[llvm-dev] TBAA

[Victor Yodaiken via llvm-dev]

On Fri, Jun 18, 2021 at 9:40 AM Richard Kenner <kenner at adacore.com> wrote:

> > Are there any studies or examples of TBAA optimization for C showing
> > substantive performance wins? I have not been able to find any
> > papers or even compelling examples.
>
> Consider:
>
> void
> square (float *arr, int *bound)
> {
>   int i;
>
>   for (i = 0; i <= *bound; i++)
>     arr[i] = arr[i] * arr[i];
> }
>
> without TBAA you can't hoist the reference to *bound out of the loop.  That
> will prevent almost all loop optimizations.
>


The programmer can get this with a simple change, no?

 void
square (float *arr, int *bound)
{
  int i;
  const int b = *bound;

  for (i = 0; i <= b ; i++)
    arr[i] = arr[i] * arr[i];
}
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-dev/attachments/20210618/34d9465b/attachment-0001.html>