This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: REG_EQUIV question


Jeffrey A Law <law@cygnus.com> writes:

>   In message <200001131702.CAA25179@svan18.sannet.ne.jp>you write:
>   > Reg137 (reg/v:DI 137) is a copy of an argument passed by reg3.
>   > The reg137 is first used at insn 571 and has a REG_EQUIV in its note.
>   > 
>   > The unrecognized instruction was:
>   > 
>   > (insn 977 570 571 (set (reg:QI 5 $s5)
>   >         (mem/s:QI (plus:DI (plus:DI (mem/f:DI (plus:DI (reg:DI 3 $s3)
>   >                             (const_int 40 [0x28])) 0)
>   >                     (reg:DI 5 $s5))
>   >                 (const_int 1 [0x1])) 0)) -1 (nil)
>   >     (nil))
>   > 
>   > This was perhaps a result of reload of the rhs of insn 571 and
>   > substitution because of REG_EQUIV, but its addressing mode 
>   > was illegal.
> Then it should have been identified as an invalid address and reloaded.  The
> (plus:DI (plus:DI (...))) should have been passed to find_reloads_address,
> identified as an invalid address and reloaded.

May be this port have invalid LEGITIMIZE_RELOAD_ADDRESS.

Denis.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]