https://gcc.gnu.org/bugzilla/show_bug.cgi?id=115571
Bug ID: 115571 Summary: `cmp * float` vs `tmp_bool * float` and better vectorization for `cmp * float` Product: gcc Version: 15.0 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 Target Milestone: --- Take: ``` void f(float *c, int *a) { for (int i = 0; i < 1024; i++) { bool t = a[i] == 10; c[i] = t * c[i]; } } void f1(float *c, int *a) { for (int i = 0; i < 1024; i++) { c[i] = (a[i] == 10) * c[i]; } } ``` The second one is vectorized better than the first one at -Ofast due to the second one being converted into `(a[i] == 10) ? c : 0` in fold before gimplification. This was inspired by https://github.com/llvm/llvm-project/pull/96216 .