https://gcc.gnu.org/bugzilla/show_bug.cgi?id=125065
--- Comment #1 from Hunter X <bug_hunters at yeah dot net> --- **Correction/Update:** The original description states that GCC 15.2.0 vectorizes this loop while GCC trunk does not, implying the behavior is independent of `-fvect-cost-model=unlimited`. This is inaccurate. Further testing reveals: - **Without** `-fvect-cost-model=unlimited`: GCC 15.2.0 also **fails** to vectorize this loop, but for a different reason — the cost model rejects it as unprofitable (feasibility analysis passes, but cost model says no). - **With** `-fvect-cost-model=unlimited`: GCC 15.2.0 vectorizes it successfully, while GCC trunk still fails with "unsupported SLP instances" (feasibility analysis fails outright). This confirms the regression is specifically in the SLP feasibility analysis (`vect_slp_analyze_operations`), not in the cost model. GCC trunk's SLP cannot handle the strided access pattern at all, while GCC 15.2.0's SLP could (and only needed `-fvect-cost-model=unlimited` to override cost model rejection). Sorry for the imprecise wording in the original description — please use this clarification for your analysis. Thank you!
