[llvm-commits] [llvm] r168320 - in /llvm/trunk: lib/CodeGen/LiveInterval.cpp test/CodeGen/X86/inline-asm.ll

[Chris Lattner]

On Nov 21, 2012, at 11:14 PM, Kai <kai at redstar.de> wrote:

> On 22.11.2012 07:48, Anton Korobeynikov wrote:
>> Is this a regression? Code owner should decide whether this patch will
>> be considered for inclusion to release branch or not.
> 
> Yes, it is a regression. Worked in LLVM 3.0 and 3.1 but not 3.2rc1.

Jakob, do you think it makes sense to pull your patch into the 3.2 release?

-Chris

> 
> Kai
> 
>> 
>> On Wed, Nov 21, 2012 at 11:33 PM, Kai <kai at redstar.de> wrote:
>>> Hi!
>>> 
>>> Will this commit be merged into the release_32 stream? PR14376 was raised to
>>> importance "release blocker".
>>> 
>>> Thanks!!!!!
>>> 
>>> Regards
>>> Kai
>>> 
>>> 
>>> 
>>> On 19.11.2012 20:31, Jakob Stoklund Olesen wrote:
>>>> 
>>>> Author: stoklund
>>>> Date: Mon Nov 19 13:31:10 2012
>>>> New Revision: 168320
>>>> 
>>>> URL: http://llvm.org/viewvc/llvm-project?rev=168320&view=rev
>>>> Log:
>>>> Handle mixed normal and early-clobber defs on inline asm.
>>>> 
>>>> PR14376.
>>>> 
>>>> Modified:
>>>>      llvm/trunk/lib/CodeGen/LiveInterval.cpp
>>>>      llvm/trunk/test/CodeGen/X86/inline-asm.ll
>>>> 
>>>> Modified: llvm/trunk/lib/CodeGen/LiveInterval.cpp
>>>> URL:
>>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/CodeGen/LiveInterval.cpp?rev=168320&r1=168319&r2=168320&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/lib/CodeGen/LiveInterval.cpp (original)
>>>> +++ llvm/trunk/lib/CodeGen/LiveInterval.cpp Mon Nov 19 13:31:10 2012
>>>> @@ -59,8 +59,16 @@
>>>>       return VNI;
>>>>     }
>>>>     if (SlotIndex::isSameInstr(Def, I->start)) {
>>>> -    assert(I->start == Def && "Cannot insert def, already live");
>>>> -    assert(I->valno->def == Def && "Inconsistent existing value def");
>>>> +    assert(I->valno->def == I->start && "Inconsistent existing value
>>>> def");
>>>> +
>>>> +    // It is possible to have both normal and early-clobber defs of the
>>>> same
>>>> +    // register on an instruction. It doesn't make a lot of sense, but it
>>>> is
>>>> +    // possible to specify in inline assembly.
>>>> +    //
>>>> +    // Just convert everything to early-clobber.
>>>> +    Def = std::min(Def, I->start);
>>>> +    if (Def != I->start)
>>>> +      I->start = I->valno->def = Def;
>>>>       return I->valno;
>>>>     }
>>>>     assert(SlotIndex::isEarlierInstr(Def, I->start) && "Already live at
>>>> def");
>>>> 
>>>> Modified: llvm/trunk/test/CodeGen/X86/inline-asm.ll
>>>> URL:
>>>> http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/X86/inline-asm.ll?rev=168320&r1=168319&r2=168320&view=diff
>>>> 
>>>> ==============================================================================
>>>> --- llvm/trunk/test/CodeGen/X86/inline-asm.ll (original)
>>>> +++ llvm/trunk/test/CodeGen/X86/inline-asm.ll Mon Nov 19 13:31:10 2012
>>>> @@ -52,3 +52,10 @@
>>>>     %0 = call { i32, i32, i32, i32, i32 } asm sideeffect "",
>>>> "=&r,=&r,=&r,=&r,=&q,r,~{ecx},~{memory},~{dirflag},~{fpsr},~{flags}"(i8* %h)
>>>> nounwind
>>>>     ret void
>>>>   }
>>>> +
>>>> +; Mix normal and EC defs of the same register.
>>>> +define i32 @pr14376() nounwind noinline {
>>>> +entry:
>>>> +  %asm = tail call i32 asm sideeffect "",
>>>> "={ax},i,~{eax},~{flags},~{rax}"(i64 61) nounwind
>>>> +  ret i32 %asm
>>>> +}
>>>> 
>>> 
>>> 
>>> _______________________________________________
>>> llvm-commits mailing list
>>> llvm-commits at cs.uiuc.edu
>>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>> 
>> 
>> 
> 
> 
> _______________________________________________
> llvm-commits mailing list
> llvm-commits at cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits