https://gcc.gnu.org/bugzilla/show_bug.cgi?id=123476
Bug ID: 123476
Summary: check for sqrt call not removed with loops
Product: gcc
Version: 16.0
Status: UNCONFIRMED
Keywords: missed-optimization
Severity: enhancement
Priority: P3
Component: tree-optimization
Assignee: unassigned at gcc dot gnu.org
Reporter: pinskia at gcc dot gnu.org
Blocks: 85316
Target Milestone: ---
Target: aarch64 x86_64
Take:
```
#include <cmath>
double CompareDistmats(const double* const distmat1,
const double* const distmat2, int l){
double RMSD = 0.0;
for (int i=0; i<l; i++){
RMSD += (distmat1[i]-distmat2[i])
*(distmat1[i]-distmat2[i]);
}
return std::sqrt(RMSD);
}
```
For aarch64 and x86_64, we should be able to optimize away the check for RMSD
being less than 0 here. at both -O2 and -O3.
We do handle the case where l is a small number say 1 or 2 (4 on aarch64
because not vectoring the loop). But we should be able to detect the range of
RMSD is always positive (or NAN or +inf) which we don't currently.
Note unlike LLVM we handle the case where the loop unrolls (but only if we
don't vectorize the loop).
Referenced Bugs:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=85316
[Bug 85316] [meta-bug] VRP range propagation missed cases
