Re: [dpdk-dev] [PATCH v1 01/12] test/pmd_perf: use compiler atomic builtins for polling sync

[Joyce Kong <Joyce.Kong@arm.com>]

> <snip>
> 
> >
> > Convert rte_atomic usages to compiler atomic built-ins for polling
> > sync in pmd_perf test cases.
> >
> > Signed-off-by: Joyce Kong <joyce.kong@arm.com>
> > Reviewed-by: Ruifeng Wang <ruifeng.wang@arm.com>
> > ---
> >  app/test/test_pmd_perf.c | 12 +++++-------
> >  1 file changed, 5 insertions(+), 7 deletions(-)
> >
> > diff --git a/app/test/test_pmd_perf.c b/app/test/test_pmd_perf.c index
> > 3a248d512c..41ef001b90 100644
> > --- a/app/test/test_pmd_perf.c
> > +++ b/app/test/test_pmd_perf.c
> > @@ -10,7 +10,6 @@
> >  #include <rte_cycles.h>
> >  #include <rte_ethdev.h>
> >  #include <rte_byteorder.h>
> > -#include <rte_atomic.h>
> >  #include <rte_malloc.h>
> >  #include "packet_burst_generator.h"
> >  #include "test.h"
> > @@ -526,7 +525,7 @@ main_loop(__rte_unused void *args)
> >  	return 0;
> >  }
> >
> > -static rte_atomic64_t start;
> > +static uint64_t start;
> >
> >  static inline int
> >  poll_burst(void *args)
> > @@ -564,8 +563,7 @@ poll_burst(void *args)
> >  		num[portid] = pkt_per_port;
> >  	}
> >
> > -	while (!rte_atomic64_read(&start))
> > -		;
> > +	rte_wait_until_equal_64(&start, 1, __ATOMIC_RELAXED);
> This needs to be ACQUIRE. Please see comments below for function
> 'exec_burst'.
> 
> >
> >  	cur_tsc = rte_rdtsc();
> >  	while (total) {
> > @@ -617,7 +615,7 @@ exec_burst(uint32_t flags, int lcore)
> >  	pkt_per_port = MAX_TRAFFIC_BURST;
> >  	num = pkt_per_port * conf->nb_ports;
> >
> > -	rte_atomic64_init(&start);
> > +	__atomic_store_n(&start, 0, __ATOMIC_RELAXED);
> >
> >  	/* start polling thread, but not actually poll yet */
> >  	rte_eal_remote_launch(poll_burst,
> > @@ -625,7 +623,7 @@ exec_burst(uint32_t flags, int lcore)
> >
> >  	/* Only when polling first */
> >  	if (flags == SC_BURST_POLL_FIRST)
> > -		rte_atomic64_set(&start, 1);
> > +		__atomic_store_n(&start, 1, __ATOMIC_RELAXED);
> This line can be combined with initialization of 'start'. i.e.
> if (flags == SC_BURST_POLL_FIRST) {
> 	__atomic_store_n(&start, 1, __ATOMIC_RELAXED); } else {
> 	__atomic_store_n(&start, 0, __ATOMIC_RELAXED); } It is ok to use
> RELAXED here as it is part of initialization and 'rte_eal_remote_launch' will
> ensure that the store to 'start' is visible to the worker thread.
> 
> >
> >  	/* start xmit */
> >  	i = 0;
> > @@ -642,7 +640,7 @@ exec_burst(uint32_t flags, int lcore)
> >
> >  	/* only when polling second  */
> >  	if (flags == SC_BURST_XMIT_FIRST)
> > -		rte_atomic64_set(&start, 1);
> > +		__atomic_store_n(&start, 1, __ATOMIC_RELAXED);
> This store should happen after transmitting the packets. It is better to use
> RELEASE here.

Thanks, Honnappa, will make the changes in next version.
Joyce

> 
> >
> >  	/* wait for polling finished */
> >  	diff_tsc = rte_eal_wait_lcore(lcore);
> > --
> > 2.17.1