Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-12-07 Thread Måns Rullgård
Christophe Gisquet christophe.gisq...@gmail.com writes: 2012/11/27 Justin Ruggles justin.rugg...@gmail.com: The patch itself looks good. Only the neon version would need to be modified. The altivec version already handles multiples of 8. The (yet to be ported) tak decoder could also use

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-12-07 Thread Justin Ruggles
On 12/07/2012 07:20 AM, Måns Rullgård wrote: Christophe Gisquet christophe.gisq...@gmail.com writes: 2012/11/27 Justin Ruggles justin.rugg...@gmail.com: The patch itself looks good. Only the neon version would need to be modified. The altivec version already handles multiples of 8. The

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-12-07 Thread Christophe Gisquet
Hi, 2012/12/7 Måns Rullgård m...@mansr.com: Can the buffers be padded to allow some overread? Indeed, and the same can be asked about scalarprod_and_madd_int16 and wma lossless, once I can find back a sample that triggers the interesting cases. -- Christophe

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-12-06 Thread Christophe Gisquet
2012/11/27 Justin Ruggles justin.rugg...@gmail.com: The patch itself looks good. Only the neon version would need to be modified. The altivec version already handles multiples of 8. The (yet to be ported) tak decoder could also use this. This tak decoder seems to require down to multiples of

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-11-27 Thread Justin Ruggles
On 11/24/2012 09:43 AM, Christophe Gisquet wrote: Hi, here's an updated version, as well as an example of use with real audio 14.4. But: - On the bigger sample here http://samples.mplayerhq.hu/real/RA/14_4/ I was not able to notice any kind of improvement - It requires every

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-11-25 Thread Diego Biurrun
On Sat, Nov 24, 2012 at 03:43:04PM +0100, Christophe Gisquet wrote: PS: no idea why firefox/gmail treats those patches as octet-stream, and how to fix that. Ronald wrote about this a while back, but there seems to be no easy solution:

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-11-25 Thread Ronald S. Bultje
Hi, On Sun, Nov 25, 2012 at 1:18 AM, Diego Biurrun di...@biurrun.de wrote: On Sat, Nov 24, 2012 at 03:43:04PM +0100, Christophe Gisquet wrote: PS: no idea why firefox/gmail treats those patches as octet-stream, and how to fix that. Ronald wrote about this a while back, but there seems to be

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-11-25 Thread Ronald S. Bultje
Hi, On Sun, Nov 25, 2012 at 8:43 AM, Ronald S. Bultje rsbul...@gmail.com wrote: (This is kind of a pain on Fedora Whoops, I meant Mac here. Ronald ___ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-11-25 Thread Christophe Gisquet
2012/11/25 Ronald S. Bultje rsbul...@gmail.com: The solution is basically to use git send-email, not gmail. git send-email can use your gmail account as smtp, if wanted. I can help you set that up if this doesn't help enough: Windows here, which means the msmtp way. I think I managed to set

Re: [libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-11-24 Thread Christophe Gisquet
Hi, here's an updated version, as well as an example of use with real audio 14.4. But: - On the bigger sample here http://samples.mplayerhq.hu/real/RA/14_4/ I was not able to notice any kind of improvement - It requires every implementation of the DSP code to be able to work on batches of 8

[libav-devel] [PATCH] dsputil: modify scalarproduct_int16 to handle mod8 numbers of loops.

2012-03-04 Thread Christophe Gisquet
Hi, (note: it is based on code where the shift parameter has already been removed) there aren't many users at the moment (not sure what the requirements for acelp are), but at least one future candidate exist in ra144, which uses precisely 40 elements (and thus is not a multiple of 16). Another