loads of diff alignment should not share the same coderep. WOPT should not need to deal with alignment period. That belong elsewhere Whether I miscontrue Mike's point is not relevent here. Two different places in symtab and type_idx to deal with alignment is just bad. Sun
On Fri, Aug 26, 2011 at 4:59 AM, shuxin yang <[email protected]> wrote: >>alignment is not just for performance, and the compiler does alignment for >> all kinds of reasons > I'm interested to know the reasons for the alignment other than correctness > and performance both in terms of run time and space. >>What Mike said is that the alignment should have been set in the TY. Adding >> alignment in symbol will only confuse things. > I believe you completely misconstrue what Mike is talking about. If I > remember right ty_align is 4-bit quantity stored in load or store WN nodes. > It is *NOT* part of TY. > I believe having alignment for each memory access is great. Unfortunately, > the WOPT doesn't respect it, at least in some places, it is so. > On the other hand, the nature of DAG makes it hard to annotate alignment. > Do the following two loads, which share a single CODEREP, have same > alignment? > if (p & 15) > = ... *p; > else > = ... *p; > CG doesn't respect it at all, it has its own way in handling alignment... > and it is extremely hard > to pass the any information from upstream down to CG. > ------------------------------------------------------------------------------ EMC VNX: the world's simplest storage, starting under $10K The only unified storage solution that offers unified management Up to 160% more powerful than alternatives and 25% more efficient. Guaranteed. http://p.sf.net/sfu/emc-vnx-dev2dev _______________________________________________ Open64-devel mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/open64-devel
