X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fvaapi_encode.c;h=7510816ec193fe096dae66926e7adc1b2026bc17;hb=d07534b5f5f20b4f780f5b0284aca6354da00695;hp=313f55118e09773a2dfe042e44fa008cdaff97d6;hpb=1b98bfb932ad36667ea7f18e24c54978623f6654;p=ffmpeg diff --git a/libavcodec/vaapi_encode.c b/libavcodec/vaapi_encode.c index 313f55118e0..7510816ec19 100644 --- a/libavcodec/vaapi_encode.c +++ b/libavcodec/vaapi_encode.c @@ -21,12 +21,19 @@ #include "libavutil/avassert.h" #include "libavutil/common.h" +#include "libavutil/internal.h" #include "libavutil/log.h" #include "libavutil/pixdesc.h" #include "vaapi_encode.h" +#include "encode.h" #include "avcodec.h" +const AVCodecHWConfigInternal *const ff_vaapi_encode_hw_configs[] = { + HW_CONFIG_ENCODER_FRAMES(VAAPI, VAAPI), + NULL, +}; + static const char * const picture_type_name[] = { "IDR", "I", "P", "B" }; static int vaapi_encode_make_packed_header(AVCodecContext *avctx, @@ -103,6 +110,29 @@ static int vaapi_encode_make_param_buffer(AVCodecContext *avctx, return 0; } +static int vaapi_encode_make_misc_param_buffer(AVCodecContext *avctx, + VAAPIEncodePicture *pic, + int type, + const void *data, size_t len) +{ + // Construct the buffer on the stack - 1KB is much larger than any + // current misc parameter buffer type (the largest is EncQuality at + // 224 bytes). + uint8_t buffer[1024]; + VAEncMiscParameterBuffer header = { + .type = type, + }; + size_t buffer_size = sizeof(header) + len; + av_assert0(buffer_size <= sizeof(buffer)); + + memcpy(buffer, &header, sizeof(header)); + memcpy(buffer + sizeof(header), data, len); + + return vaapi_encode_make_param_buffer(avctx, pic, + VAEncMiscParameterBufferType, + buffer, buffer_size); +} + static int vaapi_encode_wait(AVCodecContext *avctx, VAAPIEncodePicture *pic) { @@ -134,6 +164,88 @@ static int vaapi_encode_wait(AVCodecContext *avctx, return 0; } +static int vaapi_encode_make_row_slice(AVCodecContext *avctx, + VAAPIEncodePicture *pic) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + VAAPIEncodeSlice *slice; + int i, rounding; + + for (i = 0; i < pic->nb_slices; i++) + pic->slices[i].row_size = ctx->slice_size; + + rounding = ctx->slice_block_rows - ctx->nb_slices * ctx->slice_size; + if (rounding > 0) { + // Place rounding error at top and bottom of frame. + av_assert0(rounding < pic->nb_slices); + // Some Intel drivers contain a bug where the encoder will fail + // if the last slice is smaller than the one before it. Since + // that's straightforward to avoid here, just do so. + if (rounding <= 2) { + for (i = 0; i < rounding; i++) + ++pic->slices[i].row_size; + } else { + for (i = 0; i < (rounding + 1) / 2; i++) + ++pic->slices[pic->nb_slices - i - 1].row_size; + for (i = 0; i < rounding / 2; i++) + ++pic->slices[i].row_size; + } + } else if (rounding < 0) { + // Remove rounding error from last slice only. + av_assert0(rounding < ctx->slice_size); + pic->slices[pic->nb_slices - 1].row_size += rounding; + } + + for (i = 0; i < pic->nb_slices; i++) { + slice = &pic->slices[i]; + slice->index = i; + if (i == 0) { + slice->row_start = 0; + slice->block_start = 0; + } else { + const VAAPIEncodeSlice *prev = &pic->slices[i - 1]; + slice->row_start = prev->row_start + prev->row_size; + slice->block_start = prev->block_start + prev->block_size; + } + slice->block_size = slice->row_size * ctx->slice_block_cols; + + av_log(avctx, AV_LOG_DEBUG, "Slice %d: %d-%d (%d rows), " + "%d-%d (%d blocks).\n", i, slice->row_start, + slice->row_start + slice->row_size - 1, slice->row_size, + slice->block_start, slice->block_start + slice->block_size - 1, + slice->block_size); + } + + return 0; +} + +static int vaapi_encode_make_tile_slice(AVCodecContext *avctx, + VAAPIEncodePicture *pic) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + VAAPIEncodeSlice *slice; + int i, j, index; + + for (i = 0; i < ctx->tile_cols; i++) { + for (j = 0; j < ctx->tile_rows; j++) { + index = j * ctx->tile_cols + i; + slice = &pic->slices[index]; + slice->index = index; + + pic->slices[index].block_start = ctx->col_bd[i] + + ctx->row_bd[j] * ctx->slice_block_cols; + pic->slices[index].block_size = ctx->row_height[j] * ctx->col_width[i]; + + av_log(avctx, AV_LOG_DEBUG, "Slice %2d: (%2d, %2d) start at: %4d " + "width:%2d height:%2d (%d blocks).\n", index, ctx->col_bd[i], + ctx->row_bd[j], slice->block_start, ctx->col_width[i], + ctx->row_height[j], slice->block_size); + } + } + + return 0; +} + static int vaapi_encode_issue(AVCodecContext *avctx, VAAPIEncodePicture *pic) { @@ -143,6 +255,7 @@ static int vaapi_encode_issue(AVCodecContext *avctx, int err, i; char data[MAX_PARAM_BUFFER_SIZE]; size_t bit_len; + av_unused AVFrameSideData *sd; av_log(avctx, AV_LOG_DEBUG, "Issuing encode for pic %"PRId64"/%"PRId64" " "as type %s.\n", pic->display_order, pic->encode_order, @@ -158,16 +271,10 @@ static int vaapi_encode_issue(AVCodecContext *avctx, av_log(avctx, AV_LOG_DEBUG, ".\n"); } - av_assert0(pic->input_available && !pic->encode_issued); + av_assert0(!pic->encode_issued); for (i = 0; i < pic->nb_refs; i++) { av_assert0(pic->refs[i]); - // If we are serialised then the references must have already - // completed. If not, they must have been issued but need not - // have completed yet. - if (ctx->issue_mode == ISSUE_MODE_SERIALISE_EVERYTHING) - av_assert0(pic->refs[i]->encode_complete); - else - av_assert0(pic->refs[i]->encode_issued); + av_assert0(pic->refs[i]->encode_issued); } av_log(avctx, AV_LOG_DEBUG, "Input surface is %#x.\n", pic->input_surface); @@ -207,19 +314,6 @@ static int vaapi_encode_issue(AVCodecContext *avctx, pic->nb_param_buffers = 0; - if (pic->encode_order == 0) { - // Global parameter buffers are set on the first picture only. - - for (i = 0; i < ctx->nb_global_params; i++) { - err = vaapi_encode_make_param_buffer(avctx, pic, - VAEncMiscParameterBufferType, - (char*)ctx->global_params[i], - ctx->global_params_size[i]); - if (err < 0) - goto fail; - } - } - if (pic->type == PICTURE_TYPE_IDR && ctx->codec->init_sequence_params) { err = vaapi_encode_make_param_buffer(avctx, pic, VAEncSequenceParameterBufferType, @@ -229,6 +323,17 @@ static int vaapi_encode_issue(AVCodecContext *avctx, goto fail; } + if (pic->type == PICTURE_TYPE_IDR) { + for (i = 0; i < ctx->nb_global_params; i++) { + err = vaapi_encode_make_misc_param_buffer(avctx, pic, + ctx->global_params_type[i], + ctx->global_params[i], + ctx->global_params_size[i]); + if (err < 0) + goto fail; + } + } + if (ctx->codec->init_picture_params) { err = ctx->codec->init_picture_params(avctx, pic); if (err < 0) { @@ -321,16 +426,23 @@ static int vaapi_encode_issue(AVCodecContext *avctx, } } + if (pic->nb_slices == 0) + pic->nb_slices = ctx->nb_slices; if (pic->nb_slices > 0) { pic->slices = av_mallocz_array(pic->nb_slices, sizeof(*pic->slices)); if (!pic->slices) { err = AVERROR(ENOMEM); goto fail; } + + if (ctx->tile_rows && ctx->tile_cols) + vaapi_encode_make_tile_slice(avctx, pic); + else + vaapi_encode_make_row_slice(avctx, pic); } + for (i = 0; i < pic->nb_slices; i++) { slice = &pic->slices[i]; - slice->index = i; if (ctx->codec->slice_params_size > 0) { slice->codec_slice_params = av_mallocz(ctx->codec->slice_params_size); @@ -376,6 +488,71 @@ static int vaapi_encode_issue(AVCodecContext *avctx, } } +#if VA_CHECK_VERSION(1, 0, 0) + sd = av_frame_get_side_data(pic->input_image, + AV_FRAME_DATA_REGIONS_OF_INTEREST); + if (sd && ctx->roi_allowed) { + const AVRegionOfInterest *roi; + uint32_t roi_size; + VAEncMiscParameterBufferROI param_roi; + int nb_roi, i, v; + + roi = (const AVRegionOfInterest*)sd->data; + roi_size = roi->self_size; + av_assert0(roi_size && sd->size % roi_size == 0); + nb_roi = sd->size / roi_size; + if (nb_roi > ctx->roi_max_regions) { + if (!ctx->roi_warned) { + av_log(avctx, AV_LOG_WARNING, "More ROIs set than " + "supported by driver (%d > %d).\n", + nb_roi, ctx->roi_max_regions); + ctx->roi_warned = 1; + } + nb_roi = ctx->roi_max_regions; + } + + pic->roi = av_mallocz_array(nb_roi, sizeof(*pic->roi)); + if (!pic->roi) { + err = AVERROR(ENOMEM); + goto fail; + } + // For overlapping regions, the first in the array takes priority. + for (i = 0; i < nb_roi; i++) { + roi = (const AVRegionOfInterest*)(sd->data + roi_size * i); + + av_assert0(roi->qoffset.den != 0); + v = roi->qoffset.num * ctx->roi_quant_range / roi->qoffset.den; + av_log(avctx, AV_LOG_DEBUG, "ROI: (%d,%d)-(%d,%d) -> %+d.\n", + roi->top, roi->left, roi->bottom, roi->right, v); + + pic->roi[i] = (VAEncROI) { + .roi_rectangle = { + .x = roi->left, + .y = roi->top, + .width = roi->right - roi->left, + .height = roi->bottom - roi->top, + }, + .roi_value = av_clip_int8(v), + }; + } + + param_roi = (VAEncMiscParameterBufferROI) { + .num_roi = nb_roi, + .max_delta_qp = INT8_MAX, + .min_delta_qp = INT8_MIN, + .roi = pic->roi, + .roi_flags.bits.roi_value_is_qp_delta = 1, + }; + + err = vaapi_encode_make_misc_param_buffer(avctx, pic, + VAEncMiscParameterTypeROI, + ¶m_roi, + sizeof(param_roi)); + if (err < 0) + goto fail; + } +#endif + vas = vaBeginPicture(ctx->hwctx->display, ctx->va_context, pic->input_surface); if (vas != VA_STATUS_SUCCESS) { @@ -424,26 +601,22 @@ static int vaapi_encode_issue(AVCodecContext *avctx, pic->encode_issued = 1; - if (ctx->issue_mode == ISSUE_MODE_SERIALISE_EVERYTHING) - return vaapi_encode_wait(avctx, pic); - else - return 0; + return 0; fail_with_picture: vaEndPicture(ctx->hwctx->display, ctx->va_context); fail: for(i = 0; i < pic->nb_param_buffers; i++) vaDestroyBuffer(ctx->hwctx->display, pic->param_buffers[i]); - for (i = 0; i < pic->nb_slices; i++) { - if (pic->slices) { - av_freep(&pic->slices[i].priv_data); + if (pic->slices) { + for (i = 0; i < pic->nb_slices; i++) av_freep(&pic->slices[i].codec_slice_params); - } } fail_at_end: av_freep(&pic->codec_picture_params); av_freep(&pic->param_buffers); av_freep(&pic->slices); + av_freep(&pic->roi); av_frame_free(&pic->recon_image); av_buffer_unref(&pic->output_buffer_ref); pic->output_buffer = VA_INVALID_ID; @@ -456,6 +629,8 @@ static int vaapi_encode_output(AVCodecContext *avctx, VAAPIEncodeContext *ctx = avctx->priv_data; VACodedBufferSegment *buf_list, *buf; VAStatus vas; + int total_size = 0; + uint8_t *ptr; int err; err = vaapi_encode_wait(avctx, pic); @@ -472,15 +647,21 @@ static int vaapi_encode_output(AVCodecContext *avctx, goto fail; } + for (buf = buf_list; buf; buf = buf->next) + total_size += buf->size; + + err = ff_get_encode_buffer(avctx, pkt, total_size, 0); + ptr = pkt->data; + + if (err < 0) + goto fail_mapped; + for (buf = buf_list; buf; buf = buf->next) { av_log(avctx, AV_LOG_DEBUG, "Output buffer: %u bytes " "(status %08x).\n", buf->size, buf->status); - err = av_new_packet(pkt, buf->size); - if (err < 0) - goto fail_mapped; - - memcpy(pkt->data, buf->buf, buf->size); + memcpy(ptr, buf->buf, buf->size); + ptr += buf->size; } if (pic->type == PICTURE_TYPE_IDR) @@ -528,14 +709,23 @@ static int vaapi_encode_discard(AVCodecContext *avctx, return 0; } -static VAAPIEncodePicture *vaapi_encode_alloc(void) +static VAAPIEncodePicture *vaapi_encode_alloc(AVCodecContext *avctx) { + VAAPIEncodeContext *ctx = avctx->priv_data; VAAPIEncodePicture *pic; pic = av_mallocz(sizeof(*pic)); if (!pic) return NULL; + if (ctx->codec->picture_priv_data_size > 0) { + pic->priv_data = av_mallocz(ctx->codec->picture_priv_data_size); + if (!pic->priv_data) { + av_freep(&pic); + return NULL; + } + } + pic->input_surface = VA_INVALID_ID; pic->recon_surface = VA_INVALID_ID; pic->output_buffer = VA_INVALID_ID; @@ -551,11 +741,9 @@ static int vaapi_encode_free(AVCodecContext *avctx, if (pic->encode_issued) vaapi_encode_discard(avctx, pic); - for (i = 0; i < pic->nb_slices; i++) { - if (pic->slices) { - av_freep(&pic->slices[i].priv_data); + if (pic->slices) { + for (i = 0; i < pic->nb_slices; i++) av_freep(&pic->slices[i].codec_slice_params); - } } av_freep(&pic->codec_picture_params); @@ -569,321 +757,366 @@ static int vaapi_encode_free(AVCodecContext *avctx, av_freep(&pic->priv_data); av_freep(&pic->codec_picture_params); + av_freep(&pic->roi); av_free(pic); return 0; } -static int vaapi_encode_step(AVCodecContext *avctx, - VAAPIEncodePicture *target) +static void vaapi_encode_add_ref(AVCodecContext *avctx, + VAAPIEncodePicture *pic, + VAAPIEncodePicture *target, + int is_ref, int in_dpb, int prev) { - VAAPIEncodeContext *ctx = avctx->priv_data; - VAAPIEncodePicture *pic; - int i, err; - - if (ctx->issue_mode == ISSUE_MODE_SERIALISE_EVERYTHING || - ctx->issue_mode == ISSUE_MODE_MINIMISE_LATENCY) { - // These two modes are equivalent, except that we wait for - // immediate completion on each operation if serialised. - - if (!target) { - // No target, nothing to do yet. - return 0; - } - - if (target->encode_complete) { - // Already done. - return 0; - } - - pic = target; - for (i = 0; i < pic->nb_refs; i++) { - if (!pic->refs[i]->encode_complete) { - err = vaapi_encode_step(avctx, pic->refs[i]); - if (err < 0) - return err; - } - } - - err = vaapi_encode_issue(avctx, pic); - if (err < 0) - return err; + int refs = 0; - } else if (ctx->issue_mode == ISSUE_MODE_MAXIMISE_THROUGHPUT) { - int activity; - - // Run through the list of all available pictures repeatedly - // and issue the first one found which has all dependencies - // available (including previously-issued but not necessarily - // completed pictures). - do { - activity = 0; - for (pic = ctx->pic_start; pic; pic = pic->next) { - if (!pic->input_available || pic->encode_issued) - continue; - for (i = 0; i < pic->nb_refs; i++) { - if (!pic->refs[i]->encode_issued) - break; - } - if (i < pic->nb_refs) - continue; - err = vaapi_encode_issue(avctx, pic); - if (err < 0) - return err; - activity = 1; - // Start again from the beginning of the list, - // because issuing this picture may have satisfied - // forward dependencies of earlier ones. - break; - } - } while(activity); + if (is_ref) { + av_assert0(pic != target); + av_assert0(pic->nb_refs < MAX_PICTURE_REFERENCES); + pic->refs[pic->nb_refs++] = target; + ++refs; + } - // If we had a defined target for this step then it will - // always have been issued by now. - if (target) { - av_assert0(target->encode_issued && "broken dependencies?"); - } + if (in_dpb) { + av_assert0(pic->nb_dpb_pics < MAX_DPB_SIZE); + pic->dpb[pic->nb_dpb_pics++] = target; + ++refs; + } - } else { - av_assert0(0); + if (prev) { + av_assert0(!pic->prev); + pic->prev = target; + ++refs; } - return 0; + target->ref_count[0] += refs; + target->ref_count[1] += refs; } -static int vaapi_encode_get_next(AVCodecContext *avctx, - VAAPIEncodePicture **pic_out) +static void vaapi_encode_remove_refs(AVCodecContext *avctx, + VAAPIEncodePicture *pic, + int level) { - VAAPIEncodeContext *ctx = avctx->priv_data; - VAAPIEncodePicture *start, *end, *pic; int i; - for (pic = ctx->pic_start; pic; pic = pic->next) { - if (pic->next) - av_assert0(pic->display_order + 1 == pic->next->display_order); - if (pic->display_order == ctx->input_order) { - *pic_out = pic; - return 0; - } + if (pic->ref_removed[level]) + return; + + for (i = 0; i < pic->nb_refs; i++) { + av_assert0(pic->refs[i]); + --pic->refs[i]->ref_count[level]; + av_assert0(pic->refs[i]->ref_count[level] >= 0); } - pic = vaapi_encode_alloc(); - if (!pic) - return AVERROR(ENOMEM); + for (i = 0; i < pic->nb_dpb_pics; i++) { + av_assert0(pic->dpb[i]); + --pic->dpb[i]->ref_count[level]; + av_assert0(pic->dpb[i]->ref_count[level] >= 0); + } - if (ctx->input_order == 0 || ctx->force_idr || - ctx->gop_counter >= avctx->gop_size) { - pic->type = PICTURE_TYPE_IDR; - ctx->force_idr = 0; - ctx->gop_counter = 1; - ctx->p_counter = 0; - } else if (ctx->p_counter >= ctx->p_per_i) { - pic->type = PICTURE_TYPE_I; - ++ctx->gop_counter; - ctx->p_counter = 0; - } else { - pic->type = PICTURE_TYPE_P; - pic->refs[0] = ctx->pic_end; - pic->nb_refs = 1; - ++ctx->gop_counter; - ++ctx->p_counter; + av_assert0(pic->prev || pic->type == PICTURE_TYPE_IDR); + if (pic->prev) { + --pic->prev->ref_count[level]; + av_assert0(pic->prev->ref_count[level] >= 0); } - start = end = pic; - if (pic->type != PICTURE_TYPE_IDR) { - // If that was not an IDR frame, add B-frames display-before and - // encode-after it, but not exceeding the GOP size. + pic->ref_removed[level] = 1; +} - for (i = 0; i < ctx->b_per_p && - ctx->gop_counter < avctx->gop_size; i++) { - pic = vaapi_encode_alloc(); - if (!pic) - goto fail; +static void vaapi_encode_set_b_pictures(AVCodecContext *avctx, + VAAPIEncodePicture *start, + VAAPIEncodePicture *end, + VAAPIEncodePicture *prev, + int current_depth, + VAAPIEncodePicture **last) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + VAAPIEncodePicture *pic, *next, *ref; + int i, len; - pic->type = PICTURE_TYPE_B; - pic->refs[0] = ctx->pic_end; - pic->refs[1] = end; - pic->nb_refs = 2; + av_assert0(start && end && start != end && start->next != end); - pic->next = start; - pic->display_order = ctx->input_order + ctx->b_per_p - i - 1; - pic->encode_order = pic->display_order + 1; - start = pic; + // If we are at the maximum depth then encode all pictures as + // non-referenced B-pictures. Also do this if there is exactly one + // picture left, since there will be nothing to reference it. + if (current_depth == ctx->max_b_depth || start->next->next == end) { + for (pic = start->next; pic; pic = pic->next) { + if (pic == end) + break; + pic->type = PICTURE_TYPE_B; + pic->b_depth = current_depth; - ++ctx->gop_counter; + vaapi_encode_add_ref(avctx, pic, start, 1, 1, 0); + vaapi_encode_add_ref(avctx, pic, end, 1, 1, 0); + vaapi_encode_add_ref(avctx, pic, prev, 0, 0, 1); + + for (ref = end->refs[1]; ref; ref = ref->refs[1]) + vaapi_encode_add_ref(avctx, pic, ref, 0, 1, 0); } - } + *last = prev; - if (ctx->input_order == 0) { - pic->display_order = 0; - pic->encode_order = 0; + } else { + // Split the current list at the midpoint with a referenced + // B-picture, then descend into each side separately. + len = 0; + for (pic = start->next; pic != end; pic = pic->next) + ++len; + for (pic = start->next, i = 1; 2 * i < len; pic = pic->next, i++); - ctx->pic_start = ctx->pic_end = pic; + pic->type = PICTURE_TYPE_B; + pic->b_depth = current_depth; - } else { - for (i = 0, pic = start; pic; i++, pic = pic->next) { - pic->display_order = ctx->input_order + i; - if (end->type == PICTURE_TYPE_IDR) - pic->encode_order = ctx->input_order + i; - else if (pic == end) - pic->encode_order = ctx->input_order; - else - pic->encode_order = ctx->input_order + i + 1; - } + pic->is_reference = 1; - av_assert0(ctx->pic_end); - ctx->pic_end->next = start; - ctx->pic_end = end; - } - *pic_out = start; + vaapi_encode_add_ref(avctx, pic, pic, 0, 1, 0); + vaapi_encode_add_ref(avctx, pic, start, 1, 1, 0); + vaapi_encode_add_ref(avctx, pic, end, 1, 1, 0); + vaapi_encode_add_ref(avctx, pic, prev, 0, 0, 1); - av_log(avctx, AV_LOG_DEBUG, "Pictures:"); - for (pic = ctx->pic_start; pic; pic = pic->next) { - av_log(avctx, AV_LOG_DEBUG, " %s (%"PRId64"/%"PRId64")", - picture_type_name[pic->type], - pic->display_order, pic->encode_order); - } - av_log(avctx, AV_LOG_DEBUG, "\n"); + for (ref = end->refs[1]; ref; ref = ref->refs[1]) + vaapi_encode_add_ref(avctx, pic, ref, 0, 1, 0); - return 0; + if (i > 1) + vaapi_encode_set_b_pictures(avctx, start, pic, pic, + current_depth + 1, &next); + else + next = pic; -fail: - while (start) { - pic = start->next; - vaapi_encode_free(avctx, start); - start = pic; + vaapi_encode_set_b_pictures(avctx, pic, end, next, + current_depth + 1, last); } - return AVERROR(ENOMEM); } -static int vaapi_encode_truncate_gop(AVCodecContext *avctx) +static int vaapi_encode_pick_next(AVCodecContext *avctx, + VAAPIEncodePicture **pic_out) { VAAPIEncodeContext *ctx = avctx->priv_data; - VAAPIEncodePicture *pic, *last_pic, *next; - - av_assert0(!ctx->pic_start || ctx->pic_start->input_available); + VAAPIEncodePicture *pic = NULL, *next, *start; + int i, b_counter, closed_gop_end; - // Find the last picture we actually have input for. + // If there are any B-frames already queued, the next one to encode + // is the earliest not-yet-issued frame for which all references are + // available. for (pic = ctx->pic_start; pic; pic = pic->next) { - if (!pic->input_available) + if (pic->encode_issued) + continue; + if (pic->type != PICTURE_TYPE_B) + continue; + for (i = 0; i < pic->nb_refs; i++) { + if (!pic->refs[i]->encode_issued) + break; + } + if (i == pic->nb_refs) break; - last_pic = pic; } if (pic) { - if (last_pic->type == PICTURE_TYPE_B) { - // Some fixing up is required. Change the type of this - // picture to P, then modify preceding B references which - // point beyond it to point at it instead. - - last_pic->type = PICTURE_TYPE_P; - last_pic->encode_order = last_pic->refs[1]->encode_order; - - for (pic = ctx->pic_start; pic != last_pic; pic = pic->next) { - if (pic->type == PICTURE_TYPE_B && - pic->refs[1] == last_pic->refs[1]) - pic->refs[1] = last_pic; - } + av_log(avctx, AV_LOG_DEBUG, "Pick B-picture at depth %d to " + "encode next.\n", pic->b_depth); + *pic_out = pic; + return 0; + } + + // Find the B-per-Pth available picture to become the next picture + // on the top layer. + start = NULL; + b_counter = 0; + closed_gop_end = ctx->closed_gop || + ctx->idr_counter == ctx->gop_per_idr; + for (pic = ctx->pic_start; pic; pic = next) { + next = pic->next; + if (pic->encode_issued) { + start = pic; + continue; + } + // If the next available picture is force-IDR, encode it to start + // a new GOP immediately. + if (pic->force_idr) + break; + if (b_counter == ctx->b_per_p) + break; + // If this picture ends a closed GOP or starts a new GOP then it + // needs to be in the top layer. + if (ctx->gop_counter + b_counter + closed_gop_end >= ctx->gop_size) + break; + // If the picture after this one is force-IDR, we need to encode + // this one in the top layer. + if (next && next->force_idr) + break; + ++b_counter; + } + + // At the end of the stream the last picture must be in the top layer. + if (!pic && ctx->end_of_stream) { + --b_counter; + pic = ctx->pic_end; + if (pic->encode_issued) + return AVERROR_EOF; + } + + if (!pic) { + av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - " + "need more input for reference pictures.\n"); + return AVERROR(EAGAIN); + } + if (ctx->input_order <= ctx->decode_delay && !ctx->end_of_stream) { + av_log(avctx, AV_LOG_DEBUG, "Pick nothing to encode next - " + "need more input for timestamps.\n"); + return AVERROR(EAGAIN); + } + + if (pic->force_idr) { + av_log(avctx, AV_LOG_DEBUG, "Pick forced IDR-picture to " + "encode next.\n"); + pic->type = PICTURE_TYPE_IDR; + ctx->idr_counter = 1; + ctx->gop_counter = 1; - last_pic->nb_refs = 1; - last_pic->refs[1] = NULL; + } else if (ctx->gop_counter + b_counter >= ctx->gop_size) { + if (ctx->idr_counter == ctx->gop_per_idr) { + av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP IDR-picture to " + "encode next.\n"); + pic->type = PICTURE_TYPE_IDR; + ctx->idr_counter = 1; } else { - // We can use the current structure (no references point - // beyond the end), but there are unused pics to discard. + av_log(avctx, AV_LOG_DEBUG, "Pick new-GOP I-picture to " + "encode next.\n"); + pic->type = PICTURE_TYPE_I; + ++ctx->idr_counter; } + ctx->gop_counter = 1; - // Discard all following pics, they will never be used. - for (pic = last_pic->next; pic; pic = next) { - next = pic->next; - vaapi_encode_free(avctx, pic); + } else { + if (ctx->gop_counter + b_counter + closed_gop_end == ctx->gop_size) { + av_log(avctx, AV_LOG_DEBUG, "Pick group-end P-picture to " + "encode next.\n"); + } else { + av_log(avctx, AV_LOG_DEBUG, "Pick normal P-picture to " + "encode next.\n"); } + pic->type = PICTURE_TYPE_P; + av_assert0(start); + ctx->gop_counter += 1 + b_counter; + } + pic->is_reference = 1; + *pic_out = pic; - last_pic->next = NULL; - ctx->pic_end = last_pic; + vaapi_encode_add_ref(avctx, pic, pic, 0, 1, 0); + if (pic->type != PICTURE_TYPE_IDR) { + vaapi_encode_add_ref(avctx, pic, start, + pic->type == PICTURE_TYPE_P, + b_counter > 0, 0); + vaapi_encode_add_ref(avctx, pic, ctx->next_prev, 0, 0, 1); + } + if (ctx->next_prev) + --ctx->next_prev->ref_count[0]; + if (b_counter > 0) { + vaapi_encode_set_b_pictures(avctx, start, pic, pic, 1, + &ctx->next_prev); } else { - // Input is available for all pictures, so we don't need to - // mangle anything. + ctx->next_prev = pic; + } + ++ctx->next_prev->ref_count[0]; + return 0; +} + +static int vaapi_encode_clear_old(AVCodecContext *avctx) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + VAAPIEncodePicture *pic, *prev, *next; + + av_assert0(ctx->pic_start); + + // Remove direct references once each picture is complete. + for (pic = ctx->pic_start; pic; pic = pic->next) { + if (pic->encode_complete && pic->next) + vaapi_encode_remove_refs(avctx, pic, 0); } - av_log(avctx, AV_LOG_DEBUG, "Pictures ending truncated GOP:"); + // Remove indirect references once a picture has no direct references. for (pic = ctx->pic_start; pic; pic = pic->next) { - av_log(avctx, AV_LOG_DEBUG, " %s (%"PRId64"/%"PRId64")", - picture_type_name[pic->type], - pic->display_order, pic->encode_order); + if (pic->encode_complete && pic->ref_count[0] == 0) + vaapi_encode_remove_refs(avctx, pic, 1); + } + + // Clear out all complete pictures with no remaining references. + prev = NULL; + for (pic = ctx->pic_start; pic; pic = next) { + next = pic->next; + if (pic->encode_complete && pic->ref_count[1] == 0) { + av_assert0(pic->ref_removed[0] && pic->ref_removed[1]); + if (prev) + prev->next = next; + else + ctx->pic_start = next; + vaapi_encode_free(avctx, pic); + } else { + prev = pic; + } } - av_log(avctx, AV_LOG_DEBUG, "\n"); return 0; } -static int vaapi_encode_clear_old(AVCodecContext *avctx) +static int vaapi_encode_check_frame(AVCodecContext *avctx, + const AVFrame *frame) { VAAPIEncodeContext *ctx = avctx->priv_data; - VAAPIEncodePicture *pic, *old; - int i; - while (ctx->pic_start != ctx->pic_end) { - old = ctx->pic_start; - if (old->encode_order > ctx->output_order) - break; + if ((frame->crop_top || frame->crop_bottom || + frame->crop_left || frame->crop_right) && !ctx->crop_warned) { + av_log(avctx, AV_LOG_WARNING, "Cropping information on input " + "frames ignored due to lack of API support.\n"); + ctx->crop_warned = 1; + } - for (pic = old->next; pic; pic = pic->next) { - if (pic->encode_complete) - continue; - for (i = 0; i < pic->nb_refs; i++) { - if (pic->refs[i] == old) { - // We still need this picture because it's referred to - // directly by a later one, so it and all following - // pictures have to stay. - return 0; - } - } - } + if (!ctx->roi_allowed) { + AVFrameSideData *sd = + av_frame_get_side_data(frame, AV_FRAME_DATA_REGIONS_OF_INTEREST); - pic = ctx->pic_start; - ctx->pic_start = pic->next; - vaapi_encode_free(avctx, pic); + if (sd && !ctx->roi_warned) { + av_log(avctx, AV_LOG_WARNING, "ROI side data on input " + "frames ignored due to lack of driver support.\n"); + ctx->roi_warned = 1; + } } return 0; } -int ff_vaapi_encode2(AVCodecContext *avctx, AVPacket *pkt, - const AVFrame *input_image, int *got_packet) +static int vaapi_encode_send_frame(AVCodecContext *avctx, AVFrame *frame) { VAAPIEncodeContext *ctx = avctx->priv_data; VAAPIEncodePicture *pic; int err; - if (input_image) { - av_log(avctx, AV_LOG_DEBUG, "Encode frame: %ux%u (%"PRId64").\n", - input_image->width, input_image->height, input_image->pts); - - if (input_image->pict_type == AV_PICTURE_TYPE_I) { - err = vaapi_encode_truncate_gop(avctx); - if (err < 0) - goto fail; - ctx->force_idr = 1; - } + if (frame) { + av_log(avctx, AV_LOG_DEBUG, "Input frame: %ux%u (%"PRId64").\n", + frame->width, frame->height, frame->pts); - err = vaapi_encode_get_next(avctx, &pic); - if (err) { - av_log(avctx, AV_LOG_ERROR, "Input setup failed: %d.\n", err); + err = vaapi_encode_check_frame(avctx, frame); + if (err < 0) return err; - } + + pic = vaapi_encode_alloc(avctx); + if (!pic) + return AVERROR(ENOMEM); pic->input_image = av_frame_alloc(); if (!pic->input_image) { err = AVERROR(ENOMEM); goto fail; } - err = av_frame_ref(pic->input_image, input_image); - if (err < 0) - goto fail; - pic->input_surface = (VASurfaceID)(uintptr_t)input_image->data[3]; - pic->pts = input_image->pts; + + if (ctx->input_order == 0 || frame->pict_type == AV_PICTURE_TYPE_I) + pic->force_idr = 1; + + pic->input_surface = (VASurfaceID)(uintptr_t)frame->data[3]; + pic->pts = frame->pts; + + av_frame_move_ref(pic->input_image, frame); if (ctx->input_order == 0) ctx->first_pts = pic->pts; @@ -892,320 +1125,678 @@ int ff_vaapi_encode2(AVCodecContext *avctx, AVPacket *pkt, if (ctx->output_delay > 0) ctx->ts_ring[ctx->input_order % (3 * ctx->output_delay)] = pic->pts; - pic->input_available = 1; - - } else { - if (!ctx->end_of_stream) { - err = vaapi_encode_truncate_gop(avctx); - if (err < 0) - goto fail; - ctx->end_of_stream = 1; - } - } - - ++ctx->input_order; - ++ctx->output_order; - av_assert0(ctx->output_order + ctx->output_delay + 1 == ctx->input_order); - - for (pic = ctx->pic_start; pic; pic = pic->next) - if (pic->encode_order == ctx->output_order) - break; - - // pic can be null here if we don't have a specific target in this - // iteration. We might still issue encodes if things can be overlapped, - // even though we don't intend to output anything. - - err = vaapi_encode_step(avctx, pic); - if (err < 0) { - av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err); - goto fail; - } - - if (!pic) { - *got_packet = 0; - } else { - err = vaapi_encode_output(avctx, pic, pkt); - if (err < 0) { - av_log(avctx, AV_LOG_ERROR, "Output failed: %d.\n", err); - goto fail; - } + pic->display_order = ctx->input_order; + ++ctx->input_order; - if (ctx->output_delay == 0) { - pkt->dts = pkt->pts; - } else if (ctx->output_order < ctx->decode_delay) { - if (ctx->ts_ring[ctx->output_order] < INT64_MIN + ctx->dts_pts_diff) - pkt->dts = INT64_MIN; - else - pkt->dts = ctx->ts_ring[ctx->output_order] - ctx->dts_pts_diff; + if (ctx->pic_start) { + ctx->pic_end->next = pic; + ctx->pic_end = pic; } else { - pkt->dts = ctx->ts_ring[(ctx->output_order - ctx->decode_delay) % - (3 * ctx->output_delay)]; + ctx->pic_start = pic; + ctx->pic_end = pic; } - *got_packet = 1; - } + } else { + ctx->end_of_stream = 1; - err = vaapi_encode_clear_old(avctx); - if (err < 0) { - av_log(avctx, AV_LOG_ERROR, "List clearing failed: %d.\n", err); - goto fail; + // Fix timestamps if we hit end-of-stream before the initial decode + // delay has elapsed. + if (ctx->input_order < ctx->decode_delay) + ctx->dts_pts_diff = ctx->pic_end->pts - ctx->first_pts; } return 0; fail: - // Unclear what to clean up on failure. There are probably some things we - // could do usefully clean up here, but for now just leave them for uninit() - // to do instead. + vaapi_encode_free(avctx, pic); return err; } -static av_cold void vaapi_encode_add_global_param(AVCodecContext *avctx, - VAEncMiscParameterBuffer *buffer, - size_t size) +int ff_vaapi_encode_receive_packet(AVCodecContext *avctx, AVPacket *pkt) { VAAPIEncodeContext *ctx = avctx->priv_data; + VAAPIEncodePicture *pic; + AVFrame *frame = ctx->frame; + int err; - av_assert0(ctx->nb_global_params < MAX_GLOBAL_PARAMS); + err = ff_encode_get_frame(avctx, frame); + if (err < 0 && err != AVERROR_EOF) + return err; - ctx->global_params [ctx->nb_global_params] = buffer; - ctx->global_params_size[ctx->nb_global_params] = size; + if (err == AVERROR_EOF) + frame = NULL; - ++ctx->nb_global_params; -} + err = vaapi_encode_send_frame(avctx, frame); + if (err < 0) + return err; -static av_cold int vaapi_encode_config_attributes(AVCodecContext *avctx) -{ - VAAPIEncodeContext *ctx = avctx->priv_data; + if (!ctx->pic_start) { + if (ctx->end_of_stream) + return AVERROR_EOF; + else + return AVERROR(EAGAIN); + } + + pic = NULL; + err = vaapi_encode_pick_next(avctx, &pic); + if (err < 0) + return err; + av_assert0(pic); + + pic->encode_order = ctx->encode_order++; + + err = vaapi_encode_issue(avctx, pic); + if (err < 0) { + av_log(avctx, AV_LOG_ERROR, "Encode failed: %d.\n", err); + return err; + } + + err = vaapi_encode_output(avctx, pic, pkt); + if (err < 0) { + av_log(avctx, AV_LOG_ERROR, "Output failed: %d.\n", err); + return err; + } + + if (ctx->output_delay == 0) { + pkt->dts = pkt->pts; + } else if (pic->encode_order < ctx->decode_delay) { + if (ctx->ts_ring[pic->encode_order] < INT64_MIN + ctx->dts_pts_diff) + pkt->dts = INT64_MIN; + else + pkt->dts = ctx->ts_ring[pic->encode_order] - ctx->dts_pts_diff; + } else { + pkt->dts = ctx->ts_ring[(pic->encode_order - ctx->decode_delay) % + (3 * ctx->output_delay)]; + } + av_log(avctx, AV_LOG_DEBUG, "Output packet: pts %"PRId64" dts %"PRId64".\n", + pkt->pts, pkt->dts); + + ctx->output_order = pic->encode_order; + vaapi_encode_clear_old(avctx); + + return 0; +} + + +static av_cold void vaapi_encode_add_global_param(AVCodecContext *avctx, int type, + void *buffer, size_t size) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + + av_assert0(ctx->nb_global_params < MAX_GLOBAL_PARAMS); + + ctx->global_params_type[ctx->nb_global_params] = type; + ctx->global_params [ctx->nb_global_params] = buffer; + ctx->global_params_size[ctx->nb_global_params] = size; + + ++ctx->nb_global_params; +} + +typedef struct VAAPIEncodeRTFormat { + const char *name; + unsigned int value; + int depth; + int nb_components; + int log2_chroma_w; + int log2_chroma_h; +} VAAPIEncodeRTFormat; + +static const VAAPIEncodeRTFormat vaapi_encode_rt_formats[] = { + { "YUV400", VA_RT_FORMAT_YUV400, 8, 1, }, + { "YUV420", VA_RT_FORMAT_YUV420, 8, 3, 1, 1 }, + { "YUV422", VA_RT_FORMAT_YUV422, 8, 3, 1, 0 }, +#if VA_CHECK_VERSION(1, 2, 0) + { "YUV422_10", VA_RT_FORMAT_YUV422_10, 10, 3, 1, 0 }, +#endif + { "YUV444", VA_RT_FORMAT_YUV444, 8, 3, 0, 0 }, + { "YUV411", VA_RT_FORMAT_YUV411, 8, 3, 2, 0 }, +#if VA_CHECK_VERSION(0, 38, 1) + { "YUV420_10", VA_RT_FORMAT_YUV420_10BPP, 10, 3, 1, 1 }, +#endif +}; + +static const VAEntrypoint vaapi_encode_entrypoints_normal[] = { + VAEntrypointEncSlice, + VAEntrypointEncPicture, +#if VA_CHECK_VERSION(0, 39, 2) + VAEntrypointEncSliceLP, +#endif + 0 +}; +#if VA_CHECK_VERSION(0, 39, 2) +static const VAEntrypoint vaapi_encode_entrypoints_low_power[] = { + VAEntrypointEncSliceLP, + 0 +}; +#endif + +static av_cold int vaapi_encode_profile_entrypoint(AVCodecContext *avctx) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + VAProfile *va_profiles = NULL; + VAEntrypoint *va_entrypoints = NULL; VAStatus vas; - int i, n, err; - VAProfile *profiles = NULL; - VAEntrypoint *entrypoints = NULL; - VAConfigAttrib attr[] = { - { VAConfigAttribRTFormat }, - { VAConfigAttribRateControl }, - { VAConfigAttribEncMaxRefFrames }, - { VAConfigAttribEncPackedHeaders }, - }; + const VAEntrypoint *usable_entrypoints; + const VAAPIEncodeProfile *profile; + const AVPixFmtDescriptor *desc; + VAConfigAttrib rt_format_attr; + const VAAPIEncodeRTFormat *rt_format; + const char *profile_string, *entrypoint_string; + int i, j, n, depth, err; + + + if (ctx->low_power) { +#if VA_CHECK_VERSION(0, 39, 2) + usable_entrypoints = vaapi_encode_entrypoints_low_power; +#else + av_log(avctx, AV_LOG_ERROR, "Low-power encoding is not " + "supported with this VAAPI version.\n"); + return AVERROR(EINVAL); +#endif + } else { + usable_entrypoints = vaapi_encode_entrypoints_normal; + } + + desc = av_pix_fmt_desc_get(ctx->input_frames->sw_format); + if (!desc) { + av_log(avctx, AV_LOG_ERROR, "Invalid input pixfmt (%d).\n", + ctx->input_frames->sw_format); + return AVERROR(EINVAL); + } + depth = desc->comp[0].depth; + for (i = 1; i < desc->nb_components; i++) { + if (desc->comp[i].depth != depth) { + av_log(avctx, AV_LOG_ERROR, "Invalid input pixfmt (%s).\n", + desc->name); + return AVERROR(EINVAL); + } + } + av_log(avctx, AV_LOG_VERBOSE, "Input surface format is %s.\n", + desc->name); n = vaMaxNumProfiles(ctx->hwctx->display); - profiles = av_malloc_array(n, sizeof(VAProfile)); - if (!profiles) { + va_profiles = av_malloc_array(n, sizeof(VAProfile)); + if (!va_profiles) { err = AVERROR(ENOMEM); goto fail; } - vas = vaQueryConfigProfiles(ctx->hwctx->display, profiles, &n); + vas = vaQueryConfigProfiles(ctx->hwctx->display, va_profiles, &n); if (vas != VA_STATUS_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "Failed to query profiles: %d (%s).\n", + av_log(avctx, AV_LOG_ERROR, "Failed to query profiles: %d (%s).\n", vas, vaErrorStr(vas)); - err = AVERROR(ENOSYS); + err = AVERROR_EXTERNAL; goto fail; } - for (i = 0; i < n; i++) { - if (profiles[i] == ctx->va_profile) - break; + + av_assert0(ctx->codec->profiles); + for (i = 0; (ctx->codec->profiles[i].av_profile != + FF_PROFILE_UNKNOWN); i++) { + profile = &ctx->codec->profiles[i]; + if (depth != profile->depth || + desc->nb_components != profile->nb_components) + continue; + if (desc->nb_components > 1 && + (desc->log2_chroma_w != profile->log2_chroma_w || + desc->log2_chroma_h != profile->log2_chroma_h)) + continue; + if (avctx->profile != profile->av_profile && + avctx->profile != FF_PROFILE_UNKNOWN) + continue; + +#if VA_CHECK_VERSION(1, 0, 0) + profile_string = vaProfileStr(profile->va_profile); +#else + profile_string = "(no profile names)"; +#endif + + for (j = 0; j < n; j++) { + if (va_profiles[j] == profile->va_profile) + break; + } + if (j >= n) { + av_log(avctx, AV_LOG_VERBOSE, "Compatible profile %s (%d) " + "is not supported by driver.\n", profile_string, + profile->va_profile); + continue; + } + + ctx->profile = profile; + break; } - if (i >= n) { - av_log(ctx, AV_LOG_ERROR, "Encoding profile not found (%d).\n", - ctx->va_profile); + if (!ctx->profile) { + av_log(avctx, AV_LOG_ERROR, "No usable encoding profile found.\n"); err = AVERROR(ENOSYS); goto fail; } + avctx->profile = profile->av_profile; + ctx->va_profile = profile->va_profile; + av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI profile %s (%d).\n", + profile_string, ctx->va_profile); + n = vaMaxNumEntrypoints(ctx->hwctx->display); - entrypoints = av_malloc_array(n, sizeof(VAEntrypoint)); - if (!entrypoints) { + va_entrypoints = av_malloc_array(n, sizeof(VAEntrypoint)); + if (!va_entrypoints) { err = AVERROR(ENOMEM); goto fail; } vas = vaQueryConfigEntrypoints(ctx->hwctx->display, ctx->va_profile, - entrypoints, &n); + va_entrypoints, &n); if (vas != VA_STATUS_SUCCESS) { - av_log(ctx, AV_LOG_ERROR, "Failed to query entrypoints for " - "profile %u: %d (%s).\n", ctx->va_profile, - vas, vaErrorStr(vas)); - err = AVERROR(ENOSYS); + av_log(avctx, AV_LOG_ERROR, "Failed to query entrypoints for " + "profile %s (%d): %d (%s).\n", profile_string, + ctx->va_profile, vas, vaErrorStr(vas)); + err = AVERROR_EXTERNAL; goto fail; } + for (i = 0; i < n; i++) { - if (entrypoints[i] == ctx->va_entrypoint) + for (j = 0; usable_entrypoints[j]; j++) { + if (va_entrypoints[i] == usable_entrypoints[j]) + break; + } + if (usable_entrypoints[j]) break; } if (i >= n) { - av_log(ctx, AV_LOG_ERROR, "Encoding entrypoint not found " - "(%d / %d).\n", ctx->va_profile, ctx->va_entrypoint); + av_log(avctx, AV_LOG_ERROR, "No usable encoding entrypoint found " + "for profile %s (%d).\n", profile_string, ctx->va_profile); + err = AVERROR(ENOSYS); + goto fail; + } + + ctx->va_entrypoint = va_entrypoints[i]; +#if VA_CHECK_VERSION(1, 0, 0) + entrypoint_string = vaEntrypointStr(ctx->va_entrypoint); +#else + entrypoint_string = "(no entrypoint names)"; +#endif + av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI entrypoint %s (%d).\n", + entrypoint_string, ctx->va_entrypoint); + + for (i = 0; i < FF_ARRAY_ELEMS(vaapi_encode_rt_formats); i++) { + rt_format = &vaapi_encode_rt_formats[i]; + if (rt_format->depth == depth && + rt_format->nb_components == profile->nb_components && + rt_format->log2_chroma_w == profile->log2_chroma_w && + rt_format->log2_chroma_h == profile->log2_chroma_h) + break; + } + if (i >= FF_ARRAY_ELEMS(vaapi_encode_rt_formats)) { + av_log(avctx, AV_LOG_ERROR, "No usable render target format " + "found for profile %s (%d) entrypoint %s (%d).\n", + profile_string, ctx->va_profile, + entrypoint_string, ctx->va_entrypoint); err = AVERROR(ENOSYS); goto fail; } + rt_format_attr = (VAConfigAttrib) { VAConfigAttribRTFormat }; vas = vaGetConfigAttributes(ctx->hwctx->display, ctx->va_profile, ctx->va_entrypoint, - attr, FF_ARRAY_ELEMS(attr)); + &rt_format_attr, 1); if (vas != VA_STATUS_SUCCESS) { - av_log(avctx, AV_LOG_ERROR, "Failed to fetch config " - "attributes: %d (%s).\n", vas, vaErrorStr(vas)); - return AVERROR(EINVAL); + av_log(avctx, AV_LOG_ERROR, "Failed to query RT format " + "config attribute: %d (%s).\n", vas, vaErrorStr(vas)); + err = AVERROR_EXTERNAL; + goto fail; } - for (i = 0; i < FF_ARRAY_ELEMS(attr); i++) { - if (attr[i].value == VA_ATTRIB_NOT_SUPPORTED) { - // Unfortunately we have to treat this as "don't know" and hope - // for the best, because the Intel MJPEG encoder returns this - // for all the interesting attributes. - av_log(avctx, AV_LOG_DEBUG, "Attribute (%d) is not supported.\n", - attr[i].type); - continue; - } - switch (attr[i].type) { - case VAConfigAttribRTFormat: - if (!(ctx->va_rt_format & attr[i].value)) { - av_log(avctx, AV_LOG_ERROR, "Surface RT format %#x " - "is not supported (mask %#x).\n", - ctx->va_rt_format, attr[i].value); - err = AVERROR(EINVAL); - goto fail; - } - ctx->config_attributes[ctx->nb_config_attributes++] = - (VAConfigAttrib) { - .type = VAConfigAttribRTFormat, - .value = ctx->va_rt_format, - }; - break; - case VAConfigAttribRateControl: - // Hack for backward compatibility: CBR was the only - // usable RC mode for a long time, so old drivers will - // only have it. Normal default options may now choose - // VBR and then fail, however, so override it here with - // CBR if that is the only supported mode. - if (ctx->va_rc_mode == VA_RC_VBR && - !(attr[i].value & VA_RC_VBR) && - (attr[i].value & VA_RC_CBR)) { - av_log(avctx, AV_LOG_WARNING, "VBR rate control is " - "not supported with this driver version; " - "using CBR instead.\n"); - ctx->va_rc_mode = VA_RC_CBR; - } - if (!(ctx->va_rc_mode & attr[i].value)) { - av_log(avctx, AV_LOG_ERROR, "Rate control mode %#x " - "is not supported (mask: %#x).\n", - ctx->va_rc_mode, attr[i].value); - err = AVERROR(EINVAL); - goto fail; - } - ctx->config_attributes[ctx->nb_config_attributes++] = - (VAConfigAttrib) { - .type = VAConfigAttribRateControl, - .value = ctx->va_rc_mode, - }; - break; - 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); - err = AVERROR(EINVAL); - goto fail; - } - 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 - // preferable because they are under our control. - // When absent, the driver is generating them and some - // features may not work (e.g. VUI or SEI in H.264). - av_log(avctx, AV_LOG_WARNING, "Warning: some packed " - "headers are not supported (want %#x, got %#x).\n", - ctx->va_packed_headers, attr[i].value); - ctx->va_packed_headers &= attr[i].value; - } - ctx->config_attributes[ctx->nb_config_attributes++] = - (VAConfigAttrib) { - .type = VAConfigAttribEncPackedHeaders, - .value = ctx->va_packed_headers, - }; - break; - default: - av_assert0(0 && "Unexpected config attribute."); - } + if (rt_format_attr.value == VA_ATTRIB_NOT_SUPPORTED) { + av_log(avctx, AV_LOG_VERBOSE, "RT format config attribute not " + "supported by driver: assuming surface RT format %s " + "is valid.\n", rt_format->name); + } else if (!(rt_format_attr.value & rt_format->value)) { + av_log(avctx, AV_LOG_ERROR, "Surface RT format %s not supported " + "by driver for encoding profile %s (%d) entrypoint %s (%d).\n", + rt_format->name, profile_string, ctx->va_profile, + entrypoint_string, ctx->va_entrypoint); + err = AVERROR(ENOSYS); + goto fail; + } else { + av_log(avctx, AV_LOG_VERBOSE, "Using VAAPI render target " + "format %s (%#x).\n", rt_format->name, rt_format->value); + ctx->config_attributes[ctx->nb_config_attributes++] = + (VAConfigAttrib) { + .type = VAConfigAttribRTFormat, + .value = rt_format->value, + }; } err = 0; fail: - av_freep(&profiles); - av_freep(&entrypoints); + av_freep(&va_profiles); + av_freep(&va_entrypoints); return err; } +static const VAAPIEncodeRCMode vaapi_encode_rc_modes[] = { + // Bitrate Quality + // | Maxrate | HRD/VBV + { 0 }, // | | | | + { RC_MODE_CQP, "CQP", 1, VA_RC_CQP, 0, 0, 1, 0 }, + { RC_MODE_CBR, "CBR", 1, VA_RC_CBR, 1, 0, 0, 1 }, + { RC_MODE_VBR, "VBR", 1, VA_RC_VBR, 1, 1, 0, 1 }, +#if VA_CHECK_VERSION(1, 1, 0) + { RC_MODE_ICQ, "ICQ", 1, VA_RC_ICQ, 0, 0, 1, 0 }, +#else + { RC_MODE_ICQ, "ICQ", 0 }, +#endif +#if VA_CHECK_VERSION(1, 3, 0) + { RC_MODE_QVBR, "QVBR", 1, VA_RC_QVBR, 1, 1, 1, 1 }, + { RC_MODE_AVBR, "AVBR", 0, VA_RC_AVBR, 1, 0, 0, 0 }, +#else + { RC_MODE_QVBR, "QVBR", 0 }, + { RC_MODE_AVBR, "AVBR", 0 }, +#endif +}; + 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; + uint32_t supported_va_rc_modes; + const VAAPIEncodeRCMode *rc_mode; + int64_t rc_bits_per_second; + int rc_target_percentage; + int rc_window_size; + int rc_quality; + int64_t hrd_buffer_size; + int64_t hrd_initial_buffer_fullness; int fr_num, fr_den; + VAConfigAttrib rc_attr = { VAConfigAttribRateControl }; + VAStatus vas; + char supported_rc_modes_string[64]; - if (avctx->bit_rate > INT32_MAX) { - av_log(avctx, AV_LOG_ERROR, "Target bitrate of 2^31 bps or " - "higher is not supported.\n"); - return AVERROR(EINVAL); + 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 (rc_attr.value == VA_ATTRIB_NOT_SUPPORTED) { + av_log(avctx, AV_LOG_VERBOSE, "Driver does not report any " + "supported rate control modes: assuming CQP only.\n"); + supported_va_rc_modes = VA_RC_CQP; + strcpy(supported_rc_modes_string, "unknown"); + } else { + char *str = supported_rc_modes_string; + size_t len = sizeof(supported_rc_modes_string); + int i, first = 1, res; + + supported_va_rc_modes = rc_attr.value; + for (i = 0; i < FF_ARRAY_ELEMS(vaapi_encode_rc_modes); i++) { + rc_mode = &vaapi_encode_rc_modes[i]; + if (supported_va_rc_modes & rc_mode->va_mode) { + res = snprintf(str, len, "%s%s", + first ? "" : ", ", rc_mode->name); + first = 0; + if (res < 0) { + *str = 0; + break; + } + len -= res; + str += res; + if (len == 0) + break; + } + } + + av_log(avctx, AV_LOG_DEBUG, "Driver supports RC modes %s.\n", + supported_rc_modes_string); + } + + // Rate control mode selection: + // * If the user has set a mode explicitly with the rc_mode option, + // use it and fail if it is not available. + // * If an explicit QP option has been set, use CQP. + // * If the codec is CQ-only, use CQP. + // * If the QSCALE avcodec option is set, use CQP. + // * If bitrate and quality are both set, try QVBR. + // * If quality is set, try ICQ, then CQP. + // * If bitrate and maxrate are set and have the same value, try CBR. + // * If a bitrate is set, try AVBR, then VBR, then CBR. + // * If no bitrate is set, try ICQ, then CQP. + +#define TRY_RC_MODE(mode, fail) do { \ + rc_mode = &vaapi_encode_rc_modes[mode]; \ + if (!(rc_mode->va_mode & supported_va_rc_modes)) { \ + if (fail) { \ + av_log(avctx, AV_LOG_ERROR, "Driver does not support %s " \ + "RC mode (supported modes: %s).\n", rc_mode->name, \ + supported_rc_modes_string); \ + return AVERROR(EINVAL); \ + } \ + av_log(avctx, AV_LOG_DEBUG, "Driver does not support %s " \ + "RC mode.\n", rc_mode->name); \ + rc_mode = NULL; \ + } else { \ + goto rc_mode_found; \ + } \ + } while (0) + + if (ctx->explicit_rc_mode) + TRY_RC_MODE(ctx->explicit_rc_mode, 1); + + if (ctx->explicit_qp) + TRY_RC_MODE(RC_MODE_CQP, 1); + + if (ctx->codec->flags & FLAG_CONSTANT_QUALITY_ONLY) + TRY_RC_MODE(RC_MODE_CQP, 1); + + if (avctx->flags & AV_CODEC_FLAG_QSCALE) + TRY_RC_MODE(RC_MODE_CQP, 1); + + if (avctx->bit_rate > 0 && avctx->global_quality > 0) + TRY_RC_MODE(RC_MODE_QVBR, 0); + + if (avctx->global_quality > 0) { + TRY_RC_MODE(RC_MODE_ICQ, 0); + TRY_RC_MODE(RC_MODE_CQP, 0); + } + + if (avctx->bit_rate > 0 && avctx->rc_max_rate == avctx->bit_rate) + TRY_RC_MODE(RC_MODE_CBR, 0); + + if (avctx->bit_rate > 0) { + TRY_RC_MODE(RC_MODE_AVBR, 0); + TRY_RC_MODE(RC_MODE_VBR, 0); + TRY_RC_MODE(RC_MODE_CBR, 0); + } else { + TRY_RC_MODE(RC_MODE_ICQ, 0); + TRY_RC_MODE(RC_MODE_CQP, 0); } - if (avctx->rc_buffer_size) - hrd_buffer_size = avctx->rc_buffer_size; - else - hrd_buffer_size = avctx->bit_rate; - if (avctx->rc_initial_buffer_occupancy) - hrd_initial_buffer_fullness = avctx->rc_initial_buffer_occupancy; - else - hrd_initial_buffer_fullness = hrd_buffer_size * 3 / 4; + av_log(avctx, AV_LOG_ERROR, "Driver does not support any " + "RC mode compatible with selected options " + "(supported modes: %s).\n", supported_rc_modes_string); + return AVERROR(EINVAL); - 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. +rc_mode_found: + if (rc_mode->bitrate) { + if (avctx->bit_rate <= 0) { + av_log(avctx, AV_LOG_ERROR, "Bitrate must be set for %s " + "RC mode.\n", rc_mode->name); + return AVERROR(EINVAL); + } + + if (rc_mode->mode == RC_MODE_AVBR) { + // For maximum confusion AVBR is hacked into the existing API + // by overloading some of the fields with completely different + // meanings. + + // Target percentage does not apply in AVBR mode. + rc_bits_per_second = avctx->bit_rate; + + // Accuracy tolerance range for meeting the specified target + // bitrate. It's very unclear how this is actually intended + // to work - since we do want to get the specified bitrate, + // set the accuracy to 100% for now. + rc_target_percentage = 100; + + // Convergence period in frames. The GOP size reflects the + // user's intended block size for cutting, so reusing that + // as the convergence period seems a reasonable default. + rc_window_size = avctx->gop_size > 0 ? avctx->gop_size : 60; + + } else if (rc_mode->maxrate) { + if (avctx->rc_max_rate > 0) { + if (avctx->rc_max_rate < avctx->bit_rate) { + av_log(avctx, AV_LOG_ERROR, "Invalid bitrate settings: " + "bitrate (%"PRId64") must not be greater than " + "maxrate (%"PRId64").\n", avctx->bit_rate, + avctx->rc_max_rate); + return AVERROR(EINVAL); + } + rc_bits_per_second = avctx->rc_max_rate; + rc_target_percentage = (avctx->bit_rate * 100) / + avctx->rc_max_rate; + } else { + // We only have a target bitrate, but this mode requires + // that a maximum rate be supplied as well. Since the + // user does not want this to be a constraint, arbitrarily + // pick a maximum rate of double the target rate. + rc_bits_per_second = 2 * avctx->bit_rate; + rc_target_percentage = 50; + } + } else { + if (avctx->rc_max_rate > avctx->bit_rate) { + av_log(avctx, AV_LOG_WARNING, "Max bitrate is ignored " + "in %s RC mode.\n", rc_mode->name); + } rc_bits_per_second = avctx->bit_rate; rc_target_percentage = 100; + } + } else { + rc_bits_per_second = 0; + rc_target_percentage = 100; + } + + if (rc_mode->quality) { + if (ctx->explicit_qp) { + rc_quality = ctx->explicit_qp; + } else if (avctx->global_quality > 0) { + rc_quality = avctx->global_quality; } else { - rc_bits_per_second = avctx->rc_max_rate; - rc_target_percentage = (avctx->bit_rate * 100) / rc_bits_per_second; + rc_quality = ctx->codec->default_quality; + av_log(avctx, AV_LOG_WARNING, "No quality level set; " + "using default (%d).\n", rc_quality); } - rc_window_size = (hrd_buffer_size * 1000) / avctx->bit_rate; + } else { + rc_quality = 0; } - ctx->rc_params.misc.type = VAEncMiscParameterTypeRateControl; - ctx->rc_params.rc = (VAEncMiscParameterRateControl) { - .bits_per_second = rc_bits_per_second, - .target_percentage = rc_target_percentage, - .window_size = rc_window_size, - .initial_qp = 0, - .min_qp = (avctx->qmin > 0 ? avctx->qmin : 0), - .basic_unit_size = 0, - }; - vaapi_encode_add_global_param(avctx, &ctx->rc_params.misc, - sizeof(ctx->rc_params)); + if (rc_mode->hrd) { + 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 > 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 { + hrd_initial_buffer_fullness = hrd_buffer_size * 3 / 4; + } - ctx->hrd_params.misc.type = VAEncMiscParameterTypeHRD; - ctx->hrd_params.hrd = (VAEncMiscParameterHRD) { - .initial_buffer_fullness = hrd_initial_buffer_fullness, - .buffer_size = hrd_buffer_size, - }; - vaapi_encode_add_global_param(avctx, &ctx->hrd_params.misc, - sizeof(ctx->hrd_params)); + rc_window_size = (hrd_buffer_size * 1000) / rc_bits_per_second; + } else { + if (avctx->rc_buffer_size || avctx->rc_initial_buffer_occupancy) { + av_log(avctx, AV_LOG_WARNING, "Buffering settings are ignored " + "in %s RC mode.\n", rc_mode->name); + } + + hrd_buffer_size = 0; + hrd_initial_buffer_fullness = 0; + + if (rc_mode->mode != RC_MODE_AVBR) { + // Already set (with completely different meaning) for AVBR. + rc_window_size = 1000; + } + } + + 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->rc_mode = rc_mode; + ctx->rc_quality = rc_quality; + ctx->va_rc_mode = rc_mode->va_mode; + ctx->va_bit_rate = rc_bits_per_second; + + av_log(avctx, AV_LOG_VERBOSE, "RC mode: %s.\n", rc_mode->name); + if (rc_attr.value == VA_ATTRIB_NOT_SUPPORTED) { + // This driver does not want the RC mode attribute to be set. + } else { + ctx->config_attributes[ctx->nb_config_attributes++] = + (VAConfigAttrib) { + .type = VAConfigAttribRateControl, + .value = ctx->va_rc_mode, + }; + } + + if (rc_mode->quality) + av_log(avctx, AV_LOG_VERBOSE, "RC quality: %d.\n", rc_quality); + + if (rc_mode->va_mode != VA_RC_CQP) { + if (rc_mode->mode == RC_MODE_AVBR) { + av_log(avctx, AV_LOG_VERBOSE, "RC target: %"PRId64" bps " + "converging in %d frames with %d%% accuracy.\n", + rc_bits_per_second, rc_window_size, + rc_target_percentage); + } else if (rc_mode->bitrate) { + av_log(avctx, AV_LOG_VERBOSE, "RC target: %d%% of " + "%"PRId64" bps over %d ms.\n", rc_target_percentage, + rc_bits_per_second, rc_window_size); + } + + ctx->rc_params = (VAEncMiscParameterRateControl) { + .bits_per_second = rc_bits_per_second, + .target_percentage = rc_target_percentage, + .window_size = rc_window_size, + .initial_qp = 0, + .min_qp = (avctx->qmin > 0 ? avctx->qmin : 0), + .basic_unit_size = 0, +#if VA_CHECK_VERSION(1, 1, 0) + .ICQ_quality_factor = av_clip(rc_quality, 1, 51), + .max_qp = (avctx->qmax > 0 ? avctx->qmax : 0), +#endif +#if VA_CHECK_VERSION(1, 3, 0) + .quality_factor = rc_quality, +#endif + }; + vaapi_encode_add_global_param(avctx, + VAEncMiscParameterTypeRateControl, + &ctx->rc_params, + sizeof(ctx->rc_params)); + } + + if (rc_mode->hrd) { + av_log(avctx, AV_LOG_VERBOSE, "RC buffer: %"PRId64" bits, " + "initial fullness %"PRId64" bits.\n", + hrd_buffer_size, hrd_initial_buffer_fullness); + + ctx->hrd_params = (VAEncMiscParameterHRD) { + .initial_buffer_fullness = hrd_initial_buffer_fullness, + .buffer_size = hrd_buffer_size, + }; + vaapi_encode_add_global_param(avctx, + VAEncMiscParameterTypeHRD, + &ctx->hrd_params, + sizeof(ctx->hrd_params)); + } if (avctx->framerate.num > 0 && avctx->framerate.den > 0) av_reduce(&fr_num, &fr_den, @@ -1214,17 +1805,421 @@ static av_cold int vaapi_encode_init_rate_control(AVCodecContext *avctx) av_reduce(&fr_num, &fr_den, avctx->time_base.den, avctx->time_base.num, 65535); - ctx->fr_params.misc.type = VAEncMiscParameterTypeFrameRate; - ctx->fr_params.fr.framerate = (unsigned int)fr_den << 16 | fr_num; + av_log(avctx, AV_LOG_VERBOSE, "RC framerate: %d/%d (%.2f fps).\n", + fr_num, fr_den, (double)fr_num / fr_den); + ctx->fr_params = (VAEncMiscParameterFrameRate) { + .framerate = (unsigned int)fr_den << 16 | fr_num, + }; #if VA_CHECK_VERSION(0, 40, 0) - vaapi_encode_add_global_param(avctx, &ctx->fr_params.misc, + vaapi_encode_add_global_param(avctx, + VAEncMiscParameterTypeFrameRate, + &ctx->fr_params, sizeof(ctx->fr_params)); #endif 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 (ctx->codec->flags & FLAG_INTRA_ONLY || + 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 (!(ctx->codec->flags & FLAG_B_PICTURES) || + 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; + 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; + if (ctx->codec->flags & FLAG_B_PICTURE_REFERENCES) { + ctx->max_b_depth = FFMIN(ctx->desired_b_depth, + av_log2(ctx->b_per_p) + 1); + } else { + ctx->max_b_depth = 1; + } + } + + if (ctx->codec->flags & FLAG_NON_IDR_KEY_PICTURES) { + ctx->closed_gop = !!(avctx->flags & AV_CODEC_FLAG_CLOSED_GOP); + ctx->gop_per_idr = ctx->idr_interval + 1; + } else { + ctx->closed_gop = 1; + ctx->gop_per_idr = 1; + } + + return 0; +} + +static av_cold int vaapi_encode_init_row_slice_structure(AVCodecContext *avctx, + uint32_t slice_structure) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + int req_slices; + + // For fixed-size slices currently we only support whole rows, making + // rectangular slices. This could be extended to arbitrary runs of + // blocks, but since slices tend to be a conformance requirement and + // most cases (such as broadcast or bluray) want rectangular slices + // only it would need to be gated behind another option. + if (avctx->slices > ctx->slice_block_rows) { + av_log(avctx, AV_LOG_WARNING, "Not enough rows to use " + "configured number of slices (%d < %d); using " + "maximum.\n", ctx->slice_block_rows, avctx->slices); + req_slices = ctx->slice_block_rows; + } else { + req_slices = avctx->slices; + } + if (slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS || + slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS) { + ctx->nb_slices = req_slices; + ctx->slice_size = ctx->slice_block_rows / ctx->nb_slices; + } else if (slice_structure & VA_ENC_SLICE_STRUCTURE_POWER_OF_TWO_ROWS) { + int k; + for (k = 1;; k *= 2) { + if (2 * k * (req_slices - 1) + 1 >= ctx->slice_block_rows) + break; + } + ctx->nb_slices = (ctx->slice_block_rows + k - 1) / k; + ctx->slice_size = k; +#if VA_CHECK_VERSION(1, 0, 0) + } else if (slice_structure & VA_ENC_SLICE_STRUCTURE_EQUAL_ROWS) { + ctx->nb_slices = ctx->slice_block_rows; + ctx->slice_size = 1; +#endif + } else { + av_log(avctx, AV_LOG_ERROR, "Driver does not support any usable " + "slice structure modes (%#x).\n", slice_structure); + return AVERROR(EINVAL); + } + + return 0; +} + +static av_cold int vaapi_encode_init_tile_slice_structure(AVCodecContext *avctx, + uint32_t slice_structure) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + int i, req_tiles; + + if (!(slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_MACROBLOCKS || + (slice_structure & VA_ENC_SLICE_STRUCTURE_ARBITRARY_ROWS && + ctx->tile_cols == 1))) { + av_log(avctx, AV_LOG_ERROR, "Supported slice structure (%#x) doesn't work for " + "current tile requirement.\n", slice_structure); + return AVERROR(EINVAL); + } + + if (ctx->tile_rows > ctx->slice_block_rows || + ctx->tile_cols > ctx->slice_block_cols) { + av_log(avctx, AV_LOG_WARNING, "Not enough block rows/cols (%d x %d) " + "for configured number of tile (%d x %d); ", + ctx->slice_block_rows, ctx->slice_block_cols, + ctx->tile_rows, ctx->tile_cols); + ctx->tile_rows = ctx->tile_rows > ctx->slice_block_rows ? + ctx->slice_block_rows : ctx->tile_rows; + ctx->tile_cols = ctx->tile_cols > ctx->slice_block_cols ? + ctx->slice_block_cols : ctx->tile_cols; + av_log(avctx, AV_LOG_WARNING, "using allowed maximum (%d x %d).\n", + ctx->tile_rows, ctx->tile_cols); + } + + req_tiles = ctx->tile_rows * ctx->tile_cols; + + // Tile slice is not allowed to cross the boundary of a tile due to + // the constraints of media-driver. Currently we support one slice + // per tile. This could be extended to multiple slices per tile. + if (avctx->slices != req_tiles) + av_log(avctx, AV_LOG_WARNING, "The number of requested slices " + "mismatches with configured number of tile (%d != %d); " + "using requested tile number for slice.\n", + avctx->slices, req_tiles); + + ctx->nb_slices = req_tiles; + + // Default in uniform spacing + // 6-3, 6-5 + for (i = 0; i < ctx->tile_cols; i++) { + ctx->col_width[i] = ( i + 1 ) * ctx->slice_block_cols / ctx->tile_cols - + i * ctx->slice_block_cols / ctx->tile_cols; + ctx->col_bd[i + 1] = ctx->col_bd[i] + ctx->col_width[i]; + } + // 6-4, 6-6 + for (i = 0; i < ctx->tile_rows; i++) { + ctx->row_height[i] = ( i + 1 ) * ctx->slice_block_rows / ctx->tile_rows - + i * ctx->slice_block_rows / ctx->tile_rows; + ctx->row_bd[i + 1] = ctx->row_bd[i] + ctx->row_height[i]; + } + + av_log(avctx, AV_LOG_VERBOSE, "Encoding pictures with %d x %d tile.\n", + ctx->tile_rows, ctx->tile_cols); + + return 0; +} + +static av_cold int vaapi_encode_init_slice_structure(AVCodecContext *avctx) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + VAConfigAttrib attr[3] = { { VAConfigAttribEncMaxSlices }, + { VAConfigAttribEncSliceStructure }, +#if VA_CHECK_VERSION(1, 1, 0) + { VAConfigAttribEncTileSupport }, +#endif + }; + VAStatus vas; + uint32_t max_slices, slice_structure; + int ret; + + if (!(ctx->codec->flags & FLAG_SLICE_CONTROL)) { + if (avctx->slices > 0) { + av_log(avctx, AV_LOG_WARNING, "Multiple slices were requested " + "but this codec does not support controlling slices.\n"); + } + return 0; + } + + ctx->slice_block_rows = (avctx->height + ctx->slice_block_height - 1) / + ctx->slice_block_height; + ctx->slice_block_cols = (avctx->width + ctx->slice_block_width - 1) / + ctx->slice_block_width; + + if (avctx->slices <= 1 && !ctx->tile_rows && !ctx->tile_cols) { + ctx->nb_slices = 1; + ctx->slice_size = ctx->slice_block_rows; + return 0; + } + + vas = vaGetConfigAttributes(ctx->hwctx->display, + ctx->va_profile, + ctx->va_entrypoint, + attr, FF_ARRAY_ELEMS(attr)); + if (vas != VA_STATUS_SUCCESS) { + av_log(avctx, AV_LOG_ERROR, "Failed to query slice " + "attributes: %d (%s).\n", vas, vaErrorStr(vas)); + return AVERROR_EXTERNAL; + } + max_slices = attr[0].value; + slice_structure = attr[1].value; + if (max_slices == VA_ATTRIB_NOT_SUPPORTED || + slice_structure == VA_ATTRIB_NOT_SUPPORTED) { + av_log(avctx, AV_LOG_ERROR, "Driver does not support encoding " + "pictures as multiple slices.\n."); + return AVERROR(EINVAL); + } + + if (ctx->tile_rows && ctx->tile_cols) { +#if VA_CHECK_VERSION(1, 1, 0) + uint32_t tile_support = attr[2].value; + if (tile_support == VA_ATTRIB_NOT_SUPPORTED) { + av_log(avctx, AV_LOG_ERROR, "Driver does not support encoding " + "pictures as multiple tiles.\n."); + return AVERROR(EINVAL); + } +#else + av_log(avctx, AV_LOG_ERROR, "Tile encoding option is " + "not supported with this VAAPI version.\n"); + return AVERROR(EINVAL); +#endif + } + + if (ctx->tile_rows && ctx->tile_cols) + ret = vaapi_encode_init_tile_slice_structure(avctx, slice_structure); + else + ret = vaapi_encode_init_row_slice_structure(avctx, slice_structure); + if (ret < 0) + return ret; + + if (ctx->nb_slices > avctx->slices) { + av_log(avctx, AV_LOG_WARNING, "Slice count rounded up to " + "%d (from %d) due to driver constraints on slice " + "structure.\n", ctx->nb_slices, avctx->slices); + } + if (ctx->nb_slices > max_slices) { + av_log(avctx, AV_LOG_ERROR, "Driver does not support " + "encoding with %d slices (max %"PRIu32").\n", + ctx->nb_slices, max_slices); + return AVERROR(EINVAL); + } + + av_log(avctx, AV_LOG_VERBOSE, "Encoding pictures with %d slices.\n", + ctx->nb_slices); + return 0; +} + +static av_cold int vaapi_encode_init_packed_headers(AVCodecContext *avctx) +{ + VAAPIEncodeContext *ctx = avctx->priv_data; + VAStatus vas; + VAConfigAttrib attr = { VAConfigAttribEncPackedHeaders }; + + 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 packed headers " + "attribute: %d (%s).\n", vas, vaErrorStr(vas)); + return AVERROR_EXTERNAL; + } + + if (attr.value == VA_ATTRIB_NOT_SUPPORTED) { + if (ctx->desired_packed_headers) { + av_log(avctx, AV_LOG_WARNING, "Driver does not support any " + "packed headers (wanted %#x).\n", + ctx->desired_packed_headers); + } else { + av_log(avctx, AV_LOG_VERBOSE, "Driver does not support any " + "packed headers (none wanted).\n"); + } + ctx->va_packed_headers = 0; + } else { + if (ctx->desired_packed_headers & ~attr.value) { + av_log(avctx, AV_LOG_WARNING, "Driver does not support some " + "wanted packed headers (wanted %#x, found %#x).\n", + ctx->desired_packed_headers, attr.value); + } else { + av_log(avctx, AV_LOG_VERBOSE, "All wanted packed headers " + "available (wanted %#x, found %#x).\n", + ctx->desired_packed_headers, attr.value); + } + ctx->va_packed_headers = ctx->desired_packed_headers & attr.value; + } + + if (ctx->va_packed_headers) { + ctx->config_attributes[ctx->nb_config_attributes++] = + (VAConfigAttrib) { + .type = VAConfigAttribEncPackedHeaders, + .value = ctx->va_packed_headers, + }; + } + + if ( (ctx->desired_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE) && + !(ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE) && + (avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER)) { + av_log(avctx, AV_LOG_WARNING, "Driver does not support packed " + "sequence headers, but a global header is requested.\n"); + av_log(avctx, AV_LOG_WARNING, "No global header will be written: " + "this may result in a stream which is not usable for some " + "purposes (e.g. not muxable to some containers).\n"); + } + + return 0; +} + +static av_cold int vaapi_encode_init_quality(AVCodecContext *avctx) +{ +#if VA_CHECK_VERSION(0, 36, 0) + VAAPIEncodeContext *ctx = avctx->priv_data; + VAStatus vas; + VAConfigAttrib attr = { VAConfigAttribEncQualityRange }; + int quality = avctx->compression_level; + + 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 quality " + "config attribute: %d (%s).\n", vas, vaErrorStr(vas)); + return AVERROR_EXTERNAL; + } + + if (attr.value == VA_ATTRIB_NOT_SUPPORTED) { + if (quality != 0) { + av_log(avctx, AV_LOG_WARNING, "Quality attribute is not " + "supported: will use default quality level.\n"); + } + } else { + if (quality > attr.value) { + av_log(avctx, AV_LOG_WARNING, "Invalid quality level: " + "valid range is 0-%d, using %d.\n", + attr.value, attr.value); + quality = attr.value; + } + + ctx->quality_params = (VAEncMiscParameterBufferQualityLevel) { + .quality_level = quality, + }; + vaapi_encode_add_global_param(avctx, + VAEncMiscParameterTypeQualityLevel, + &ctx->quality_params, + sizeof(ctx->quality_params)); + } +#else + av_log(avctx, AV_LOG_WARNING, "The encode quality option is " + "not supported with this VAAPI version.\n"); +#endif + + return 0; +} + +static av_cold int vaapi_encode_init_roi(AVCodecContext *avctx) +{ +#if VA_CHECK_VERSION(1, 0, 0) + VAAPIEncodeContext *ctx = avctx->priv_data; + VAStatus vas; + VAConfigAttrib attr = { VAConfigAttribEncROI }; + + 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 ROI " + "config attribute: %d (%s).\n", vas, vaErrorStr(vas)); + return AVERROR_EXTERNAL; + } + + if (attr.value == VA_ATTRIB_NOT_SUPPORTED) { + ctx->roi_allowed = 0; + } else { + VAConfigAttribValEncROI roi = { + .value = attr.value, + }; + + ctx->roi_max_regions = roi.bits.num_roi_regions; + ctx->roi_allowed = ctx->roi_max_regions > 0 && + (ctx->va_rc_mode == VA_RC_CQP || + roi.bits.roi_rc_qp_delta_support); + } +#endif + return 0; +} + static void vaapi_encode_free_output_buffer(void *opaque, uint8_t *data) { @@ -1240,7 +2235,7 @@ static void vaapi_encode_free_output_buffer(void *opaque, } static AVBufferRef *vaapi_encode_alloc_output_buffer(void *opaque, - int size) + size_t size) { AVCodecContext *avctx = opaque; VAAPIEncodeContext *ctx = avctx->priv_data; @@ -1349,9 +2344,6 @@ static av_cold int vaapi_encode_create_recon_frames(AVCodecContext *avctx) ctx->recon_frames->sw_format = recon_format; ctx->recon_frames->width = ctx->surface_width; 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; err = av_hwframe_ctx_init(ctx->recon_frames_ref); if (err < 0) { @@ -1374,6 +2366,11 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) VAStatus vas; int err; + ctx->frame = av_frame_alloc(); + if (!ctx->frame) { + return AVERROR(ENOMEM); + } + if (!avctx->hw_frames_ctx) { av_log(avctx, AV_LOG_ERROR, "A hardware frames reference is " "required to associate the encoding device.\n"); @@ -1398,10 +2395,36 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) ctx->device = (AVHWDeviceContext*)ctx->device_ref->data; ctx->hwctx = ctx->device->hwctx; - err = vaapi_encode_config_attributes(avctx); + err = vaapi_encode_profile_entrypoint(avctx); + if (err < 0) + goto fail; + + err = vaapi_encode_init_rate_control(avctx); if (err < 0) goto fail; + err = vaapi_encode_init_gop_structure(avctx); + if (err < 0) + goto fail; + + err = vaapi_encode_init_slice_structure(avctx); + if (err < 0) + goto fail; + + err = vaapi_encode_init_packed_headers(avctx); + if (err < 0) + goto fail; + + err = vaapi_encode_init_roi(avctx); + if (err < 0) + goto fail; + + if (avctx->compression_level >= 0) { + err = vaapi_encode_init_quality(avctx); + if (err < 0) + goto fail; + } + vas = vaCreateConfig(ctx->hwctx->display, ctx->va_profile, ctx->va_entrypoint, ctx->config_attributes, ctx->nb_config_attributes, @@ -1439,59 +2462,14 @@ 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) goto fail; } - if (avctx->compression_level >= 0) { -#if VA_CHECK_VERSION(0, 36, 0) - VAConfigAttrib attr = { VAConfigAttribEncQualityRange }; - - vas = vaGetConfigAttributes(ctx->hwctx->display, - ctx->va_profile, - ctx->va_entrypoint, - &attr, 1); - if (vas != VA_STATUS_SUCCESS) { - av_log(avctx, AV_LOG_WARNING, "Failed to query quality " - "attribute: will use default compression level.\n"); - } else { - if (avctx->compression_level > attr.value) { - av_log(avctx, AV_LOG_WARNING, "Invalid compression " - "level: valid range is 0-%d, using %d.\n", - attr.value, attr.value); - avctx->compression_level = attr.value; - } - - ctx->quality_params.misc.type = - VAEncMiscParameterTypeQualityLevel; - ctx->quality_params.quality.quality_level = - avctx->compression_level; - - vaapi_encode_add_global_param(avctx, &ctx->quality_params.misc, - sizeof(ctx->quality_params)); - } -#else - av_log(avctx, AV_LOG_WARNING, "The encode compression level " - "option is not supported with this VAAPI version.\n"); -#endif - } - - ctx->input_order = 0; - ctx->output_delay = avctx->max_b_frames; - 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; + ctx->output_delay = ctx->b_per_p; + ctx->decode_delay = ctx->max_b_depth; if (ctx->codec->sequence_params_size > 0) { ctx->codec_sequence_params = @@ -1519,12 +2497,9 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) } } - // This should be configurable somehow. (Needs testing on a machine - // where it actually overlaps properly, though.) - ctx->issue_mode = ISSUE_MODE_MAXIMISE_THROUGHPUT; - if (ctx->va_packed_headers & VA_ENC_PACKED_HEADER_SEQUENCE && - ctx->codec->write_sequence_header) { + ctx->codec->write_sequence_header && + avctx->flags & AV_CODEC_FLAG_GLOBAL_HEADER) { char data[MAX_PARAM_BUFFER_SIZE]; size_t bit_len = 8 * sizeof(data); @@ -1548,7 +2523,6 @@ av_cold int ff_vaapi_encode_init(AVCodecContext *avctx) return 0; fail: - ff_vaapi_encode_close(avctx); return err; } @@ -1574,6 +2548,8 @@ av_cold int ff_vaapi_encode_close(AVCodecContext *avctx) ctx->va_config = VA_INVALID_ID; } + av_frame_free(&ctx->frame); + av_freep(&ctx->codec_sequence_params); av_freep(&ctx->codec_picture_params);