Message ID | c2cf08f6-a81c-bdeb-afa4-3598597a0383@mediaarea.net |
---|---|
State | Accepted |
Commit | fb580731c1febec44e0f1ec1433f6b3ee89cfe65 |
Headers | show |
On Wed, Feb 14, 2018 at 09:48:20AM +0100, Jerome Martinez wrote: > Add support for 10- and 12-bit/component RGB with Alpha encoding and > decoding in FFV1. > > Benched with START/STOP_TIMER around "for (x = 0; x < w; x++)" part during > decoding, before the previous patch and with the previous patch + this > patch, no obvious impact (+/-1%), e.g. with 1 second of gbrp12 4K content: > ffmpeg -i a.mkv -f framemd5 a.framemd5 -y > 325549 UNITS in decode_rgb_frame, 130899 runs, 173 skips > frame= 24 fps=1.3 q=-0.0 Lsize= 2kB time=00:00:01.00 bitrate= > 17.1kbits/s speed=0.0544x > > The 2 deleted lines are the ones I inadvertently kept after testing in my > previous patch, they are actually never used. > > Some input test files used for testing (input file framemd5 = FFV1 file > framemd5): > https://samples.ffmpeg.org/ffmpeg-bugs/trac/ticket2392/converted_image_gets_skewed.dpx > https://samples.ffmpeg.org/image-samples/dpx_samples.zip > > For reference, this leads to the following array of supported pix_fmt > (underscore means not supported): > Y : 8/_/10/12/__/16 > YA : 8/_/__/__/__/__ > YUV 420: 8/9/10/12/__/16 > YUVA 420: 8/9/10/__/__/16 > YUV 422: 8/9/10/12/__/16 > YUVA 422: 8/9/10/__/__/16 > YUV 444: 8/9/10/12/__/16 > YUVA 444: 8/9/10/__/__/16 > RGB : 8/9/10/12/14/16 > RGBA : 8/_/10/12/__/16 > and 8-bit for YUV 410/411/440 > it could be interesting for coherency in the listed supported pix_fmt to > fill some gaps when corresponding pix_fmt exists in FFmpeg e.g. YUVA 12-bit, > GRAY9 or YA16 > > ffv1dec.c | 4 ++++ > ffv1dec_template.c | 4 +--- > ffv1enc.c | 3 +++ > ffv1enc_template.c | 2 +- > 4 files changed, 9 insertions(+), 4 deletions(-) > 156fd44f16601376ba5d5dcc9f8ab935b6f83726 0001-avcodec-ffv1-Support-for-GBRAP10-and-GBRAP12.patch > From 3d24b30f2f23a2624e00333911e82fb48cf6d35c Mon Sep 17 00:00:00 2001 > From: =?UTF-8?q?J=C3=A9r=C3=B4me=20Martinez?= <jerome@mediaarea.net> > Date: Wed, 14 Feb 2018 08:39:15 +0100 > Subject: [PATCH] avcodec/ffv1: Support for GBRAP10 and GBRAP12 will apply please add fate tests thx [...]
diff --git a/libavcodec/ffv1dec.c b/libavcodec/ffv1dec.c index 923b79f3ab..3d2ee2569f 100644 --- a/libavcodec/ffv1dec.c +++ b/libavcodec/ffv1dec.c @@ -688,8 +688,12 @@ static int read_header(FFV1Context *f) f->avctx->pix_fmt = AV_PIX_FMT_GBRP9; else if (f->avctx->bits_per_raw_sample == 10 && !f->transparency) f->avctx->pix_fmt = AV_PIX_FMT_GBRP10; + else if (f->avctx->bits_per_raw_sample == 10 && f->transparency) + f->avctx->pix_fmt = AV_PIX_FMT_GBRAP10; else if (f->avctx->bits_per_raw_sample == 12 && !f->transparency) f->avctx->pix_fmt = AV_PIX_FMT_GBRP12; + else if (f->avctx->bits_per_raw_sample == 12 && f->transparency) + f->avctx->pix_fmt = AV_PIX_FMT_GBRAP12; else if (f->avctx->bits_per_raw_sample == 14 && !f->transparency) f->avctx->pix_fmt = AV_PIX_FMT_GBRP14; else if (f->avctx->bits_per_raw_sample == 16 && !f->transparency) { diff --git a/libavcodec/ffv1dec_template.c b/libavcodec/ffv1dec_template.c index a25b3384f1..f8a42a6d44 100644 --- a/libavcodec/ffv1dec_template.c +++ b/libavcodec/ffv1dec_template.c @@ -155,7 +155,7 @@ static void RENAME(decode_rgb_frame)(FFV1Context *s, uint8_t *src[4], int w, int if (lbd) *((uint32_t*)(src[0] + x*4 + stride[0]*y)) = b + ((unsigned)g<<8) + ((unsigned)r<<16) + ((unsigned)a<<24); - else if (sizeof(TYPE) == 4) { + else if (sizeof(TYPE) == 4 || transparency) { *((uint16_t*)(src[0] + x*2 + stride[0]*y)) = g; *((uint16_t*)(src[1] + x*2 + stride[1]*y)) = b; *((uint16_t*)(src[2] + x*2 + stride[2]*y)) = r; @@ -165,8 +165,6 @@ static void RENAME(decode_rgb_frame)(FFV1Context *s, uint8_t *src[4], int w, int *((uint16_t*)(src[0] + x*2 + stride[0]*y)) = b; *((uint16_t*)(src[1] + x*2 + stride[1]*y)) = g; *((uint16_t*)(src[2] + x*2 + stride[2]*y)) = r; - if (transparency) - *((uint16_t*)(src[3] + x*2 + stride[3]*y)) = a; } } } diff --git a/libavcodec/ffv1enc.c b/libavcodec/ffv1enc.c index 49b8d590a3..d71d952c6d 100644 --- a/libavcodec/ffv1enc.c +++ b/libavcodec/ffv1enc.c @@ -648,9 +648,11 @@ FF_ENABLE_DEPRECATION_WARNINGS if (!avctx->bits_per_raw_sample) s->bits_per_raw_sample = 9; case AV_PIX_FMT_GBRP10: + case AV_PIX_FMT_GBRAP10: if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample) s->bits_per_raw_sample = 10; case AV_PIX_FMT_GBRP12: + case AV_PIX_FMT_GBRAP12: if (!avctx->bits_per_raw_sample && !s->bits_per_raw_sample) s->bits_per_raw_sample = 12; case AV_PIX_FMT_GBRP14: @@ -1326,6 +1328,7 @@ AVCodec ff_ffv1_encoder = { AV_PIX_FMT_YUVA444P9, AV_PIX_FMT_YUVA422P9, AV_PIX_FMT_YUVA420P9, AV_PIX_FMT_GRAY16, AV_PIX_FMT_GRAY8, AV_PIX_FMT_GBRP9, AV_PIX_FMT_GBRP10, AV_PIX_FMT_GBRP12, AV_PIX_FMT_GBRP14, + AV_PIX_FMT_GBRAP10, AV_PIX_FMT_GBRAP12, AV_PIX_FMT_YA8, AV_PIX_FMT_GRAY10, AV_PIX_FMT_GRAY12, AV_PIX_FMT_GBRP16, AV_PIX_FMT_RGB48, diff --git a/libavcodec/ffv1enc_template.c b/libavcodec/ffv1enc_template.c index b93336538b..bc0add5ed7 100644 --- a/libavcodec/ffv1enc_template.c +++ b/libavcodec/ffv1enc_template.c @@ -160,7 +160,7 @@ static int RENAME(encode_rgb_frame)(FFV1Context *s, const uint8_t *src[4], b = p[2]; if (transparency) a = p[3]; - } else if (sizeof(TYPE) == 4) { + } else if (sizeof(TYPE) == 4 || transparency) { g = *((const uint16_t *)(src[0] + x*2 + stride[0]*y)); b = *((const uint16_t *)(src[1] + x*2 + stride[1]*y)); r = *((const uint16_t *)(src[2] + x*2 + stride[2]*y));
Add support for 10- and 12-bit/component RGB with Alpha encoding and decoding in FFV1. Benched with START/STOP_TIMER around "for (x = 0; x < w; x++)" part during decoding, before the previous patch and with the previous patch + this patch, no obvious impact (+/-1%), e.g. with 1 second of gbrp12 4K content: ffmpeg -i a.mkv -f framemd5 a.framemd5 -y 325549 UNITS in decode_rgb_frame, 130899 runs, 173 skips frame= 24 fps=1.3 q=-0.0 Lsize= 2kB time=00:00:01.00 bitrate= 17.1kbits/s speed=0.0544x The 2 deleted lines are the ones I inadvertently kept after testing in my previous patch, they are actually never used. Some input test files used for testing (input file framemd5 = FFV1 file framemd5): https://samples.ffmpeg.org/ffmpeg-bugs/trac/ticket2392/converted_image_gets_skewed.dpx https://samples.ffmpeg.org/image-samples/dpx_samples.zip For reference, this leads to the following array of supported pix_fmt (underscore means not supported): Y : 8/_/10/12/__/16 YA : 8/_/__/__/__/__ YUV 420: 8/9/10/12/__/16 YUVA 420: 8/9/10/__/__/16 YUV 422: 8/9/10/12/__/16 YUVA 422: 8/9/10/__/__/16 YUV 444: 8/9/10/12/__/16 YUVA 444: 8/9/10/__/__/16 RGB : 8/9/10/12/14/16 RGBA : 8/_/10/12/__/16 and 8-bit for YUV 410/411/440 it could be interesting for coherency in the listed supported pix_fmt to fill some gaps when corresponding pix_fmt exists in FFmpeg e.g. YUVA 12-bit, GRAY9 or YA16 From 3d24b30f2f23a2624e00333911e82fb48cf6d35c Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?J=C3=A9r=C3=B4me=20Martinez?= <jerome@mediaarea.net> Date: Wed, 14 Feb 2018 08:39:15 +0100 Subject: [PATCH] avcodec/ffv1: Support for GBRAP10 and GBRAP12 --- libavcodec/ffv1dec.c | 4 ++++ libavcodec/ffv1dec_template.c | 4 +--- libavcodec/ffv1enc.c | 3 +++ libavcodec/ffv1enc_template.c | 2 +- 4 files changed, 9 insertions(+), 4 deletions(-)