m68k-elf local register allocation

[Joel Sherrill]

Ray Pimm wrote:
> 
> I am noticing suboptimal register allocation produced by gcc built for the
> m68k-elf target.

The best thing you can do is prepare as small a test case as possible
that
demonstrates the problem still applies to 3.2 and submit a gcc PR.

> In some functions the compiler uses registers other than d0/d1/a0/a1, and
> goes to the trouble of saving and restoring them to/from the stack, while
> leaving unused one or more of d0/d1/a0/a1, when it could have achieved the
> same result using only registers from the d0/d1/a0/a1 set.
> 
> Then in other functions, it will use d0/d1/a0/a1 only. I cannot see any
> pattern to the register allocation.
> 
> I am using primarily gcc-2.95.3, but I notice that the same phenomenon
> occurs on gcc-3.2. I generally use optimization level -O2, but higher
> optimization levels do not seem to affect the result.
> 
> My compiler options are as follows:
> 
> CXXFLAGS =  -O2                 \
>              -mcpu32             \
>              -fno-exceptions     \
>              -fno-for-scope      \
>              -fshort-enums       \
>              -Wall               \
>              -Wno-uninitialized  \
>              -Wno-switch         \
> 
> BTW, all of my code is C++, if this is relevant.
> 
> Thanks - Ray Pimm
> 
> ------
> Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
> Want to unsubscribe? Send a note to crossgcc-unsubscribe@sources.redhat.com

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
Support Available                (256) 722-9985

------
Want more information?  See the CrossGCC FAQ, http://www.objsw.com/CrossGCC/
Want to unsubscribe? Send a note to crossgcc-unsubscribe@sources.redhat.com