hevcdec: add a CUVID hwaccel
authorAnton Khirnov <anton@khirnov.net>
Sat, 11 Feb 2017 15:49:34 +0000 (16:49 +0100)
committerAnton Khirnov <anton@khirnov.net>
Fri, 28 Jul 2017 12:12:56 +0000 (14:12 +0200)
Changelog
configure
libavcodec/Makefile
libavcodec/allcodecs.c
libavcodec/cuvid.c
libavcodec/cuvid_hevc.c [new file with mode: 0644]
libavcodec/hevcdec.c

index 7115066..92a7249 100644 (file)
--- a/Changelog
+++ b/Changelog
@@ -18,7 +18,7 @@ version <next>:
 - support for decoding through D3D11VA in avconv
 - Cinepak encoder
 - Intel QSV-accelerated MJPEG encoding
-- NVIDIA CUVID-accelerated H.264 decoding
+- NVIDIA CUVID-accelerated H.264 and HEVC decoding
 
 
 version 12:
index 7a7fbb2..4510100 100755 (executable)
--- a/configure
+++ b/configure
@@ -2210,6 +2210,8 @@ h264_vda_old_hwaccel_deps="vda"
 h264_vda_old_hwaccel_select="h264_decoder"
 h264_vdpau_hwaccel_deps="vdpau"
 h264_vdpau_hwaccel_select="h264_decoder"
+hevc_cuvid_hwaccel_deps="cuvid CUVIDHEVCPICPARAMS"
+hevc_cuvid_hwaccel_select="hevc_decoder"
 hevc_d3d11va_hwaccel_deps="d3d11va DXVA_PicParams_HEVC"
 hevc_d3d11va_hwaccel_select="hevc_decoder"
 hevc_d3d11va2_hwaccel_deps="d3d11va DXVA_PicParams_HEVC"
@@ -4700,6 +4702,7 @@ check_lib psapi    "windows.h psapi.h"    GetProcessMemoryInfo -lpsapi
 check_struct "sys/time.h sys/resource.h" "struct rusage" ru_maxrss
 
 check_type "cuviddec.h" "CUVIDH264PICPARAMS"
+check_type "cuviddec.h" "CUVIDHEVCPICPARAMS"
 check_struct "cuviddec.h" "CUVIDDECODECREATEINFO" bitDepthMinus8
 
 check_type "windows.h dxva.h" "DXVA_PicParams_HEVC" -DWINAPI_FAMILY=WINAPI_FAMILY_DESKTOP_APP -D_CRT_BUILD_DESKTOP_APP=0
index eba651f..bb568dd 100644 (file)
@@ -641,6 +641,7 @@ OBJS-$(CONFIG_H264_QSV_HWACCEL)           += qsvdec_h2645.o
 OBJS-$(CONFIG_H264_VAAPI_HWACCEL)         += vaapi_h264.o
 OBJS-$(CONFIG_H264_VDA_HWACCEL)           += vda_h264.o
 OBJS-$(CONFIG_H264_VDPAU_HWACCEL)         += vdpau_h264.o
+OBJS-$(CONFIG_HEVC_CUVID_HWACCEL)         += cuvid_hevc.o
 OBJS-$(CONFIG_HEVC_D3D11VA_HWACCEL)       += dxva2_hevc.o
 OBJS-$(CONFIG_HEVC_DXVA2_HWACCEL)         += dxva2_hevc.o
 OBJS-$(CONFIG_HEVC_QSV_HWACCEL)           += qsvdec_h2645.o
index ea3a6b9..4ece430 100644 (file)
@@ -78,6 +78,7 @@ void avcodec_register_all(void)
     REGISTER_HWACCEL(H264_VDA,          h264_vda);
     REGISTER_HWACCEL(H264_VDA_OLD,      h264_vda_old);
     REGISTER_HWACCEL(H264_VDPAU,        h264_vdpau);
+    REGISTER_HWACCEL(HEVC_CUVID,        hevc_cuvid);
     REGISTER_HWACCEL(HEVC_D3D11VA,      hevc_d3d11va);
     REGISTER_HWACCEL(HEVC_D3D11VA2,     hevc_d3d11va2);
     REGISTER_HWACCEL(HEVC_DXVA2,        hevc_dxva2);
index 69f624c..2d35e92 100644 (file)
@@ -53,6 +53,7 @@ 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;
     }
     return -1;
 }
diff --git a/libavcodec/cuvid_hevc.c b/libavcodec/cuvid_hevc.c
new file mode 100644 (file)
index 0000000..5de9bca
--- /dev/null
@@ -0,0 +1,280 @@
+/*
+ * HEVC HW decode acceleration through CUVID
+ *
+ * Copyright (c) 2017 Anton Khirnov
+ *
+ * This file is part of Libav.
+ *
+ * Libav 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.
+ *
+ * Libav 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 Libav; if not, write to the Free Software Foundation,
+ * Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
+ */
+
+#include <cuviddec.h>
+#include <stdint.h>
+#include <string.h>
+
+#include "avcodec.h"
+#include "cuvid.h"
+#include "decode.h"
+#include "internal.h"
+#include "hevcdec.h"
+#include "hevc_data.h"
+
+static void dpb_add(CUVIDHEVCPICPARAMS *pp, int idx, const HEVCFrame *src)
+{
+    FrameDecodeData *fdd = (FrameDecodeData*)src->frame->opaque_ref->data;
+    const CUVIDFrame *cf = fdd->hwaccel_priv;
+
+    pp->RefPicIdx[idx]      = cf ? cf->idx : -1;
+    pp->PicOrderCntVal[idx] = src->poc;
+    pp->IsLongTerm[idx]     = !!(src->flags & HEVC_FRAME_FLAG_LONG_REF);
+}
+
+static void fill_scaling_lists(CUVIDHEVCPICPARAMS *ppc, const HEVCContext *s)
+{
+    const ScalingList *sl = s->ps.pps->scaling_list_data_present_flag ?
+                            &s->ps.pps->scaling_list : &s->ps.sps->scaling_list;
+    int i, j, pos;
+
+    for (i = 0; i < 6; i++) {
+        for (j = 0; j < 16; j++) {
+            pos = 4 * ff_hevc_diag_scan4x4_y[j] + ff_hevc_diag_scan4x4_x[j];
+            ppc->ScalingList4x4[i][j] = sl->sl[0][i][pos];
+        }
+
+        for (j = 0; j < 64; j++) {
+            pos = 8 * ff_hevc_diag_scan8x8_y[j] + ff_hevc_diag_scan8x8_x[j];
+            ppc->ScalingList8x8[i][j]   = sl->sl[1][i][pos];
+            ppc->ScalingList16x16[i][j] = sl->sl[2][i][pos];
+
+            if (i < 2)
+                ppc->ScalingList32x32[i][j] = sl->sl[3][i][pos];
+        }
+    }
+
+    memcpy(ppc->ScalingListDCCoeff16x16, sl->sl_dc[0], sizeof(ppc->ScalingListDCCoeff16x16));
+    memcpy(ppc->ScalingListDCCoeff32x32, sl->sl_dc[1], sizeof(ppc->ScalingListDCCoeff32x32));
+}
+
+static int cuvid_hevc_start_frame(AVCodecContext *avctx,
+                                  const uint8_t *buffer, uint32_t size)
+{
+    const HEVCContext *s = avctx->priv_data;
+    const HEVCPPS *pps = s->ps.pps;
+    const HEVCSPS *sps = s->ps.sps;
+
+    CUVIDContext       *ctx = avctx->internal->hwaccel_priv_data;
+    CUVIDPICPARAMS      *pp = &ctx->pic_params;
+    CUVIDHEVCPICPARAMS *ppc = &pp->CodecSpecific.hevc;
+    FrameDecodeData *fdd;
+    CUVIDFrame *cf;
+
+    int i, j, dpb_size, ret;
+
+    ret = ff_cuvid_start_frame(avctx, s->ref->frame);
+    if (ret < 0)
+        return ret;
+
+    fdd = (FrameDecodeData*)s->ref->frame->opaque_ref->data;
+    cf  = (CUVIDFrame*)fdd->hwaccel_priv;
+
+    *pp = (CUVIDPICPARAMS) {
+        .PicWidthInMbs     = sps->width  / 16,
+        .FrameHeightInMbs  = sps->height / 16,
+        .CurrPicIdx        = cf->idx,
+        .ref_pic_flag      = 1,
+        .intra_pic_flag    = 0,
+
+        .CodecSpecific.hevc = {
+            .pic_width_in_luma_samples                    = sps->width,
+            .pic_height_in_luma_samples                   = sps->height,
+            .log2_min_luma_coding_block_size_minus3       = sps->log2_min_cb_size - 3,
+            .log2_diff_max_min_luma_coding_block_size     = sps->log2_diff_max_min_coding_block_size,
+            .log2_min_transform_block_size_minus2         = sps->log2_min_tb_size - 2,
+            .log2_diff_max_min_transform_block_size       = sps->log2_max_trafo_size - sps->log2_min_tb_size,
+            .pcm_enabled_flag                             = sps->pcm_enabled_flag,
+            .log2_min_pcm_luma_coding_block_size_minus3   = sps->pcm_enabled_flag ? sps->pcm.log2_min_pcm_cb_size - 3 : 0,
+            .log2_diff_max_min_pcm_luma_coding_block_size = sps->pcm.log2_max_pcm_cb_size - sps->pcm.log2_min_pcm_cb_size,
+            .pcm_sample_bit_depth_luma_minus1             = sps->pcm_enabled_flag ? sps->pcm.bit_depth - 1 : 0,
+            .pcm_sample_bit_depth_chroma_minus1           = sps->pcm_enabled_flag ? sps->pcm.bit_depth_chroma - 1 : 0,
+            .pcm_loop_filter_disabled_flag                = sps->pcm.loop_filter_disable_flag,
+            .strong_intra_smoothing_enabled_flag          = sps->sps_strong_intra_smoothing_enable_flag,
+            .max_transform_hierarchy_depth_intra          = sps->max_transform_hierarchy_depth_intra,
+            .max_transform_hierarchy_depth_inter          = sps->max_transform_hierarchy_depth_inter,
+            .amp_enabled_flag                             = sps->amp_enabled_flag,
+            .separate_colour_plane_flag                   = sps->separate_colour_plane_flag,
+            .log2_max_pic_order_cnt_lsb_minus4            = sps->log2_max_poc_lsb - 4,
+            .num_short_term_ref_pic_sets                  = sps->nb_st_rps,
+            .long_term_ref_pics_present_flag              = sps->long_term_ref_pics_present_flag,
+            .num_long_term_ref_pics_sps                   = sps->num_long_term_ref_pics_sps,
+            .sps_temporal_mvp_enabled_flag                = sps->sps_temporal_mvp_enabled_flag,
+            .sample_adaptive_offset_enabled_flag          = sps->sao_enabled,
+            .scaling_list_enable_flag                     = sps->scaling_list_enable_flag,
+            .IrapPicFlag                                  = IS_IRAP(s),
+            .IdrPicFlag                                   = IS_IDR(s),
+            .bit_depth_luma_minus8                        = sps->bit_depth - 8,
+            .bit_depth_chroma_minus8                      = sps->bit_depth - 8,
+
+            .dependent_slice_segments_enabled_flag        = pps->dependent_slice_segments_enabled_flag,
+            .slice_segment_header_extension_present_flag  = pps->slice_header_extension_present_flag,
+            .sign_data_hiding_enabled_flag                = pps->sign_data_hiding_flag,
+            .cu_qp_delta_enabled_flag                     = pps->cu_qp_delta_enabled_flag,
+            .diff_cu_qp_delta_depth                       = pps->diff_cu_qp_delta_depth,
+            .init_qp_minus26                              = pps->pic_init_qp_minus26,
+            .pps_cb_qp_offset                             = pps->cb_qp_offset,
+            .pps_cr_qp_offset                             = pps->cr_qp_offset,
+            .constrained_intra_pred_flag                  = pps->constrained_intra_pred_flag,
+            .weighted_pred_flag                           = pps->weighted_pred_flag,
+            .weighted_bipred_flag                         = pps->weighted_bipred_flag,
+            .transform_skip_enabled_flag                  = pps->transform_skip_enabled_flag,
+            .transquant_bypass_enabled_flag               = pps->transquant_bypass_enable_flag,
+            .entropy_coding_sync_enabled_flag             = pps->entropy_coding_sync_enabled_flag,
+            .log2_parallel_merge_level_minus2             = pps->log2_parallel_merge_level - 2,
+            .num_extra_slice_header_bits                  = pps->num_extra_slice_header_bits,
+            .loop_filter_across_tiles_enabled_flag        = pps->loop_filter_across_tiles_enabled_flag,
+            .loop_filter_across_slices_enabled_flag       = pps->seq_loop_filter_across_slices_enabled_flag,
+            .output_flag_present_flag                     = pps->output_flag_present_flag,
+            .num_ref_idx_l0_default_active_minus1         = pps->num_ref_idx_l0_default_active - 1,
+            .num_ref_idx_l1_default_active_minus1         = pps->num_ref_idx_l1_default_active - 1,
+            .lists_modification_present_flag              = pps->lists_modification_present_flag,
+            .cabac_init_present_flag                      = pps->cabac_init_present_flag,
+            .pps_slice_chroma_qp_offsets_present_flag     = pps->pic_slice_level_chroma_qp_offsets_present_flag,
+            .deblocking_filter_override_enabled_flag      = pps->deblocking_filter_override_enabled_flag,
+            .pps_deblocking_filter_disabled_flag          = pps->disable_dbf,
+            .pps_beta_offset_div2                         = pps->beta_offset / 2,
+            .pps_tc_offset_div2                           = pps->tc_offset / 2,
+            .tiles_enabled_flag                           = pps->tiles_enabled_flag,
+            .uniform_spacing_flag                         = pps->uniform_spacing_flag,
+            .num_tile_columns_minus1                      = pps->num_tile_columns - 1,
+            .num_tile_rows_minus1                         = pps->num_tile_rows - 1,
+
+            .NumBitsForShortTermRPSInSlice                = s->sh.short_term_rps ? s->sh.short_term_ref_pic_set_size : 0,
+            .NumDeltaPocsOfRefRpsIdx                      = s->sh.short_term_rps ? s->sh.short_term_rps->rps_idx_num_delta_pocs : 0,
+            .NumPocTotalCurr                              = s->rps[ST_CURR_BEF].nb_refs + s->rps[ST_CURR_AFT].nb_refs +
+                                                            s->rps[LT_CURR].nb_refs,
+            .NumPocStCurrBefore                           = s->rps[ST_CURR_BEF].nb_refs,
+            .NumPocStCurrAfter                            = s->rps[ST_CURR_AFT].nb_refs,
+            .NumPocLtCurr                                 = s->rps[LT_CURR].nb_refs,
+            .CurrPicOrderCntVal                           = s->ref->poc,
+        },
+    };
+
+    if (pps->num_tile_columns > FF_ARRAY_ELEMS(ppc->column_width_minus1) ||
+        pps->num_tile_rows    > FF_ARRAY_ELEMS(ppc->row_height_minus1)) {
+        av_log(avctx, AV_LOG_ERROR, "Too many tiles\n");
+        return AVERROR(ENOSYS);
+    }
+    for (i = 0; i < pps->num_tile_columns; i++)
+        ppc->column_width_minus1[i] = pps->column_width[i] - 1;
+    for (i = 0; i < pps->num_tile_rows; i++)
+        ppc->row_height_minus1[i] = pps->row_height[i] - 1;
+
+    if (s->rps[LT_CURR].nb_refs     > FF_ARRAY_ELEMS(ppc->RefPicSetLtCurr)       ||
+        s->rps[ST_CURR_BEF].nb_refs > FF_ARRAY_ELEMS(ppc->RefPicSetStCurrBefore) ||
+        s->rps[ST_CURR_AFT].nb_refs > FF_ARRAY_ELEMS(ppc->RefPicSetStCurrAfter)) {
+        av_log(avctx, AV_LOG_ERROR, "Too many reference frames\n");
+        return AVERROR(ENOSYS);
+    }
+
+    dpb_size = 0;
+    for (i = 0; i < FF_ARRAY_ELEMS(s->DPB); i++) {
+        const HEVCFrame *ref = &s->DPB[i];
+        if (!(ref->flags & (HEVC_FRAME_FLAG_SHORT_REF | HEVC_FRAME_FLAG_LONG_REF)))
+            continue;
+        if (dpb_size >= FF_ARRAY_ELEMS(ppc->RefPicIdx)) {
+            av_log(avctx, AV_LOG_ERROR, "Too many reference frames\n");
+            return AVERROR_INVALIDDATA;
+        }
+        dpb_add(ppc, dpb_size++, ref);
+
+    }
+    for (i = dpb_size; i < FF_ARRAY_ELEMS(ppc->RefPicIdx); i++)
+        ppc->RefPicIdx[i] = -1;
+
+    for (i = 0; i < s->rps[ST_CURR_BEF].nb_refs; i++) {
+        for (j = 0; j < dpb_size; j++) {
+            if (ppc->PicOrderCntVal[j] == s->rps[ST_CURR_BEF].list[i]) {
+                ppc->RefPicSetStCurrBefore[i] = j;
+                break;
+            }
+        }
+    }
+    for (i = 0; i < s->rps[ST_CURR_AFT].nb_refs; i++) {
+        for (j = 0; j < dpb_size; j++) {
+            if (ppc->PicOrderCntVal[j] == s->rps[ST_CURR_AFT].list[i]) {
+                ppc->RefPicSetStCurrAfter[i] = j;
+                break;
+            }
+        }
+    }
+    for (i = 0; i < s->rps[LT_CURR].nb_refs; i++) {
+        for (j = 0; j < dpb_size; j++) {
+            if (ppc->PicOrderCntVal[j] == s->rps[LT_CURR].list[i]) {
+                ppc->RefPicSetLtCurr[i] = j;
+                break;
+            }
+        }
+    }
+
+    fill_scaling_lists(ppc, s);
+
+    return 0;
+}
+
+static int cuvid_hevc_decode_slice(AVCodecContext *avctx, const uint8_t *buffer,
+                                   uint32_t size)
+{
+    CUVIDContext *ctx = avctx->internal->hwaccel_priv_data;
+    void *tmp;
+
+    tmp = av_fast_realloc(ctx->bitstream, &ctx->bitstream_allocated,
+                          ctx->bitstream_len + size + 3);
+    if (!tmp)
+        return AVERROR(ENOMEM);
+    ctx->bitstream = 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;
+
+    AV_WB24(ctx->bitstream + ctx->bitstream_len, 1);
+    memcpy(ctx->bitstream + ctx->bitstream_len + 3, buffer, size);
+    ctx->slice_offsets[ctx->nb_slices] = ctx->bitstream_len ;
+    ctx->bitstream_len += size + 3;
+    ctx->nb_slices++;
+
+    return 0;
+}
+
+static int cuvid_hevc_decode_init(AVCodecContext *avctx)
+{
+    const HEVCContext *s = avctx->priv_data;
+    const HEVCSPS *sps = s->ps.sps;
+    return ff_cuvid_decode_init(avctx, sps->temporal_layer[sps->max_sub_layers - 1].max_dec_pic_buffering + 1);
+}
+
+AVHWAccel ff_hevc_cuvid_hwaccel = {
+    .name                 = "hevc_cuvid",
+    .type                 = AVMEDIA_TYPE_VIDEO,
+    .id                   = AV_CODEC_ID_HEVC,
+    .pix_fmt              = AV_PIX_FMT_CUDA,
+    .start_frame          = cuvid_hevc_start_frame,
+    .end_frame            = ff_cuvid_end_frame,
+    .decode_slice         = cuvid_hevc_decode_slice,
+    .init                 = cuvid_hevc_decode_init,
+    .uninit               = ff_cuvid_decode_uninit,
+    .priv_data_size       = sizeof(CUVIDContext),
+};
index 664e4ac..a1619cf 100644 (file)
@@ -384,7 +384,8 @@ static void export_stream_params(AVCodecContext *avctx, const HEVCParamSets *ps,
 static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps)
 {
     #define HWACCEL_MAX (CONFIG_HEVC_DXVA2_HWACCEL + CONFIG_HEVC_D3D11VA_HWACCEL * 2 + \
-                         CONFIG_HEVC_VAAPI_HWACCEL + CONFIG_HEVC_VDPAU_HWACCEL)
+                         CONFIG_HEVC_VAAPI_HWACCEL + CONFIG_HEVC_VDPAU_HWACCEL + \
+                         CONFIG_HEVC_CUVID_HWACCEL)
     enum AVPixelFormat pix_fmts[HWACCEL_MAX + 2], *fmt = pix_fmts;
 
     if (sps->pix_fmt == AV_PIX_FMT_YUV420P || sps->pix_fmt == AV_PIX_FMT_YUVJ420P ||
@@ -399,8 +400,14 @@ static enum AVPixelFormat get_format(HEVCContext *s, const HEVCSPS *sps)
 #if CONFIG_HEVC_VAAPI_HWACCEL
         *fmt++ = AV_PIX_FMT_VAAPI;
 #endif
+#if CONFIG_HEVC_CUVID_HWACCEL && HAVE_CUVIDDECODECREATEINFO_BITDEPTHMINUS8
+        *fmt++ = AV_PIX_FMT_CUDA;
+#endif
     }
     if (sps->pix_fmt == AV_PIX_FMT_YUV420P || sps->pix_fmt == AV_PIX_FMT_YUVJ420P) {
+#if CONFIG_HEVC_CUVID_HWACCEL && !HAVE_CUVIDDECODECREATEINFO_BITDEPTHMINUS8
+        *fmt++ = AV_PIX_FMT_CUDA;
+#endif
 #if CONFIG_HEVC_VDPAU_HWACCEL
         *fmt++ = AV_PIX_FMT_VDPAU;
 #endif