Hi Bruce, > -----Original Message----- > From: Richardson, Bruce > Sent: Wednesday, April 5, 2017 10:37 AM > To: Thomas Monjalon <thomas.monja...@6wind.com>; Singh, Jasvinder > <jasvinder.si...@intel.com> > Cc: dev@dpdk.org; olivier.m...@6wind.com; Doherty, Declan > <declan.dohe...@intel.com>; De Lara Guarch, Pablo > <pablo.de.lara.gua...@intel.com> > Subject: RE: [dpdk-dev] [PATCH v9 0/3] librte_net: add crc computation > support > > > > > -----Original Message----- > > From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Thomas Monjalon > > Sent: Wednesday, April 5, 2017 10:01 AM > > To: Singh, Jasvinder <jasvinder.si...@intel.com> > > Cc: dev@dpdk.org; olivier.m...@6wind.com; Doherty, Declan > > <declan.dohe...@intel.com>; De Lara Guarch, Pablo > > <pablo.de.lara.gua...@intel.com> > > Subject: Re: [dpdk-dev] [PATCH v9 0/3] librte_net: add crc computation > > support > > > > 2017-04-05 08:34, Singh, Jasvinder: > > > Hi Thomas, > > > > > > From: Thomas Monjalon [mailto:thomas.monja...@6wind.com] > > > > 2017-03-30 17:15, Jasvinder Singh: > > > > > In some applications, CRC (Cyclic Redundancy Check) needs to be > > > > > computed or updated during packet processing operations. This > > > > > patchset adds software implementation of some common standard > > > > > CRCs (32-bit Ethernet CRC as per Ethernet/[ISO/IEC 8802-3] and > > > > > 16-bit CCITT-CRC [ITU-T > > > > X.25]). > > > > > Two versions of each 32-bit and 16-bit CRC calculation are proposed. > > > > > > > > > > The first version presents a fast and efficient CRC generation > > > > > on IA processors by using the carry-less multiplication > > > > > instruction PCLMULQDQ (i.e SSE4.2 instrinsics). In this > > > > > implementation, a parallelized folding approach has been used to > > > > > first reduce an arbitrary length buffer to a small fixed size > > > > > length buffer (16 > > > > > bytes) with the > > > > help of precomputed constants. > > > > > The resultant single 16-bytes chunk is further reduced by > > > > > Barrett reduction method to generate final CRC value. For more > > > > > details on the implementation, see reference [1]. > > > > > > > > > > The second version presents the fallback solution to support the > > > > > CRC generation without needing any specific support from CPU > > > > > (for > > > > > examples- > > > > > SSE4.2 intrinsics). It is based on generic Look-Up Table(LUT) > > > > > algorithm that uses precomputed 256 element table as explained > > > > > in > > > > reference[2]. > > > > > > > > > > During intialisation, all the data structures required for CRC > > > > > computation are initialised. Also, x86 specific crc > > > > > implementation (if supported by the platform) or scalar version is > enabled. > > > > > > > > As you can see in patchwork, it does not compile on FreeBSD: > > > > http://dpdk.org/ml/archives/test-report/2017-April/016943.html > > > > > > As I stated in the cover letter notes as well that The patchset > > > build fails on clang version earlier than 3.7.0 due to missing > > > intrinsics and > > this issue is listed in DPDK known issue section. FreeBSD build on gcc > > target should work fine. > > > > Ah, I have not seen this explanation. > > > > However, we cannot let the build fails. > > It is a blocker for patch admission. > > > > Can you, at least, disable the code for some compiler versions? > > Hi Jasvinder, > > Any chance a work-around for this issue. The default compiler on BSD is > clang, and the BSD 10 series of releases uses v3.4. This means this > functionality will be unavailable for anyone using DPDK from BSD ports on > BSD 10. > > /Bruce
I will have a look at this and send another version with fix. Jasvinder
