From mboxrd@z Thu Jan 1 00:00:00 1970 From: Adrien Mazarguil Subject: Re: [PATCH v5 0/8] net/mlx4: Tx path improvements Date: Fri, 3 Nov 2017 10:48:00 +0100 Message-ID: <20171103094800.GL24849@6wind.com> References: <1509358049-18854-1-git-send-email-matan@mellanox.com> <1509640971-8637-1-git-send-email-matan@mellanox.com> <3d518b9d-15dd-48b7-ae10-95f9de3c1cf1@intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Matan Azrad , dev@dpdk.org, Ophir Munk To: Ferruh Yigit Return-path: Received: from mail-wm0-f52.google.com (mail-wm0-f52.google.com [74.125.82.52]) by dpdk.org (Postfix) with ESMTP id 2B84B1B5FC for ; Fri, 3 Nov 2017 10:48:13 +0100 (CET) Received: by mail-wm0-f52.google.com with SMTP id p75so251627wmg.3 for ; Fri, 03 Nov 2017 02:48:13 -0700 (PDT) Content-Disposition: inline In-Reply-To: <3d518b9d-15dd-48b7-ae10-95f9de3c1cf1@intel.com> List-Id: DPDK patches and discussions List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: dev-bounces@dpdk.org Sender: "dev" On Thu, Nov 02, 2017 at 01:41:39PM -0700, Ferruh Yigit wrote: > On 11/2/2017 9:42 AM, Matan Azrad wrote: > > v1: > > This series is a follow-up of the new datapath implementation introduced in RC1. > > It is dedicated to Tx and improves the performance. > > > > v2: > > Change uint32_t be_lkey to rte_be32_t lkey > > > > v3: > > Rebase to 17.11-rc2. > > Remove "net/mlx4: save lkey in big-endian format" patch which degrade performance. > > Merge separate segment cases patches "net/mlx4: improve performance of one Tx segment" and "net/mlx4: separate Tx for multi-segments". > > "net/mlx4: inline more Tx functions" renamed to "net/mlx4: associate MR to MP in a short function". > > Leave bytes counter as old version. > > Add memory barrier improvement patch. > > Remove empty segment support. > > > > v4: > > Remove "net/mlx4: remove empty Tx segment support" patch. > > Add "fix ring wraparound compiler hint" patch. > > Add "fix HW memory optimizations careless" patch. > > Remove unnecessary IO memory barrier in completion function. > > Remove inline declaration of completion function. > > Fix unnecessary empty lines. > > Fix indetations. > > Move max=0 check removal to patch "remove duplicate handling in Tx burst" and rename the patch. > > Add performace improvement in patch "mitigate Tx path memory barriers" > > > > v5: > > Fix compilation issue in debug mode for patch "net/mlx4: associate MR to MP in a short function". > > Add missing includes in "net/mlx4: associate MR to MP in a short function". > > Add cq_uar as volatile too in "net/mlx4: fix HW memory optimizations careless". > > Fix indentation issues in "net/mlx4: separate Tx segment cases". > > > > > > Matan Azrad (7): > > net/mlx4: remove error flows from Tx fast path > > net/mlx4: fix ring wraparound compiler hint > > net/mlx4: merge Tx path functions > > net/mlx4: remove duplicate handling in Tx burst > > net/mlx4: separate Tx segment cases > > net/mlx4: fix HW memory optimizations careless > > net/mlx4: mitigate Tx path memory barriers > > > > Ophir Munk (1): > > net/mlx4: associate MR to MP in a short function > > Patches has been sent and acked, so I already have my answer, but just to double > check, these are not just fixes which normally rc3 should target, there are data > path updates and performance improvements. > > Can you please confirm you want these patches in at rc3 phase? > I will wait for your explicit approval before getting them. Yes, since they address a bunch of remaining performance issues, we'd like them included in v17.11 (otherwise we'll have to backport them for v17.11.1 anyway). Thanks Ferruh. -- Adrien Mazarguil 6WIND