[PATCH] D42216: Use New Module Metadata String "AvoidPLT" to avoid calls via PLT
Sriraman Tallam via llvm-commits
llvm-commits at lists.llvm.org
Mon Jan 29 12:21:35 PST 2018
On Mon, Jan 29, 2018 at 11:47 AM, Rafael Avila de Espindola
<rafael.espindola at gmail.com> wrote:
> The attached patch includes my suggestions and cleans up the testcase to
> have less unrelated information.
Looks good to me, updating with your patch.
>
> Cheers,
> Rafael
>
>
> diff --git a/include/llvm/IR/Module.h b/include/llvm/IR/Module.h
> index 196e32e3615..49bdb577a62 100644
> --- a/include/llvm/IR/Module.h
> +++ b/include/llvm/IR/Module.h
> @@ -838,6 +838,12 @@ public:
> Metadata *getProfileSummary();
> /// @}
>
> + /// Returns true if PLT should be avoided for RTLib calls.
> + bool getRtLibUseGOT() const;
> +
> + /// Set that PLT should be avoid for RTLib calls.
> + void setRtLibUseGOT();
> +
> /// Take ownership of the given memory buffer.
> void setOwnedMemoryBuffer(std::unique_ptr<MemoryBuffer> MB);
> };
> diff --git a/lib/IR/Module.cpp b/lib/IR/Module.cpp
> index c230a50044c..9f1e6096932 100644
> --- a/lib/IR/Module.cpp
> +++ b/lib/IR/Module.cpp
> @@ -510,6 +510,15 @@ void Module::setOwnedMemoryBuffer(std::unique_ptr<MemoryBuffer> MB) {
> OwnedMemoryBuffer = std::move(MB);
> }
>
> +bool Module::getRtLibUseGOT() const {
> + auto *Val = cast_or_null<ConstantAsMetadata>(getModuleFlag("RtLibUseGOT"));
> + return Val && (cast<ConstantInt>(Val->getValue())->getZExtValue() > 0);
> +}
> +
> +void Module::setRtLibUseGOT() {
> + addModuleFlag(ModFlagBehavior::Max, "RtLibUseGOT", 1);
> +}
> +
> GlobalVariable *llvm::collectUsedGlobalVariables(
> const Module &M, SmallPtrSetImpl<GlobalValue *> &Set, bool CompilerUsed) {
> const char *Name = CompilerUsed ? "llvm.compiler.used" : "llvm.used";
> diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
> index 925796cae7a..f4d503b9934 100644
> --- a/lib/Target/X86/X86ISelLowering.cpp
> +++ b/lib/Target/X86/X86ISelLowering.cpp
> @@ -3761,6 +3761,14 @@ X86TargetLowering::LowerCall(TargetLowering::CallLoweringInfo &CLI,
>
> Callee = DAG.getTargetExternalSymbol(
> S->getSymbol(), getPointerTy(DAG.getDataLayout()), OpFlags);
> +
> + if (OpFlags == X86II::MO_GOTPCREL) {
> + Callee = DAG.getNode(X86ISD::WrapperRIP, dl,
> + getPointerTy(DAG.getDataLayout()), Callee);
> + Callee = DAG.getLoad(
> + getPointerTy(DAG.getDataLayout()), dl, DAG.getEntryNode(), Callee,
> + MachinePointerInfo::getGOT(DAG.getMachineFunction()));
> + }
> } else if (Subtarget.isTarget64BitILP32() &&
> Callee->getValueType(0) == MVT::i32) {
> // Zero-extend the 32-bit Callee address into a 64-bit according to x32 ABI
> diff --git a/lib/Target/X86/X86Subtarget.cpp b/lib/Target/X86/X86Subtarget.cpp
> index 217b22c2bf7..9cfcd1540fc 100644
> --- a/lib/Target/X86/X86Subtarget.cpp
> +++ b/lib/Target/X86/X86Subtarget.cpp
> @@ -157,7 +157,10 @@ X86Subtarget::classifyGlobalFunctionReference(const GlobalValue *GV,
> // In Regcall calling convention those registers are used for passing
> // parameters. Thus we need to prevent lazy binding in Regcall.
> return X86II::MO_GOTPCREL;
> - if (F && F->hasFnAttribute(Attribute::NonLazyBind) && is64Bit())
> + // If PLT must be avoided then the call should be via GOTPCREL.
> + if (((F && F->hasFnAttribute(Attribute::NonLazyBind)) ||
> + (!F && M.getRtLibUseGOT())) &&
> + is64Bit())
> return X86II::MO_GOTPCREL;
> return X86II::MO_PLT;
> }
> diff --git a/test/CodeGen/X86/no-plt.ll b/test/CodeGen/X86/no-plt.ll
> index 9688a2e6f5b..8aa60f60169 100644
> --- a/test/CodeGen/X86/no-plt.ll
> +++ b/test/CodeGen/X86/no-plt.ll
> @@ -3,6 +3,18 @@
> ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnu \
> ; RUN: | FileCheck -check-prefix=X64 --check-prefix=STATIC %s
>
> +define void @memset_call(i8* nocapture %a, i8 %c, i32 %n) {
> +; X64: callq *memset at GOTPCREL(%rip)
> + call void @llvm.memset.p0i8.i32(i8* %a, i8 %c, i32 %n, i1 false)
> + ret void
> +}
> +
> +define void @memcpy_call(i8* nocapture %a, i8* nocapture readonly %b, i64 %n) {
> +; X64: callq *memcpy at GOTPCREL(%rip)
> + call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %b, i64 %n, i32 1, i1 false)
> + ret void
> +}
> +
> define i32 @main() {
> ; X64: callq *foo at GOTPCREL(%rip)
> ; PIC: callq bar at PLT
> @@ -20,3 +32,8 @@ define i32 @main() {
> declare i32 @foo() nonlazybind
> declare i32 @bar()
> declare hidden i32 @baz() nonlazybind
> +declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1)
> +declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i1)
> +
> +!llvm.module.flags = !{!1}
> +!1 = !{i32 7, !"RtLibUseGOT", i32 1}
>
>
> Sriraman Tallam via Phabricator <reviews at reviews.llvm.org> writes:
>
>> tmsriram updated this revision to Diff 131514.
>> tmsriram added a comment.
>>
>> Rafael's change included and test case updated. "Avoid PLT" changed to "RtLibUseGOT".
>>
>>
>> https://reviews.llvm.org/D42216
>>
>> Files:
>> include/llvm/IR/Module.h
>> lib/IR/Module.cpp
>> lib/Target/X86/X86ISelLowering.cpp
>> test/CodeGen/X86/no-plt.ll
>>
>> Index: test/CodeGen/X86/no-plt.ll
>> ===================================================================
>> --- test/CodeGen/X86/no-plt.ll
>> +++ test/CodeGen/X86/no-plt.ll
>> @@ -3,7 +3,23 @@
>> ; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnu \
>> ; RUN: | FileCheck -check-prefix=X64 %s
>>
>> -define i32 @main() #0 {
>> +; Function Attrs: noinline nounwind uwtable
>> +define void @memset_call(i8* nocapture %a, i8 %c, i32 %n) local_unnamed_addr #0 {
>> +; X64: callq *memset at GOTPCREL(%rip)
>> +entry:
>> + call void @llvm.memset.p0i8.i32(i8* %a, i8 %c, i32 %n, i1 false)
>> + ret void
>> +}
>> +
>> +; Function Attrs: noinline nounwind uwtable
>> +define void @memcpy_call(i8* nocapture %a, i8* nocapture readonly %b, i64 %n) local_unnamed_addr #0 {
>> +; X64: callq *memcpy at GOTPCREL(%rip)
>> +entry:
>> + call void @llvm.memcpy.p0i8.p0i8.i64(i8* %a, i8* %b, i64 %n, i32 1, i1 false)
>> + ret void
>> +}
>> +
>> +define i32 @main() #1 {
>> ; X64: callq *_Z3foov at GOTPCREL(%rip)
>> ; X64: callq _Z3barv
>> ; X64: callq _Z3bazv
>> @@ -18,13 +34,21 @@
>> }
>>
>> ; Function Attrs: nonlazybind
>> -declare i32 @_Z3foov() #1
>> +declare i32 @_Z3foov() #2
>>
>> -declare i32 @_Z3barv() #2
>> +declare i32 @_Z3barv() #3
>>
>> ; Function Attrs: nonlazybind
>> -declare hidden i32 @_Z3bazv() #3
>> +declare hidden i32 @_Z3bazv() #4
>> +
>> +; Function Attrs: argmemonly nounwind
>> +declare void @llvm.memcpy.p0i8.p0i8.i64(i8* nocapture writeonly, i8* nocapture readonly, i64, i32, i1)
>> +; Function Attrs: argmemonly nounwind
>> +declare void @llvm.memset.p0i8.i32(i8* nocapture, i8, i32, i1) #0
>> +
>>
>> +attributes #2 = { nonlazybind }
>> +attributes #4 = { nonlazybind }
>>
>> -attributes #1 = { nonlazybind }
>> -attributes #3 = { nonlazybind }
>> +!llvm.module.flags = !{!1}
>> +!1 = !{i32 7, !"RtLibUseGOT", i32 1}
>> Index: lib/Target/X86/X86ISelLowering.cpp
>> ===================================================================
>> --- lib/Target/X86/X86ISelLowering.cpp
>> +++ lib/Target/X86/X86ISelLowering.cpp
>> @@ -3756,11 +3756,23 @@
>> }
>> } else if (ExternalSymbolSDNode *S = dyn_cast<ExternalSymbolSDNode>(Callee)) {
>> const Module *Mod = DAG.getMachineFunction().getFunction().getParent();
>> - unsigned char OpFlags =
>> - Subtarget.classifyGlobalFunctionReference(nullptr, *Mod);
>> + // If PLT must be avoided then the call should be via GOTPCREL.
>> + unsigned char OpFlags = X86II::MO_GOTPCREL;
>> +
>> + if (!Mod->getRtLibUseGOT()) {
>> + OpFlags = Subtarget.classifyGlobalFunctionReference(nullptr, *Mod);
>> + }
>>
>> Callee = DAG.getTargetExternalSymbol(
>> S->getSymbol(), getPointerTy(DAG.getDataLayout()), OpFlags);
>> +
>> + if (OpFlags == X86II::MO_GOTPCREL) {
>> + Callee = DAG.getNode(X86ISD::WrapperRIP, dl,
>> + getPointerTy(DAG.getDataLayout()), Callee);
>> + Callee = DAG.getLoad(
>> + getPointerTy(DAG.getDataLayout()), dl, DAG.getEntryNode(), Callee,
>> + MachinePointerInfo::getGOT(DAG.getMachineFunction()));
>> + }
>> } else if (Subtarget.isTarget64BitILP32() &&
>> Callee->getValueType(0) == MVT::i32) {
>> // Zero-extend the 32-bit Callee address into a 64-bit according to x32 ABI
>> Index: lib/IR/Module.cpp
>> ===================================================================
>> --- lib/IR/Module.cpp
>> +++ lib/IR/Module.cpp
>> @@ -510,6 +510,19 @@
>> OwnedMemoryBuffer = std::move(MB);
>> }
>>
>> +bool Module::getRtLibUseGOT() const {
>> + auto *Val = cast_or_null<ConstantAsMetadata>(getModuleFlag("RtLibUseGOT"));
>> +
>> + if (Val && (cast<ConstantInt>(Val->getValue())->getZExtValue() > 0))
>> + return true;
>> +
>> + return false;
>> +}
>> +
>> +void Module::setRtLibUseGOT() {
>> + addModuleFlag(ModFlagBehavior::Max, "RtLibUseGOT", 1);
>> +}
>> +
>> GlobalVariable *llvm::collectUsedGlobalVariables(
>> const Module &M, SmallPtrSetImpl<GlobalValue *> &Set, bool CompilerUsed) {
>> const char *Name = CompilerUsed ? "llvm.compiler.used" : "llvm.used";
>> Index: include/llvm/IR/Module.h
>> ===================================================================
>> --- include/llvm/IR/Module.h
>> +++ include/llvm/IR/Module.h
>> @@ -838,6 +838,16 @@
>> Metadata *getProfileSummary();
>> /// @}
>>
>> + /// @name Utility functions for querying and setting if PLT must be avoided.
>> +
>> + /// \brief Returns true if PLT should be avoided for RTLib calls.
>> + bool getRtLibUseGOT() const;
>> +
>> + /// \brief Set that PLT should be avoid for RTLib calls.
>> + void setRtLibUseGOT();
>> + /// @}
>> +
>> +
>> /// Take ownership of the given memory buffer.
>> void setOwnedMemoryBuffer(std::unique_ptr<MemoryBuffer> MB);
>> };
>
More information about the llvm-commits
mailing list