@@ -43,6 +43,7 @@ version <next>:
- adenorm filter
- ADPCM IMA AMV encoder
- AMV muxer
+- NVDEC AV1 hwaccel
version 4.3:
@@ -2918,6 +2918,8 @@ videotoolbox_hwaccel_deps="videotoolbox pthreads"
videotoolbox_hwaccel_extralibs="-framework QuartzCore"
xvmc_deps="X11_extensions_XvMClib_h"
+av1_nvdec_hwaccel_deps="nvdec CUVIDAV1PICPARAMS"
+av1_nvdec_hwaccel_select="av1_decoder"
av1_vaapi_hwaccel_deps="vaapi VADecPictureParameterBufferAV1_bit_depth_idx"
av1_vaapi_hwaccel_select="av1_decoder"
h263_vaapi_hwaccel_deps="vaapi"
@@ -912,6 +912,7 @@ OBJS-$(CONFIG_VAAPI) += vaapi_decode.o
OBJS-$(CONFIG_VIDEOTOOLBOX) += videotoolbox.o
OBJS-$(CONFIG_VDPAU) += vdpau.o
+OBJS-$(CONFIG_AV1_NVDEC_HWACCEL) += nvdec_av1.o
OBJS-$(CONFIG_AV1_VAAPI_HWACCEL) += vaapi_av1.o
OBJS-$(CONFIG_H263_VAAPI_HWACCEL) += vaapi_mpeg4.o
OBJS-$(CONFIG_H263_VIDEOTOOLBOX_HWACCEL) += videotoolbox.o
@@ -324,7 +324,7 @@ static int get_pixel_format(AVCodecContext *avctx)
uint8_t bit_depth;
int ret;
enum AVPixelFormat pix_fmt = AV_PIX_FMT_NONE;
-#define HWACCEL_MAX (CONFIG_AV1_VAAPI_HWACCEL)
+#define HWACCEL_MAX (CONFIG_AV1_NVDEC_HWACCEL + CONFIG_AV1_VAAPI_HWACCEL)
enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmtp = pix_fmts;
if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
@@ -387,11 +387,17 @@ static int get_pixel_format(AVCodecContext *avctx)
switch (s->pix_fmt) {
case AV_PIX_FMT_YUV420P:
+#if CONFIG_AV1_NVDEC_HWACCEL
+ *fmtp++ = AV_PIX_FMT_CUDA;
+#endif
#if CONFIG_AV1_VAAPI_HWACCEL
*fmtp++ = AV_PIX_FMT_VAAPI;
#endif
break;
case AV_PIX_FMT_YUV420P10:
+#if CONFIG_AV1_NVDEC_HWACCEL
+ *fmtp++ = AV_PIX_FMT_CUDA;
+#endif
#if CONFIG_AV1_VAAPI_HWACCEL
*fmtp++ = AV_PIX_FMT_VAAPI;
#endif
@@ -972,6 +978,9 @@ AVCodec ff_av1_decoder = {
.flush = av1_decode_flush,
.profiles = NULL_IF_CONFIG_SMALL(ff_av1_profiles),
.hw_configs = (const AVCodecHWConfigInternal * []) {
+#if CONFIG_AV1_NVDEC_HWACCEL
+ HWACCEL_NVDEC(av1),
+#endif
#if CONFIG_AV1_VAAPI_HWACCEL
HWACCEL_VAAPI(av1),
#endif
@@ -21,6 +21,7 @@
#include "avcodec.h"
+extern const AVHWAccel ff_av1_nvdec_hwaccel;
extern const AVHWAccel ff_av1_vaapi_hwaccel;
extern const AVHWAccel ff_h263_vaapi_hwaccel;
extern const AVHWAccel ff_h263_videotoolbox_hwaccel;
@@ -62,6 +62,9 @@ typedef struct NVDECFramePool {
static int map_avcodec_id(enum AVCodecID id)
{
switch (id) {
+#if CONFIG_AV1_NVDEC_HWACCEL
+ case AV_CODEC_ID_AV1: return cudaVideoCodec_AV1;
+#endif
case AV_CODEC_ID_H264: return cudaVideoCodec_H264;
case AV_CODEC_ID_HEVC: return cudaVideoCodec_HEVC;
case AV_CODEC_ID_MJPEG: return cudaVideoCodec_JPEG;
@@ -258,6 +261,7 @@ int ff_nvdec_decode_uninit(AVCodecContext *avctx)
NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
av_freep(&ctx->bitstream);
+ av_freep(&ctx->bitstream_internal);
ctx->bitstream_len = 0;
ctx->bitstream_allocated = 0;
@@ -59,6 +59,7 @@ typedef struct NVDECContext {
uint8_t *bitstream;
int bitstream_len;
unsigned int bitstream_allocated;
+ uint8_t *bitstream_internal;
unsigned *slice_offsets;
int nb_slices;
new file mode 100644
@@ -0,0 +1,347 @@
+/*
+ * VP9 HW decode acceleration through NVDEC
+ *
+ * Copyright (c) 2020 Timo Rothenpieler
+ *
+ * This file is part of FFmpeg.
+ *
+ * FFmpeg is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License as published by the Free Software Foundation; either
+ * version 2.1 of the License, or (at your option) any later version.
+ *
+ * FFmpeg is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
+ * Lesser General Public License for more details.
+ *
+ * You should have received a copy of the GNU Lesser General Public
+ * License along with FFmpeg; if not, write to the Free Software
+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include "avcodec.h"
+#include "nvdec.h"
+#include "decode.h"
+#include "internal.h"
+#include "av1dec.h"
+
+
+static int get_bit_depth_from_seq(const AV1RawSequenceHeader *seq)
+{
+ if (seq->seq_profile == 2 && seq->color_config.high_bitdepth)
+ return seq->color_config.twelve_bit ? 12 : 10;
+ else if (seq->seq_profile <= 2 && seq->color_config.high_bitdepth)
+ return 10;
+ else
+ return 8;
+}
+
+static int nvdec_av1_start_frame(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
+{
+ const AV1DecContext *s = avctx->priv_data;
+ const AV1RawSequenceHeader *seq = s->raw_seq;
+ const AV1RawFrameHeader *frame_header = s->raw_frame_header;
+
+ NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
+ CUVIDPICPARAMS *pp = &ctx->pic_params;
+ CUVIDAV1PICPARAMS *ppc = &pp->CodecSpecific.av1;
+ FrameDecodeData *fdd;
+ NVDECFrame *cf;
+ AVFrame *cur_frame = s->cur_frame.tf.f;
+
+ unsigned char remap_lr_type[4] = { AV1_RESTORE_NONE, AV1_RESTORE_SWITCHABLE, AV1_RESTORE_WIENER, AV1_RESTORE_SGRPROJ };
+
+ int ret, i, j;
+
+ ret = ff_nvdec_start_frame_sep_ref(avctx, cur_frame, frame_header->apply_grain);
+ if (ret < 0)
+ return ret;
+
+ fdd = (FrameDecodeData*)cur_frame->private_ref->data;
+ cf = (NVDECFrame*)fdd->hwaccel_priv;
+
+ *pp = (CUVIDPICPARAMS) {
+ .PicWidthInMbs = (cur_frame->width + 15) / 16,
+ .FrameHeightInMbs = (cur_frame->height + 15) / 16,
+ .CurrPicIdx = cf->idx,
+ .ref_pic_flag = !!frame_header->refresh_frame_flags,
+ .intra_pic_flag = frame_header->frame_type == AV1_FRAME_INTRA_ONLY ||
+ frame_header->frame_type == AV1_FRAME_KEY,
+
+ .CodecSpecific.av1 = {
+ .width = cur_frame->width,
+ .height = cur_frame->height,
+
+ .frame_offset = frame_header->order_hint,
+ .decodePicIdx = cf->ref_idx,
+
+ /* Sequence Header */
+ .profile = seq->seq_profile,
+ .use_128x128_superblock = seq->use_128x128_superblock,
+ .subsampling_x = seq->color_config.subsampling_x,
+ .subsampling_y = seq->color_config.subsampling_y,
+ .mono_chrome = seq->color_config.mono_chrome,
+ .bit_depth_minus8 = get_bit_depth_from_seq(seq) - 8,
+ .enable_filter_intra = seq->enable_filter_intra,
+ .enable_intra_edge_filter = seq->enable_intra_edge_filter,
+ .enable_interintra_compound = seq->enable_interintra_compound,
+ .enable_masked_compound = seq->enable_masked_compound,
+ .enable_dual_filter = seq->enable_dual_filter,
+ .enable_order_hint = seq->enable_order_hint,
+ .order_hint_bits_minus1 = seq->order_hint_bits_minus_1,
+ .enable_jnt_comp = seq->enable_jnt_comp,
+ .enable_superres = seq->enable_superres,
+ .enable_cdef = seq->enable_cdef,
+ .enable_restoration = seq->enable_restoration,
+ .enable_fgs = seq->film_grain_params_present,
+
+ /* Frame Header */
+ .frame_type = frame_header->frame_type,
+ .show_frame = frame_header->show_frame,
+ .disable_cdf_update = frame_header->disable_cdf_update,
+ .allow_screen_content_tools = frame_header->allow_screen_content_tools,
+ .force_integer_mv = frame_header->force_integer_mv ||
+ frame_header->frame_type == AV1_FRAME_INTRA_ONLY ||
+ frame_header->frame_type == AV1_FRAME_KEY,
+ .coded_denom = frame_header->coded_denom,
+ .allow_intrabc = frame_header->allow_intrabc,
+ .allow_high_precision_mv = frame_header->allow_high_precision_mv,
+ .interp_filter = frame_header->interpolation_filter,
+ .switchable_motion_mode = frame_header->is_motion_mode_switchable,
+ .use_ref_frame_mvs = frame_header->use_ref_frame_mvs,
+ .disable_frame_end_update_cdf = frame_header->disable_frame_end_update_cdf,
+ .delta_q_present = frame_header->delta_q_present,
+ .delta_q_res = frame_header->delta_q_res,
+ .using_qmatrix = frame_header->using_qmatrix,
+ .coded_lossless = s->cur_frame.coded_lossless,
+ .use_superres = frame_header->use_superres,
+ .tx_mode = frame_header->tx_mode,
+ .reference_mode = frame_header->reference_select,
+ .allow_warped_motion = frame_header->allow_warped_motion,
+ .reduced_tx_set = frame_header->reduced_tx_set,
+ .skip_mode = frame_header->skip_mode_present,
+
+ /* Tiling Info */
+ .num_tile_cols = frame_header->tile_cols,
+ .num_tile_rows = frame_header->tile_rows,
+ .context_update_tile_id = frame_header->context_update_tile_id,
+
+ /* CDEF */
+ .cdef_damping_minus_3 = frame_header->cdef_damping_minus_3,
+ .cdef_bits = frame_header->cdef_bits,
+
+ /* SkipModeFrames */
+ .SkipModeFrame0 = s->cur_frame.skip_mode_frame_idx[0],
+ .SkipModeFrame1 = s->cur_frame.skip_mode_frame_idx[1],
+
+ /* QP Information */
+ .base_qindex = frame_header->base_q_idx,
+ .qp_y_dc_delta_q = frame_header->delta_q_y_dc,
+ .qp_u_dc_delta_q = frame_header->delta_q_u_dc,
+ .qp_v_dc_delta_q = frame_header->delta_q_v_dc,
+ .qp_u_ac_delta_q = frame_header->delta_q_u_ac,
+ .qp_v_ac_delta_q = frame_header->delta_q_v_ac,
+ .qm_y = frame_header->qm_y,
+ .qm_u = frame_header->qm_u,
+ .qm_v = frame_header->qm_v,
+
+ /* Segmentation */
+ .segmentation_enabled = frame_header->segmentation_enabled,
+ .segmentation_update_map = frame_header->segmentation_update_map,
+ .segmentation_update_data = frame_header->segmentation_update_data,
+ .segmentation_temporal_update = frame_header->segmentation_temporal_update,
+
+ /* Loopfilter */
+ .loop_filter_level[0] = frame_header->loop_filter_level[0],
+ .loop_filter_level[1] = frame_header->loop_filter_level[1],
+ .loop_filter_level_u = frame_header->loop_filter_level[2],
+ .loop_filter_level_v = frame_header->loop_filter_level[3],
+ .loop_filter_sharpness = frame_header->loop_filter_sharpness,
+ .loop_filter_delta_enabled = frame_header->loop_filter_delta_enabled,
+ .loop_filter_delta_update = frame_header->loop_filter_delta_update,
+ .loop_filter_mode_deltas[0] = frame_header->loop_filter_mode_deltas[0],
+ .loop_filter_mode_deltas[1] = frame_header->loop_filter_mode_deltas[1],
+ .delta_lf_present = frame_header->delta_lf_present,
+ .delta_lf_res = frame_header->delta_lf_res,
+ .delta_lf_multi = frame_header->delta_lf_multi,
+
+ /* Restoration */
+ .lr_type[0] = remap_lr_type[frame_header->lr_type[0]],
+ .lr_type[1] = remap_lr_type[frame_header->lr_type[1]],
+ .lr_type[2] = remap_lr_type[frame_header->lr_type[2]],
+ .lr_unit_size[0] = 1 + frame_header->lr_unit_shift,
+ .lr_unit_size[1] = 1 + frame_header->lr_unit_shift - frame_header->lr_uv_shift,
+ .lr_unit_size[2] = 1 + frame_header->lr_unit_shift - frame_header->lr_uv_shift,
+
+ /* Reference Frames */
+ .temporal_layer_id = s->cur_frame.temporal_id,
+ .spatial_layer_id = s->cur_frame.spatial_id,
+
+ /* Film Grain Params */
+ .apply_grain = frame_header->apply_grain,
+ .overlap_flag = frame_header->overlap_flag,
+ .scaling_shift_minus8 = frame_header->grain_scaling_minus_8,
+ .chroma_scaling_from_luma = frame_header->chroma_scaling_from_luma,
+ .ar_coeff_lag = frame_header->ar_coeff_lag,
+ .ar_coeff_shift_minus6 = frame_header->ar_coeff_shift_minus_6,
+ .grain_scale_shift = frame_header->grain_scale_shift,
+ .clip_to_restricted_range = frame_header->clip_to_restricted_range,
+ .num_y_points = frame_header->num_y_points,
+ .num_cb_points = frame_header->num_cb_points,
+ .num_cr_points = frame_header->num_cr_points,
+ .random_seed = frame_header->grain_seed,
+ .cb_mult = frame_header->cb_mult,
+ .cb_luma_mult = frame_header->cb_luma_mult,
+ .cb_offset = frame_header->cb_offset,
+ .cr_mult = frame_header->cr_mult,
+ .cr_luma_mult = frame_header->cr_luma_mult,
+ .cr_offset = frame_header->cr_offset
+ }
+ };
+
+ /* Tiling Info */
+ for (i = 0; i < frame_header->tile_cols; ++i) {
+ ppc->tile_widths[i] = frame_header->width_in_sbs_minus_1[i] + 1;
+ }
+ for (i = 0; i < frame_header->tile_rows; ++i) {
+ ppc->tile_heights[i] = frame_header->height_in_sbs_minus_1[i] + 1;
+ }
+
+ /* CDEF */
+ for (i = 0; i < (1 << frame_header->cdef_bits); ++i) {
+ ppc->cdef_y_strength[i] = (frame_header->cdef_y_pri_strength[i] & 0x0F) | (frame_header->cdef_y_sec_strength[i] << 4);
+ ppc->cdef_uv_strength[i] = (frame_header->cdef_uv_pri_strength[i] & 0x0F) | (frame_header->cdef_uv_sec_strength[i] << 4);
+ }
+
+ /* Segmentation */
+ for (i = 0; i < AV1_MAX_SEGMENTS; ++i) {
+ ppc->segmentation_feature_mask[i] = 0;
+ for (j = 0; j < AV1_SEG_LVL_MAX; ++j) {
+ ppc->segmentation_feature_mask[i] |= frame_header->feature_enabled[i][j] << j;
+ ppc->segmentation_feature_data[i][j] = frame_header->feature_value[i][j];
+ }
+ }
+
+ for (i = 0; i < AV1_NUM_REF_FRAMES; ++i) {
+ /* Loopfilter */
+ ppc->loop_filter_ref_deltas[i] = frame_header->loop_filter_ref_deltas[i];
+
+ /* Reference Frames */
+ ppc->ref_frame_map[i] = ff_nvdec_get_ref_idx(s->ref[i].tf.f);
+ }
+
+ if (frame_header->primary_ref_frame == AV1_PRIMARY_REF_NONE) {
+ ppc->primary_ref_frame = -1;
+ } else {
+ int8_t pri_ref_idx = frame_header->ref_frame_idx[frame_header->primary_ref_frame];
+ ppc->primary_ref_frame = ppc->ref_frame_map[pri_ref_idx];
+ }
+
+ for (i = 0; i < AV1_REFS_PER_FRAME; ++i) {
+ /* Ref Frame List */
+ int8_t ref_idx = frame_header->ref_frame_idx[i];
+ AVFrame *ref_frame = s->ref[ref_idx].tf.f;
+
+ ppc->ref_frame[i].index = ppc->ref_frame_map[ref_idx];
+ ppc->ref_frame[i].width = ref_frame->width;
+ ppc->ref_frame[i].height = ref_frame->height;
+
+ /* Global Motion */
+ ppc->global_motion[i].invalid = !frame_header->is_global[AV1_REF_FRAME_LAST + i];
+ ppc->global_motion[i].wmtype = s->cur_frame.gm_type[AV1_REF_FRAME_LAST + i];
+ for (j = 0; j < 6; ++j) {
+ ppc->global_motion[i].wmmat[j] = s->cur_frame.gm_params[AV1_REF_FRAME_LAST + i][j];
+ }
+ }
+
+ /* Film Grain Params */
+ if (frame_header->apply_grain) {
+ for (i = 0; i < 14; ++i) {
+ ppc->scaling_points_y[i][0] = frame_header->point_y_value[i];
+ ppc->scaling_points_y[i][1] = frame_header->point_y_scaling[i];
+ }
+ for (i = 0; i < 10; ++i) {
+ ppc->scaling_points_cb[i][0] = frame_header->point_cb_value[i];
+ ppc->scaling_points_cb[i][1] = frame_header->point_cb_scaling[i];
+ ppc->scaling_points_cr[i][0] = frame_header->point_cr_value[i];
+ ppc->scaling_points_cr[i][1] = frame_header->point_cr_scaling[i];
+ }
+ for (i = 0; i < 24; ++i) {
+ ppc->ar_coeffs_y[i] = (short)frame_header->ar_coeffs_y_plus_128[i] - 128;
+ }
+ for (i = 0; i < 25; ++i) {
+ ppc->ar_coeffs_cb[i] = (short)frame_header->ar_coeffs_cb_plus_128[i] - 128;
+ ppc->ar_coeffs_cr[i] = (short)frame_header->ar_coeffs_cr_plus_128[i] - 128;
+ }
+ }
+
+ return 0;
+}
+
+static int nvdec_av1_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)
+{
+ const AV1DecContext *s = avctx->priv_data;
+ const AV1RawFrameHeader *frame_header = s->raw_frame_header;
+ NVDECContext *ctx = avctx->internal->hwaccel_priv_data;
+ void *tmp;
+
+ ctx->nb_slices = frame_header->tile_cols * frame_header->tile_rows;
+
+ tmp = av_fast_realloc(ctx->slice_offsets, &ctx->slice_offsets_allocated,
+ ctx->nb_slices * 2 * sizeof(*ctx->slice_offsets));
+ if (!tmp) {
+ return AVERROR(ENOMEM);
+ }
+ ctx->slice_offsets = tmp;
+
+ /* Shortcut if all tiles are in the same buffer */
+ if (ctx->nb_slices == s->tg_end - s->tg_start + 1) {
+ ctx->bitstream = (uint8_t*)buffer;
+ ctx->bitstream_len = size;
+
+ for (int i = 0; i < ctx->nb_slices; ++i) {
+ ctx->slice_offsets[i*2 ] = s->tile_group_info[i].tile_offset;
+ ctx->slice_offsets[i*2 + 1] = ctx->slice_offsets[i*2] + s->tile_group_info[i].tile_size;
+ }
+
+ return 0;
+ }
+
+ tmp = av_fast_realloc(ctx->bitstream_internal, &ctx->bitstream_allocated,
+ ctx->bitstream_len + size);
+ if (!tmp) {
+ return AVERROR(ENOMEM);
+ }
+ ctx->bitstream = ctx->bitstream_internal = tmp;
+
+ memcpy(ctx->bitstream + ctx->bitstream_len, buffer, size);
+
+ for (uint32_t tile_num = s->tg_start; tile_num <= s->tg_end; ++tile_num) {
+ ctx->slice_offsets[tile_num*2 ] = ctx->bitstream_len + s->tile_group_info[tile_num].tile_offset;
+ ctx->slice_offsets[tile_num*2 + 1] = ctx->slice_offsets[tile_num*2] + s->tile_group_info[tile_num].tile_size;
+ }
+ ctx->bitstream_len += size;
+
+ return 0;
+}
+
+static int nvdec_av1_frame_params(AVCodecContext *avctx, AVBufferRef *hw_frames_ctx)
+{
+ /* Maximum of 8 reference frames, but potentially stored twice due to film grain */
+ return ff_nvdec_frame_params(avctx, hw_frames_ctx, 8 * 2, 0);
+}
+
+const AVHWAccel ff_av1_nvdec_hwaccel = {
+ .name = "av1_nvdec",
+ .type = AVMEDIA_TYPE_VIDEO,
+ .id = AV_CODEC_ID_AV1,
+ .pix_fmt = AV_PIX_FMT_CUDA,
+ .start_frame = nvdec_av1_start_frame,
+ .end_frame = ff_nvdec_simple_end_frame,
+ .decode_slice = nvdec_av1_decode_slice,
+ .frame_params = nvdec_av1_frame_params,
+ .init = ff_nvdec_decode_init,
+ .uninit = ff_nvdec_decode_uninit,
+ .priv_data_size = sizeof(NVDECContext),
+};
@@ -29,7 +29,7 @@
#define LIBAVCODEC_VERSION_MAJOR 58
#define LIBAVCODEC_VERSION_MINOR 112
-#define LIBAVCODEC_VERSION_MICRO 101
+#define LIBAVCODEC_VERSION_MICRO 102
#define LIBAVCODEC_VERSION_INT AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \
LIBAVCODEC_VERSION_MINOR, \
Signed-off-by: Timo Rothenpieler <timo@rothenpieler.org> Co-authored-by: James Almer <jamrial@gmail.com> --- Changelog | 1 + configure | 2 + libavcodec/Makefile | 1 + libavcodec/av1dec.c | 11 +- libavcodec/hwaccels.h | 1 + libavcodec/nvdec.c | 4 + libavcodec/nvdec.h | 1 + libavcodec/nvdec_av1.c | 347 +++++++++++++++++++++++++++++++++++++++++ libavcodec/version.h | 2 +- 9 files changed, 368 insertions(+), 2 deletions(-) create mode 100644 libavcodec/nvdec_av1.c