[LLVMdev] Doubt with GVNPRE

[Owen Anderson]

GVNPRE is unmaintained, slow, and has been removed from trunk.

--Owen

On Mar 4, 2010, at 12:46 PM, ambika wrote:

> Hi,
> 
> I have a program as:
> 
> int main(int argc, char **argv)
> {
>        int a,b,k,l;
>    scanf("%d%d",&a,&b);
>        if (argc > 1) {
>          k=a+b;
>        } else {
>           k=5;
>        }
>        l=a+b;
>        printf("%d,%d",k,l);
>        return 0;
> }
> 
> 
> Now i run the following on it:
> 
>    llvm-gcc -O1 -emit-llvm -c -o 1.bc 1.c
>    llvm-dis 1.bc -o 1.ll
>    opt -gvnpre 1.bc -o 1p.bc            // I believe this line writes 
> the code to 1p.bc after applying gvnpre to 1.bc
>    llvm-dis 1p.bc -o 1p.ll
> 
> Now 1.ll is as :
> ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> entry:
>  %b = alloca i32, align 4                        ; <i32*> [#uses=3]
>  %a = alloca i32, align 4                        ; <i32*> [#uses=3]
>  %0 = call i32 (i8*, ...)* @scanf(i8* getelementptr inbounds ([5 x i8]* 
> @.str, i32 0, i32 0), i32* %a, i32* %b) nounwind ; <i32> [#uses=0]
>  %1 = icmp sgt i32 %argc, 1                      ; <i1> [#uses=1]
>  br i1 %1, label %bb, label %bb2
> 
> bb:                                               ; preds = %entry
>  %2 = load i32* %a, align 4                      ; <i32> [#uses=1]
>  %3 = load i32* %b, align 4                      ; <i32> [#uses=1]
>  %4 = add nsw i32 %3, %2                         ; <i32> [#uses=1]
>  br label %bb2
> 
> bb2:                                              ; preds = %bb, %entry
>  %k.0 = phi i32 [ %4, %bb ], [ 5, %entry ]       ; <i32> [#uses=1]
>  %5 = load i32* %a, align 4                      ; <i32> [#uses=1]
>  %6 = load i32* %b, align 4                      ; <i32> [#uses=1]
>  %7 = add nsw i32 %6, %5                         ; <i32> [#uses=1]
>  %8 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([6 x 
> i8]* @.str1, i32 0, i32 0), i32 %k.0, i32 %7) nounwind ; <i32> [#uses=0]
>  ret i32 0
> ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> 
> 
> And 1p.ll as:
> ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> entry:
>  %b = alloca i32, align 4                        ; <i32*> [#uses=3]
>  %a = alloca i32, align 4                        ; <i32*> [#uses=3]
>  %0 = call i32 (i8*, ...)* @scanf(i8* getelementptr inbounds ([5 x i8]* 
> @.str, i32 0, i32 0), i32* %a, i32* %b) nounwind ; <i32> [#uses=0]
>  %1 = icmp sgt i32 %argc, 1                      ; <i1> [#uses=1]
>  br i1 %1, label %bb, label %entry.bb2_crit_edge
> 
> entry.bb2_crit_edge:                              ; preds = %entry
>  br label %bb2
> 
> bb:                                               ; preds = %entry
>  %2 = load i32* %a, align 4                      ; <i32> [#uses=1]
>  %3 = load i32* %b, align 4                      ; <i32> [#uses=1]
>  %4 = add nsw i32 %3, %2                         ; <i32> [#uses=1]
>  br label %bb2
> 
> bb2:                                              ; preds = %bb, 
> %entry.bb2_crit_edge
>  %k.0 = phi i32 [ %4, %bb ], [ 5, %entry.bb2_crit_edge ] ; <i32> [#uses=1]
>  %5 = load i32* %a, align 4                      ; <i32> [#uses=1]
>  %6 = load i32* %b, align 4                      ; <i32> [#uses=1]
>  %7 = add nsw i32 %6, %5                         ; <i32> [#uses=1]
>  %8 = call i32 (i8*, ...)* @printf(i8* getelementptr inbounds ([6 x 
> i8]* @.str1, i32 0, i32 0), i32 %k.0, i32 %7) nounwind ; <i32> [#uses=0]
>  ret i32 0
> ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
> 
> I believe GVNPRE should have remove partial redundancy by introducing an 
> extra computation in entry.bb2_crit_edge.
> But this does not happen, redundancy still remains.
> 
> Can someone tell me where am I going wrong.
> 
> Thanks and Regards,
> Ambika
> 
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev

-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 2620 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20100304/97900a0d/attachment.bin>