On 14/06/18 07:48, Xiang, Haihao wrote:
> On Fri, 2018-06-08 at 00:43 +0100, Mark Thompson wrote:
>> Choose what types of reference frames will be used based on what types
>> are available, and make the intra-only mode explicit (GOP size one,
>> which must be used for MJPEG).
>> ---
>> libavcodec/vaapi_encode.c | 83
>> +++++++++++++++++++++++++++-------------
>> -
>> libavcodec/vaapi_encode.h | 1 +
>> libavcodec/vaapi_encode_h264.c | 4 +-
>> libavcodec/vaapi_encode_h265.c | 4 +-
>> libavcodec/vaapi_encode_mjpeg.c | 1 +
>> libavcodec/vaapi_encode_mpeg2.c | 2 +-
>> libavcodec/vaapi_encode_vp8.c | 7 +---
>> libavcodec/vaapi_encode_vp9.c | 7 ++--
>> 8 files changed, 68 insertions(+), 41 deletions(-)
>>
>> diff --git a/libavcodec/vaapi_encode.c b/libavcodec/vaapi_encode.c
>> index 0e806a29e3..af9224c98f 100644
>> --- a/libavcodec/vaapi_encode.c
>> +++ b/libavcodec/vaapi_encode.c
>> @@ -680,7 +680,7 @@ static int vaapi_encode_get_next(AVCodecContext *avctx,
>> return AVERROR(ENOMEM);
>>
>> if (ctx->input_order == 0 || ctx->force_idr ||
>> - ctx->gop_counter >= avctx->gop_size) {
>> + ctx->gop_counter >= ctx->gop_size) {
>> pic->type = PICTURE_TYPE_IDR;
>> ctx->force_idr = 0;
>> ctx->gop_counter = 1;
>> @@ -703,7 +703,7 @@ static int vaapi_encode_get_next(AVCodecContext *avctx,
>> // encode-after it, but not exceeding the GOP size.
>>
>> for (i = 0; i < ctx->b_per_p &&
>> - ctx->gop_counter < avctx->gop_size; i++) {
>> + ctx->gop_counter < ctx->gop_size; i++) {
>> pic = vaapi_encode_alloc(avctx);
>> if (!pic)
>> goto fail;
>> @@ -1217,7 +1217,6 @@ static av_cold int
>> vaapi_encode_config_attributes(AVCodecContext *avctx)
>> int i;
>>
>> VAConfigAttrib attr[] = {
>> - { VAConfigAttribEncMaxRefFrames },
>> { VAConfigAttribEncPackedHeaders },
>> };
>>
>> @@ -1240,24 +1239,6 @@ static av_cold int
>> vaapi_encode_config_attributes(AVCodecContext *avctx)
>> continue;
>> }
>> switch (attr[i].type) {
>> - case VAConfigAttribEncMaxRefFrames:
>> - {
>> - unsigned int ref_l0 = attr[i].value & 0xffff;
>> - unsigned int ref_l1 = (attr[i].value >> 16) & 0xffff;
>> -
>> - if (avctx->gop_size > 1 && ref_l0 < 1) {
>> - av_log(avctx, AV_LOG_ERROR, "P frames are not "
>> - "supported (%#x).\n", attr[i].value);
>> - return AVERROR(EINVAL);
>> - }
>> - if (avctx->max_b_frames > 0 && ref_l1 < 1) {
>> - av_log(avctx, AV_LOG_WARNING, "B frames are not "
>> - "supported (%#x) by the underlying driver.\n",
>> - attr[i].value);
>> - avctx->max_b_frames = 0;
>> - }
>> - }
>> - break;
>> case VAConfigAttribEncPackedHeaders:
>> if (ctx->va_packed_headers & ~attr[i].value) {
>> // This isn't fatal, but packed headers are always
>> @@ -1465,6 +1446,54 @@ static av_cold int
>> vaapi_encode_init_rate_control(AVCodecContext *avctx)
>> return 0;
>> }
>>
>> +static av_cold int vaapi_encode_init_gop_structure(AVCodecContext *avctx)
>> +{
>> + VAAPIEncodeContext *ctx = avctx->priv_data;
>> + VAStatus vas;
>> + VAConfigAttrib attr = { VAConfigAttribEncMaxRefFrames };
>> + uint32_t ref_l0, ref_l1;
>> +
>> + vas = vaGetConfigAttributes(ctx->hwctx->display,
>> + ctx->va_profile,
>> + ctx->va_entrypoint,
>> + &attr, 1);
>> + if (vas != VA_STATUS_SUCCESS) {
>> + av_log(avctx, AV_LOG_ERROR, "Failed to query reference frames "
>> + "attribute: %d (%s).\n", vas, vaErrorStr(vas));
>> + return AVERROR_EXTERNAL;
>> + }
>> +
>> + if (attr.value == VA_ATTRIB_NOT_SUPPORTED) {
>> + ref_l0 = ref_l1 = 0;
>> + } else {
>> + ref_l0 = attr.value & 0xffff;
>> + ref_l1 = attr.value >> 16 & 0xffff;
>> + }
>> +
>> + if (avctx->gop_size <= 1) {
>> + av_log(avctx, AV_LOG_VERBOSE, "Using intra frames only.\n");
>> + ctx->gop_size = 1;
>> + } else if (ref_l0 < 1) {
>> + av_log(avctx, AV_LOG_ERROR, "Driver does not support any "
>> + "reference frames.\n");
>> + return AVERROR(EINVAL);
>> + } else if (ref_l1 < 1 || avctx->max_b_frames < 1) {
>> + av_log(avctx, AV_LOG_VERBOSE, "Using intra and P-frames "
>> + "(supported references: %d / %d).\n", ref_l0, ref_l1);
>> + ctx->gop_size = avctx->gop_size;
>> + ctx->p_per_i = INT_MAX;
>
> How about removing p_per_i? I see p_per_i is used only in the following 'else
> if' statement and I don't think ctx->p_counter can be INT_MAX in reality.
>
> } else if (ctx->p_counter >= ctx->p_per_i) {...}
It's not currently user-visible, but changing these values does work to test
the open-GOP handling in the codec-specific code. (E.g. in H.264, I frames as
recovery points rather than IDR frames.)
I do intend to expose this at some point.
>> + ctx->b_per_p = 0;
>> + } else {
>> + av_log(avctx, AV_LOG_VERBOSE, "Using intra, P- and B-frames "
>> + "(supported references: %d / %d).\n", ref_l0, ref_l1);
>> + ctx->gop_size = avctx->gop_size;
>> + ctx->p_per_i = INT_MAX;
>> + ctx->b_per_p = avctx->max_b_frames;
>> + }
>> +
>> + return 0;
>> +}
>> +
>> static av_cold int vaapi_encode_init_quality(AVCodecContext *avctx)
>> {
>> #if VA_CHECK_VERSION(0, 36, 0)
>> @@ -1636,7 +1665,7 @@ static av_cold int
>> vaapi_encode_create_recon_frames(AVCodecContext *avctx)
>> ctx->recon_frames->height = ctx->surface_height;
>> // At most three IDR/I/P frames and two runs of B frames can be in
>> // flight at any one time.
>> - ctx->recon_frames->initial_pool_size = 3 + 2 * avctx->max_b_frames;
>> + ctx->recon_frames->initial_pool_size = 3 + 2 * ctx->b_per_p;
>>
>> err = av_hwframe_ctx_init(ctx->recon_frames_ref);
>> if (err < 0) {
>> @@ -1691,6 +1720,10 @@ av_cold int ff_vaapi_encode_init(AVCodecContext
>> *avctx)
>> if (err < 0)
>> goto fail;
>>
>> + err = vaapi_encode_init_gop_structure(avctx);
>> + if (err < 0)
>> + goto fail;
>> +
>> err = vaapi_encode_config_attributes(avctx);
>> if (err < 0)
>> goto fail;
>> @@ -1745,14 +1778,10 @@ av_cold int ff_vaapi_encode_init(AVCodecContext
>> *avctx)
>> }
>>
>> ctx->input_order = 0;
>> - ctx->output_delay = avctx->max_b_frames;
>> + ctx->output_delay = ctx->b_per_p;
>> ctx->decode_delay = 1;
>> ctx->output_order = - ctx->output_delay - 1;
>>
>> - // Currently we never generate I frames, only IDR.
>> - ctx->p_per_i = INT_MAX;
>> - ctx->b_per_p = avctx->max_b_frames;
>> -
>> if (ctx->codec->sequence_params_size > 0) {
>> ctx->codec_sequence_params =
>> av_mallocz(ctx->codec->sequence_params_size);
Thanks,
- Mark
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
