From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mails.dpdk.org (mails.dpdk.org [217.70.189.124]) by inbox.dpdk.org (Postfix) with ESMTP id 361E241D49; Thu, 23 Feb 2023 09:26:36 +0100 (CET) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 15B42430D9; Thu, 23 Feb 2023 09:26:36 +0100 (CET) Received: from us-smtp-delivery-124.mimecast.com (us-smtp-delivery-124.mimecast.com [170.10.129.124]) by mails.dpdk.org (Postfix) with ESMTP id 093FC43081 for ; Thu, 23 Feb 2023 09:26:33 +0100 (CET) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=redhat.com; s=mimecast20190719; t=1677140793; h=from:from:reply-to:subject:subject:date:date:message-id:message-id: to:to:cc:cc:mime-version:mime-version:content-type:content-type: content-transfer-encoding:content-transfer-encoding: in-reply-to:in-reply-to:references:references; bh=lp9wmern+Gn47p4LiN7FazhqXnMiyjl+Rk2W9gdY9TI=; b=DNnTiQbZ7StUzn2gULTrn0Ips8Xyiyp6wDGJcdtMkT/i7Qfow84eJxTenWrV6ebpRNhHc8 AplQSnLB+bZWS3F+ulcQosIektnAtnvr8RZDa8AI6wVQLUPYDrLz6oTvEGhhp2zJb+SWkp Vw52NHAQUErmYEQjfAmiofs5MQObJPk= Received: from mimecast-mx02.redhat.com (mimecast-mx02.redhat.com [66.187.233.88]) by relay.mimecast.com with ESMTP with STARTTLS (version=TLSv1.2, cipher=TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384) id us-mta-395-0rHday5bNl-_2R0v_4y5oA-1; Thu, 23 Feb 2023 03:26:32 -0500 X-MC-Unique: 0rHday5bNl-_2R0v_4y5oA-1 Received: from smtp.corp.redhat.com (int-mx03.intmail.prod.int.rdu2.redhat.com [10.11.54.3]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by mimecast-mx02.redhat.com (Postfix) with ESMTPS id E24D285A588; Thu, 23 Feb 2023 08:26:31 +0000 (UTC) Received: from [10.39.208.21] (unknown [10.39.208.21]) by smtp.corp.redhat.com (Postfix) with ESMTPS id 9570A1121314; Thu, 23 Feb 2023 08:26:30 +0000 (UTC) Message-ID: Date: Thu, 23 Feb 2023 09:26:28 +0100 MIME-Version: 1.0 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101 Thunderbird/102.7.1 Subject: Re: [PATCH v1 07/13] test/bbdev: add support for BLER for 4G To: "Vargas, Hernan" , "dev@dpdk.org" , "gakhil@marvell.com" , "Rix, Tom" Cc: "Chautru, Nicolas" , "Zhang, Qi Z" References: <20230117165023.20567-1-hernan.vargas@intel.com> <20230117165023.20567-8-hernan.vargas@intel.com> <597165ba-5e2c-a453-fbb3-e71ab456200c@redhat.com> From: Maxime Coquelin In-Reply-To: X-Scanned-By: MIMEDefang 3.1 on 10.11.54.3 X-Mimecast-Spam-Score: 0 X-Mimecast-Originator: redhat.com Content-Language: en-US Content-Type: text/plain; charset=UTF-8; format=flowed Content-Transfer-Encoding: 7bit X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.29 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Hi Hernan, On 2/22/23 22:55, Vargas, Hernan wrote: > Hi Maxime, > >> -----Original Message----- >> From: Maxime Coquelin >> Sent: Monday, February 20, 2023 9:44 AM >> To: Vargas, Hernan ; dev@dpdk.org; >> gakhil@marvell.com; Rix, Tom >> Cc: Chautru, Nicolas ; Zhang, Qi Z >> >> Subject: Re: [PATCH v1 07/13] test/bbdev: add support for BLER for 4G >> >> >> >> On 2/13/23 21:59, Vargas, Hernan wrote: >>> >>> >>>> -----Original Message----- >>>> From: Maxime Coquelin >>>> Sent: Tuesday, January 31, 2023 4:20 AM >>>> To: Vargas, Hernan ; dev@dpdk.org; >>>> gakhil@marvell.com; Rix, Tom >>>> Cc: Chautru, Nicolas ; Zhang, Qi Z >>>> >>>> Subject: Re: [PATCH v1 07/13] test/bbdev: add support for BLER for 4G >>>> >>>> >>>> >>>> On 1/17/23 17:50, Hernan Vargas wrote: >>>>> New feature to add BLER support for 4G in bbdev-test. >>>>> >>>>> Signed-off-by: Hernan Vargas >>>>> --- >>>>> app/test-bbdev/test_bbdev_perf.c | 184 >>>> ++++++++++++++++++++++++++++++- >>>>> 1 file changed, 183 insertions(+), 1 deletion(-) >>>>> >>>>> diff --git a/app/test-bbdev/test_bbdev_perf.c >>>>> b/app/test-bbdev/test_bbdev_perf.c >>>>> index ffb2de0604..69b86cdeb1 100644 >>>>> --- a/app/test-bbdev/test_bbdev_perf.c >>>>> +++ b/app/test-bbdev/test_bbdev_perf.c >>>>> @@ -1756,6 +1756,30 @@ gen_qm2_llr(int8_t *llrs, uint32_t j, double >>>>> N0, >>>> double llr_max) >>>>> llrs[j] = (int8_t) b; >>>>> } >>>>> >>>>> +/* Simple LLR generation assuming AWGN and QPSK */ static void >>>>> +gen_turbo_llr(int8_t *llrs, uint32_t j, double N0, double llr_max) { >>>>> + double b, b1, n; >>>>> + double coeff = 2.0 * sqrt(N0); >>>>> + >>>>> + /* Ignore in vectors null LLRs not to be saturated */ >>>>> + if (llrs[j] == 0) >>>>> + return; >>>>> + >>>>> + /* Note don't change sign here */ >>>>> + n = randn(j % 2); >>>>> + b1 = ((llrs[j] > 0 ? 2.0 : -2.0) >>>>> + + coeff * n) / N0; >>>>> + b = b1 * (1 << 4); >>>>> + b = round(b); >>>>> + if (b > llr_max) >>>>> + b = llr_max; >>>>> + if (b < -llr_max) >>>>> + b = -llr_max; >>>>> + llrs[j] = (int8_t) b; >>>>> +} >>>>> + >>>>> /* Generate LLR for a given SNR */ >>>>> static void >>>>> generate_llr_input(uint16_t n, struct rte_bbdev_op_data *inputs, >>>>> @@ >>>>> -1791,6 +1815,27 @@ generate_llr_input(uint16_t n, struct >>>> rte_bbdev_op_data *inputs, >>>>> } >>>>> } >>>>> >>>>> +/* Generate LLR for turbo decoder for a given SNR */ static void >>>>> +generate_turbo_llr_input(uint16_t n, struct rte_bbdev_op_data >> *inputs, >>>>> + struct rte_bbdev_dec_op *ref_op) >>>>> +{ >>>>> + struct rte_mbuf *m; >>>>> + uint32_t i, j, range; >>>>> + double N0, llr_max; >>>>> + >>>>> + llr_max = 127; >>>>> + range = ref_op->turbo_dec.input.length; >>>>> + N0 = 1.0 / pow(10.0, get_snr() / 10.0); >>>>> + >>>>> + for (i = 0; i < n; ++i) { >>>>> + m = inputs[i].data; >>>>> + int8_t *llrs = rte_pktmbuf_mtod_offset(m, int8_t *, 0); >>>>> + for (j = 0; j < range; ++j) >>>>> + gen_turbo_llr(llrs, j, N0, llr_max); >>>>> + } >>>>> +} >>>>> + >>>>> static void >>>>> copy_reference_ldpc_dec_op(struct rte_bbdev_dec_op **ops, >>>>> unsigned >>>> int n, >>>>> unsigned int start_idx, >>>>> @@ -2301,6 +2346,31 @@ validate_ldpc_bler(struct rte_bbdev_dec_op >>>> **ops, const uint16_t n) >>>>> return errors; >>>>> } >>>>> >>>>> +/* Check Number of code blocks errors */ static int >>>>> +validate_turbo_bler(struct rte_bbdev_dec_op **ops, const uint16_t >>>>> +n) { >>>>> + unsigned int i; >>>>> + struct op_data_entries *hard_data_orig = >>>>> + &test_vector.entries[DATA_HARD_OUTPUT]; >>>>> + struct rte_bbdev_op_turbo_dec *ops_td; >>>>> + struct rte_bbdev_op_data *hard_output; >>>>> + int errors = 0; >>>>> + struct rte_mbuf *m; >>>>> + >>>>> + for (i = 0; i < n; ++i) { >>>>> + ops_td = &ops[i]->turbo_dec; >>>>> + hard_output = &ops_td->hard_output; >>>>> + m = hard_output->data; >>>>> + if (memcmp(rte_pktmbuf_mtod_offset(m, uint32_t *, 0), >>>>> + hard_data_orig->segments[0].addr, >>>>> + hard_data_orig->segments[0].length)) >>>> >>>> No need to check mbuf is at least as large as segment 0 length? >>>> >>>>> + errors++; >>>>> + } >>>>> + return errors; >>>>> +} >>>>> + >>>>> + >>>>> static int >>>>> validate_ldpc_dec_op(struct rte_bbdev_dec_op **ops, const uint16_t >> n, >>>>> struct rte_bbdev_dec_op *ref_op, const int vector_mask) >>>> @@ -3736,6 >>>>> +3806,114 @@ bler_pmd_lcore_ldpc_dec(void *arg) >>>>> return TEST_SUCCESS; >>>>> } >>>>> >>>>> + >>>>> +static int >>>>> +bler_pmd_lcore_turbo_dec(void *arg) { >>>>> + struct thread_params *tp = arg; >>>>> + uint16_t enq, deq; >>>>> + uint64_t total_time = 0, start_time; >>>>> + const uint16_t queue_id = tp->queue_id; >>>>> + const uint16_t burst_sz = tp->op_params->burst_sz; >>>>> + const uint16_t num_ops = tp->op_params->num_to_process; >>>>> + struct rte_bbdev_dec_op *ops_enq[num_ops]; >>>>> + struct rte_bbdev_dec_op *ops_deq[num_ops]; >>>>> + struct rte_bbdev_dec_op *ref_op = tp->op_params->ref_dec_op; >>>>> + struct test_buffers *bufs = NULL; >>>>> + int i, j, ret; >>>>> + struct rte_bbdev_info info; >>>>> + uint16_t num_to_enq; >>>>> + >>>>> + TEST_ASSERT_SUCCESS((burst_sz > MAX_BURST), >>>>> + "BURST_SIZE should be <= %u", MAX_BURST); >>>>> + >>>>> + rte_bbdev_info_get(tp->dev_id, &info); >>>>> + >>>>> + TEST_ASSERT_SUCCESS((num_ops > info.drv.queue_size_lim), >>>>> + "NUM_OPS cannot exceed %u for this device", >>>>> + info.drv.queue_size_lim); >>>>> + >>>>> + bufs = &tp->op_params- >>>>> q_bufs[GET_SOCKET(info.socket_id)][queue_id]; >>>>> + >>>>> + rte_wait_until_equal_16(&tp->op_params->sync, SYNC_START, >>>>> +__ATOMIC_RELAXED); >>>>> + >>>>> + ret = rte_bbdev_dec_op_alloc_bulk(tp->op_params->mp, ops_enq, >>>> num_ops); >>>>> + TEST_ASSERT_SUCCESS(ret, "Allocation failed for %d ops", >>>> num_ops); >>>>> + >>>>> + /* For BLER tests we need to enable early termination */ >>>>> + if (!check_bit(ref_op->turbo_dec.op_flags, >>>>> + RTE_BBDEV_TURBO_EARLY_TERMINATION)) >>>>> + ref_op->turbo_dec.op_flags += >>>>> + RTE_BBDEV_TURBO_EARLY_TERMINATION; >>>>> + ref_op->turbo_dec.iter_max = get_iter_max(); >>>>> + ref_op->turbo_dec.iter_count = ref_op->turbo_dec.iter_max; >>>>> + >>>>> + if (test_vector.op_type != RTE_BBDEV_OP_NONE) >>>>> + copy_reference_dec_op(ops_enq, num_ops, 0, bufs- >>>>> inputs, >>>>> + bufs->hard_outputs, bufs->soft_outputs, >>>>> + ref_op); >>>>> + generate_turbo_llr_input(num_ops, bufs->inputs, ref_op); >>>>> + >>>>> + /* Set counter to validate the ordering */ >>>>> + for (j = 0; j < num_ops; ++j) >>>>> + ops_enq[j]->opaque_data = (void *)(uintptr_t)j; >>>>> + >>>>> + for (i = 0; i < 1; ++i) { /* Could add more iterations */ >>>>> + for (j = 0; j < num_ops; ++j) { >>>>> + mbuf_reset( >>>>> + ops_enq[j]->turbo_dec.hard_output.data); >>>>> + } >>>>> + >>>>> + start_time = rte_rdtsc_precise(); >>>>> + >>>>> + for (enq = 0, deq = 0; enq < num_ops;) { >>>>> + num_to_enq = burst_sz; >>>>> + >>>>> + if (unlikely(num_ops - enq < num_to_enq)) >>>>> + num_to_enq = num_ops - enq; >>>>> + >>>>> + enq += rte_bbdev_enqueue_dec_ops(tp->dev_id, >>>>> + queue_id, &ops_enq[enq], >>>> num_to_enq); >>>> >>>> If for some reason the operation fails (looks like it can fail in at >>>> least ACC200 I checked), it ends up doing an endless loop as enc won't get >> incremented. >>> The python wrapper script used to invoke test-bbdev has a configurable >> timeout (default 600 seconds) that would stop the test in case it ends up in >> an infinite loop. >> >> Ok, but having a timeout in the test script does not prevent to do a proprer >> error propagation/handling in the application. >> >> It is happens in when running in a CI, having errors properly checked would >> provide useful information to investigate without having to reproduce, or >> when failing to reproduce. >> >> Regards, >> Maxime >> > This is a good point for future improvement but not related to this series. > We can implement it in the future. What do you think? This patch is introducing a new feature, there is a bug in this exact feature. So this should be fixed in this patch, not in a future release. Thanks, Maxime > Thanks, > Hernan >>> >>>>> + >>>>> + deq += rte_bbdev_dequeue_dec_ops(tp->dev_id, >>>>> + queue_id, &ops_deq[deq], enq - >>>> deq); >>>>> + } >>>>> + >>>>> + /* dequeue the remaining */ >>>>> + while (deq < enq) { >>>>> + deq += rte_bbdev_dequeue_dec_ops(tp->dev_id, >>>>> + queue_id, &ops_deq[deq], enq - >>>> deq); >>>>> + } >>>>> + >>>>> + total_time += rte_rdtsc_precise() - start_time; >>>>> + } >>>>> + >>>>> + tp->iter_count = 0; >>>>> + tp->iter_average = 0; >>>>> + /* get the max of iter_count for all dequeued ops */ >>>>> + for (i = 0; i < num_ops; ++i) { >>>>> + tp->iter_count = RTE_MAX(ops_enq[i]- >>>>> turbo_dec.iter_count, >>>>> + tp->iter_count); >>>>> + tp->iter_average += (double) ops_enq[i]- >>>>> turbo_dec.iter_count; >>>>> + } >>>>> + >>>>> + tp->iter_average /= num_ops; >>>>> + tp->bler = (double) validate_turbo_bler(ops_deq, num_ops) / >>>> num_ops; >>>>> + >>>>> + rte_bbdev_dec_op_free_bulk(ops_enq, num_ops); >>>>> + >>>>> + double tb_len_bits = calc_dec_TB_size(ref_op); >>>>> + tp->ops_per_sec = ((double)num_ops * 1) / >>>>> + ((double)total_time / (double)rte_get_tsc_hz()); >>>>> + tp->mbps = (((double)(num_ops * 1 * tb_len_bits)) / >>>>> + 1000000.0) / ((double)total_time / >>>>> + (double)rte_get_tsc_hz()); >>>>> + printf("TBS %.0f Time %.0f\n", tb_len_bits, 1000000.0 * >>>>> + ((double)total_time / (double)rte_get_tsc_hz())); >>>>> + >>>>> + return TEST_SUCCESS; >>>>> +} >>>>> + >>>>> static int >>>>> throughput_pmd_lcore_ldpc_dec(void *arg) >>>>> { >>>>> @@ -4193,7 +4371,7 @@ print_dec_bler(struct thread_params >> *t_params, >>>> unsigned int used_cores) >>>>> total_bler /= used_cores; >>>>> total_iter /= used_cores; >>>>> >>>>> - printf("SNR %.2f BLER %.1f %% - Iterations %.1f %d - Tp %.1f Mbps >>>> %s\n", >>>>> + printf("SNR %.2f BLER %.1f %% - Iterations %.1f %d - Tp %.3f Mbps >>>>> +%s\n", >>>>> snr, total_bler * 100, total_iter, get_iter_max(), >>>>> total_mbps, get_vector_filename()); >>>>> } >>>>> @@ -4245,6 +4423,10 @@ bler_test(struct active_device *ad, >>>>> && !check_bit(test_vector.ldpc_dec.op_flags, >>>>> RTE_BBDEV_LDPC_LLR_COMPRESSION)) >>>>> bler_function = bler_pmd_lcore_ldpc_dec; >>>>> + else if ((test_vector.op_type == RTE_BBDEV_OP_TURBO_DEC) && >>>>> + !check_bit(test_vector.turbo_dec.op_flags, >>>>> + RTE_BBDEV_TURBO_SOFT_OUTPUT)) >>>>> + bler_function = bler_pmd_lcore_turbo_dec; >>>>> else >>>>> return TEST_SKIPPED; >>>>> >>> >