Re: GCC 3.0 Status Report (the ARM bits)

[Richard Earnshaw <rearnsha at arm dot com>]

> ARM: There are still multiple ARM regressions from GCC 2.95.  There
>      is still time to fix them -- but we must act extremely quickly.
>      These are bugs 3037, 3053, 2878, and 817.

#817: I've approved Phil's proposed patch for the branch only.  It's not 
the cleanest way to fix the problem, but it does fix it for now.

#2878: I posted a patch last night.  A bootstrap of the gcc directory with 
BOOT_CFLAGS=-g passed with the patch applied, but I've had no further time 
for testing it.

#3037: I suspect fixing this fully may be hard.  One possibility may be to 
(ab)use LOCAL_REGNO to achieve the result we want.  Something like
#define LOCAL_REGNO(X) \
	((reload_in_progress || reload_completed) \
	 && (regs_ever_live[X] && ! call_used_regs[X]))

But I haven't been able to test this much due to other commitments.

#3053: I've not had a chance to try RTH's patch yet.