Re: [lttng-dev] [PATCH] QSBR: Use xor operation to replace add operation when changing rcu_gp.ctr value

[Mathieu Desnoyers via lttng-dev <lttng-dev@lists.lttng.org>]

----- On Feb 16, 2022, at 2:35 AM, lttng-dev lttng-dev@lists.lttng.org wrote:

> It is enough to have three values of rcu_gp.ctr, 00 for INACTIVE,
> 01 or 11 for ACTIVE. So it is possible to replace add operation
> with xor operation when changing rcu_gp.ctr value.

What is missing here is a description justifying why this change is useful.

What is inherently better about XOR compared to ADD or even binary-OR ?

If it's about performance, then a benchmark on relevant architectures
would be useful. But I suspect that if end users care that much about the
performance of urcu_qsbr_synchronize_rcu(), they might be doing something
wrong.

Thanks,

Mathieu

> 
> Signed-off-by: yaowenbin <yaowenbin1@huawei.com>
> ---
> src/urcu-qsbr.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/src/urcu-qsbr.c b/src/urcu-qsbr.c
> index 3709412..46135f9 100644
> --- a/src/urcu-qsbr.c
> +++ b/src/urcu-qsbr.c
> @@ -391,7 +391,7 @@ void urcu_qsbr_synchronize_rcu(void)
> 		goto out;
> 
> 	/* Increment current G.P. */
> -	CMM_STORE_SHARED(urcu_qsbr_gp.ctr, urcu_qsbr_gp.ctr + URCU_QSBR_GP_CTR);
> +	CMM_STORE_SHARED(urcu_qsbr_gp.ctr, urcu_qsbr_gp.ctr ^ URCU_QSBR_GP_CTR);
> 
> 	/*
> 	 * Must commit urcu_qsbr_gp.ctr update to memory before waiting for
> --
> 2.27.0
> _______________________________________________
> lttng-dev mailing list
> lttng-dev@lists.lttng.org
> https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
_______________________________________________
lttng-dev mailing list
lttng-dev@lists.lttng.org
https://lists.lttng.org/cgi-bin/mailman/listinfo/lttng-dev