Steven Bosscher wrote: > On 6/15/06, Joern RENNECKE <[EMAIL PROTECTED]> wrote: >> In http://gcc.gnu.org/ml/gcc/2006-06/msg00549.html, you wrote: >> >> > -ftree-live-range-split splits live ranges on exit from ssa. This is >> > on by default, and in fact it is more work to NOT split unrelated live >> > ranges, and creates extra register pressure. >> >> I've found that in 4.1, -fno-web is often needed in order to generate >> auto-increment addressing. >> I suppose that will be similar for any live range splitting that >> happens before flow. > > Well, is that something you have to blame live range splitting for, or > is perhaps flow just too stupid to see through the split live ranges? > > Actually, Zadeck is working on a new auto-increment pass anyway IIUC. > > Gr. > Steven
My pass is redoing the code in flow so that it is not in flow, and does the premodify and postmodify properly. It does not attempt to do what Joern is doing, i.e. changing a reference from one base register to another. Kenny
