[LLVMdev] optimizer returning wrong variable?

[Andrew Ferguson]

On Mon, 2011-07-04 at 23:43 -0700, Nick Lewycky wrote:
> Andrew Ferguson wrote:
> >   I'm having some trouble trying to workout how to form functions from
> > the c interface
> > I thought I had it sorted but I guess I'm missing something or haven't
> > understood
> > the requirements, a case of trial and error and not really having a clue
> > to start with!
> >
> > I've got binary ops, cmps, for loops, while loops working but then it
> > hit the
> > wall with a tail cmp loop.
> >
> > looking at the output from the c interface it looks ok to me
> > but the optimizer is returning the wrong variable
> > I don't know what to do to avoid it?
> 
> I don't see the problem. In your input .ll, the function's return is:
> 
>    %c16 = load i32* %c
>    ret i32 %c16
> 
> The last store to %c is the one in Repeat2
> 
>    %add6 = add i32 %c, %add
>    store i32 %add6, i32* %c
> 
> which postdominates (ie., all routes from entry to exit must encounter 
> that store). Therefore we conclude that the return value is in fact %add6.
> 
> Are you actually seeing different behaviour for any particular set of 
> inputs on some target?
> 
> Nick

Thanks, that make sense, it's just that the returned value from the
example is coming out as zero, when it should be 325   
Should or does the optimized code return the correct value? 
Or should I be looking at the JIT?

I've only been able to test what I'm doing on (32 bit Ubuntu 11.04)
I did try to build a dll on windows with MinGW, 2.9 failed, 2.8 built ok
but was broken, so I deferred it to a later date. I couldn't satisfy the
linker with the static libs. So I'm using the shared object if that
makes any difference.  

Regarding the JIT none of the .ll I'm generating runs without having
been optimized first. I assumed it would run if it's been validated but
apparently it doesn't, so I concluded that the JIT only accepts the
proper SSA representation, I don't know if that's supposed to be the
case or not. 
     

Cheers Andrew

> >
> > source language input
> >
> > Procedure TRU(a,b)
> > Protected x,y,c
> > Repeat
> > y=0
> > Repeat
> > c+x+y
> > y+1
> > Until y >= b
> > x+1
> > Until x >= a
> > ProcedureReturn c
> > EndProcedure
> >
> > output generated via the c Interface
> >
> > define i32 @TRU(i32 %a, i32 %b) {
> > Entry:
> > %a1 = alloca i32
> > store i32 %a, i32* %a1
> > %b2 = alloca i32
> > store i32 %b, i32* %b2
> > %c = alloca i32
> > store i32 0, i32* %c
> > %x = alloca i32
> > store i32 0, i32* %x
> > %y = alloca i32
> > store i32 0, i32* %y
> > br label %Repeat1
> >
> > Repeat1: ; preds = %Until1, %Entry
> > store i32 0, i32* %y
> > br label %Repeat2
> >
> > Repeat2: ; preds = %Until2, %Repeat1
> > %x3 = load i32* %x
> > %y4 = load i32* %y
> > %add = add i32 %x3, %y4
> > %c5 = load i32* %c
> > %add6 = add i32 %c5, %add
> > store i32 %add6, i32* %c
> > %y7 = load i32* %y
> > %add8 = add i32 %y7, 1
> > store i32 %add8, i32* %y
> > br label %Until2
> >
> > Until2: ; preds = %Repeat2
> > %y9 = load i32* %y
> > %b10 = load i32* %b2
> > %cond = icmp sge i32 %y9, %b10
> > br i1 %cond, label %Repeat2, label %Untilcmp2
> >
> > Untilcmp2: ; preds = %Until2
> > %x11 = load i32* %x
> > %add12 = add i32 %x11, 1
> > store i32 %add12, i32* %x
> > br label %Until1
> >
> > Until1: ; preds = %Untilcmp2
> > %x13 = load i32* %x
> > %a14 = load i32* %a1
> > %cond15 = icmp sge i32 %x13, %a14
> > br i1 %cond15, label %Repeat1, label %Untilcmp1
> >
> > Untilcmp1: ; preds = %Until1
> > %c16 = load i32* %c
> > ret i32 %c16
> > }
> >
> > verifying module
> > Optimize Function
> > Run in JIT
> > Jit Result: = 0
> > should = 325
> >
> > dump the optimized module
> >
> > define i32 @TRU(i32 %a, i32 %b) {
> > Entry:
> > br label %Repeat1
> >
> > Repeat1: ; preds = %Untilcmp2, %Entry
> > %c.0 = phi i32 [ 0, %Entry ], [ %add6, %Untilcmp2 ]
> > %x.0 = phi i32 [ 0, %Entry ], [ %add12, %Untilcmp2 ]
> > br label %Repeat2
> >
> > Repeat2: ; preds = %Repeat2, %Repeat1
> > %c.1 = phi i32 [ %c.0, %Repeat1 ], [ %add6, %Repeat2 ]
> > %y.0 = phi i32 [ 0, %Repeat1 ], [ %add8, %Repeat2 ]
> > %add = add i32 %c.1, %x.0
> > %add6 = add i32 %add, %y.0
> > %add8 = add i32 %y.0, 1
> > %cond = icmp slt i32 %add8, %b
> > br i1 %cond, label %Untilcmp2, label %Repeat2
> >
> > Untilcmp2: ; preds = %Repeat2
> > %add12 = add i32 %x.0, 1
> > %cond15 = icmp slt i32 %add12, %a
> > br i1 %cond15, label %Untilcmp1, label %Repeat1
> >
> > Untilcmp1: ; preds = %Untilcmp2
> > ret i32 %add6
> > }
> >
> > --
> > Andrew Ferguson
> >
> >
> >
> >
> > _______________________________________________
> > LLVM Developers mailing list
> > LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> > http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev
> 

-- 
Andrew Ferguson