Re: [FFmpeg-devel] swscale/input : avoid float calc for GrayFloat to Gray16 conv
On Wed, Aug 22, 2018 at 09:02:57PM +0200, Martin Vignali wrote: > > > > well then it should be ok > > not sure if the ff_ prefix is ideal for a static inline function > > > > > Do you think flt_2_uint16 is a better name ? yes thx > (can probably be interesting to also add same kind of func for float to > uint8 in unscaled part) > > Martin > ___ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB Let us carefully observe those good qualities wherein our enemies excel us and endeavor to excel them, by avoiding what is faulty, and imitating what is excellent in them. -- Plutarch signature.asc Description: PGP signature ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
Re: [FFmpeg-devel] swscale/input : avoid float calc for GrayFloat to Gray16 conv
> > well then it should be ok > not sure if the ff_ prefix is ideal for a static inline function > > Do you think flt_2_uint16 is a better name ? (can probably be interesting to also add same kind of func for float to uint8 in unscaled part) Martin ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
Re: [FFmpeg-devel] swscale/input : avoid float calc for GrayFloat to Gray16 conv
On Wed, Aug 22, 2018 at 12:45:50PM +0200, Martin Vignali wrote: > > > --- > > > libswscale/input.c| 10 +- > > > libswscale/swscale_internal.h | 20 > > > 2 files changed, 25 insertions(+), 5 deletions(-) > > > > please provide benchmark, what is the impact on speed from this ? > > > > > In my tests, the patch increase speed by around 20%. well then it should be ok not sure if the ff_ prefix is ideal for a static inline function thx [...] -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB Does the universe only have a finite lifespan? No, its going to go on forever, its just that you wont like living in it. -- Hiranya Peiri signature.asc Description: PGP signature ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
Re: [FFmpeg-devel] swscale/input : avoid float calc for GrayFloat to Gray16 conv
> > --- > > libswscale/input.c| 10 +- > > libswscale/swscale_internal.h | 20 > > 2 files changed, 25 insertions(+), 5 deletions(-) > > please provide benchmark, what is the impact on speed from this ? > > In my tests, the patch increase speed by around 20%. Test with this sample (UHD Gray float PSD file) : https://we.tl/t-Ty9hR20gUw And this command line (test the bswap func : grayf32ToY16_bswap_c) ./ffmpeg -benchmark -loop 1 -t 30 -i TEST_PSD_GRAY_0.psd -pix_fmt gray16be -f null - Without the patch : (using float calc) frame= 750 fps= 23 q=-0.0 Lsize=N/A time=00:00:30.00 bitrate=N/A speed=0.909x bench: utime=26.577s stime=10.061s rtime=33.063s bench: maxrss=697888768kB With the patch : (not use float calc) frame= 750 fps= 27 q=-0.0 Lsize=N/A time=00:00:30.00 bitrate=N/A speed=1.07x bench: utime=21.512s stime=10.136s rtime=27.992s bench: maxrss=697942016kB Martin ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
Re: [FFmpeg-devel] swscale/input : avoid float calc for GrayFloat to Gray16 conv
On Mon, Aug 20, 2018 at 04:41:02PM +0200, Martin Vignali wrote: > Hello, > > Patch in attach modify GrayFloat to Gray16 conversion > using the same method currently use inside exr decoder (no float calc) > > duplicate the float_to_uint func inside swscale_internal > > Martin > input.c| 10 +- > swscale_internal.h | 20 > 2 files changed, 25 insertions(+), 5 deletions(-) > 075a95ff08c3976a0b804422c515e49bea3a529b > 0004-swscale-input-avoid-float-calc-for-grayFloat-to-uint.patch > From 742160d2f161b316ad923b0ea0c60d79193c3357 Mon Sep 17 00:00:00 2001 > From: Martin Vignali > Date: Mon, 20 Aug 2018 16:35:13 +0200 > Subject: [PATCH 4/4] swscale/input : avoid float calc for grayFloat to uint16 > conversion > > reuse the float to uint16 func currently use inside exr decoder > --- > libswscale/input.c| 10 +- > libswscale/swscale_internal.h | 20 > 2 files changed, 25 insertions(+), 5 deletions(-) please provide benchmark, what is the impact on speed from this ? thx [...] -- Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB During times of universal deceit, telling the truth becomes a revolutionary act. -- George Orwell signature.asc Description: PGP signature ___ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel