On 10/24/07, Diego Novillo <[EMAIL PROTECTED]> wrote: > We are also thinking about lowering GIMPLE a bit further and delay > the transition into RTL.
Diego, can you be more specific about what parts you think have to be lowered more from GIMPLE? Together with Richard Guenther, we're planing to send a proposal for having higher level vector constructs in GIMPLE. More precisely, we want to capture Fortran's vector operations in GIMPLE and feed loop optimizers with this higher level representation. The lowering will be performed either by the simdization or by a scalarizer just after loop transforms. I'm still reading through Fortran's lower-front-end to figure out how the scalarizer is working, and what can be improved if we move it in the LNO or in the simdizer. Sebastian