Re: [RFC] Avoid unnecessary futex_wake syscalls when all threads are busy waiting

[Richard Henderson <rth at redhat dot com>]

On 07/15/2011 02:37 AM, Jakub Jelinek wrote:
> Any comments?  Can anyone see meassurable differences on some benchmark?
> 
> 2011-07-15  Jakub Jelinek  <jakub@redhat.com>
> 
> 	* config/linux/wait.h (do_spin): New inline, largely copied
> 	from do_wait, just don't do futex_wait here, instead return true if
> 	it should be done.
> 	(do_wait): Implement using do_spin.
> 	* config/linux/mutex.h (gomp_mutex_lock_slow): Add an int argument
> 	to prototype.
> 	(gomp_mutex_lock): Use __sync_val_compare_and_swap instead of
> 	__sync_bool_compare_and_swap, pass the oldval to
> 	gomp_mutex_lock_slow.
> 	* config/linux/mutex.c (gomp_mutex_lock_slow): Add oldval argument.
> 	If all mutex contenders are just spinning and not sleeping, don't
> 	change state to 2 unnecessarily.  Optimize the loop when state has
> 	already become 2 to use just one atomic operation per loop instead
> 	of two.
> 	* config/linux/ia64/mutex.h (gomp_mutex_lock_slow): Add an int argument
> 	to prototype.
> 	(gomp_mutex_lock): Use __sync_val_compare_and_swap instead of
> 	__sync_bool_compare_and_swap, pass the oldval to
> 	gomp_mutex_lock_slow.

I think that even if the only measurable difference is a
reduction in the number of syscalls, that's still an improvement.

The patch is ok.


r~