------- Comment #5 from dorit at il dot ibm dot com 2005-10-06 15:45 ------- For some targets the vect_no_align keyword means that the target intrinsically cannot support the feature, and for other targets it means that the feature can potentially be supported but something is still missing in the compiler/target-md-files. This is probably true for other vect_* keywords as well.
To address this we'd need to introduce separate duplicate keywords for the case the feature is intrinsically missing and for the case the feature is temporarily missing, and change all the testcases accordingly, and open tens of PRs for the XFAILS. I think instead it would be more productive to open a PR per target for each keyword that is not intrinsically unsupported. i.e - open one target PR for ia64 for being a vect_no_align target (i.e. for missing alignment support for ia64), instead of opening a PR for each testcase that fails for vect_no_align targets. We can also open a testsuite PR for converting all the XFAILS on target-keyword in the vectorizer testcases to UNSUPPORTED (whether the target intrinsically can't support the feature or not). This PR however I think can be closed cause there's nothing special about vect-105.c. -- http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23119
