Return-Path: MIME-Version: 1.0 In-Reply-To: <20121030034633.2242d317@i7> References: <1351272036-4875-1-git-send-email-frederic.dalleau@linux.intel.com> <1351272036-4875-7-git-send-email-frederic.dalleau@linux.intel.com> <20121028020621.3653ac8e@i7> <20121028022942.3bcd1bd7@i7> <20121030034633.2242d317@i7> Date: Tue, 30 Oct 2012 06:26:57 +0100 Message-ID: Subject: Re: [PATCH v3 06/10] sbc: Add mmx primitive for 1b 8s analyse From: "Dalleau, Frederic" To: Siarhei Siamashka Cc: =?ISO-8859-1?Q?Fr=E9d=E9ric_Dalleau?= , linux-bluetooth@vger.kernel.org Content-Type: text/plain; charset=ISO-8859-1 Sender: linux-bluetooth-owner@vger.kernel.org List-ID: Hi Siarhei, On Tue, Oct 30, 2012 at 2:46 AM, Siarhei Siamashka wrote: > On Mon, 29 Oct 2012 08:42:08 +0100 > "Dalleau, Frederic" wrote: > So I still think that it is safer and cleaner to have > "sbc_init_primitives" function performing the following in order: > 1. Initialize function pointers with C implementations. > 2. Allow to override them with various platform specific > implementations (which would work fine for old SBC formats). > 3. At the end of the function have a check if we are actually dealing > with mSBC format and restore all the function pointers back to C > implementations in this case. That's only ok for simd, but as soon as mmx gets in there would be a 4th category where implementation supporting msbc would override the already defined pointer. Imho, the definitive clean solution is to add the missing primitives. > Well, I just don't quite like that after the patches > [PATCH 04/10] sbc: Add msbc flag and generic C primitive > [PATCH 05/10] sbc: Add support for mSBC frame header > we already have a complete mSBC support in C code, which is still > unusable (as in buggy) if run on mmx, iwmmxt, armv6 or arm neon capable > systems. This is experimental feature and the time of commit, it is unused. And btw, you just suggested to split this patch, and this will also lead to a transitional state where some new functionality can be broken. > The sbc-1.0 library is already starting to get packaged in some linux > distributions. If somebody tries to run some application to do mSBC > encoding/decoding, but happens to have an old sbc-1.0 library in his I may be mistaking but most projects uses sbc statically, so checking is mostly done at compile time. So as I often hear in open source : we'll cross the bridge when we get to it :) Regards, Fr?d?ric