This filter does HDR(HDR10/HLG) to SDR conversion with tone-mapping.
An example command to use this filter with vaapi codecs:
FFMPEG -init_hw_device vaapi=va:/dev/dri/renderD128 -init_hw_device \
opencl=ocl@va -hwaccel vaapi -hwaccel_device va -hwaccel_output_format \
vaapi -i INPUT -filter_hw_device ocl -filter_complex \
'[0:v]hwmap,tonemap_opencl=t=bt2020:tonemap=linear:format=p010[x1]; \
[x1]hwmap=derive_device=vaapi:reverse=1' -c:v hevc_vaapi -profile 2 OUTPUT
Signed-off-by: Ruiling Song <ruiling.song@intel.com>
HMS is formatted as HH:MM:SS.mmm, but, HH part is not limited to
24 hours. For example, the the drawn text may look like this:
243029:20:30.342. To present the timestamp in more readable and
user friendly format, this patch provides an additional option
to limit the hour part in the range 0-23.
Note: Actually the above required format can be obtained with
format options 'localtime' and 'gmtime', but, milliseconds part
is not supported in those formats.
These files depend on libavformat, and the vf_srcnn filter
currently is the only thing utilizing these dnn_* files and
already happens to have a dependency on libavformat.
This fixes compilation in cases where libavformat is not a
dependency for libavfilter.
Reported by Kam_ on IRC.
Generates color bar test patterns based on EBU PAL recommendations.
Reviewed-by: Paul B Mahol <onemda@gmail.com>
Signed-off-by: Tobias Rapp <t.rapp@noa-archive.com>
This helps figuring out where the filter is slow:
70.53% ffmpeg_g ffmpeg_g [.] nlmeans_slice
25.73% ffmpeg_g ffmpeg_g [.] compute_safe_ssd_integral_image_c
1.74% ffmpeg_g ffmpeg_g [.] compute_unsafe_ssd_integral_image
0.82% ffmpeg_g ffmpeg_g [.] ff_mjpeg_decode_sos
0.51% ffmpeg_g [unknown] [k] 0xffffffff91800a80
0.24% ffmpeg_g ffmpeg_g [.] weight_averages
(Tested with a large image that takes several seconds to process)
Since this function is irrelevant speed wise, the file's TODO is
updated.
before: ssd_integral_image_c: 49204.6
after: ssd_integral_image_c: 44272.8
Unrolling by 4 made the biggest difference on odroid-c2 (aarch64);
unrolling by 2 or 8 both raised 46k cycles vs 44k for 4.
Additionally, this is a much better reference when writing SIMD (SIMD
vectorization will just target 16 instead of 4).
SIMD code will not have to deal with padding itself. Overwriting in that
function may have been possible but involve large overreading of the
sources. Instead, we simply make sure the width to process is always a
multiple of 16. Additionally, there must be some actual area to process
so the SIMD code can have its boundary checks after processing the first
pixels.
