Re: Problem with auto-inc code (ARM)

[Franz Sirl <Franz dot Sirl-kernel at lauterbach dot com>]

At 15:45 08.11.99 , Richard Earnshaw wrote:

>I thought I'd have a go at building gcc on the ARM with --enable-checking
>to find out just how slow it is... :-)
>
>Unfortunately, it has already tripped twice (and I've only reached
>building cccp.o with the stage1 compiler :-()
>
>The bug I've just found is occuring in combine_instructions, but can be
>tracked back to the flow pass when we set up an auto_inc but don't clean
>up properly after doing so.

You didn't tell which exact failure you see, but is it similar to the 
followin I see on powerpc-linux-gnu?

../../../cvsx/gccm/gcc/reload.c: In function `find_reloads_address':
../../../cvsx/gccm/gcc/reload.c:4812: RTL check: access of elt 5 of `note' 
with last elt 4
../../../cvsx/gccm/gcc/reload.c:4812: Internal compiler error in 
`combine_instructions', at combine.c:620
Please submit a full bug report.
See <URL:http://www.gnu.org/software/gcc/faq.html#bugreport> for instructions.

Strange thing is that I don't get this one if I bootstrap with 
enable-checking on a machine with egcs-1.1.2 as the bootstrap compiler. If 
I bootstrap with gcc-2.95.2 I see the above error. Unfortunately the 
machines are quite different in other respects too, that's why I didn't 
report that yet.

Franz.