[ GCC 3.0.x] Significant regression on OpenBSD i386

[Richard Henderson]

On Wed, Aug 22, 2001 at 02:38:47PM +0200, Marc Espie wrote:
> I'm not sure I understand precisely what's going on, but it seems there
> should be a distinction between @GOT (for global symbols) and @GOTOFF
> (for local symbols) that gcc is not making, and that this is the problem
> that is killing gas here...

Yep.  You'll have to add an ENCODE_SECTION_INFO hook that sets
SYMBOL_REF_FLAG, and we'll have to modify sjlj_emit_function_enter
to gen up a fake VAR_DECL to pass to ENCODE_SECTION_INFO.

I could give it a shot later this week, but I don't have openbsd
installed anywhere to test it.

I suspect that profile directed feedback is broken similarly.


r~