[LLVMdev] another SCEV surprise

[Hal Finkel]

On Wed, 20 Jun 2012 02:18:49 -0700
Eli Friedman <eli.friedman at gmail.com> wrote:

> On Tue, Jun 19, 2012 at 10:21 PM, Preston Briggs
> <preston.briggs at gmail.com> wrote:
> > When compile the following case and look at the SCEV analysis, I
> > notice that the first two loops don't have a
> > LoopInvariantBackedgeTakenCount (surprising) and the last one does
> > (not surprising, except in the context of the first two examples).
> >
> > void p4(int *A, int *B, long int n) {
> >   for (char i = 0; i < n; i++) {
> >     A[i + 2] = i;
> >     *B++ = A[i];
> >   }
> > }
> >
> > void p5(int *A, int *B, long int n) {
> >   for (short i = 0; i < n; i++) {
> >     A[i + 2] = i;
> >     *B++ = A[i];
> >   }
> > }
> >
> > void p6(int *A, int *B, long int n) {
> >   for (int i = 0; i < n; i++) {
> >     A[i + 2] = i;
> >     *B++ = A[i];
> >   }
> > }
> >
> >
> > Why the difference? Is there a reason we don't implement all the
> > loops using 64-bit values?
> 
> The third loop is provably not an infinite loop, thanks to C's signed
> overflow rules; we can't tell whether the first two loops might be
> infinite.

This seems like a perfect example of where LLVM might want to emit a
note to the user (especially when we have loop vectorization and other
transformations).

 -Hal

> 
> -Eli
> 
> _______________________________________________
> LLVM Developers mailing list
> LLVMdev at cs.uiuc.edu         http://llvm.cs.uiuc.edu
> http://lists.cs.uiuc.edu/mailman/listinfo/llvmdev



-- 
Hal Finkel
Postdoctoral Appointee
Leadership Computing Facility
Argonne National Laboratory