On Jan 27, 2014, at 8:59 AM, Paulo Matos <pma...@broadcom.com> wrote:
>> -----Original Message----- >> From: Richard Sandiford [mailto:rsand...@linux.vnet.ibm.com] >> Sent: 27 January 2014 16:50 >> To: Paulo Matos >> Cc: gcc@gcc.gnu.org >> Subject: Re: Mode change for bswap pattern expansion >> >> Sorry, I meant we use an unspec for the first ("V2HI") rotate. >> I.e. rather than: >> >> (set (subreg:V2HI (match_dup 2) 0) >> (rotate:V2HI (subreg:V2HI (match_dup 1) 0) >> (const_int 8))) >> (set (match_dup 0) >> (rotate:SI (match_dup 2) >> (const_int 16))) >> >> we have: >> >> (set (match_dup 2) (unspec:SI [(match_dup 1)] UNSPEC_FOO)) >> (set (match_dup 0) >> (rotate:SI (match_dup 2) >> (const_int 16))) >> >> In your case the define_insn for the UNSPEC_FOO pattern would have the >> same attributes as a V2HI rotate, so it should get scheduled in the same way. > > In that case it would work. My only concern would then be if it prevents > further optimizations. On the other hand I am not sure if GCC would try to > optimize a rotate with vector V2HI mode... > Might give both solutions a try and see what results I get. Maybe do a split after reload instead? Thanks, Andrew > > Thanks, > > Paulo Matos > >> Thanks, >> Richard >
