> On Mon, Feb 04, 2008 at 12:24:33PM +0100, Jan Hubicka wrote: > > > > > > -malign-double is (was?) indeed a performance improvement for > > > numerical applications on 32bits. But DImode is still not 8 bytes aligned > > > there (which makes a next-gen 32bit ABI for 64bit x86 difficult there, > > > if you want to retain DImode/DFmode 8 byte alignment and re-use the > > > kernel 32bit compatibility layer). > > > > Perhaps it was intended for numerical applications, but it is > > implemented by increasing maximal alignment that affect long long too. > > > > I think H. J. hit it with this auto alignment branch and I would say > > that decreasing DImode alignment there to avoid long long triggering the > > prologue is good thing to do. > > > > Yes, I run into it on the stack alignment branch. It is not a big > deal. But it is a waste of space for no performance gain. However, > it does break i386 psABI. We align DF scalar to 8 byte and DF > field in struct to 4 byte for i386. Can we just align DI scalar > to 4 byte and leave DI field in struct unchanged for i386?
I don't quite follow. Where do we break psABI? We set alignemnt to DImode but with maximal alignment bound to 4 bytes, we won't require alignment here. I think all we should do is to align static variables that might be small performance win by not crossing cache boundary or just small space loss. In this case changing the behaviour for when automatic alignemnt is performed is not big deal and we should be psABI compatible. Honza > > Thanks. > > H.J.
