From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from dpdk.org (dpdk.org [92.243.14.124]) by inbox.dpdk.org (Postfix) with ESMTP id 8C3BCA04B1; Wed, 26 Aug 2020 10:13:45 +0200 (CEST) Received: from [92.243.14.124] (localhost [127.0.0.1]) by dpdk.org (Postfix) with ESMTP id AEE851C0BD; Wed, 26 Aug 2020 10:13:32 +0200 (CEST) Received: from mga14.intel.com (mga14.intel.com [192.55.52.115]) by dpdk.org (Postfix) with ESMTP id 17E5D1C0B0 for ; Wed, 26 Aug 2020 10:13:30 +0200 (CEST) IronPort-SDR: C6qjoH8r/Kqn1/SAEHaOKLUCG29PzrJI5JUwDLkcnmJqIceXl6Z5Li+YKq8zrE59Q52AVISQz3 w5oeQJLqXVSw== X-IronPort-AV: E=McAfee;i="6000,8403,9724"; a="155512881" X-IronPort-AV: E=Sophos;i="5.76,354,1592895600"; d="scan'208";a="155512881" X-Amp-Result: SKIPPED(no attachment in message) X-Amp-File-Uploaded: False Received: from orsmga002.jf.intel.com ([10.7.209.21]) by fmsmga103.fm.intel.com with ESMTP/TLS/ECDHE-RSA-AES256-GCM-SHA384; 26 Aug 2020 01:13:30 -0700 IronPort-SDR: fzvyu8JgpXU4Kzn00qEb/itEJm04+vTXPVksXS2m2Q//od3iuepGYCNAFIuoMMc2azSIJdZrq+ hvDLHPhUVNEQ== X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.76,354,1592895600"; d="scan'208";a="312816185" Received: from dpdk.sh.intel.com ([10.239.255.12]) by orsmga002.jf.intel.com with ESMTP; 26 Aug 2020 01:13:28 -0700 From: Guinan Sun To: dev@dpdk.org Cc: Qi Zhang , Qiming Yang , Junyu Jiang Date: Wed, 26 Aug 2020 07:54:56 +0000 Message-Id: <20200826075501.50052-3-guinanx.sun@intel.com> X-Mailer: git-send-email 2.17.1 In-Reply-To: <20200826075501.50052-1-guinanx.sun@intel.com> References: <20200826075501.50052-1-guinanx.sun@intel.com> Subject: [dpdk-dev] [PATCH 2/7] net/ice: change RSS hash parsing in SSE path 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" From: Junyu Jiang Change RSS hash parsing from Flex Rx descriptor in SSE data path. Signed-off-by: Junyu Jiang --- drivers/net/ice/ice_rxtx_vec_sse.c | 89 ++++++++++++++++++++++++------ 1 file changed, 73 insertions(+), 16 deletions(-) diff --git a/drivers/net/ice/ice_rxtx_vec_sse.c b/drivers/net/ice/ice_rxtx_vec_sse.c index 382ef31f3..fffb27138 100644 --- a/drivers/net/ice/ice_rxtx_vec_sse.c +++ b/drivers/net/ice/ice_rxtx_vec_sse.c @@ -230,7 +230,8 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts, const __m128i zero = _mm_setzero_si128(); /* mask to shuffle from desc. to mbuf */ const __m128i shuf_msk = _mm_set_epi8 - (15, 14, 13, 12, /* octet 12~15, 32 bits rss */ + (0xFF, 0xFF, + 0xFF, 0xFF, /* rss hash parsed separately */ 11, 10, /* octet 10~11, 16 bits vlan_macip */ 5, 4, /* octet 4~5, 16 bits data_len */ 0xFF, 0xFF, /* skip high 16 bits pkt_len, zero out */ @@ -321,7 +322,7 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts, pos += ICE_DESCS_PER_LOOP, rxdp += ICE_DESCS_PER_LOOP) { __m128i descs[ICE_DESCS_PER_LOOP]; - __m128i pkt_mb1, pkt_mb2, pkt_mb3, pkt_mb4; + __m128i pkt_mb0, pkt_mb1, pkt_mb2, pkt_mb3; __m128i staterr, sterr_tmp1, sterr_tmp2; /* 2 64 bit or 4 32 bit mbuf pointers in one XMM reg. */ __m128i mbp1; @@ -367,8 +368,12 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts, rte_compiler_barrier(); /* D.1 pkt 3,4 convert format from desc to pktmbuf */ - pkt_mb4 = _mm_shuffle_epi8(descs[3], shuf_msk); - pkt_mb3 = _mm_shuffle_epi8(descs[2], shuf_msk); + pkt_mb3 = _mm_shuffle_epi8(descs[3], shuf_msk); + pkt_mb2 = _mm_shuffle_epi8(descs[2], shuf_msk); + + /* D.1 pkt 1,2 convert format from desc to pktmbuf */ + pkt_mb1 = _mm_shuffle_epi8(descs[1], shuf_msk); + pkt_mb0 = _mm_shuffle_epi8(descs[0], shuf_msk); /* C.1 4=>2 filter staterr info only */ sterr_tmp2 = _mm_unpackhi_epi32(descs[3], descs[2]); @@ -378,12 +383,68 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts, ice_rx_desc_to_olflags_v(rxq, descs, &rx_pkts[pos]); /* D.2 pkt 3,4 set in_port/nb_seg and remove crc */ - pkt_mb4 = _mm_add_epi16(pkt_mb4, crc_adjust); pkt_mb3 = _mm_add_epi16(pkt_mb3, crc_adjust); + pkt_mb2 = _mm_add_epi16(pkt_mb2, crc_adjust); - /* D.1 pkt 1,2 convert format from desc to pktmbuf */ - pkt_mb2 = _mm_shuffle_epi8(descs[1], shuf_msk); - pkt_mb1 = _mm_shuffle_epi8(descs[0], shuf_msk); + /* D.2 pkt 1,2 set in_port/nb_seg and remove crc */ + pkt_mb1 = _mm_add_epi16(pkt_mb1, crc_adjust); + pkt_mb0 = _mm_add_epi16(pkt_mb0, crc_adjust); + +#ifndef RTE_LIBRTE_ICE_16BYTE_RX_DESC + /** + * needs to load 2nd 16B of each desc for RSS hash parsing, + * will cause performance drop to get into this context. + */ + if (rxq->vsi->adapter->eth_dev->data->dev_conf.rxmode.offloads & + DEV_RX_OFFLOAD_RSS_HASH) { + /* load bottom half of every 32B desc */ + const __m128i raw_desc_bh3 = + _mm_load_si128 + ((void *)(&rxdp[3].wb.status_error1)); + rte_compiler_barrier(); + const __m128i raw_desc_bh2 = + _mm_load_si128 + ((void *)(&rxdp[2].wb.status_error1)); + rte_compiler_barrier(); + const __m128i raw_desc_bh1 = + _mm_load_si128 + ((void *)(&rxdp[1].wb.status_error1)); + rte_compiler_barrier(); + const __m128i raw_desc_bh0 = + _mm_load_si128 + ((void *)(&rxdp[0].wb.status_error1)); + + /** + * to shift the 32b RSS hash value to the + * highest 32b of each 128b before mask + */ + __m128i rss_hash3 = + _mm_slli_epi64(raw_desc_bh3, 32); + __m128i rss_hash2 = + _mm_slli_epi64(raw_desc_bh2, 32); + __m128i rss_hash1 = + _mm_slli_epi64(raw_desc_bh1, 32); + __m128i rss_hash0 = + _mm_slli_epi64(raw_desc_bh0, 32); + + __m128i rss_hash_msk = + _mm_set_epi32(0xFFFFFFFF, 0, 0, 0); + + rss_hash3 = _mm_and_si128 + (rss_hash3, rss_hash_msk); + rss_hash2 = _mm_and_si128 + (rss_hash2, rss_hash_msk); + rss_hash1 = _mm_and_si128 + (rss_hash1, rss_hash_msk); + rss_hash0 = _mm_and_si128 + (rss_hash0, rss_hash_msk); + + pkt_mb3 = _mm_or_si128(pkt_mb3, rss_hash3); + pkt_mb2 = _mm_or_si128(pkt_mb2, rss_hash2); + pkt_mb1 = _mm_or_si128(pkt_mb1, rss_hash1); + pkt_mb0 = _mm_or_si128(pkt_mb0, rss_hash0); + } /* if() on RSS hash parsing */ +#endif /* C.2 get 4 pkts staterr value */ staterr = _mm_unpacklo_epi32(sterr_tmp1, sterr_tmp2); @@ -391,14 +452,10 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts, /* D.3 copy final 3,4 data to rx_pkts */ _mm_storeu_si128 ((void *)&rx_pkts[pos + 3]->rx_descriptor_fields1, - pkt_mb4); + pkt_mb3); _mm_storeu_si128 ((void *)&rx_pkts[pos + 2]->rx_descriptor_fields1, - pkt_mb3); - - /* D.2 pkt 1,2 set in_port/nb_seg and remove crc */ - pkt_mb2 = _mm_add_epi16(pkt_mb2, crc_adjust); - pkt_mb1 = _mm_add_epi16(pkt_mb1, crc_adjust); + pkt_mb2); /* C* extract and record EOP bit */ if (split_packet) { @@ -422,9 +479,9 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts, /* D.3 copy final 1,2 data to rx_pkts */ _mm_storeu_si128 ((void *)&rx_pkts[pos + 1]->rx_descriptor_fields1, - pkt_mb2); + pkt_mb1); _mm_storeu_si128((void *)&rx_pkts[pos]->rx_descriptor_fields1, - pkt_mb1); + pkt_mb0); ice_rx_desc_to_ptype_v(descs, &rx_pkts[pos], ptype_tbl); /* C.4 calc avaialbe number of desc */ var = __builtin_popcountll(_mm_cvtsi128_si64(staterr)); -- 2.17.1