https://gcc.gnu.org/bugzilla/show_bug.cgi?id=120805
--- Comment #14 from Segher Boessenkool <segher at gcc dot gnu.org> --- Hi! (In reply to Avinash Jayakar from comment #12) > (In reply to Segher Boessenkool from comment #10) > > As a meta-comment: almost everything using scan-assembler-times is > > obfuscated. > > > > It should always say in comments *why* it expects it as many times as that, > > so that when the test starts failing it is somewhat clear what happened :-) > > I had a hard time understanding what the test case does for these vector > test case. It looks like it just checks whether the load and store > instructions are generated in the epilogue for each line in the loop. That is the goal maybe, yeah. > Ideally since there are 7 additions and 10 types, there should be a load > vector and store vector instruction for each addition roughly. > But the test looks for 120 lxv/lxvx instruction which also sees the > instruction in the main loop as well as some loads happening from TOC, and > is very confusing to figure out what the check is doing. Yup, it should at a minimum be documented what it expects, or if possible be written in such a way that things are clear and not so mixed up. > Would it be better to just check what is generated in the epilogue instead > of seeing the main loop as well for these tests? But how can you do that? It isn't obvious that this test wants to test just the loop epilogue code, btw!
