From mboxrd@z Thu Jan 1 00:00:00 1970 From: Bruce Richardson Subject: Re: [PATCH 3/3] net: run-time function selection Date: Mon, 22 Jan 2018 17:36:34 +0000 Message-ID: <20180122173634.GA7528@bricha3-MOBL3.ger.corp.intel.com> References: <1509991543-26521-1-git-send-email-elza.mathew@intel.com> <20171211153116.GA12544@bricha3-MOBL3.ger.corp.intel.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: jasvinder.singh@intel.com, dev@dpdk.org To: Elza Mathew Return-path: Received: from mga11.intel.com (mga11.intel.com [192.55.52.93]) by dpdk.org (Postfix) with ESMTP id A0D7728F3 for ; Mon, 22 Jan 2018 18:36:39 +0100 (CET) Content-Disposition: inline In-Reply-To: <20171211153116.GA12544@bricha3-MOBL3.ger.corp.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 Mon, Dec 11, 2017 at 03:31:16PM +0000, Bruce Richardson wrote: > On Mon, Nov 06, 2017 at 10:05:43AM -0800, Elza Mathew wrote: > > Compile-time function selection can potentially lead to > > lower performance on generic builds done by distros. > > Replaced compile time flag checks with run-time > > function selection. > > > > Signed-off-by: Elza Mathew > > --- > > lib/librte_net/rte_net_crc.c | 22 +++++++++++++--------- > > 1 file changed, 13 insertions(+), 9 deletions(-) > > > Patch looks good to me, but unfortunately the compilation testing shows > an issue with clang compliation on FreeBSD 10. I've also tested with gcc > on both BSD 10 and 11 and saw no issues, and BSD 11 clang compilation is > also fine. > > The compilation error is due to missing _mm_clmulepi64_si128 instrinsic: > In file included from /home/patchWorkOrg/compilation/lib/librte_net/rte_net_crc.c:43: > /home/patchWorkOrg/compilation/lib/librte_net/net_crc_sse.h:81:17: error: implicit declaration of function '_mm_clmulepi64_si128' is invalid in C99 [-Werror,-Wimplicit-function-declaration] > __m128i tmp0 = _mm_clmulepi64_si128(fold, precomp, 0x01); > > /Bruce Checking with latest images for 10.3 or 10.4 no longer shows this problem. I think the patch can be merged as-is. Acked-by: Bruce Richardson Thomas, do you need a resubmit to appear as a new patch in patchwork?