egcs/gcc internal errors in the recent snapshot

[Richard Henderson]

On Sun, Mar 14, 1999 at 02:53:39AM -0700, Jeffrey A Law wrote:
> Any thoughts on why we haven't seen this fail during a bootstrap before --
> it's not like people haven't been using the alpha port.

Previously fp-int reg moves were not possible -- they had to go
through memory.  So fp regs were not selected for just hanging
on to random data.

> Presumably TARGET_CIX is an ev6?

In effect.  CIX is a specific instruction set extension, but only
ev6 has it.

> I'll probably include it.  I've certainly seen quite a few bug reports
> that looked similar.  Though for some reason I didn't remember any of them
> being reported against egcs-1.1.  

Tege's was; I've also one from a Compaq engineer that was.

What clued me is that I finally got such a report against current
cvs -- before I'd just been thinking random reload losage, and that
it was fixed in the rewrite.


r~