x86 19991121 ICE

[Jan Hubicka]

> Once again glibc proves fertile ground for FP related ICEs
> Maybe the compiler I tested only has part of your changes Jan?
I can reproduce this with current CVS. The problem looks more like
somewhat weird problem in reload. The insn in question is newly
created by reload and it ought not to do that, because of:
#define SECONDARY_MEMORY_NEEDED(CLASS1,CLASS2,MODE) \
  (FLOAT_CLASS_P (CLASS1) != FLOAT_CLASS_P (CLASS2))

My understanding of reload don't seems to be enought to track this down.
I give up for now. I will try once again at the evening..

Honza