On Mon, Nov 30, 2009 at 11:02 AM, Toon Moene <t...@moene.org> wrote: >> Maybe this is more/also relevant: >> http://gcc.gnu.org/ml/gcc-patches/2004-11/msg01405.html >> >> dorit > > Jeepers creepers - why has this never been included in the trunk (it might > be quite a bit of work to get it going again, and I certainly do not have > time before the Christmas holidays ....)
Many reasons, let's see... The pass didn't bring any benefit at the time for any benchmark, because there were no passes who could use the result of the analysis. The vectorizer was still very primitive back then (already 5 years ago!). The analysis as implemented is intraprocedural only, and most of the benefit is expected from intraprocedural alignment propagation (FWIW: http://webhome.cs.uvic.ca/~nigelh/Publications/align_MSP5.pdf for a primitive implementation). There was no infrastructure for interprocedural analysis at the time. LNO was more-or-less abandoned in favor of Graphite. I'll see if I can make the intraprocedural version work again before Christmass. It shouldn't be that much work, actually. But it's be nice to have some test cases to demonstrate that the intraprocedural version would do any good. Ciao! Steven
