[FFmpeg-devel] [PATCH v2 16/36] vaapi_encode: Clean up rate control configuration

Mark Thompson sw at jkqxz.net
Sun Jun 17 16:17:48 EEST 2018


On 14/06/18 07:07, Xiang, Haihao wrote:
> On Wed, 2018-06-13 at 23:42 +0100, Mark Thompson wrote:
>> On 13/06/18 08:03, Xiang, Haihao wrote:
>>> On Fri, 2018-06-08 at 00:43 +0100, Mark Thompson wrote:
>>>> Query which modes are supported and select between VBR and CBR based
>>>> on that - this removes all of the codec-specific rate control mode
>>>> selection code.
>>>> ---
>>>>  doc/encoders.texi               |   2 -
>>>>  libavcodec/vaapi_encode.c       | 173 ++++++++++++++++++++++++++++-------
>>>> ----
>>>> -
>>>>  libavcodec/vaapi_encode.h       |   6 +-
>>>>  libavcodec/vaapi_encode_h264.c  |  18 +----
>>>>  libavcodec/vaapi_encode_h265.c  |  14 +---
>>>>  libavcodec/vaapi_encode_mjpeg.c |   3 +-
>>>>  libavcodec/vaapi_encode_mpeg2.c |   9 +--
>>>>  libavcodec/vaapi_encode_vp8.c   |  13 +--
>>>>  libavcodec/vaapi_encode_vp9.c   |  13 +--
>>>>  9 files changed, 137 insertions(+), 114 deletions(-)
>>>>
>>>> ...
>>>> diff --git a/libavcodec/vaapi_encode.c b/libavcodec/vaapi_encode.c
>>>> index f4c063734c..5de5483454 100644
>>>> --- a/libavcodec/vaapi_encode.c
>>>> +++ b/libavcodec/vaapi_encode.c
>>>> ...
>>>> @@ -1313,44 +1286,144 @@ static av_cold int
>>>> vaapi_encode_config_attributes(AVCodecContext *avctx)
>>>>  static av_cold int vaapi_encode_init_rate_control(AVCodecContext *avctx)
>>>>  {
>>>>      VAAPIEncodeContext *ctx = avctx->priv_data;
>>>> -    int rc_bits_per_second;
>>>> -    int rc_target_percentage;
>>>> -    int rc_window_size;
>>>> -    int hrd_buffer_size;
>>>> -    int hrd_initial_buffer_fullness;
>>>> +    int64_t rc_bits_per_second;
>>>> +    int     rc_target_percentage;
>>>> +    int     rc_window_size;
>>>> +    int64_t hrd_buffer_size;
>>>> +    int64_t hrd_initial_buffer_fullness;
>>>>      int fr_num, fr_den;
>>>> +    VAConfigAttrib rc_attr = { VAConfigAttribRateControl };
>>>> +    VAStatus vas;
>>>> +
>>>> +    vas = vaGetConfigAttributes(ctx->hwctx->display,
>>>> +                                ctx->va_profile, ctx->va_entrypoint,
>>>> +                                &rc_attr, 1);
>>>> +    if (vas != VA_STATUS_SUCCESS) {
>>>> +        av_log(avctx, AV_LOG_ERROR, "Failed to query rate control "
>>>> +               "config attribute: %d (%s).\n", vas, vaErrorStr(vas));
>>>> +        return AVERROR_EXTERNAL;
>>>> +    }
>>>>  
>>>> -    if (avctx->bit_rate > INT32_MAX) {
>>>> -        av_log(avctx, AV_LOG_ERROR, "Target bitrate of 2^31 bps or "
>>>> -               "higher is not supported.\n");
>>>> +    if (rc_attr.value == VA_ATTRIB_NOT_SUPPORTED) {
>>>> +        av_log(avctx, AV_LOG_VERBOSE, "Driver does not report any "
>>>> +               "supported rate control modes: assuming constant-
>>>> quality.\n");
>>>
>>> How about logging it as warning message?
>>
>> What would a user do about it?
>>
> 
> User may know what is not supported in the driver and he/she might get wrong
> result, he/she may file an issue to the driver.

If it's broken then the verbose logging provides the answer.  If it isn't broken, as in

>> (Note that it gets logged for JPEG encoding on all versions of the i965 driver
>> except the most recent, so if it were a warning it would be seen by everyone
>> using those versions.)

then the warning is unhelpful.

>>>> +        ctx->va_rc_mode = VA_RC_CQP;
>>>> +        return 0;
>>>> +    }
>>>> +    if (avctx->flags & AV_CODEC_FLAG_QSCALE ||
>>>> +        avctx->bit_rate <= 0) {
>>>> +        if (rc_attr.value & VA_RC_CQP) {
>>>> +            av_log(avctx, AV_LOG_VERBOSE, "Using constant-quality
>>>> mode.\n");
>>>> +            ctx->va_rc_mode = VA_RC_CQP;
>>>> +            return 0;
>>>> +        } else {
>>>> +            av_log(avctx, AV_LOG_ERROR, "Driver does not support "
>>>> +                   "constant-quality mode (%#x).\n", rc_attr.value);
>>>> +            return AVERROR(EINVAL);
>>>> +        }
>>>> +    }
>>>> +
>>>> +    if (!(rc_attr.value & (VA_RC_CBR | VA_RC_VBR))) {
>>>> +        av_log(avctx, AV_LOG_ERROR, "Driver does not support any "
>>>> +               "bitrate-targetted rate control modes.\n");
>>>>          return AVERROR(EINVAL);
>>>>      }
>>>>  
>>>>      if (avctx->rc_buffer_size)
>>>>          hrd_buffer_size = avctx->rc_buffer_size;
>>>> +    else if (avctx->rc_max_rate > 0)
>>>> +        hrd_buffer_size = avctx->rc_max_rate;
>>>>      else
>>>>          hrd_buffer_size = avctx->bit_rate;
>>>> -    if (avctx->rc_initial_buffer_occupancy)
>>>> +    if (avctx->rc_initial_buffer_occupancy) {
>>>> +        if (avctx->rc_initial_buffer_occupancy > hrd_buffer_size) {
>>>> +            av_log(avctx, AV_LOG_ERROR, "Invalid RC buffer settings: "
>>>> +                   "must have initial buffer size (%d) < "
>>>> +                   "buffer size (%"PRId64").\n",
>>>> +                   avctx->rc_initial_buffer_occupancy, hrd_buffer_size);
>>>> +            return AVERROR(EINVAL);
>>>> +        }
>>>>          hrd_initial_buffer_fullness = avctx->rc_initial_buffer_occupancy;
>>>> -    else
>>>> +    } else {
>>>>          hrd_initial_buffer_fullness = hrd_buffer_size * 3 / 4;
>>>> +    }
>>>> +
>>>> +    if (avctx->rc_max_rate && avctx->rc_max_rate < avctx->bit_rate) {
>>>> +        av_log(avctx, AV_LOG_ERROR, "Invalid bitrate settings: must have
>>>> "
>>>> +               "bitrate (%"PRId64") <= maxrate (%"PRId64").\n",
>>>> +               avctx->bit_rate, avctx->rc_max_rate);
>>>> +        return AVERROR(EINVAL);
>>>> +    }
>>>> +
>>>> +    if (avctx->rc_max_rate > avctx->bit_rate) {
>>>> +        if (!(rc_attr.value & VA_RC_VBR)) {
>>>> +            av_log(avctx, AV_LOG_WARNING, "Driver does not support "
>>>> +                   "VBR mode (%#x), using CBR mode instead.\n",
>>>> +                   rc_attr.value);
>>>> +            ctx->va_rc_mode = VA_RC_CBR;
>>>> +        } else {
>>>> +            ctx->va_rc_mode = VA_RC_VBR;
>>>> +        }
>>>> +
>>>> +        rc_bits_per_second   = avctx->rc_max_rate;
>>>> +        rc_target_percentage = (avctx->bit_rate * 100) / avctx-
>>>>> rc_max_rate;
>>>
>>> I think rc_target_percentage should be 100 for CBR case.
>>
>> Yes; fixed.
>>
>> The rc_bits_per_second value is also wrong in that case (shouldn't be
>> rc_max_rate if we can't use it), fixed similarly.
>>
>>>> +
>>>> +    } else if (avctx->rc_max_rate == avctx->bit_rate) {
>>>> +        if (!(rc_attr.value & VA_RC_CBR)) {
>>>> +            av_log(avctx, AV_LOG_WARNING, "Driver does not support "
>>>> +                   "CBR mode (%#x), using VBR mode instead.\n",
>>>> +                   rc_attr.value);
>>>> +            ctx->va_rc_mode = VA_RC_VBR;
>>>> +        } else {
>>>> +            ctx->va_rc_mode = VA_RC_CBR;
>>>> +        }
>>>>  
>>>> -    if (ctx->va_rc_mode == VA_RC_CBR) {
>>>>          rc_bits_per_second   = avctx->bit_rate;
>>>>          rc_target_percentage = 100;
>>>> -        rc_window_size       = 1000;
>>>> +
>>>>      } else {
>>>> -        if (avctx->rc_max_rate < avctx->bit_rate) {
>>>> -            // Max rate is unset or invalid, just use the normal bitrate.
>>>> +        if (rc_attr.value & VA_RC_VBR) {
>>>> +            ctx->va_rc_mode = VA_RC_VBR;
>>>
>>> Is it better to take it as CBR when avctx->rc_max_rate is 0 and CBR is
>>> supported
>>> by driver?
>>
>> I don't think so?  VBR with the specified target is probably what you want in
>> most cases, and I think anyone with specific constraints that want constant
>> bitrate should expect to set maxrate to achieve that.
>>
> 
> I agree VBR is probably what an user wants in most case, however target percent
> set to 50% is not suitable for most case. To get a specific target percent, user
> should set both target bitrate and max bitrate, so it is reasonable to ask user
> must set both target bitrate and max bitrate for VBR cases, and for CBR user may
> set target bitrate only. 
> 
> I just saw it is also VBR in QSV when max bitrate is not set so I'm fine to keep
> consistency with QSV for this case.

Max bitrate has to be set because of the API limitations (because the target bitrate is an integer percentage of max bitrate rather than a standalone value as found in pretty much every other API).

Here we'd like to say "no max bitrate constraint", but that isn't possible.  I picked 50% because it's high enough to be mostly unconstrained, while not making numbers too much larger so that you might run into rounding or overflow issues.  (E.g. picking 1% here might feel better somehow, but that overflows the 2^32 limit at only 40Mbps.)

If you prefer different settings for this case then can you explain what they would be?

>>>> +
>>>> +            // We only have a target bitrate, but VAAPI requires that a
>>>> +            // maximum rate be supplied as well.  Since the user has
>>>> +            // offered no particular constraint, arbitrarily pick a
>>>> +            // maximum rate of double the target rate.
>>>> +            rc_bits_per_second   = 2 * avctx->bit_rate;
>>>> +            rc_target_percentage = 50;
>>>> +        } else {
>>>> +            ctx->va_rc_mode = VA_RC_CBR;
>>>> +
>>>>              rc_bits_per_second   = avctx->bit_rate;
>>>>              rc_target_percentage = 100;
>>>> -        } else {
>>>> -            rc_bits_per_second   = avctx->rc_max_rate;
>>>> -            rc_target_percentage = (avctx->bit_rate * 100) /
>>>> rc_bits_per_second;
>>>>          }
>>>> -        rc_window_size = (hrd_buffer_size * 1000) / avctx->bit_rate;
>>>>      }
>>>>  
>>>> +    rc_window_size = (hrd_buffer_size * 1000) / rc_bits_per_second;
>>>> +
>>>> +    av_log(avctx, AV_LOG_VERBOSE, "RC mode: %s, %d%% of %"PRId64" bps "
>>>> +           "over %d ms.\n", ctx->va_rc_mode == VA_RC_VBR ? "VBR" : "CBR",
>>>> +           rc_target_percentage, rc_bits_per_second, rc_window_size);
>>>> +    av_log(avctx, AV_LOG_VERBOSE, "RC buffer: %"PRId64" bits, "
>>>> +           "initial fullness %"PRId64" bits.\n",
>>>> +           hrd_buffer_size, hrd_initial_buffer_fullness);
>>>> +
>>>> +    if (rc_bits_per_second          > UINT32_MAX ||
>>>> +        hrd_buffer_size             > UINT32_MAX ||
>>>> +        hrd_initial_buffer_fullness > UINT32_MAX) {
>>>> +        av_log(avctx, AV_LOG_ERROR, "RC parameters of 2^32 or "
>>>> +               "greater are not supported by VAAPI.\n");
>>>> +        return AVERROR(EINVAL);
>>>> +    }
>>>> +
>>>> +    ctx->va_bit_rate = rc_bits_per_second;
>>>> +
>>>> +    ctx->config_attributes[ctx->nb_config_attributes++] =
>>>> +        (VAConfigAttrib) {
>>>> +        .type  = VAConfigAttribRateControl,
>>>> +        .value = ctx->va_rc_mode,
>>>> +    };
>>>> +
>>>>      ctx->rc_params.misc.type = VAEncMiscParameterTypeRateControl;
>>>>      ctx->rc_params.rc = (VAEncMiscParameterRateControl) {
>>>>          .bits_per_second   = rc_bits_per_second,
>>>> @@ -1611,6 +1684,10 @@ av_cold int ff_vaapi_encode_init(AVCodecContext
>>>> *avctx)
>>>>      if (err < 0)
>>>>          goto fail;
>>>>  
>>>> +    err = vaapi_encode_init_rate_control(avctx);
>>>> +    if (err < 0)
>>>> +        goto fail;
>>>> +
>>>>      err = vaapi_encode_config_attributes(avctx);
>>>>      if (err < 0)
>>>>          goto fail;
>>>> @@ -1658,12 +1735,6 @@ av_cold int ff_vaapi_encode_init(AVCodecContext
>>>> *avctx)
>>>>          goto fail;
>>>>      }
>>>>  
>>>> -    if (ctx->va_rc_mode & ~VA_RC_CQP) {
>>>> -        err = vaapi_encode_init_rate_control(avctx);
>>>> -        if (err < 0)
>>>> -            goto fail;
>>>> -    }
>>>> -
>>>>      if (ctx->codec->configure) {
>>>>          err = ctx->codec->configure(avctx);
>>>>          if (err < 0)
>>>> ...


More information about the ffmpeg-devel mailing list