From patchwork Thu Sep 17 07:58:32 2020 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: "Guo, Jia" X-Patchwork-Id: 78010 X-Patchwork-Delegate: qi.z.zhang@intel.com Return-Path: X-Original-To: patchwork@inbox.dpdk.org Delivered-To: patchwork@inbox.dpdk.org Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id E65DFA04B6; Thu, 17 Sep 2020 10:02:05 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id 273B61D594; Thu, 17 Sep 2020 10:01:41 +0200 (CEST) Received: from mga01.intel.com (mga01.intel.com [192.55.52.88]) by dpdk.org (Postfix) with ESMTP id B76411D577 for ; Thu, 17 Sep 2020 10:01:37 +0200 (CEST) IronPort-SDR: V/XFI/HY+71IOU3YQPUI8h8BAGU+dRVG+TkEtPbtFbiD+3TDdkVZzEbp28k/1M7hpyKh2q96YP 4b2jva9n+pvg== X-IronPort-AV: E=McAfee;i="6000,8403,9746"; a="177733481" X-IronPort-AV: E=Sophos;i="5.76,436,1592895600"; d="scan'208";a="177733481" X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga002.jf.intel.com ([10.7.209.21]) by fmsmga101.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 17 Sep 2020 01:01:37 -0700 IronPort-SDR: r5t3DZOKBNcdDG4Ya19F7bQdtZT/N8+KlltdSAMZMOBuMkbJ9wfjdxeBE8JZCRDseAPL22CyNU SxXYVmyAIQlw== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.76,436,1592895600"; d="scan'208";a="320136159" Received: from npg-dpdk-cvl-jeffguo-01.sh.intel.com ([10.67.111.128]) by orsmga002.jf.intel.com with ESMTP; 17 Sep 2020 01:01:33 -0700 From: Jeff Guo To: qiming.yang@intel.com, beilei.xing@intel.com, qi.z.zhang@intel.com, jingjing.wu@intel.com, haiyue.wang@intel.com Cc: wei.zhao1@intel.com, bruce.richardson@intel.com, dev@dpdk.org, jia.guo@intel.com, helin.zhang@intel.com, mb@smartsharesystems.com, ferruh.yigit@intel.com, stephen@networkplumber.org, barbette@kth.se Date: Thu, 17 Sep 2020 15:58:32 +0800 Message-Id: <20200917075834.60034-4-jia.guo@intel.com> X-Mailer: git-send-email 2.20.1 In-Reply-To: <20200917075834.60034-1-jia.guo@intel.com> References: <20200827075452.1751-1-jia.guo@intel.com> <20200917075834.60034-1-jia.guo@intel.com> MIME-Version: 1.0 Subject: [dpdk-dev] [PATCH v4 3/5] net/i40e: fix vector rx burst for i40e X-BeenThere: dev@dpdk.org X-Mailman-Version: 2.1.15 Precedence: list List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" The limitation of burst size in vector rx was removed, since it should retrieve as much received packets as possible. And also the scattered receive path should use a wrapper function to achieve the goal of burst maximizing. And do some code cleaning for vector rx path. Bugzilla ID: 516 Fixes: 5b463eda8d26 ("net/i40e: make vector driver filenames consistent") Fixes: ae0eb310f253 ("net/i40e: implement vector PMD for ARM") Fixes: dafadd73762e ("net/i40e: add AVX2 Rx function") Fixes: c3def6a8724c ("net/i40e: implement vector PMD for altivec") Signed-off-by: Jeff Guo --- drivers/net/i40e/i40e_rxtx.h | 1 + drivers/net/i40e/i40e_rxtx_vec_altivec.c | 64 ++++++++++++++++-------- drivers/net/i40e/i40e_rxtx_vec_avx2.c | 29 ++++++----- drivers/net/i40e/i40e_rxtx_vec_neon.c | 58 +++++++++++++-------- drivers/net/i40e/i40e_rxtx_vec_sse.c | 58 +++++++++++++-------- 5 files changed, 133 insertions(+), 77 deletions(-) diff --git a/drivers/net/i40e/i40e_rxtx.h b/drivers/net/i40e/i40e_rxtx.h index 57d7b4160..01d4609f9 100644 --- a/drivers/net/i40e/i40e_rxtx.h +++ b/drivers/net/i40e/i40e_rxtx.h @@ -14,6 +14,7 @@ #define RTE_I40E_MAX_RX_BURST RTE_I40E_RXQ_REARM_THRESH #define RTE_I40E_TX_MAX_FREE_BUF_SZ 64 #define RTE_I40E_DESCS_PER_LOOP 4 +#define RTE_I40E_DESCS_PER_LOOP_AVX 8 #define I40E_RXBUF_SZ_1024 1024 #define I40E_RXBUF_SZ_2048 2048 diff --git a/drivers/net/i40e/i40e_rxtx_vec_altivec.c b/drivers/net/i40e/i40e_rxtx_vec_altivec.c index 6862a017e..345c63aa7 100644 --- a/drivers/net/i40e/i40e_rxtx_vec_altivec.c +++ b/drivers/net/i40e/i40e_rxtx_vec_altivec.c @@ -188,11 +188,13 @@ desc_to_ptype_v(vector unsigned long descs[4], struct rte_mbuf **rx_pkts, ptype_tbl[(*(vector unsigned char *)&ptype1)[8]]; } - /* Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits - */ +/** + * vPMD raw receive routine, only accept(nb_pkts >= RTE_I40E_DESCS_PER_LOOP) + * + * Notice: + * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet + * - floor align nb_pkts to a RTE_I40E_DESCS_PER_LOOP power-of-two + */ static inline uint16_t _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts, uint8_t *split_packet) @@ -214,9 +216,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, }; vector unsigned long dd_check, eop_check; - /* nb_pkts shall be less equal than RTE_I40E_MAX_RX_BURST */ - nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_MAX_RX_BURST); - /* nb_pkts has to be floor-aligned to RTE_I40E_DESCS_PER_LOOP */ nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_I40E_DESCS_PER_LOOP); @@ -447,11 +446,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, return nb_pkts_recd; } - /* Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits - */ uint16_t i40e_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) @@ -459,19 +453,19 @@ i40e_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, return _recv_raw_pkts_vec(rx_queue, rx_pkts, nb_pkts, NULL); } - /* vPMD receive routine that reassembles scattered packets - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits - */ -uint16_t -i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, - uint16_t nb_pkts) +/** + * vPMD receive routine that reassembles single burst of 32 scattered packets + */ +static uint16_t +i40e_recv_scattered_burst_vec(void *rx_queue, struct rte_mbuf **rx_pkts, + uint16_t nb_pkts) { struct i40e_rx_queue *rxq = rx_queue; uint8_t split_flags[RTE_I40E_VPMD_RX_BURST] = {0}; + /* split_flags only can support max of RTE_I40E_VPMD_RX_BURST */ + nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_VPMD_RX_BURST); + /* get some new buffers */ uint16_t nb_bufs = _recv_raw_pkts_vec(rxq, rx_pkts, nb_pkts, split_flags); @@ -500,6 +494,32 @@ i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, &split_flags[i]); } +/** + * vPMD receive routine that reassembles scattered packets. + */ +uint16_t +i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, + uint16_t nb_pkts) +{ + uint16_t retval = 0; + + while (nb_pkts > RTE_I40E_VPMD_RX_BURST) { + uint16_t burst; + + burst = i40e_recv_scattered_burst_vec(rx_queue, + rx_pkts + retval, + RTE_I40E_VPMD_RX_BURST); + retval += burst; + nb_pkts -= burst; + if (burst < RTE_I40E_VPMD_RX_BURST) + return retval; + } + + return retval + i40e_recv_scattered_burst_vec(rx_queue, + rx_pkts + retval, + nb_pkts); +} + static inline void vtx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf *pkt, uint64_t flags) diff --git a/drivers/net/i40e/i40e_rxtx_vec_avx2.c b/drivers/net/i40e/i40e_rxtx_vec_avx2.c index 3bcef1363..b5e6867d0 100644 --- a/drivers/net/i40e/i40e_rxtx_vec_avx2.c +++ b/drivers/net/i40e/i40e_rxtx_vec_avx2.c @@ -36,7 +36,7 @@ i40e_rxq_rearm(struct i40e_rx_queue *rxq) rxq->nb_rx_desc) { __m128i dma_addr0; dma_addr0 = _mm_setzero_si128(); - for (i = 0; i < RTE_I40E_DESCS_PER_LOOP; i++) { + for (i = 0; i < RTE_I40E_DESCS_PER_LOOP_AVX; i++) { rxep[i].mbuf = &rxq->fake_mbuf; _mm_store_si128((__m128i *)&rxdp[i].read, dma_addr0); @@ -219,13 +219,18 @@ desc_fdir_processing_32b(volatile union i40e_rx_desc *rxdp, #define PKTLEN_SHIFT 10 -/* Force inline as some compilers will not inline by default. */ +/** + * vPMD raw receive routine, only accept(nb_pkts >= RTE_I40E_DESCS_PER_LOOP_AVX) + * + * Notice: + * - nb_pkts < RTE_I40E_DESCS_PER_LOOP_AVX, just return no packet + * - floor align nb_pkts to a RTE_I40E_DESCS_PER_LOOP_AVX power-of-two + * - force inline as some compilers will not inline by default + */ static __rte_always_inline uint16_t _recv_raw_pkts_vec_avx2(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, uint16_t nb_pkts, uint8_t *split_packet) { -#define RTE_I40E_DESCS_PER_LOOP_AVX 8 - const uint32_t *ptype_tbl = rxq->vsi->adapter->ptype_tbl; const __m256i mbuf_init = _mm256_set_epi64x(0, 0, 0, rxq->mbuf_initializer); @@ -729,10 +734,6 @@ _recv_raw_pkts_vec_avx2(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, return received; } -/* - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - */ uint16_t i40e_recv_pkts_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) @@ -740,10 +741,8 @@ i40e_recv_pkts_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts, return _recv_raw_pkts_vec_avx2(rx_queue, rx_pkts, nb_pkts, NULL); } -/* +/** * vPMD receive routine that reassembles single burst of 32 scattered packets - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet */ static uint16_t i40e_recv_scattered_burst_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts, @@ -752,6 +751,9 @@ i40e_recv_scattered_burst_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts, struct i40e_rx_queue *rxq = rx_queue; uint8_t split_flags[RTE_I40E_VPMD_RX_BURST] = {0}; + /* split_flags only can support max of RTE_I40E_VPMD_RX_BURST */ + nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_VPMD_RX_BURST); + /* get some new buffers */ uint16_t nb_bufs = _recv_raw_pkts_vec_avx2(rxq, rx_pkts, nb_pkts, split_flags); @@ -781,11 +783,8 @@ i40e_recv_scattered_burst_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts, &split_flags[i]); } -/* +/** * vPMD receive routine that reassembles scattered packets. - * Main receive routine that can handle arbitrary burst sizes - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet */ uint16_t i40e_recv_scattered_pkts_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts, diff --git a/drivers/net/i40e/i40e_rxtx_vec_neon.c b/drivers/net/i40e/i40e_rxtx_vec_neon.c index 6f874e45b..143cdf4a5 100644 --- a/drivers/net/i40e/i40e_rxtx_vec_neon.c +++ b/drivers/net/i40e/i40e_rxtx_vec_neon.c @@ -187,11 +187,12 @@ desc_to_ptype_v(uint64x2_t descs[4], struct rte_mbuf **__rte_restrict rx_pkts, } - /* +/** + * vPMD raw receive routine, only accept(nb_pkts >= RTE_I40E_DESCS_PER_LOOP) + * * Notice: * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits + * - floor align nb_pkts to a RTE_I40E_DESCS_PER_LOOP power-of-two */ static inline uint16_t _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq, @@ -230,9 +231,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq, 0, 0, 0 /* ignore non-length fields */ }; - /* nb_pkts shall be less equal than RTE_I40E_MAX_RX_BURST */ - nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_MAX_RX_BURST); - /* nb_pkts has to be floor-aligned to RTE_I40E_DESCS_PER_LOOP */ nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_I40E_DESCS_PER_LOOP); @@ -426,12 +424,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq, return nb_pkts_recd; } - /* - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits - */ uint16_t i40e_recv_pkts_vec(void *__rte_restrict rx_queue, struct rte_mbuf **__rte_restrict rx_pkts, uint16_t nb_pkts) @@ -439,20 +431,20 @@ i40e_recv_pkts_vec(void *__rte_restrict rx_queue, return _recv_raw_pkts_vec(rx_queue, rx_pkts, nb_pkts, NULL); } - /* vPMD receive routine that reassembles scattered packets - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits +/** + * vPMD receive routine that reassembles single burst of 32 scattered packets */ -uint16_t -i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, - uint16_t nb_pkts) +static uint16_t +i40e_recv_scattered_burst_vec(void *rx_queue, struct rte_mbuf **rx_pkts, + uint16_t nb_pkts) { struct i40e_rx_queue *rxq = rx_queue; uint8_t split_flags[RTE_I40E_VPMD_RX_BURST] = {0}; + /* split_flags only can support max of RTE_I40E_VPMD_RX_BURST */ + nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_VPMD_RX_BURST); + /* get some new buffers */ uint16_t nb_bufs = _recv_raw_pkts_vec(rxq, rx_pkts, nb_pkts, split_flags); @@ -482,6 +474,32 @@ i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, &split_flags[i]); } +/** + * vPMD receive routine that reassembles scattered packets. + */ +uint16_t +i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, + uint16_t nb_pkts) +{ + uint16_t retval = 0; + + while (nb_pkts > RTE_I40E_VPMD_RX_BURST) { + uint16_t burst; + + burst = i40e_recv_scattered_burst_vec(rx_queue, + rx_pkts + retval, + RTE_I40E_VPMD_RX_BURST); + retval += burst; + nb_pkts -= burst; + if (burst < RTE_I40E_VPMD_RX_BURST) + return retval; + } + + return retval + i40e_recv_scattered_burst_vec(rx_queue, + rx_pkts + retval, + nb_pkts); +} + static inline void vtx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf *pkt, uint64_t flags) diff --git a/drivers/net/i40e/i40e_rxtx_vec_sse.c b/drivers/net/i40e/i40e_rxtx_vec_sse.c index 698518349..605912246 100644 --- a/drivers/net/i40e/i40e_rxtx_vec_sse.c +++ b/drivers/net/i40e/i40e_rxtx_vec_sse.c @@ -342,11 +342,12 @@ desc_to_ptype_v(__m128i descs[4], struct rte_mbuf **rx_pkts, rx_pkts[3]->packet_type = ptype_tbl[_mm_extract_epi8(ptype1, 8)]; } - /* +/** + * vPMD raw receive routine, only accept(nb_pkts >= RTE_I40E_DESCS_PER_LOOP) + * * Notice: * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits + * - floor align nb_pkts to a RTE_I40E_DESCS_PER_LOOP power-of-two */ static inline uint16_t _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, @@ -378,9 +379,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, offsetof(struct rte_mbuf, rx_descriptor_fields1) + 8); __m128i dd_check, eop_check; - /* nb_pkts shall be less equal than RTE_I40E_MAX_RX_BURST */ - nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_MAX_RX_BURST); - /* nb_pkts has to be floor-aligned to RTE_I40E_DESCS_PER_LOOP */ nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_I40E_DESCS_PER_LOOP); @@ -592,12 +590,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts, return nb_pkts_recd; } - /* - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits - */ uint16_t i40e_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts) @@ -605,20 +597,20 @@ i40e_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, return _recv_raw_pkts_vec(rx_queue, rx_pkts, nb_pkts, NULL); } - /* vPMD receive routine that reassembles scattered packets - * Notice: - * - nb_pkts < RTE_I40E_DESCS_PER_LOOP, just return no packet - * - nb_pkts > RTE_I40E_VPMD_RX_BURST, only scan RTE_I40E_VPMD_RX_BURST - * numbers of DD bits +/** + * vPMD receive routine that reassembles single burst of 32 scattered packets */ -uint16_t -i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, - uint16_t nb_pkts) +static uint16_t +i40e_recv_scattered_burst_vec(void *rx_queue, struct rte_mbuf **rx_pkts, + uint16_t nb_pkts) { struct i40e_rx_queue *rxq = rx_queue; uint8_t split_flags[RTE_I40E_VPMD_RX_BURST] = {0}; + /* split_flags only can support max of RTE_I40E_VPMD_RX_BURST */ + nb_pkts = RTE_MIN(nb_pkts, RTE_I40E_VPMD_RX_BURST); + /* get some new buffers */ uint16_t nb_bufs = _recv_raw_pkts_vec(rxq, rx_pkts, nb_pkts, split_flags); @@ -648,6 +640,32 @@ i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, &split_flags[i]); } +/** + * vPMD receive routine that reassembles scattered packets. + */ +uint16_t +i40e_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts, + uint16_t nb_pkts) +{ + uint16_t retval = 0; + + while (nb_pkts > RTE_I40E_VPMD_RX_BURST) { + uint16_t burst; + + burst = i40e_recv_scattered_burst_vec(rx_queue, + rx_pkts + retval, + RTE_I40E_VPMD_RX_BURST); + retval += burst; + nb_pkts -= burst; + if (burst < RTE_I40E_VPMD_RX_BURST) + return retval; + } + + return retval + i40e_recv_scattered_burst_vec(rx_queue, + rx_pkts + retval, + nb_pkts); +} + static inline void vtx1(volatile struct i40e_tx_desc *txdp, struct rte_mbuf *pkt, uint64_t flags)