow, Count is automatically
> set to false. Perhaps InShadow would be a better name?
>
>> You've got a code emitter you're creating in the class - why?
>
> The generic streamer interface doesn't provide any mechanism for getting an
> instruction's length. After all - we might be streaming asm. To reliably
> determine instruction length on x86 we have to encode the instruction.
> Unfortunately that means redundantly encoding it in the case where we're
> streaming objects, but the Count/IsShadow latch will keep the overhead of
> that to a minimum (proportional to the number of stack maps used in your
> code).
>
Probably wouldn't need to run this in that case as well (or we do, at
which point you'd think that the main assembler interface would still
want this code.
Also MCAssembler does this sort of computation when it's figuring out
if the instruction needs relaxation so we might be able to do
something there as well?
Just having two of these data structures seems messy in some way. :\
>> You've made changes here, but it's not obvious what you did. Perhaps
>> this refactor should have gone in first? (Along with the other ones).
>
> Nailed it. Sorry again. :)
>
>> This looks reasonably heavyweight to be calling on each instruction
>> emission.
>
> As covered above, this will only be activated following a stackmap, where it
> will actually do some useful work. Since I don't have a large body of code
> that uses stackmaps handy, I haven't tested the impact there. On x86 code
> without stack maps the overhead of this code should be one bool test per
> instruction emitted.
OK. That's better than it seemed. I seem to have missed the turning it
on only when processing a stack map. Due to the bit above the diff
itself is a little messy. :)
-eric
>
> - Lang.
>
> On Thu, Jul 24, 2014 at 3:49 PM, Eric Christopher <echristo at gmail.com>
> wrote:
>>
>> Hi Lang,
>>
>> Some comments and many questions:
>>
>> So you've create an EmitAndCountInstruction that you use for
>> everything except nop... since it looks like this stuff is more
>> multiplatform than just x86, why not generalize this out to the
>> AsmPrinter level (other than the fact that x86 is the only variable
>> length instruction set architecture we have?). Also, why not nop?
>>
>> + bool Count
>>
>> Ahem. A boolean called Count? Especially when you have a function called
>> count.
>>
>> + TargetMachine &TM;
>> + std::unique_ptr<MCCodeEmitter> CodeEmitter;
>>
>> You've got a code emitter you're creating in the class - why? This is
>> theoretically run at the same time that the rest of asm printing is
>> going on, Also you wouldn't need the TargetMachine stored either.
>>
>> + unsigned RequiredShadowSize, CurrentShadowSize;
>>
>> Comments for these please.
>>
>> - return LowerTlsAddr(OutStreamer, MCInstLowering, *MI,
>> getSubtargetInfo());
>> + return LowerTlsAddr(MCInstLowering, *MI);
>>
>> You've made changes here, but it's not obvious what you did. Perhaps
>> this refactor should have gone in first? (Along with the other ones).
>>
>> + X86AsmPrinter::StackMapShadowTracker::~StackMapShadowTracker() {}
>>
>> Empty non-default destructor?
>>
>> + if (Count) {
>> + SmallString<256> Code;
>> + SmallVector<MCFixup, 4> Fixups;
>> + raw_svector_ostream VecOS(Code);
>> + CodeEmitter->EncodeInstruction(Inst, VecOS, Fixups, STI);
>> + VecOS.flush();
>> + CurrentShadowSize += Code.size();
>> + if (CurrentShadowSize >= RequiredShadowSize)
>> + Count = false; // The shadow is big enough. Stop counting.
>> + }
>>
>> This looks reasonably heavyweight to be calling on each instruction
>> emission - have you timed the performance on assembling some large
>> files? Also this means that you're calling EncodeInstruction twice on
>> every instruction emission?
>>
>> -eric
>>
>>
>> On Thu, Jul 24, 2014 at 1:40 PM, Lang Hames <lhames at gmail.com> wrote:
>> > Author: lhames
>> > Date: Thu Jul 24 15:40:55 2014
>> > New Revision: 213892
>> >
>> > URL: http://llvm.org/viewvc/llvm-project?rev=213892&view=rev
>> > Log:
>> > [X86] Optimize stackmap shadows on X86.
>> >
>> > This patch minimizes the number of nops that must be emitted on X86 to
>> > satisfy
>> > stackmap shadow constraints.
>> >
>> > To minimize the number of nops inserted, the X86AsmPrinter now records
>> > the
>> > size of the most recent stackmap's shadow in the StackMapShadowTracker
>> > class,
>> > and tracks the number of instruction bytes emitted since the that
>> > stackmap
>> > instruction was encountered. Padding is emitted (if it is required at
>> > all)
>> > immediately before the next stackmap/patchpoint instruction, or at the
>> > end of
>> > the basic block.
>> >
>> > This optimization should reduce code-size and improve performance for
>> > people
>> > using the llvm stackmap intrinsic on X86.
>> >
>> > <rdar://problem/14959522>
>> >
>> >
>> > Added:
>> > llvm/trunk/test/CodeGen/X86/stackmap-shadow-optimization.ll
>> > Modified:
>> > llvm/trunk/include/llvm/CodeGen/AsmPrinter.h
>> > llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
>> > llvm/trunk/lib/Target/X86/X86AsmPrinter.cpp
>> > llvm/trunk/lib/Target/X86/X86AsmPrinter.h
>> > llvm/trunk/lib/Target/X86/X86MCInstLower.cpp
>> > llvm/trunk/test/CodeGen/X86/stackmap-nops.ll
>> > llvm/trunk/test/MC/X86/stackmap-nops.ll
>> >
>> > Modified: llvm/trunk/include/llvm/CodeGen/AsmPrinter.h
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/include/llvm/CodeGen/AsmPrinter.h?rev=213892&r1=213891&r2=213892&view=diff
>> >
>> > ==============================================================================
>> > --- llvm/trunk/include/llvm/CodeGen/AsmPrinter.h (original)
>> > +++ llvm/trunk/include/llvm/CodeGen/AsmPrinter.h Thu Jul 24 15:40:55
>> > 2014
>> > @@ -264,6 +264,9 @@ public:
>> > /// function.
>> > virtual void EmitFunctionBodyEnd() {}
>> >
>> > + /// Targets can override this to emit stuff at the end of a basic
>> > block.
>> > + virtual void EmitBasicBlockEnd(const MachineBasicBlock &MBB) {}
>> > +
>> > /// Targets should implement this to emit instructions.
>> > virtual void EmitInstruction(const MachineInstr *) {
>> > llvm_unreachable("EmitInstruction not implemented");
>> >
>> > Modified: llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp?rev=213892&r1=213891&r2=213892&view=diff
>> >
>> > ==============================================================================
>> > --- llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp (original)
>> > +++ llvm/trunk/lib/CodeGen/AsmPrinter/AsmPrinter.cpp Thu Jul 24 15:40:55
>> > 2014
>> > @@ -804,6 +804,8 @@ void AsmPrinter::EmitFunctionBody() {
>> > }
>> > }
>> > }
>> > +
>> > + EmitBasicBlockEnd(MBB);
>> > }
>> >
>> > // If the last instruction was a prolog label, then we have a
>> > situation where
>> >
>> > Modified: llvm/trunk/lib/Target/X86/X86AsmPrinter.cpp
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86AsmPrinter.cpp?rev=213892&r1=213891&r2=213892&view=diff
>> >
>> > ==============================================================================
>> > --- llvm/trunk/lib/Target/X86/X86AsmPrinter.cpp (original)
>> > +++ llvm/trunk/lib/Target/X86/X86AsmPrinter.cpp Thu Jul 24 15:40:55 2014
>> > @@ -47,6 +47,8 @@ using namespace llvm;
>> > /// runOnMachineFunction - Emit the function body.
>> > ///
>> > bool X86AsmPrinter::runOnMachineFunction(MachineFunction &MF) {
>> > + SMShadowTracker.startFunction(MF);
>> > +
>> > SetupMachineFunction(MF);
>> >
>> > if (Subtarget->isTargetCOFF()) {
>> >
>> > Modified: llvm/trunk/lib/Target/X86/X86AsmPrinter.h
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86AsmPrinter.h?rev=213892&r1=213891&r2=213892&view=diff
>> >
>> > ==============================================================================
>> > --- llvm/trunk/lib/Target/X86/X86AsmPrinter.h (original)
>> > +++ llvm/trunk/lib/Target/X86/X86AsmPrinter.h Thu Jul 24 15:40:55 2014
>> > @@ -15,6 +15,11 @@
>> > #include "llvm/CodeGen/StackMaps.h"
>> > #include "llvm/Target/TargetMachine.h"
>> >
>> > +// Implemented in X86MCInstLower.cpp
>> > +namespace {
>> > + class X86MCInstLower;
>> > +}
>> > +
>> > namespace llvm {
>> > class MCStreamer;
>> > class MCSymbol;
>> > @@ -25,9 +30,52 @@ class LLVM_LIBRARY_VISIBILITY X86AsmPrin
>> >
>> > void GenerateExportDirective(const MCSymbol *Sym, bool IsData);
>> >
>> > + // This utility class tracks the length of a stackmap instruction's
>> > 'shadow'.
>> > + // It is used by the X86AsmPrinter to ensure that the stackmap shadow
>> > + // invariants (i.e. no other stackmaps, patchpoints, or control flow
>> > within
>> > + // the shadow) are met, while outputting a minimal number of NOPs for
>> > padding.
>> > + //
>> > + // To minimise the number of NOPs used, the shadow tracker counts the
>> > number
>> > + // of instruction bytes output since the last stackmap. Only if there
>> > are too
>> > + // few instruction bytes to cover the shadow are NOPs used for
>> > padding.
>> > + class StackMapShadowTracker {
>> > + public:
>> > + StackMapShadowTracker(TargetMachine &TM);
>> > + ~StackMapShadowTracker();
>> > + void startFunction(MachineFunction &MF);
>> > + void count(MCInst &Inst, const MCSubtargetInfo &STI);
>> > + void reset(unsigned RequiredSize) {
>> > + RequiredShadowSize = RequiredSize;
>> > + CurrentShadowSize = 0;
>> > + Count = true;
>> > + }
>> > + void emitShadowPadding(MCStreamer &OutStreamer, const
>> > MCSubtargetInfo &STI);
>> > + private:
>> > + TargetMachine &TM;
>> > + std::unique_ptr<MCCodeEmitter> CodeEmitter;
>> > + bool Count;
>> > + unsigned RequiredShadowSize, CurrentShadowSize;
>> > + };
>> > +
>> > + StackMapShadowTracker SMShadowTracker;
>> > +
>> > + // All instructions emitted by the X86AsmPrinter should use this
>> > helper
>> > + // method.
>> > + //
>> > + // This helper function invokes the SMShadowTracker on each
>> > instruction before
>> > + // outputting it to the OutStream. This allows the shadow tracker to
>> > minimise
>> > + // the number of NOPs used for stackmap padding.
>> > + void EmitAndCountInstruction(MCInst &Inst);
>> > +
>> > + void InsertStackMapShadows(MachineFunction &MF);
>> > + void LowerSTACKMAP(const MachineInstr &MI);
>> > + void LowerPATCHPOINT(const MachineInstr &MI);
>> > +
>> > + void LowerTlsAddr(X86MCInstLower &MCInstLowering, const MachineInstr
>> > &MI);
>> > +
>> > public:
>> > explicit X86AsmPrinter(TargetMachine &TM, MCStreamer &Streamer)
>> > - : AsmPrinter(TM, Streamer), SM(*this) {
>> > + : AsmPrinter(TM, Streamer), SM(*this), SMShadowTracker(TM) {
>> > Subtarget = &TM.getSubtarget<X86Subtarget>();
>> > }
>> >
>> > @@ -43,6 +91,10 @@ class LLVM_LIBRARY_VISIBILITY X86AsmPrin
>> >
>> > void EmitInstruction(const MachineInstr *MI) override;
>> >
>> > + void EmitBasicBlockEnd(const MachineBasicBlock &MBB) override {
>> > + SMShadowTracker.emitShadowPadding(OutStreamer, getSubtargetInfo());
>> > + }
>> > +
>> > bool PrintAsmOperand(const MachineInstr *MI, unsigned OpNo,
>> > unsigned AsmVariant, const char *ExtraCode,
>> > raw_ostream &OS) override;
>> >
>> > Modified: llvm/trunk/lib/Target/X86/X86MCInstLower.cpp
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86MCInstLower.cpp?rev=213892&r1=213891&r2=213892&view=diff
>> >
>> > ==============================================================================
>> > --- llvm/trunk/lib/Target/X86/X86MCInstLower.cpp (original)
>> > +++ llvm/trunk/lib/Target/X86/X86MCInstLower.cpp Thu Jul 24 15:40:55
>> > 2014
>> > @@ -24,12 +24,14 @@
>> > #include "llvm/IR/GlobalValue.h"
>> > #include "llvm/IR/Mangler.h"
>> > #include "llvm/MC/MCAsmInfo.h"
>> > +#include "llvm/MC/MCCodeEmitter.h"
>> > #include "llvm/MC/MCContext.h"
>> > #include "llvm/MC/MCExpr.h"
>> > #include "llvm/MC/MCInst.h"
>> > #include "llvm/MC/MCInstBuilder.h"
>> > #include "llvm/MC/MCStreamer.h"
>> > #include "llvm/MC/MCSymbol.h"
>> > +#include "llvm/Support/TargetRegistry.h"
>> > using namespace llvm;
>> >
>> > namespace {
>> > @@ -58,6 +60,52 @@ private:
>> >
>> > } // end anonymous namespace
>> >
>> > +// Emit a minimal sequence of nops spanning NumBytes bytes.
>> > +static void EmitNops(MCStreamer &OS, unsigned NumBytes, bool Is64Bit,
>> > + const MCSubtargetInfo &STI);
>> > +
>> > +namespace llvm {
>> > +
>> > X86AsmPrinter::StackMapShadowTracker::StackMapShadowTracker(TargetMachine
>> > &TM)
>> > + : TM(TM), Count(false), RequiredShadowSize(0),
>> > CurrentShadowSize(0) {}
>> > +
>> > + X86AsmPrinter::StackMapShadowTracker::~StackMapShadowTracker() {}
>> > +
>> > + void
>> > + X86AsmPrinter::StackMapShadowTracker::startFunction(MachineFunction
>> > &MF) {
>> > +
>> > CodeEmitter.reset(TM.getTarget().createMCCodeEmitter(*TM.getInstrInfo(),
>> > +
>> > *TM.getRegisterInfo(),
>> > +
>> > *TM.getSubtargetImpl(),
>> > +
>> > MF.getContext()));
>> > + }
>> > +
>> > + void X86AsmPrinter::StackMapShadowTracker::count(MCInst &Inst,
>> > + const
>> > MCSubtargetInfo &STI) {
>> > + if (Count) {
>> > + SmallString<256> Code;
>> > + SmallVector<MCFixup, 4> Fixups;
>> > + raw_svector_ostream VecOS(Code);
>> > + CodeEmitter->EncodeInstruction(Inst, VecOS, Fixups, STI);
>> > + VecOS.flush();
>> > + CurrentShadowSize += Code.size();
>> > + if (CurrentShadowSize >= RequiredShadowSize)
>> > + Count = false; // The shadow is big enough. Stop counting.
>> > + }
>> > + }
>> > +
>> > + void X86AsmPrinter::StackMapShadowTracker::emitShadowPadding(
>> > + MCStreamer &OutStreamer, const MCSubtargetInfo &STI) {
>> > + if (Count && CurrentShadowSize < RequiredShadowSize)
>> > + EmitNops(OutStreamer, RequiredShadowSize - CurrentShadowSize,
>> > + TM.getSubtarget<X86Subtarget>().is64Bit(), STI);
>> > + Count = false;
>> > + }
>> > +
>> > + void X86AsmPrinter::EmitAndCountInstruction(MCInst &Inst) {
>> > + OutStreamer.EmitInstruction(Inst, getSubtargetInfo());
>> > + SMShadowTracker.count(Inst, getSubtargetInfo());
>> > + }
>> > +} // end llvm namespace
>> > +
>> > X86MCInstLower::X86MCInstLower(const MachineFunction &mf,
>> > X86AsmPrinter &asmprinter)
>> > : Ctx(mf.getContext()), MF(mf), TM(mf.getTarget()),
>> > @@ -602,10 +650,8 @@ ReSimplify:
>> > }
>> > }
>> >
>> > -static void LowerTlsAddr(MCStreamer &OutStreamer,
>> > - X86MCInstLower &MCInstLowering,
>> > - const MachineInstr &MI,
>> > - const MCSubtargetInfo& STI) {
>> > +void X86AsmPrinter::LowerTlsAddr(X86MCInstLower &MCInstLowering,
>> > + const MachineInstr &MI) {
>> >
>> > bool is64Bits = MI.getOpcode() == X86::TLS_addr64 ||
>> > MI.getOpcode() == X86::TLS_base_addr64;
>> > @@ -615,7 +661,7 @@ static void LowerTlsAddr(MCStreamer &Out
>> > MCContext &context = OutStreamer.getContext();
>> >
>> > if (needsPadding)
>> > - OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX),
>> > STI);
>> > + EmitAndCountInstruction(MCInstBuilder(X86::DATA16_PREFIX));
>> >
>> > MCSymbolRefExpr::VariantKind SRVK;
>> > switch (MI.getOpcode()) {
>> > @@ -662,12 +708,12 @@ static void LowerTlsAddr(MCStreamer &Out
>> > LEA.addOperand(MCOperand::CreateExpr(symRef)); // disp
>> > LEA.addOperand(MCOperand::CreateReg(0)); // seg
>> > }
>> > - OutStreamer.EmitInstruction(LEA, STI);
>> > + EmitAndCountInstruction(LEA);
>> >
>> > if (needsPadding) {
>> > - OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX),
>> > STI);
>> > - OutStreamer.EmitInstruction(MCInstBuilder(X86::DATA16_PREFIX),
>> > STI);
>> > - OutStreamer.EmitInstruction(MCInstBuilder(X86::REX64_PREFIX), STI);
>> > + EmitAndCountInstruction(MCInstBuilder(X86::DATA16_PREFIX));
>> > + EmitAndCountInstruction(MCInstBuilder(X86::DATA16_PREFIX));
>> > + EmitAndCountInstruction(MCInstBuilder(X86::REX64_PREFIX));
>> > }
>> >
>> > StringRef name = is64Bits ? "__tls_get_addr" : "___tls_get_addr";
>> > @@ -677,9 +723,9 @@ static void LowerTlsAddr(MCStreamer &Out
>> > MCSymbolRefExpr::VK_PLT,
>> > context);
>> >
>> > - OutStreamer.EmitInstruction(MCInstBuilder(is64Bits ?
>> > X86::CALL64pcrel32
>> > - :
>> > X86::CALLpcrel32)
>> > - .addExpr(tlsRef), STI);
>> > + EmitAndCountInstruction(MCInstBuilder(is64Bits ? X86::CALL64pcrel32
>> > + : X86::CALLpcrel32)
>> > + .addExpr(tlsRef));
>> > }
>> >
>> > /// \brief Emit the optimal amount of multi-byte nops on X86.
>> > @@ -725,10 +771,9 @@ static void EmitNops(MCStreamer &OS, uns
>> > break;
>> > case X86::NOOPL:
>> > case X86::NOOPW:
>> > -
>> > OS.EmitInstruction(MCInstBuilder(Opc).addReg(BaseReg).addImm(ScaleVal)
>> > - .addReg(IndexReg)
>> > - .addImm(Displacement)
>> > - .addReg(SegmentReg), STI);
>> > + OS.EmitInstruction(MCInstBuilder(Opc).addReg(BaseReg)
>> > + .addImm(ScaleVal).addReg(IndexReg)
>> > + .addImm(Displacement).addReg(SegmentReg),
>> > STI);
>> > break;
>> > }
>> > } // while (NumBytes)
>> > @@ -736,22 +781,20 @@ static void EmitNops(MCStreamer &OS, uns
>> >
>> > // Lower a stackmap of the form:
>> > // <id>, <shadowBytes>, ...
>> > -static void LowerSTACKMAP(MCStreamer &OS, StackMaps &SM,
>> > - const MachineInstr &MI, bool Is64Bit, const
>> > MCSubtargetInfo& STI) {
>> > - unsigned NumBytes = MI.getOperand(1).getImm();
>> > +void X86AsmPrinter::LowerSTACKMAP(const MachineInstr &MI) {
>> > + SMShadowTracker.emitShadowPadding(OutStreamer, getSubtargetInfo());
>> > SM.recordStackMap(MI);
>> > - // Emit padding.
>> > - // FIXME: These nops ensure that the stackmap's shadow is covered by
>> > - // instructions from the same basic block, but the nops should not be
>> > - // necessary if instructions from the same block follow the stackmap.
>> > - EmitNops(OS, NumBytes, Is64Bit, STI);
>> > + unsigned NumShadowBytes = MI.getOperand(1).getImm();
>> > + SMShadowTracker.reset(NumShadowBytes);
>> > }
>> >
>> > // Lower a patchpoint of the form:
>> > // [<def>], <id>, <numBytes>, <target>, <numArgs>, <cc>, ...
>> > -static void LowerPATCHPOINT(MCStreamer &OS, StackMaps &SM,
>> > - const MachineInstr &MI, bool Is64Bit, const
>> > MCSubtargetInfo& STI) {
>> > - assert(Is64Bit && "Patchpoint currently only supports X86-64");
>> > +void X86AsmPrinter::LowerPATCHPOINT(const MachineInstr &MI) {
>> > + assert(Subtarget->is64Bit() && "Patchpoint currently only supports
>> > X86-64");
>> > +
>> > + SMShadowTracker.emitShadowPadding(OutStreamer, getSubtargetInfo());
>> > +
>> > SM.recordPatchPoint(MI);
>> >
>> > PatchPointOpers opers(&MI);
>> > @@ -766,16 +809,17 @@ static void LowerPATCHPOINT(MCStreamer &
>> > EncodedBytes = 13;
>> > else
>> > EncodedBytes = 12;
>> > - OS.EmitInstruction(MCInstBuilder(X86::MOV64ri).addReg(ScratchReg)
>> > - .addImm(CallTarget),
>> > STI);
>> > - OS.EmitInstruction(MCInstBuilder(X86::CALL64r).addReg(ScratchReg),
>> > STI);
>> > +
>> > EmitAndCountInstruction(MCInstBuilder(X86::MOV64ri).addReg(ScratchReg)
>> > +
>> > .addImm(CallTarget));
>> > +
>> > EmitAndCountInstruction(MCInstBuilder(X86::CALL64r).addReg(ScratchReg));
>> > }
>> > // Emit padding.
>> > unsigned NumBytes =
>> > opers.getMetaOper(PatchPointOpers::NBytesPos).getImm();
>> > assert(NumBytes >= EncodedBytes &&
>> > "Patchpoint can't request size less than the length of a
>> > call.");
>> >
>> > - EmitNops(OS, NumBytes - EncodedBytes, Is64Bit, STI);
>> > + EmitNops(OutStreamer, NumBytes - EncodedBytes, Subtarget->is64Bit(),
>> > + getSubtargetInfo());
>> > }
>> >
>> > void X86AsmPrinter::EmitInstruction(const MachineInstr *MI) {
>> > @@ -812,7 +856,7 @@ void X86AsmPrinter::EmitInstruction(cons
>> > case X86::TLS_addr64:
>> > case X86::TLS_base_addr32:
>> > case X86::TLS_base_addr64:
>> > - return LowerTlsAddr(OutStreamer, MCInstLowering, *MI,
>> > getSubtargetInfo());
>> > + return LowerTlsAddr(MCInstLowering, *MI);
>> >
>> > case X86::MOVPC32r: {
>> > // This is a pseudo op for a two instruction sequence with a label,
>> > which
>> > @@ -825,15 +869,15 @@ void X86AsmPrinter::EmitInstruction(cons
>> > MCSymbol *PICBase = MF->getPICBaseSymbol();
>> > // FIXME: We would like an efficient form for this, so we don't
>> > have to do a
>> > // lot of extra uniquing.
>> > - EmitToStreamer(OutStreamer, MCInstBuilder(X86::CALLpcrel32)
>> > + EmitAndCountInstruction(MCInstBuilder(X86::CALLpcrel32)
>> > .addExpr(MCSymbolRefExpr::Create(PICBase, OutContext)));
>> >
>> > // Emit the label.
>> > OutStreamer.EmitLabel(PICBase);
>> >
>> > // popl $reg
>> > - EmitToStreamer(OutStreamer, MCInstBuilder(X86::POP32r)
>> > - .addReg(MI->getOperand(0).getReg()));
>> > + EmitAndCountInstruction(MCInstBuilder(X86::POP32r)
>> > + .addReg(MI->getOperand(0).getReg()));
>> > return;
>> > }
>> >
>> > @@ -863,7 +907,7 @@ void X86AsmPrinter::EmitInstruction(cons
>> > DotExpr =
>> > MCBinaryExpr::CreateAdd(MCSymbolRefExpr::Create(OpSym,OutContext),
>> > DotExpr, OutContext);
>> >
>> > - EmitToStreamer(OutStreamer, MCInstBuilder(X86::ADD32ri)
>> > + EmitAndCountInstruction(MCInstBuilder(X86::ADD32ri)
>> > .addReg(MI->getOperand(0).getReg())
>> > .addReg(MI->getOperand(1).getReg())
>> > .addExpr(DotExpr));
>> > @@ -871,21 +915,21 @@ void X86AsmPrinter::EmitInstruction(cons
>> > }
>> >
>> > case TargetOpcode::STACKMAP:
>> > - return LowerSTACKMAP(OutStreamer, SM, *MI, Subtarget->is64Bit(),
>> > getSubtargetInfo());
>> > + return LowerSTACKMAP(*MI);
>> >
>> > case TargetOpcode::PATCHPOINT:
>> > - return LowerPATCHPOINT(OutStreamer, SM, *MI, Subtarget->is64Bit(),
>> > getSubtargetInfo());
>> > + return LowerPATCHPOINT(*MI);
>> >
>> > case X86::MORESTACK_RET:
>> > - EmitToStreamer(OutStreamer,
>> > MCInstBuilder(getRetOpcode(*Subtarget)));
>> > + EmitAndCountInstruction(MCInstBuilder(getRetOpcode(*Subtarget)));
>> > return;
>> >
>> > case X86::MORESTACK_RET_RESTORE_R10:
>> > // Return, then restore R10.
>> > - EmitToStreamer(OutStreamer,
>> > MCInstBuilder(getRetOpcode(*Subtarget)));
>> > - EmitToStreamer(OutStreamer, MCInstBuilder(X86::MOV64rr)
>> > - .addReg(X86::R10)
>> > - .addReg(X86::RAX));
>> > + EmitAndCountInstruction(MCInstBuilder(getRetOpcode(*Subtarget)));
>> > + EmitAndCountInstruction(MCInstBuilder(X86::MOV64rr)
>> > + .addReg(X86::R10)
>> > + .addReg(X86::RAX));
>> > return;
>> >
>> > case X86::SEH_PushReg:
>> > @@ -922,5 +966,5 @@ void X86AsmPrinter::EmitInstruction(cons
>> >
>> > MCInst TmpInst;
>> > MCInstLowering.Lower(MI, TmpInst);
>> > - EmitToStreamer(OutStreamer, TmpInst);
>> > + EmitAndCountInstruction(TmpInst);
>> > }
>> >
>> > Modified: llvm/trunk/test/CodeGen/X86/stackmap-nops.ll
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/stackmap-nops.ll?rev=213892&r1=213891&r2=213892&view=diff
>> >
>> > ==============================================================================
>> > --- llvm/trunk/test/CodeGen/X86/stackmap-nops.ll (original)
>> > +++ llvm/trunk/test/CodeGen/X86/stackmap-nops.ll Thu Jul 24 15:40:55
>> > 2014
>> > @@ -224,6 +224,7 @@ entry:
>> > tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 28,
>> > i32 28)
>> > tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 29,
>> > i32 29)
>> > tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 30,
>> > i32 30)
>> > + tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 31,
>> > i32 0)
>> > ret void
>> > }
>> >
>> >
>> > Added: llvm/trunk/test/CodeGen/X86/stackmap-shadow-optimization.ll
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/stackmap-shadow-optimization.ll?rev=213892&view=auto
>> >
>> > ==============================================================================
>> > --- llvm/trunk/test/CodeGen/X86/stackmap-shadow-optimization.ll (added)
>> > +++ llvm/trunk/test/CodeGen/X86/stackmap-shadow-optimization.ll Thu Jul
>> > 24 15:40:55 2014
>> > @@ -0,0 +1,21 @@
>> > +; RUN: llc < %s -mtriple=x86_64-apple-darwin -mcpu=corei7
>> > -disable-fp-elim | FileCheck %s
>> > +
>> > +; Check that the X86 stackmap shadow optimization is only outputting a
>> > 1-byte
>> > +; nop here. 8-bytes are requested, but 7 are covered by the code for
>> > the call to
>> > +; bar, the frame teardown and the return.
>> > +define void @shadow_optimization_test() {
>> > +entry:
>> > +; CHECK-LABEL: shadow_optimization_test:
>> > +; CHECK: callq _bar
>> > +; CHECK-NOT: nop
>> > +; CHECK: callq _bar
>> > +; CHECK: retq
>> > +; CHECK: nop
>> > + call void @bar()
>> > + tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 0,
>> > i32 8)
>> > + call void @bar()
>> > + ret void
>> > +}
>> > +
>> > +declare void @llvm.experimental.stackmap(i64, i32, ...)
>> > +declare void @bar()
>> > \ No newline at end of file
>> >
>> > Modified: llvm/trunk/test/MC/X86/stackmap-nops.ll
>> > URL:
>> > http://llvm.org/viewvc/llvm-project/llvm/trunk/test/MC/X86/stackmap-nops.ll?rev=213892&r1=213891&r2=213892&view=diff
>> >
>> > ==============================================================================
>> > --- llvm/trunk/test/MC/X86/stackmap-nops.ll (original)
>> > +++ llvm/trunk/test/MC/X86/stackmap-nops.ll Thu Jul 24 15:40:55 2014
>> > @@ -41,6 +41,7 @@ entry:
>> > tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 13,
>> > i32 13)
>> > tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 14,
>> > i32 14)
>> > tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 15,
>> > i32 15)
>> > + tail call void (i64, i32, ...)* @llvm.experimental.stackmap(i64 16,
>> > i32 0)
>> > ret void
>> > }
>> >
>> >
>> >
>> > _______________________________________________
>> > llvm-commits mailing list
>> > llvm-commits at cs.uiuc.edu
>> > http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>
>
More information about the llvm-commits
mailing list