From mboxrd@z Thu Jan 1 00:00:00 1970 From: Jerin Jacob Subject: Re: [PATCH 25/33] app/testeventdev: perf queue: add worker functions Date: Fri, 2 Jun 2017 17:51:48 +0530 Message-ID: <20170602122148.GB10169@jerin> References: <20170528195854.6064-1-jerin.jacob@caviumnetworks.com> <20170528195854.6064-26-jerin.jacob@caviumnetworks.com> <9184057F7FC11744A2107296B6B8EB1E01EC6753@FMSMSX108.amr.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: "dev@dpdk.org" , "Richardson, Bruce" , "Van Haaren, Harry" , "hemant.agrawal@nxp.com" , "nipun.gupta@nxp.com" , "Vangati, Narender" , "Rao, Nikhil" , "gprathyusha@caviumnetworks.com" To: "Eads, Gage" Return-path: Received: from NAM02-BL2-obe.outbound.protection.outlook.com (mail-bl2nam02on0059.outbound.protection.outlook.com [104.47.38.59]) by dpdk.org (Postfix) with ESMTP id 8D1247CFC for ; Fri, 2 Jun 2017 14:22:10 +0200 (CEST) Content-Disposition: inline In-Reply-To: <9184057F7FC11744A2107296B6B8EB1E01EC6753@FMSMSX108.amr.corp.intel.com> List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" -----Original Message----- > Date: Thu, 1 Jun 2017 21:04:15 +0000 > From: "Eads, Gage" > To: Jerin Jacob , "dev@dpdk.org" > > CC: "Richardson, Bruce" , "Van Haaren, Harry" > , "hemant.agrawal@nxp.com" > , "nipun.gupta@nxp.com" , > "Vangati, Narender" , "Rao, Nikhil" > , "gprathyusha@caviumnetworks.com" > > Subject: RE: [dpdk-dev] [PATCH 25/33] app/testeventdev: perf queue: add > worker functions > > > > > -----Original Message----- > > From: Jerin Jacob [mailto:jerin.jacob@caviumnetworks.com] > > Sent: Sunday, May 28, 2017 2:59 PM > > To: dev@dpdk.org > > Cc: Richardson, Bruce ; Van Haaren, Harry > > ; hemant.agrawal@nxp.com; Eads, Gage > > ; nipun.gupta@nxp.com; Vangati, Narender > > ; Rao, Nikhil ; > > gprathyusha@caviumnetworks.com; Jerin Jacob > > > > Subject: [dpdk-dev] [PATCH 25/33] app/testeventdev: perf queue: add worker > > functions > > > > Signed-off-by: Jerin Jacob > > --- > > app/test-eventdev/test_perf_common.h | 60 +++++++++++++++ app/test- > > eventdev/test_perf_queue.c | 137 +++++++++++++++++++++++++++++++++++ > > 2 files changed, 197 insertions(+) > > > > diff --git a/app/test-eventdev/test_perf_common.h b/app/test- > > eventdev/test_perf_common.h > > index f8246953a..9888e5078 100644 > > --- a/app/test-eventdev/test_perf_common.h > > +++ b/app/test-eventdev/test_perf_common.h > > @@ -86,6 +86,66 @@ struct perf_elt { > > uint64_t timestamp; > > } __rte_cache_aligned; > > > > +#define BURST_SIZE 16 > > + > > +#define PERF_WORKER_INIT\ > > + struct worker_data *w = arg;\ > > + struct test_perf *t = w->t;\ > > + struct evt_options *opt = t->opt;\ > > + const uint8_t dev = w->dev_id;\ > > + const uint8_t port = w->port_id;\ > > + uint8_t *const sched_type_list = &t->sched_type_list[0];\ > > + struct rte_mempool *const pool = t->pool;\ > > + const uint8_t nb_stages = t->opt->nb_stages;\ > > + const uint8_t laststage = nb_stages - 1;\ > > + uint8_t cnt = 0;\ > > + void *bufs[16] __rte_cache_aligned;\ > > + int const sz = RTE_DIM(bufs);\ > > + if (opt->verbose_level > 1)\ > > + printf("%s(): lcore %d dev_id %d port=%d\n", __func__,\ > > + rte_lcore_id(), dev, port) > > + > > +static inline __attribute__((always_inline)) int > > +perf_process_last_stage(struct rte_mempool *const pool, > > + struct rte_event *const ev, struct worker_data *const w, > > + void *bufs[], int const buf_sz, uint8_t count) { > > + bufs[count++] = ev->event_ptr; > > + w->processed_pkts++; > > + rte_smp_wmb(); > > + > > + if (unlikely(count == buf_sz)) { > > + count = 0; > > + rte_mempool_put_bulk(pool, bufs, buf_sz); > > + } > > + return count; > > +} > > + > > +static inline __attribute__((always_inline)) uint8_t > > +perf_process_last_stage_latency(struct rte_mempool *const pool, > > + struct rte_event *const ev, struct worker_data *const w, > > + void *bufs[], int const buf_sz, uint8_t count) { > > + uint64_t latency; > > + struct perf_elt *const m = ev->event_ptr; > > + > > + bufs[count++] = ev->event_ptr; > > + w->processed_pkts++; > > + > > + if (unlikely(count == buf_sz)) { > > + count = 0; > > + latency = rte_get_timer_cycles() - m->timestamp; > > + rte_mempool_put_bulk(pool, bufs, buf_sz); > > + } else { > > + latency = rte_get_timer_cycles() - m->timestamp; > > + } > > + > > + w->latency += latency; > > + rte_smp_wmb(); > > + return count; > > +} > > What purpose does the store barrier serve in these two functions? The master core(!worker core) reads w->latency and w->processed_pkts periodically from all workers.