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 D276546F14; Wed, 17 Sep 2025 07:28:01 +0200 (CEST) Received: from mails.dpdk.org (localhost [127.0.0.1]) by mails.dpdk.org (Postfix) with ESMTP id 2449E402D1; Wed, 17 Sep 2025 07:27:58 +0200 (CEST) Received: from mgamail.intel.com (mgamail.intel.com [198.175.65.19]) by mails.dpdk.org (Postfix) with ESMTP id 1CA5A4025A for ; Wed, 17 Sep 2025 07:27:55 +0200 (CEST) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/simple; d=intel.com; i=@intel.com; q=dns/txt; s=Intel; t=1758086876; x=1789622876; h=from:to:subject:date:message-id:in-reply-to:references: mime-version:content-transfer-encoding; bh=w5UnHTWVqQdtpIMYFYyJWeV65N0glXa6lgzAmqcjBN8=; b=VZj/r4GZzPlHa2btKxVCIW2McDZiWsGiCPoLJ1H736/f2pD6+c3XU+Gl EUQ4m5dvgbtZxE2sBOUvXf7+s//G8j+QNiNBaxj+LsyF6/LLXVz5j8RdP pAC7JZzyuhq0iHEpOKvHtkYf4BYsHADvIi/zHLbhrYdmasASg/Fv60oW/ 5AzjSRGq1oD942OTl4h3eXYQ8/d9d/KsDtI5jcDnZ5+Movtqbea65Ok2X FM6WPVPZ/QpuUF5/VvGvg20r+9zj1CkaR+rFGd1Q6DWpDDkDhAkpeE6Dg CawURp2G9UUebI2tnHYCNa7dWRD2Id5+Djx7NzpPRUyRqAvIR5WVgoyqR w==; X-CSE-ConnectionGUID: R+I19awnRquI4lqxUpIRjg== X-CSE-MsgGUID: Ll6N6C0TRf+tFgiY88Nb6A== X-IronPort-AV: E=McAfee;i="6800,10657,11555"; a="60250205" X-IronPort-AV: E=Sophos;i="6.18,271,1751266800"; d="scan'208";a="60250205" Received: from fmviesa001.fm.intel.com ([10.60.135.141]) by orvoesa111.jf.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 16 Sep 2025 22:27:55 -0700 X-CSE-ConnectionGUID: dC9b7mB2Rm6uS2kBs+wEGQ== X-CSE-MsgGUID: ggkzKYxpRNeYaMH8EPZSag== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="6.18,271,1751266800"; d="scan'208";a="206101417" Received: from unknown (HELO srv24..) ([10.138.182.231]) by fmviesa001.fm.intel.com with ESMTP; 16 Sep 2025 22:27:53 -0700 From: Shaiq Wani To: dev@dpdk.org, bruce.richardson@intel.com, aman.deep.singh@intel.com Subject: [PATCH 1/2] net/idpf: enable AVX2 for split queue Tx Date: Wed, 17 Sep 2025 10:56:57 +0530 Message-Id: <20250917052658.582872-2-shaiq.wani@intel.com> X-Mailer: git-send-email 2.34.1 In-Reply-To: <20250917052658.582872-1-shaiq.wani@intel.com> References: <20250917052658.582872-1-shaiq.wani@intel.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit 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 In case some CPUs don't support AVX512. Enable AVX2 for them to get better per-core performance. In the single queue model, the same descriptor queue is used by SW to post descriptors to the device and used by device to report completed descriptors to SW. While as the split queue model separates them into different queues for parallel processing and improved performance. Signed-off-by: Shaiq Wani --- drivers/net/intel/idpf/idpf_common_rxtx.h | 3 + .../net/intel/idpf/idpf_common_rxtx_avx2.c | 202 ++++++++++++++++++ drivers/net/intel/idpf/idpf_rxtx.c | 9 + 3 files changed, 214 insertions(+) diff --git a/drivers/net/intel/idpf/idpf_common_rxtx.h b/drivers/net/intel/idpf/idpf_common_rxtx.h index f84a760334..82ddcf3310 100644 --- a/drivers/net/intel/idpf/idpf_common_rxtx.h +++ b/drivers/net/intel/idpf/idpf_common_rxtx.h @@ -249,6 +249,9 @@ uint16_t idpf_dp_singleq_recv_pkts_avx2(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts); __rte_internal +uint16_t idpf_dp_splitq_xmit_pkts_avx2(void *tx_queue, struct rte_mbuf **tx_pkts, + uint16_t nb_pkts); +__rte_internal uint16_t idpf_dp_singleq_xmit_pkts_avx2(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts); diff --git a/drivers/net/intel/idpf/idpf_common_rxtx_avx2.c b/drivers/net/intel/idpf/idpf_common_rxtx_avx2.c index 1babc5114b..d0c37cbfc7 100644 --- a/drivers/net/intel/idpf/idpf_common_rxtx_avx2.c +++ b/drivers/net/intel/idpf/idpf_common_rxtx_avx2.c @@ -640,3 +640,205 @@ idpf_dp_singleq_xmit_pkts_avx2(void *tx_queue, struct rte_mbuf **tx_pkts, return nb_tx; } + +static __rte_always_inline void +idpf_splitq_scan_cq_ring(struct ci_tx_queue *cq) +{ + struct idpf_splitq_tx_compl_desc *compl_ring; + struct ci_tx_queue *txq; + uint16_t genid, txq_qid, cq_qid, i; + uint8_t ctype; + + cq_qid = cq->tx_tail; + + for (i = 0; i < IDPD_TXQ_SCAN_CQ_THRESH; i++) { + if (cq_qid == cq->nb_tx_desc) { + cq_qid = 0; + cq->expected_gen_id ^= 1; /* toggle generation bit */ + } + + compl_ring = &cq->compl_ring[cq_qid]; + + genid = (rte_le_to_cpu_16(compl_ring->qid_comptype_gen) & + IDPF_TXD_COMPLQ_GEN_M) >> IDPF_TXD_COMPLQ_GEN_S; + + if (genid != cq->expected_gen_id) + break; + + ctype = (rte_le_to_cpu_16(compl_ring->qid_comptype_gen) & + IDPF_TXD_COMPLQ_COMPL_TYPE_M) >> IDPF_TXD_COMPLQ_COMPL_TYPE_S; + + txq_qid = (rte_le_to_cpu_16(compl_ring->qid_comptype_gen) & + IDPF_TXD_COMPLQ_QID_M) >> IDPF_TXD_COMPLQ_QID_S; + + txq = cq->txqs[txq_qid - cq->tx_start_qid]; + if (ctype == IDPF_TXD_COMPLT_RS) + txq->rs_compl_count++; + + cq_qid++; + } + + cq->tx_tail = cq_qid; +} + +static __rte_always_inline void +idpf_splitq_vtx1_avx2(volatile struct idpf_flex_tx_sched_desc *txdp, + struct rte_mbuf *pkt, uint64_t flags) +{ + uint64_t high_qw = + IDPF_TX_DESC_DTYPE_FLEX_FLOW_SCHE | + ((uint64_t)flags) | + ((uint64_t)pkt->data_len << IDPF_TXD_QW1_TX_BUF_SZ_S); + + __m128i descriptor = _mm_set_epi64x(high_qw, + pkt->buf_iova + pkt->data_off); + _mm_storeu_si128(RTE_CAST_PTR(__m128i *, txdp), descriptor); +} + + +static inline void +idpf_splitq_vtx_avx2(volatile struct idpf_flex_tx_sched_desc *txdp, + struct rte_mbuf **pkt, uint16_t nb_pkts, uint64_t flags) +{ + const uint64_t hi_qw_tmpl = IDPF_TX_DESC_DTYPE_FLEX_FLOW_SCHE | + ((uint64_t)flags); + + /* align if needed */ + if (((uintptr_t)txdp & 0x1F) != 0 && nb_pkts != 0) { + idpf_splitq_vtx1_avx2(txdp, *pkt, flags); + txdp++, pkt++, nb_pkts--; + } + + for (; nb_pkts > 3; txdp += 4, pkt += 4, nb_pkts -= 4) { + uint64_t hi_qw3 = hi_qw_tmpl | + ((uint64_t)pkt[3]->data_len << IDPF_TXD_QW1_TX_BUF_SZ_S); + uint64_t hi_qw2 = hi_qw_tmpl | + ((uint64_t)pkt[2]->data_len << IDPF_TXD_QW1_TX_BUF_SZ_S); + uint64_t hi_qw1 = hi_qw_tmpl | + ((uint64_t)pkt[1]->data_len << IDPF_TXD_QW1_TX_BUF_SZ_S); + uint64_t hi_qw0 = hi_qw_tmpl | + ((uint64_t)pkt[0]->data_len << IDPF_TXD_QW1_TX_BUF_SZ_S); + + __m256i desc2_3 = _mm256_set_epi64x(hi_qw3, + pkt[3]->buf_iova + pkt[3]->data_off, + hi_qw2, + pkt[2]->buf_iova + pkt[2]->data_off); + + __m256i desc0_1 = _mm256_set_epi64x(hi_qw1, + pkt[1]->buf_iova + pkt[1]->data_off, + hi_qw0, + pkt[0]->buf_iova + pkt[0]->data_off); + + _mm256_storeu_si256(RTE_CAST_PTR(__m256i *, txdp + 2), desc2_3); + _mm256_storeu_si256(RTE_CAST_PTR(__m256i *, txdp), desc0_1); + } + + while (nb_pkts--) { + idpf_splitq_vtx1_avx2(txdp, *pkt, flags); + txdp++; + pkt++; + } +} + +static inline uint16_t +idpf_splitq_xmit_fixed_burst_vec_avx2(void *tx_queue, struct rte_mbuf **tx_pkts, + uint16_t nb_pkts) +{ + struct ci_tx_queue *txq = (struct ci_tx_queue *)tx_queue; + volatile struct idpf_flex_tx_sched_desc *txdp; + struct ci_tx_entry_vec *txep; + uint16_t n, nb_commit, tx_id; + uint64_t cmd_dtype = IDPF_TXD_FLEX_FLOW_CMD_EOP; + + tx_id = txq->tx_tail; + + /* restrict to max burst size */ + nb_pkts = RTE_MIN(nb_pkts, txq->tx_rs_thresh); + + /* make sure we have enough free space */ + if (txq->nb_tx_free < txq->tx_free_thresh) + ci_tx_free_bufs_vec(txq, idpf_tx_desc_done, false); + + nb_commit = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts); + nb_pkts = nb_commit; + if (unlikely(nb_pkts == 0)) + return 0; + + txdp = &txq->desc_ring[tx_id]; + txep = (void *)txq->sw_ring; + txep += tx_id; + + txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts); + + n = (uint16_t)(txq->nb_tx_desc - tx_id); + if (nb_commit >= n) { + ci_tx_backlog_entry_vec(txep, tx_pkts, n); + + idpf_splitq_vtx_avx2(txdp, tx_pkts, n - 1, cmd_dtype); + tx_pkts += (n - 1); + txdp += (n - 1); + + idpf_splitq_vtx1_avx2(txdp, *tx_pkts++, cmd_dtype); + + nb_commit = (uint16_t)(nb_commit - n); + + tx_id = 0; + txq->tx_next_rs = (uint16_t)(txq->tx_rs_thresh - 1); + + txdp = &txq->desc_ring[tx_id]; + txep = (void *)txq->sw_ring; + txep += tx_id; + } + + ci_tx_backlog_entry_vec(txep, tx_pkts, nb_commit); + + idpf_splitq_vtx_avx2(txdp, tx_pkts, nb_commit, cmd_dtype); + + tx_id = (uint16_t)(tx_id + nb_commit); + if (tx_id > txq->tx_next_rs) + txq->tx_next_rs = + (uint16_t)(txq->tx_next_rs + txq->tx_rs_thresh); + + txq->tx_tail = tx_id; + + IDPF_PCI_REG_WRITE(txq->qtx_tail, txq->tx_tail); + + return nb_pkts; +} + +static __rte_always_inline uint16_t +idpf_splitq_xmit_pkts_vec_avx2_cmn(void *tx_queue, struct rte_mbuf **tx_pkts, + uint16_t nb_pkts) +{ + struct ci_tx_queue *txq = (struct ci_tx_queue *)tx_queue; + uint16_t nb_tx = 0; + + while (nb_pkts) { + uint16_t ret, num; + idpf_splitq_scan_cq_ring(txq->complq); + + if (txq->rs_compl_count > txq->tx_free_thresh) { + ci_tx_free_bufs_vec(txq, idpf_tx_desc_done, false); + txq->rs_compl_count -= txq->tx_rs_thresh; + } + + num = (uint16_t)RTE_MIN(nb_pkts, txq->tx_rs_thresh); + ret = idpf_splitq_xmit_fixed_burst_vec_avx2(tx_queue, + &tx_pkts[nb_tx], + num); + nb_tx += ret; + nb_pkts -= ret; + if (ret < num) + break; + } + + return nb_tx; +} + +RTE_EXPORT_INTERNAL_SYMBOL(idpf_dp_splitq_xmit_pkts_avx2) +uint16_t +idpf_dp_splitq_xmit_pkts_avx2(void *tx_queue, struct rte_mbuf **tx_pkts, + uint16_t nb_pkts) +{ + return idpf_splitq_xmit_pkts_vec_avx2_cmn(tx_queue, tx_pkts, nb_pkts); +} diff --git a/drivers/net/intel/idpf/idpf_rxtx.c b/drivers/net/intel/idpf/idpf_rxtx.c index 5510cbd30a..7d5d8b9c48 100644 --- a/drivers/net/intel/idpf/idpf_rxtx.c +++ b/drivers/net/intel/idpf/idpf_rxtx.c @@ -919,6 +919,15 @@ idpf_set_tx_function(struct rte_eth_dev *dev) return; } #endif /* CC_AVX512_SUPPORT */ + if (vport->tx_use_avx2) { + PMD_DRV_LOG(NOTICE, + "Using Split AVX2 Vector Tx (port %d).", + dev->data->port_id); + dev->tx_pkt_burst = idpf_dp_splitq_xmit_pkts_avx2; + dev->tx_pkt_prepare = idpf_dp_prep_pkts; + return; + } + } PMD_DRV_LOG(NOTICE, "Using Split Scalar Tx (port %d).", -- 2.34.1