Richard Henderson wrote:
> I've tested this with good results on i686, x86_64, alpha, and ia64
> linux. It problem needs testing on s390 (where Ulrich is battling
> related problems) and on SH (for which Joern introduced these invariants
> back in 1998).
I've bootstrapped/regtested this patch on s390-ibm-linux and s390x-ibm-linux
without regressions. It also fixes PR 24621.
I haven't checked for performance regresions, but I don't expect any
major issues here.
Is the intent of this patch that now the various special cases in
find_reloads and gen_reload that try to handle a PLUS as 'operand'
and reload it should never trigger any more? If true, we should
remove them -- this would allow me remove one of the ugliest parts
of the s390 backend ...
Dr. Ulrich Weigand
Linux on zSeries Development