[x265] [PATCH] cleaning up CLI options, and their params
Sumalatha Polureddy
sumalatha at multicorewareinc.com
Fri Jun 28 14:20:06 CEST 2013
Answers are inline
Regards
Sumalatha/Aarthi
On Fri, Jun 28, 2013 at 12:06 AM, Steve Borho <steve at borho.org> wrote:
>
>
>
> On Thu, Jun 27, 2013 at 7:09 AM, <sumalatha at multicorewareinc.com> wrote:
>
>> # HG changeset patch
>> # User sumalatha
>> # Date 1372334962 -19800
>> # Node ID 8c818d8d27e4786cd20efe4122ece249701d679a
>> # Parent 3b93256a844bbbe5584ec1ae7b95e618bf955d69
>> cleaning up CLI options, and their params
>>
>
> Thanks for taking on this task. This work will need to happen one or two
> parameters at a time.
>
> In a single patch you will need to remove a parameter from x265.h, remove
> its CLI option from
> x265opts.h, fixup the configure() function in encoder.cpp so that a sane
> default is used in
> place of the former parameter, and also remove validation checks for that
> parameter from common.cpp.
>
> Even better is if you determine that a parameter is truly not used
> anymore, it should be removed
> from TEncCfg (including its access functions).
>
>
>> diff -r 3b93256a844b -r 8c818d8d27e4 source/x265opts.h
>> --- a/source/x265opts.h Thu Jun 27 16:14:14 2013 +0530
>> +++ b/source/x265opts.h Thu Jun 27 17:39:22 2013 +0530
>> @@ -27,12 +27,7 @@
>> HELP("Quad-Tree analysis:")
>> OPT("wpp", param->bEnableWavefront, no_argument, 0,
>> "Enable Wavefront Parallel Processing")
>> OPT("no-wpp", param->bEnableWavefront, no_argument, 0,
>> "Disable Wavefront Parallel Processing")
>> -OPT("ctu", param->maxCUSize, required_argument,
>> 's', "Maximum CU size (default: 64x64)")
>> -OPT("cu-depth", param->maxCUDepth, required_argument,
>> 'd', "Maximum CU recursion depth (default: 4)")
>> -OPT("tu-maxlog2", param->tuQTMaxLog2Size, required_argument, 0,
>> "Maximum TU size in logarithm base 2")
>> -OPT("tu-minlog2", param->tuQTMinLog2Size, required_argument, 0,
>> "Minimum TU size in logarithm base 2")
>> -OPT("tu-intra-depth", param->tuQTMaxIntraDepth, required_argument, 0,
>> "Max TU recursive depth for intra CUs")
>> -OPT("tu-inter-depth", param->tuQTMaxInterDepth, required_argument, 0,
>> "Max TU recursive depth for inter CUs")
>>
>
> maxCUSize is a param we need to keep, but maxCUDepth is not (it should be
> calculated based on maxCUSize).
> tuQTMaxLog2Size and tuQTMinLog2Size can both be calculated as well, I
> think. tuQTMaxIntraDepth and tuQTMaxInterDepth must stay, they are key
> tuning parameters.
>
1. Done the removal of maxCUDepth , tuQTMaxLog2Size and tuQTMinLog2Size
>
>
>>
>> HELP("Temporal / motion search options:")
>> OPT("me", param->searchMethod, required_argument,
>> 0, "Motion search method 0:dia 1:hex 2:umh 3:star(default) 4:hm-orig
>> 5:full")
>> @@ -44,13 +39,12 @@
>> OPT("no-amp", param->bEnableAMP, no_argument,
>> 0, "Disable asymmetric motion partitions")
>> OPT("rdo", param->bEnableRDO, no_argument,
>> 0, "Enable rate distortion-based mode decision")
>> OPT("no-rdo", param->bEnableRDO, no_argument,
>> 0, "Enable mode decision without rate distortion optimization")
>> -OPT("max-merge", param->maxNumMergeCand, required_argument,
>> 0, "Maximum number of merge candidates (default: 5)")
>>
>
> maxNumMergeCand must stay, it is a perf tuning parameter
>
>
>> +
>> OPT("fdm", param->bEnableFastMergeDecision, no_argument,
>> 0, "Enable fast decision for Merge RD Cost")
>> OPT("no-fdm", param->bEnableFastMergeDecision, no_argument,
>> 0, "Disable fast decision for Merge RD Cost")
>> OPT("early-skip", param->bEnableEarlySkip, no_argument,
>> 0, "Enable early SKIP detection")
>> OPT("no-early-skip", param->bEnableEarlySkip, no_argument,
>> 0, "Disable early SKIP detection")
>> OPT("merge-level", param->log2ParallelMergeLevel, required_argument,
>> 0, "Parallel merge estimation region")
>> -OPT("tmvp", param->TMVPModeId, required_argument,
>> 0, "TMVP mode 0:disabled 1:enabled(default) 2:auto")
>> OPT("fast-cbf", param->bEnableCbfFastMode, no_argument,
>> 0, "Enable Cbf fast mode")
>> OPT("no-fast-cbf", param->bEnableCbfFastMode, no_argument,
>> 0, "Disable Cbf fast mode")
>>
>
> All of these flags need to be tested to see if they still have any effect
> after all our mode-decision changes.
> I suspect log2ParallelMergeLevel and TMVPModeId can be hidden from the
> user - either fixed or auto-configured based on other inputs.
>
>
>>
>> @@ -60,8 +54,7 @@
>> OPT("no-tskip", param->bEnableTransformSkip, no_argument,
>> 0, "Disable intra transform skipping")
>> OPT("tskip-fast", param->bEnableTSkipFast, no_argument,
>> 0, "Enable fast intra transform skipping")
>> OPT("no-tskip-fast", param->bEnableTSkipFast, no_argument,
>> 0, "Disable fast intra transform skipping")
>> -OPT("strong-intra-smoothing", param->bEnableStrongIntraSmoothing,
>> no_argument, 0, "Enable strong intra smoothing for 32x32 blocks")
>> -OPT("no-strong-intra-smoothing", param->bEnableStrongIntraSmoothing,
>> no_argument, 0, "Disable strong intra smoothing for 32x32 blocks")
>> +
>>
>
> Does bEnableStrongIntraSmoothing not do anything?
>
bEnableStrongIntraSmoothing is required . It is used in initAdiPattern(),
it does some different filtering when it is disabled and enabled
>
>
>> OPT("constrained-intra", param->bEnableConstrainedIntra, no_argument,
>> 0, "Constrained intra prediction (use only intra coded reference pixels)")
>> OPT("no-constrained-intra", param->bEnableConstrainedIntra, no_argument,
>> 0, "Disable constrained intra prediction (use only intra coded reference
>> pixels)")
>>
>> @@ -74,7 +67,7 @@
>>
>> HELP("QP and rate distortion options:")
>> OPT("qp", param->qp, required_argument,
>> 'q', "Base QP for CQP mode (default: 30)")
>> -OPT("cbqpoffs", param->cbQpOffset, required_argument,
>> 0, "Chroma Cb QP Offset")
>> +
>>
>
> Why remove the Cb offset and not the Cr offset?
>
I doubt whether cb and cr offset is required.
At present it is set as zero
>
>
>> OPT("crqpoffs", param->crQpOffset, required_argument,
>> 0, "Chroma Cr QP Offset")
>> OPT("aqselect", param->bEnableAdaptQpSelect, no_argument,
>> 0, "Adaptive QP selection")
>> OPT("aq", param->bEnableAdaptiveQP, no_argument,
>> 0, "Enable QP adaptation based on a psycho-visual model")
>>
>
> I suspect all of AQ is broken, and needs to be hidden from the user
> interface until it is re-imlemented based on the x264 design. But I don't
> know if the Cb and Cr offsets work without AQ.
>
> --
> Steve
>
> _______________________________________________
> x265-devel mailing list
> x265-devel at videolan.org
> http://mailman.videolan.org/listinfo/x265-devel
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/x265-devel/attachments/20130628/b125bf9f/attachment-0001.html>
More information about the x265-devel
mailing list