[LLVMdev] Limitations of Alias Analysis?

[Wenzhi Tao]

Hi,
  Thanks for your advice. I have tried this way:

  opt -mem2reg test.bc -o mem2reg.bc
  llvm-dis mem2reg.bc
  opt -functionattrs -basicaa -aa-eval -print-all-alias-modref-info
mem2reg.bc

  The content of main() in mem2reg.ll:

define i32 @main() nounwind {
entry:
	%"alloca point" = bitcast i32 0 to i32		; <i32> [#uses=0]
	%0 = call %struct.Location* @getNewLocation(i32 0, i32 0) nounwind		; <
%struct.Location*> [#uses=2]
	%1 = call %struct.Location* @getNewLocation(i32 1, i32 2) nounwind		; <
%struct.Location*> [#uses=2]
	%2 = call %struct.Location* @sub(%struct.Location* %0, %
struct.Location* %1) nounwind		; <%struct.Location*> [#uses=1]
	%3 = bitcast %struct.Location* %0 to i8*		; <i8*> [#uses=1]
	call void @free(i8* %3) nounwind
	%4 = bitcast %struct.Location* %1 to i8*		; <i8*> [#uses=1]
	call void @free(i8* %4) nounwind
	%5 = bitcast %struct.Location* %2 to i8*		; <i8*> [#uses=1]
	call void @free(i8* %5) nounwind
	br label %return

return:		; preds = %entry
	ret i32 0
}

   The result of -aa-eval:

  MayAlias:	%struct.Location* %0, %struct.Location* %1
  MayAlias:	%struct.Location* %0, %struct.Location* %2
  MayAlias:	%struct.Location* %1, %struct.Location* %2

   Even when I add the AA option "-ds-aa -anders-aa", the result keeps
unchanged. en, I don't know if I miss some steps or the method doesn't
work at all.

在 2009-06-29一的 11:48 +0300，Török Edwin写道：
> On 2009-06-29 11:16, Wenzhi Tao wrote:
> > Hi, all
> >
> >   According to the document "LLVM Alias Analysis Infrastructure", I
> > evaluated the AA performance by using the paramenters '-basicaa -ds-aa
> > -anders-aa'. The source code 'test.c' is listed as follow:
> > [...] 
> > The whole process:
> >
> > llvm-gcc -emit-llvm -O0 -c test.c -o test.bc
> >
> > opt test.bc -load libLLVMDataStructure.so -basic-aa -ds-aa -anders-aa
> > -aa-eval -print-all-alias-modref-info
> >   
> 
> Try this:
> opt -mem2reg -functionattrs -basic-aa -aa-eval
> -print-all-alias-modref-info test.bc
> 
> It shows:
>   NoAlias:      %struct.Location* %0, %struct.Location* %1
> 
> > %1 and %2 are considered as MayAlias results that while using
> > llvm::MemoryDependenceAnalysis, %2 is dependent on %1(clobber). But in
> > fact it's not.
> >
> > Maybe a flow-sensitive, context-sensitive alias analysis algorithm is
> > needed to generate more precise result? Correct me, if I'm wrong. Thank
> > you.
> >   
> 
> Running the functionattrs pass should be enough to add the 'noalias'
> marker on getNewLocation(), and then even basic-aa can see that loc1 and
> loc2 are NoAlias.
> You also need to run -mem2reg, because otherwise -functionattrs doesn't
> see that the return value is coming from a malloc (which already has
> noalias attribute).
> 
> Best regards,
> --Edwin
> 
>