On Tue, Jun 18, 2013 at 10:02 PM, Oleg Endo <oleg.e...@t-online.de> wrote: > On Tue, 2013-06-18 at 18:09 +0800, Bin.Cheng wrote: >> On Tue, Jun 18, 2013 at 3:52 AM, Oleg Endo <oleg.e...@t-online.de> wrote: >> > >> > My observation is, that legitimizing addressing modes in the backend by >> > looking at one isolated address works, but doesn't give good results. >> > In the SH backend there is this a particular case with displacement >> > addressing (register + constant). On SH displacements for byte >> > addressing are 0..15, 0..31 for 16 bit words and 0..63 for 32 bit words. >> > sh_legitimize_address (or rather sh_find_mov_disp_adjust) uses a fixed >> > heuristic to satisfy the displacement constraint and splits out a plus >> > insn if needed to adjust the base address. Of course that fixed >> > heuristic doesn't work for some cases and thus sometimes results in >> > unnecessary base address adjustments. >> > I had the idea of replacing the current (partially defunct) auto-inc-dec >> > RTL pass with a more generic addressing mode selection pass: >> > http://gcc.gnu.org/bugzilla/show_bug.cgi?id=56590 >> > >> > Any suggestions/comments/... are highly appreciated. >> > >> In PR56590, is PR50749 the only one that correlate with auto-inc-dec? >> Others seem just problems of wrong addressing mode. > > Yes, PR 50749 was the initial description of auto-inc-dec defects. PR > 52049 is also related to it, as the code examples there are candidates > for post-inc addressing mode. In that case, if 'int' is replaced with > 'float' on SH post-inc is the optimal mode, because it doesn't have > displacement addressing for FPU loads, except than SH2A. Even then, > using post-inc is better as it has a more compact instruction encoding. > The current auto-inc-dec is not able to discover such opportunities if, > for example, mem accesses are reordered by preceding optimization > passes. > >> And one point on PR50749, auto-inc-dec depends on ivopt to choose >> auto-increment candidate. Since you disabled ivopt, I bet GCC will >> miss lots of auto-increment opportunities. > > No, I haven't disabled ivopt. >
But -fno-ivopts is specified in PR50749. With current implementation, auto-inc-dec iterates instructions backward, tries to find memory access and increment/decrement pairs. It will miss opportunities if instructions are interfered with each other. Thanks. bin -- Best Regards.
