[llvm] r317292 - Avoid PLT for external calls when attribute nonlazybind is used.

[Sriraman Tallam via llvm-commits]

On Thu, Nov 2, 2017 at 5:43 PM, Rafael Avila de Espindola
<rafael.espindola at gmail.com> wrote:
> Why force a call via the got when a direct one could be used?
>
> With -fno-plt -fPIC gcc compiles
>
> __attribute__((visibility("hidden"))) void foo(void);
> void bar(void);
> void zed(void) {
>   foo();
>   bar();
> }
>
> to
>
>         call    foo
>         call    *bar at GOTPCREL(%rip)
>
> But this patch will use the got for both calls, no?

I did not test this case and if it is true that foo can never be
external in this case then I got it wrong.  But what if foo was not
defined by the DSO?

>
> I think the idea of -fno-plt is to use a got when a plt would have been
> used. If a plt was not going to be used, the option has no effect.
>
> Cheers,
> Rafael
>
> Sriraman Tallam via llvm-commits <llvm-commits at lists.llvm.org> writes:
>
>> Author: tmsriram
>> Date: Thu Nov  2 17:10:19 2017
>> New Revision: 317292
>>
>> URL: http://llvm.org/viewvc/llvm-project?rev=317292&view=rev
>> Log:
>> Avoid PLT for external calls when attribute nonlazybind is used.
>>
>> Differential Revision: https://reviews.llvm.org/D39065
>>
>> Added:
>>     llvm/trunk/test/CodeGen/X86/no-plt.ll
>> Modified:
>>     llvm/trunk/lib/Target/X86/X86Subtarget.cpp
>>
>> Modified: llvm/trunk/lib/Target/X86/X86Subtarget.cpp
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/X86/X86Subtarget.cpp?rev=317292&r1=317291&r2=317292&view=diff
>> ==============================================================================
>> --- llvm/trunk/lib/Target/X86/X86Subtarget.cpp (original)
>> +++ llvm/trunk/lib/Target/X86/X86Subtarget.cpp Thu Nov  2 17:10:19 2017
>> @@ -144,6 +144,15 @@ X86Subtarget::classifyGlobalFunctionRefe
>>  unsigned char
>>  X86Subtarget::classifyGlobalFunctionReference(const GlobalValue *GV,
>>                                                const Module &M) const {
>> +  const Function *F = dyn_cast_or_null<Function>(GV);
>> +
>> +  // Do not use the PLT when explicitly told to do so for ELF 64-bit
>> +  // target.
>> +  if (isTargetELF() && is64Bit() && F &&
>> +      F->hasFnAttribute(Attribute::NonLazyBind) &&
>> +      GV->isDeclarationForLinker())
>> +    return X86II::MO_GOTPCREL;
>> +
>>    if (TM.shouldAssumeDSOLocal(M, GV))
>>      return X86II::MO_NO_FLAG;
>>
>> @@ -153,8 +162,6 @@ X86Subtarget::classifyGlobalFunctionRefe
>>      return X86II::MO_DLLIMPORT;
>>    }
>>
>> -  const Function *F = dyn_cast_or_null<Function>(GV);
>> -
>>    if (isTargetELF()) {
>>      if (is64Bit() && F && (CallingConv::X86_RegCall == F->getCallingConv()))
>>        // According to psABI, PLT stub clobbers XMM8-XMM15.
>>
>> Added: llvm/trunk/test/CodeGen/X86/no-plt.ll
>> URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/no-plt.ll?rev=317292&view=auto
>> ==============================================================================
>> --- llvm/trunk/test/CodeGen/X86/no-plt.ll (added)
>> +++ llvm/trunk/test/CodeGen/X86/no-plt.ll Thu Nov  2 17:10:19 2017
>> @@ -0,0 +1,23 @@
>> +; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnu -relocation-model=pic \
>> +; RUN:   | FileCheck -check-prefix=X64 %s
>> +; RUN: llc < %s -mcpu=generic -mtriple=x86_64-linux-gnu \
>> +; RUN:   | FileCheck -check-prefix=X64 %s
>> +
>> +define i32 @main() #0 {
>> +; X64: callq *_Z3foov at GOTPCREL(%rip)
>> +; X64: callq _Z3barv
>> +
>> +entry:
>> +  %retval = alloca i32, align 4
>> +  store i32 0, i32* %retval, align 4
>> +  %call1 = call i32 @_Z3foov()
>> +  %call2 = call i32 @_Z3barv()
>> +  ret i32 0
>> +}
>> +
>> +; Function Attrs: nonlazybind
>> +declare i32 @_Z3foov() #1
>> +
>> +declare i32 @_Z3barv() #2
>> +
>> +attributes #1 = { nonlazybind }
>>
>>
>> _______________________________________________
>> llvm-commits mailing list
>> llvm-commits at lists.llvm.org
>> http://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-commits