llvm dev - Nov 2011 - [LLVMdev] Greedy regalloc

Hi,

I get strange code when using regalloc=greedy. 

A value spill is redundant and cleared, as another spill of same value is
inserted. The former spill is however not NOP:ed, but KILL:ed, thus the operands
get a kill status. The code becomes:

    %vreg301<def> = mv32Imm 200000000, pred:0, pred:%noreg,
%CCReg<imp-def,dead>, %ac0<imp-use>, %ac1<imp-use>;
aN32_0_7:%vreg301
    Store32FI %vreg301, <fi#93>, pred:0, pred:%noreg,
%CCReg<imp-def>; mem:ST4[FixedStack93] aN32_0_7:%vreg301
    KILL %vreg301, <fi#93>, 0, %noreg, %CCReg<imp-def>;
mem:ST4[FixedStack93] aN32_0_7:%vreg301
    %a0_32<def> = COPY %vreg301; aN32_0_7:%vreg301

This results in abort:

 "Assertion `MI->allDefsAreDead() && "Def isn't really
dead"' failed."

What is the purpose of having the KILL opcode there, and do you know what might
be the cause for this error?

Are all the register allocators currently of high standard, or are one or more
less mature, meaning things might break?

/Jonas


 		 	   		  
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20111118/6806e04d/attachment.html>

Hi Jonas,
This is an unexpected failure.  I suggest filing a bug report at
http://llvm.org/bugs.  When Clang crashed it should have provided a stack trace,
preprocessed source, and a script to reproduce the error.  Please include these
items in the bug report.

 Chad


On Nov 18, 2011, at 10:03 AM, Jonas Paulsson wrote:
> Hi,
> 
> I get strange code when using regalloc=greedy. 
> 
> A value spill is redundant and cleared, as another spill of same value is
inserted. The former spill is however not NOP:ed, but KILL:ed, thus the operands
get a kill status. The code becomes:
> 
>     %vreg301<def> = mv32Imm 200000000, pred:0, pred:%noreg,
%CCReg<imp-def,dead>, %ac0<imp-use>, %ac1<imp-use>;
aN32_0_7:%vreg301
>     Store32FI %vreg301, <fi#93>, pred:0, pred:%noreg,
%CCReg<imp-def>; mem:ST4[FixedStack93] aN32_0_7:%vreg301
>     KILL %vreg301, <fi#93>, 0, %noreg, %CCReg<imp-def>;
mem:ST4[FixedStack93] aN32_0_7:%vreg301
>     %a0_32<def> = COPY %vreg301; aN32_0_7:%vreg301
> 
> This results in abort:
> 
>  "Assertion `MI->allDefsAreDead() && "Def isn't
really dead"' failed."
> 
> What is the purpose of having the KILL opcode there, and do you know what
might be the cause for this error?
> 
> Are all the register allocators currently of high standard, or are one or
more less mature, meaning things might break?
> 
> /Jonas
> 
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20111118/143a8690/attachment.html>