On Tue, Jan 03, 2012 at 11:58:09PM +0100, Christophe Gisquet wrote: > This time with the file. > > 2012/1/3 Christophe Gisquet <[email protected]>: > > Hi, > > > > 2012/1/1 Kostya Shishkov <[email protected]>: > >> I'll wait for NEON results. The patch itself looks decent. > > > > As Janne provided results for a Panda board, and I have cleaned up a > > bit my git tree, here is an updated patch generated using git > > format-patch > > > > Next patch in the series will be about implementing DC-only processing. > > > > Best regards, > > Christophe
> From 3c2343e10e42907c6b9feb29d399062d03148601 Mon Sep 17 00:00:00 2001 > From: Christophe GISQUET <[email protected]> > Date: Sun, 1 Jan 2012 15:28:47 +0100 > Subject: [PATCH 1/8] Joint coefficient decoding and dequantization > > Perform dequantization while decoding coefficients instead of performing on > the entire coefficients buffer. > > Because quantized coefficients are very sparse, this usually causes a small > speedup. Speedup of around 1% on Panda board compared to the old (removed > here) neon code. Global speedup is probably around 3%. > --- > libavcodec/arm/rv34dsp_init_neon.c | 3 - > libavcodec/arm/rv34dsp_neon.S | 24 ----------- > libavcodec/rv34.c | 80 > ++++++++++++++++++++---------------- > libavcodec/rv34data.h | 10 ---- > libavcodec/rv34dsp.c | 16 ------- > libavcodec/rv34dsp.h | 1 - > 6 files changed, 44 insertions(+), 90 deletions(-) The patch LGTM, the message needs to be reformatted though (I'll do it after I test the patch and before pushing it). _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
