| Message ID | AANLkTinfP6yXfynzTtw=Y_nWhGGRh-fzDkn7RhHYxWvQ@mail.gmail.com |
|---|---|
| State | New |
| Headers | show |
On Wed, Aug 18, 2010 at 3:46 PM, Uros Bizjak <ubizjak@gmail.com> wrote: > Hello! > > ix86_binary_operator_ok needs output operand in addition to two input operands. > > 2010-08-18 Uros Bizjak <ubizjak@gmail.com> > > PR target/45327 > * config/i386/i386.md (<any_or:code><SWI:mode>_3): Do not use > ix86_binary_operator_ok. > > Tested on x86_64-pc-linux-gnu and committed to 4.5 and SVN mainline. > Can you add the testcase? Thanks.
On Wed, 2010-08-18 at 16:06 -0700, H.J. Lu wrote: > On Wed, Aug 18, 2010 at 3:46 PM, Uros Bizjak <ubizjak@gmail.com> wrote: > > Hello! > > > > ix86_binary_operator_ok needs output operand in addition to two input operands. > > > > 2010-08-18 Uros Bizjak <ubizjak@gmail.com> > > > > PR target/45327 > > * config/i386/i386.md (<any_or:code><SWI:mode>_3): Do not use > > ix86_binary_operator_ok. > > > > Tested on x86_64-pc-linux-gnu and committed to 4.5 and SVN mainline. > > > > Can you add the testcase? Well, it was a typo, so I think we can live without. No way this failure triggers again. Uros.
Index: i386.md =================================================================== --- i386.md (revision 163353) +++ i386.md (working copy) @@ -8456,7 +8456,7 @@ (const_int 0))) (clobber (match_scratch:SWI 0 "=<r>"))] "ix86_match_ccmode (insn, CCNOmode) - && ix86_binary_operator_ok (<CODE>, <MODE>mode, operands)" + && !(MEM_P (operands[1]) && MEM_P (operands[2]))" "<logic>{<imodesuffix>}\t{%2, %0|%0, %2}" [(set_attr "type" "alu") (set_attr "mode" "<MODE>")])