Le 12/14/2016 à 11:14 AM, Savolainen, Petri (Nokia - FI/Espoo) a écrit : > > From: Mike Holmes [mailto:mike.hol...@linaro.org] > Sent: Tuesday, December 13, 2016 2:48 PM > To: Savolainen, Petri (Nokia - FI/Espoo) > <petri.savolai...@nokia-bell-labs.com> > Cc: lng-odp@lists.linaro.org > Subject: Re: [lng-odp] [PATCH 0/6] Remove Linux specifics from odp/helpers > > > > On 13 December 2016 at 05:23, Savolainen, Petri (Nokia - FI/Espoo) > <petri.savolai...@nokia-bell-labs.com> wrote: > > >> -----Original Message----- >> From: lng-odp [mailto:lng-odp-boun...@lists.linaro.org] On Behalf Of Mike >> Holmes >> Sent: Monday, December 12, 2016 4:52 PM >> To: lng-odp@lists.linaro.org >> Subject: [lng-odp] [PATCH 0/6] Remove Linux specifics from odp/helpers >> >> To allow other implementations of the helpers to exist such as those used >> on the >> Kalray system, we need to remove the hardwired Linux naming. This exists >> in the >> api, the implementation of the helper lib and the resulting library. > Opaque thread helpers may very well exist with Linux (or Windows or Unix) > specific helpers. Majority of the applications are developed against a > defined operating system, not against an operating system abstraction layer. > For example, a linux specific helper library (consuming and producing Linux > types) would be valuable to many applications, may be even more valuable than > the opaque thread helpers used by our validation suite. > > I agree, but the core reason for the helper lib is to support the tests, > performance tests and examples to keep all the implementations together and > tested uniformly in the reference implimentation. > > Anyone can use any OS specific code they want locally, we say helpers are not > mandatory for that reason. > > I am fine if we have a new alternate repo or other obvious delineation for os > specific help. However putting them in the main helper API makes portability > impossible and we want to promote that first and allow for all performance > optimizations to be done locally if needed. We do know that we have at least > two OS'es actively using ODP. > > ---- HTML stops here ---- > > The point is that with Linux and ODP an application is directly portable to > 90-95% of all networking SoCs. Also if an SoCs does not support Linux it's > typically because of lacking HW features such as MMU, which again may > restrict me to use it in the first place. So, typically 100% on the SoCs I'm > interested in do support Linux and thus I do not need an opaque OS helper but > a Linux helper. > > > -Petri >
I don't mind the helper library to have OS specific code in it. As long as: - Source and hear file are separated so the helper lib can be compiled on another OS by removing the linux specific file from the Makefile.am - The associated tests can be disabled - The ODP test suite (at least the common_plat) part does not use any of them. Nicolas