Show a patch.

GET /api/patches/74625/
HTTP 200 OK
Allow: GET, PUT, PATCH, HEAD, OPTIONS
Content-Type: application/json
Vary: Accept

{
    "id": 74625,
    "url": "https://patches.dpdk.org/api/patches/74625/",
    "web_url": "https://patches.dpdk.org/patch/74625/",
    "project": {
        "id": 1,
        "url": "https://patches.dpdk.org/api/projects/1/",
        "name": "DPDK",
        "link_name": "dpdk",
        "list_id": "dev.dpdk.org",
        "list_email": "dev@dpdk.org",
        "web_url": "http://core.dpdk.org",
        "scm_url": "git://dpdk.org/dpdk",
        "webscm_url": "http://git.dpdk.org/dpdk"
    },
    "msgid": "<20200722203238.14250-1-akozyrev@mellanox.com>",
    "date": "2020-07-22T20:32:38",
    "name": "net/mlx5: fix vectorized mini-CQE prefetching",
    "commit_ref": null,
    "pull_url": null,
    "state": "accepted",
    "archived": true,
    "hash": "4c5a71fc4418bbe8e4d209d1151fb078d1df7fdc",
    "submitter": {
        "id": 1573,
        "url": "https://patches.dpdk.org/api/people/1573/",
        "name": "Alexander Kozyrev",
        "email": "akozyrev@mellanox.com"
    },
    "delegate": {
        "id": 3268,
        "url": "https://patches.dpdk.org/api/users/3268/",
        "username": "rasland",
        "first_name": "Raslan",
        "last_name": "Darawsheh",
        "email": "rasland@mellanox.com"
    },
    "mbox": "https://patches.dpdk.org/patch/74625/mbox/",
    "series": [
        {
            "id": 11240,
            "url": "https://patches.dpdk.org/api/series/11240/",
            "web_url": "https://patches.dpdk.org/project/dpdk/list/?series=11240",
            "date": "2020-07-22T20:32:38",
            "name": "net/mlx5: fix vectorized mini-CQE prefetching",
            "version": 1,
            "mbox": "https://patches.dpdk.org/series/11240/mbox/"
        }
    ],
    "comments": "https://patches.dpdk.org/api/patches/74625/comments/",
    "check": "success",
    "checks": "https://patches.dpdk.org/api/patches/74625/checks/",
    "tags": {},
    "headers": {
        "List-Subscribe": "<https://mails.dpdk.org/listinfo/dev>,\n <mailto:dev-request@dpdk.org?subject=subscribe>",
        "X-Mailman-Version": "2.1.15",
        "Precedence": "list",
        "X-Original-To": "patchwork@inbox.dpdk.org",
        "List-Post": "<mailto:dev@dpdk.org>",
        "MIME-Version": "1.0",
        "X-BeenThere": "dev@dpdk.org",
        "List-Id": "DPDK patches and discussions <dev.dpdk.org>",
        "Subject": "[dpdk-dev] [PATCH] net/mlx5: fix vectorized mini-CQE prefetching",
        "Sender": "\"dev\" <dev-bounces@dpdk.org>",
        "From": "Alexander Kozyrev <akozyrev@mellanox.com>",
        "Received": [
            "from dpdk.org (dpdk.org [92.243.14.124])\n\tby inbox.dpdk.org (Postfix) with ESMTP id BF8BCA0526;\n\tWed, 22 Jul 2020 22:32:42 +0200 (CEST)",
            "from [92.243.14.124] (localhost [127.0.0.1])\n\tby dpdk.org (Postfix) with ESMTP id 13B2E1BFBA;\n\tWed, 22 Jul 2020 22:32:42 +0200 (CEST)",
            "from mellanox.co.il (mail-il-dmz.mellanox.com [193.47.165.129])\n by dpdk.org (Postfix) with ESMTP id DEEA22C6E\n for <dev@dpdk.org>; Wed, 22 Jul 2020 22:32:40 +0200 (CEST)",
            "from Internal Mail-Server by MTLPINE1 (envelope-from\n akozyrev@mellanox.com) with SMTP; 22 Jul 2020 23:32:39 +0300",
            "from pegasus02.mtr.labs.mlnx. (pegasus02.mtr.labs.mlnx\n [10.210.16.122])\n by labmailer.mlnx (8.13.8/8.13.8) with ESMTP id 06MKWdG8028254;\n Wed, 22 Jul 2020 23:32:39 +0300"
        ],
        "To": "dev@dpdk.org",
        "X-Mailer": "git-send-email 2.24.1",
        "List-Unsubscribe": "<https://mails.dpdk.org/options/dev>,\n <mailto:dev-request@dpdk.org?subject=unsubscribe>",
        "Date": "Wed, 22 Jul 2020 20:32:38 +0000",
        "Content-Transfer-Encoding": "8bit",
        "List-Archive": "<http://mails.dpdk.org/archives/dev/>",
        "Delivered-To": "patchwork@inbox.dpdk.org",
        "Errors-To": "dev-bounces@dpdk.org",
        "Cc": "stable@dpdk.org, rasland@mellanox.com, viacheslavo@mellanox.com",
        "List-Help": "<mailto:dev-request@dpdk.org?subject=help>",
        "Message-Id": "<20200722203238.14250-1-akozyrev@mellanox.com>",
        "Return-Path": "<dev-bounces@dpdk.org>"
    },
    "content": "There was an optimization work to prefetch all the CQEs before\ntheir invalidation. It allowed us to speed up the mini-CQE\ndecompression process by preheating the cache in the vectorized\nRx routine.\n\nPrefetching of the next mini-CQE, on the other hand, showed\nno difference in the performance on x86 platform. So, that was\nremoved. Unfortunately this caused the performance drop on ARM.\n\nPrefetch the mini-CQE as well as well as the all the soon to be\ninvalidated CQEs to get both CQE and mini-CQE on the hot path.\n\nFixes: 28a4b9632 (\"net/mlx5: prefetch CQEs for a faster decompression\")\nCc: stable@dpdk.org\n\nSigned-off-by: Alexander Kozyrev <akozyrev@mellanox.com>\nAcked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>\n---\n drivers/net/mlx5/mlx5_rxtx_vec_altivec.h | 3 ++-\n drivers/net/mlx5/mlx5_rxtx_vec_neon.h    | 3 +++\n drivers/net/mlx5/mlx5_rxtx_vec_sse.h     | 3 ++-\n 3 files changed, 7 insertions(+), 2 deletions(-)",
    "diff": "diff --git a/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h b/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h\nindex f5414eebad..cb4ce1a099 100644\n--- a/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h\n+++ b/drivers/net/mlx5/mlx5_rxtx_vec_altivec.h\n@@ -158,7 +158,6 @@ rxq_cq_decompress_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,\n \t\tfor (i = 0; i < MLX5_VPMD_DESCS_PER_LOOP; ++i)\n \t\t\tif (likely(pos + i < mcqe_n))\n \t\t\t\trte_prefetch0((void *)(cq + pos + i));\n-\n \t\t/* A.1 load mCQEs into a 128bit register. */\n \t\tmcqe1 = (vector unsigned char)vec_vsx_ld(0,\n \t\t\t(signed int const *)&mcq[pos % 8]);\n@@ -287,6 +286,8 @@ rxq_cq_decompress_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,\n \t\tpos += MLX5_VPMD_DESCS_PER_LOOP;\n \t\t/* Move to next CQE and invalidate consumed CQEs. */\n \t\tif (!(pos & 0x7) && pos < mcqe_n) {\n+\t\t\tif (pos + 8 < mcqe_n)\n+\t\t\t\trte_prefetch0((void *)(cq + pos + 8));\n \t\t\tmcq = (void *)&(cq + pos)->pkt_info;\n \t\t\tfor (i = 0; i < 8; ++i)\n \t\t\t\tcq[inv++].op_own = MLX5_CQE_INVALIDATE;\ndiff --git a/drivers/net/mlx5/mlx5_rxtx_vec_neon.h b/drivers/net/mlx5/mlx5_rxtx_vec_neon.h\nindex 555c342626..6c3149523e 100644\n--- a/drivers/net/mlx5/mlx5_rxtx_vec_neon.h\n+++ b/drivers/net/mlx5/mlx5_rxtx_vec_neon.h\n@@ -145,6 +145,7 @@ rxq_cq_decompress_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,\n \t\t\t\t    -1UL << ((mcqe_n - pos) *\n \t\t\t\t\t     sizeof(uint16_t) * 8) : 0);\n #endif\n+\n \t\tfor (i = 0; i < MLX5_VPMD_DESCS_PER_LOOP; ++i)\n \t\t\tif (likely(pos + i < mcqe_n))\n \t\t\t\trte_prefetch0((void *)(cq + pos + i));\n@@ -227,6 +228,8 @@ rxq_cq_decompress_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,\n \t\tpos += MLX5_VPMD_DESCS_PER_LOOP;\n \t\t/* Move to next CQE and invalidate consumed CQEs. */\n \t\tif (!(pos & 0x7) && pos < mcqe_n) {\n+\t\t\tif (pos + 8 < mcqe_n)\n+\t\t\t\trte_prefetch0((void *)(cq + pos + 8));\n \t\t\tmcq = (void *)&(cq + pos)->pkt_info;\n \t\t\tfor (i = 0; i < 8; ++i)\n \t\t\t\tcq[inv++].op_own = MLX5_CQE_INVALIDATE;\ndiff --git a/drivers/net/mlx5/mlx5_rxtx_vec_sse.h b/drivers/net/mlx5/mlx5_rxtx_vec_sse.h\nindex 34e3397115..554924d7fc 100644\n--- a/drivers/net/mlx5/mlx5_rxtx_vec_sse.h\n+++ b/drivers/net/mlx5/mlx5_rxtx_vec_sse.h\n@@ -135,7 +135,6 @@ rxq_cq_decompress_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,\n \t\tfor (i = 0; i < MLX5_VPMD_DESCS_PER_LOOP; ++i)\n \t\t\tif (likely(pos + i < mcqe_n))\n \t\t\t\trte_prefetch0((void *)(cq + pos + i));\n-\n \t\t/* A.1 load mCQEs into a 128bit register. */\n \t\tmcqe1 = _mm_loadu_si128((__m128i *)&mcq[pos % 8]);\n \t\tmcqe2 = _mm_loadu_si128((__m128i *)&mcq[pos % 8 + 2]);\n@@ -214,6 +213,8 @@ rxq_cq_decompress_v(struct mlx5_rxq_data *rxq, volatile struct mlx5_cqe *cq,\n \t\tpos += MLX5_VPMD_DESCS_PER_LOOP;\n \t\t/* Move to next CQE and invalidate consumed CQEs. */\n \t\tif (!(pos & 0x7) && pos < mcqe_n) {\n+\t\t\tif (pos + 8 < mcqe_n)\n+\t\t\t\trte_prefetch0((void *)(cq + pos + 8));\n \t\t\tmcq = (void *)(cq + pos);\n \t\t\tfor (i = 0; i < 8; ++i)\n \t\t\t\tcq[inv++].op_own = MLX5_CQE_INVALIDATE;\n",
    "prefixes": []
}