[PATCH] Optimized __cxa_guard_{acquire,release,abort} for Linux

[Jason Merrill]

Jakub Jelinek wrote:
> If we can trust __cxa_guard_abort and __cxa_guard_release will be always
> called by a thread which set pending_bit and guard_bit at that point is not
> set, that loop could be probably simplified.

That's the design.  I think it's appropriate to optimize this case 
rather than add extra atomic ops in order to make the locking code more 
robust in the presence of codegen bugs; do you agree?

This patch looks good to me.

Jason