@@ -13,7 +13,7 @@ version <next>:

 - PCE support for extended channel layouts in the AAC encoder

 - native aptX encoder and decoder

 - Raw aptX muxer and demuxer

-- NVIDIA NVDEC-accelerated H.264 and HEVC hwaccel decoding

+- NVIDIA NVDEC-accelerated H.264, HEVC and VP9 hwaccel decoding

 - Intel QSV-accelerated overlay filter

@@ -2764,6 +2764,8 @@ vp9_d3d11va2_hwaccel_select="vp9_decoder"

 vp9_dxva2_hwaccel_deps="dxva2 DXVA_PicParams_VP9"

 vp9_dxva2_hwaccel_select="vp9_decoder"

 vp9_mediacodec_hwaccel_deps="mediacodec"

+vp9_nvdec_hwaccel_deps="cuda nvdec"

+vp9_nvdec_hwaccel_select="vp9_decoder"

 vp9_vaapi_hwaccel_deps="vaapi VADecPictureParameterBufferVP9_bit_depth"

 vp9_vaapi_hwaccel_select="vp9_decoder"

 wmv3_d3d11va_hwaccel_select="vc1_d3d11va_hwaccel"

@@ -869,6 +869,7 @@ OBJS-$(CONFIG_VC1_VAAPI_HWACCEL)          += vaapi_vc1.o

 OBJS-$(CONFIG_VC1_VDPAU_HWACCEL)          += vdpau_vc1.o

 OBJS-$(CONFIG_VP9_D3D11VA_HWACCEL)        += dxva2_vp9.o

 OBJS-$(CONFIG_VP9_DXVA2_HWACCEL)          += dxva2_vp9.o

+OBJS-$(CONFIG_VP9_NVDEC_HWACCEL)          += nvdec_vp9.o

 OBJS-$(CONFIG_VP9_VAAPI_HWACCEL)          += vaapi_vp9.o

 OBJS-$(CONFIG_VP8_QSV_HWACCEL)            += qsvdec_other.o

@@ -123,6 +123,7 @@ static void register_all(void)

     REGISTER_HWACCEL(VP9_D3D11VA2,      vp9_d3d11va2);

     REGISTER_HWACCEL(VP9_DXVA2,         vp9_dxva2);

     REGISTER_HWACCEL(VP9_MEDIACODEC,    vp9_mediacodec);

+    REGISTER_HWACCEL(VP9_NVDEC,         vp9_nvdec);

     REGISTER_HWACCEL(VP9_VAAPI,         vp9_vaapi);

     REGISTER_HWACCEL(WMV3_D3D11VA,      wmv3_d3d11va);

     REGISTER_HWACCEL(WMV3_D3D11VA2,     wmv3_d3d11va2);

@@ -54,6 +54,7 @@ static int map_avcodec_id(enum AVCodecID id)

     switch (id) {

     case AV_CODEC_ID_H264: return cudaVideoCodec_H264;

     case AV_CODEC_ID_HEVC: return cudaVideoCodec_HEVC;

+    case AV_CODEC_ID_VP9:  return cudaVideoCodec_VP9;

     }

     return -1;

 }

new file mode 100644

@@ -0,0 +1,227 @@

+/*

+ * VP9 HW decode acceleration through NVDEC

+ *

+ * Copyright (c) 2016 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 "libavutil/pixdesc.h"

+

+#include "avcodec.h"

+#include "nvdec.h"

+#include "decode.h"

+#include "internal.h"

+#include "vp9shared.h"

+

+static unsigned char get_ref_idx(AVFrame *frame)

+{

+    FrameDecodeData *fdd;

+    NVDECFrame *cf;

+

+    if (!frame || !frame->private_ref)

+        return 255;

+

+    fdd = (FrameDecodeData*)frame->private_ref->data;

+    cf  = (NVDECFrame*)fdd->hwaccel_priv;

+

+    return cf->idx;

+}

+

+static int nvdec_vp9_start_frame(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)

+{

+    VP9SharedContext *h = avctx->priv_data;

+    const AVPixFmtDescriptor *pixdesc = av_pix_fmt_desc_get(avctx->sw_pix_fmt);

+

+    NVDECContext      *ctx = avctx->internal->hwaccel_priv_data;

+    CUVIDPICPARAMS     *pp = &ctx->pic_params;

+    CUVIDVP9PICPARAMS *ppc = &pp->CodecSpecific.vp9;

+    FrameDecodeData *fdd;

+    NVDECFrame *cf;

+    AVFrame *cur_frame = h->frames[CUR_FRAME].tf.f;

+

+    int ret, i;

+

+    ret = ff_nvdec_start_frame(avctx, cur_frame);

+    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,

+

+        .CodecSpecific.vp9 = {

+            .width                    = cur_frame->width,

+            .height                   = cur_frame->height,

+

+            .LastRefIdx               = get_ref_idx(h->refs[h->h.refidx[0]].f),

+            .GoldenRefIdx             = get_ref_idx(h->refs[h->h.refidx[1]].f),

+            .AltRefIdx                = get_ref_idx(h->refs[h->h.refidx[2]].f),

+

+            .profile                  = h->h.profile,

+            .frameContextIdx          = h->h.framectxid,

+            .frameType                = !h->h.keyframe,

+            .showFrame                = !h->h.invisible,

+            .errorResilient           = h->h.errorres,

+            .frameParallelDecoding    = h->h.parallelmode,

+            .subSamplingX             = pixdesc->log2_chroma_w,

+            .subSamplingY             = pixdesc->log2_chroma_h,

+            .intraOnly                = h->h.intraonly,

+            .allow_high_precision_mv  = h->h.keyframe ? 0 : h->h.highprecisionmvs,

+            .refreshEntropyProbs      = h->h.refreshctx,

+

+            .bitDepthMinus8Luma       = pixdesc->comp[0].depth - 8,

+            .bitDepthMinus8Chroma     = pixdesc->comp[1].depth - 8,

+

+            .loopFilterLevel          = h->h.filter.level,

+            .loopFilterSharpness      = h->h.filter.sharpness,

+            .modeRefLfEnabled         = h->h.lf_delta.enabled,

+

+            .log2_tile_columns        = h->h.tiling.log2_tile_cols,

+            .log2_tile_rows           = h->h.tiling.log2_tile_rows,

+

+            .segmentEnabled           = h->h.segmentation.enabled,

+            .segmentMapUpdate         = h->h.segmentation.update_map,

+            .segmentMapTemporalUpdate = h->h.segmentation.temporal,

+            .segmentFeatureMode       = h->h.segmentation.absolute_vals,

+

+            .qpYAc                    = h->h.yac_qi,

+            .qpYDc                    = h->h.ydc_qdelta,

+            .qpChDc                   = h->h.uvdc_qdelta,

+            .qpChAc                   = h->h.uvac_qdelta,

+

+            .resetFrameContext        = h->h.resetctx,

+            .mcomp_filter_type        = h->h.filtermode ^ (h->h.filtermode <= 1),

+

+            .frameTagSize             = h->h.uncompressed_header_size,

+            .offsetToDctParts         = h->h.compressed_header_size,

+

+            .refFrameSignBias[0]      = 0,

+        }

+    };

+

+    for (i = 0; i < 2; i++)

+        ppc->mbModeLfDelta[i] = h->h.lf_delta.mode[i];

+

+    for (i = 0; i < 4; i++)

+        ppc->mbRefLfDelta[i] = h->h.lf_delta.ref[i];

+

+    for (i = 0; i < 7; i++)

+        ppc->mb_segment_tree_probs[i] = h->h.segmentation.prob[i];

+

+    for (i = 0; i < 3; i++) {

+        ppc->activeRefIdx[i] = h->h.refidx[i];

+        ppc->segment_pred_probs[i] = h->h.segmentation.pred_prob[i];

+        ppc->refFrameSignBias[i + 1] = h->h.signbias[i];

+    }

+

+    for (i = 0; i < 8; i++) {

+        ppc->segmentFeatureEnable[i][0] = h->h.segmentation.feat[i].q_enabled;

+        ppc->segmentFeatureEnable[i][1] = h->h.segmentation.feat[i].lf_enabled;

+        ppc->segmentFeatureEnable[i][2] = h->h.segmentation.feat[i].ref_enabled;

+        ppc->segmentFeatureEnable[i][3] = h->h.segmentation.feat[i].skip_enabled;

+

+        ppc->segmentFeatureData[i][0] = h->h.segmentation.feat[i].q_val;

+        ppc->segmentFeatureData[i][1] = h->h.segmentation.feat[i].lf_val;

+        ppc->segmentFeatureData[i][2] = h->h.segmentation.feat[i].ref_val;

+        ppc->segmentFeatureData[i][3] = 0;

+    }

+

+    switch (avctx->colorspace) {

+    default:

+    case AVCOL_SPC_UNSPECIFIED:

+        ppc->colorSpace = 0;

+        break;

+    case AVCOL_SPC_BT470BG:

+        ppc->colorSpace = 1;

+        break;

+    case AVCOL_SPC_BT709:

+        ppc->colorSpace = 2;

+        break;

+    case AVCOL_SPC_SMPTE170M:

+        ppc->colorSpace = 3;

+        break;

+    case AVCOL_SPC_SMPTE240M:

+        ppc->colorSpace = 4;

+        break;

+    case AVCOL_SPC_BT2020_NCL:

+        ppc->colorSpace = 5;

+        break;

+    case AVCOL_SPC_RESERVED:

+        ppc->colorSpace = 6;

+        break;

+    case AVCOL_SPC_RGB:

+        ppc->colorSpace = 7;

+        break;

+    }

+

+    return 0;

+}

+

+static int nvdec_vp9_end_frame(AVCodecContext *avctx)

+{

+    NVDECContext *ctx = avctx->internal->hwaccel_priv_data;

+    int ret = ff_nvdec_end_frame(avctx);

+    ctx->bitstream = NULL;

+    return ret;

+}

+

+static int nvdec_vp9_decode_slice(AVCodecContext *avctx, const uint8_t *buffer, uint32_t size)

+{

+    NVDECContext *ctx = avctx->internal->hwaccel_priv_data;

+    void *tmp;

+

+    tmp = av_fast_realloc(ctx->slice_offsets, &ctx->slice_offsets_allocated,

+                          (ctx->nb_slices + 1) * sizeof(*ctx->slice_offsets));

+    if (!tmp)

+        return AVERROR(ENOMEM);

+    ctx->slice_offsets = tmp;

+

+    if (!ctx->bitstream)

+        ctx->bitstream = (uint8_t*)buffer;

+

+    ctx->slice_offsets[ctx->nb_slices] = buffer - ctx->bitstream;

+    ctx->bitstream_len += size;

+    ctx->nb_slices++;

+

+    return 0;

+}

+

+static int nvdec_vp9_frame_params(AVCodecContext *avctx,

+                                  AVBufferRef *hw_frames_ctx)

+{

+    // VP9 uses a fixed size pool of 8 possible reference frames

+    return ff_nvdec_frame_params(avctx, hw_frames_ctx, 8);

+}

+

+AVHWAccel ff_vp9_nvdec_hwaccel = {

+    .name                 = "vp9_nvdec",

+    .type                 = AVMEDIA_TYPE_VIDEO,

+    .id                   = AV_CODEC_ID_VP9,

+    .pix_fmt              = AV_PIX_FMT_CUDA,

+    .start_frame          = nvdec_vp9_start_frame,

+    .end_frame            = nvdec_vp9_end_frame,

+    .decode_slice         = nvdec_vp9_decode_slice,

+    .frame_params         = nvdec_vp9_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   3

-#define LIBAVCODEC_VERSION_MICRO 100

+#define LIBAVCODEC_VERSION_MICRO 101

 #define LIBAVCODEC_VERSION_INT  AV_VERSION_INT(LIBAVCODEC_VERSION_MAJOR, \

                                                LIBAVCODEC_VERSION_MINOR, \

@@ -169,7 +169,10 @@ fail:

 static int update_size(AVCodecContext *avctx, int w, int h)

 {

-#define HWACCEL_MAX (CONFIG_VP9_DXVA2_HWACCEL + CONFIG_VP9_D3D11VA_HWACCEL * 2 + CONFIG_VP9_VAAPI_HWACCEL)

+#define HWACCEL_MAX (CONFIG_VP9_DXVA2_HWACCEL + \

+                     CONFIG_VP9_D3D11VA_HWACCEL * 2 + \

+                     CONFIG_VP9_NVDEC_HWACCEL + \

+                     CONFIG_VP9_VAAPI_HWACCEL)

     enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmtp = pix_fmts;

     VP9Context *s = avctx->priv_data;

     uint8_t *p;

@@ -191,12 +194,18 @@ static int update_size(AVCodecContext *avctx, int w, int h)

             *fmtp++ = AV_PIX_FMT_D3D11VA_VLD;

             *fmtp++ = AV_PIX_FMT_D3D11;

 #endif

+#if CONFIG_VP9_NVDEC_HWACCEL

+            *fmtp++ = AV_PIX_FMT_CUDA;

+#endif

 #if CONFIG_VP9_VAAPI_HWACCEL

             *fmtp++ = AV_PIX_FMT_VAAPI;

 #endif

             break;

         case AV_PIX_FMT_YUV420P10:

         case AV_PIX_FMT_YUV420P12:

+#if CONFIG_VP9_NVDEC_HWACCEL

+            *fmtp++ = AV_PIX_FMT_CUDA;

+#endif

 #if CONFIG_VP9_VAAPI_HWACCEL

             *fmtp++ = AV_PIX_FMT_VAAPI;

 #endif