From: "Zhao Zhili" <quinkblack@foxmail.com>
To: "'FFmpeg development discussions and patches'" <ffmpeg-devel@ffmpeg.org>
Cc: "'jianfeng.zheng'" <jianfeng.zheng@mthreads.com>
Subject: Re: [FFmpeg-devel] [PATCH v1 1/2] vaapi: add vaapi_cavs support
Date: Sat, 20 Jan 2024 12:20:57 +0800
Message-ID: <tencent_7A2D279AA5401C9AEA9F16ABBE4E7570E708@qq.com> (raw)
In-Reply-To: <20240119154950.444144-1-jianfeng.zheng@mthreads.com>
> -----Original Message-----
> From: ffmpeg-devel <ffmpeg-devel-bounces@ffmpeg.org> On Behalf Of jianfeng.zheng
> Sent: 2024年1月19日 23:50
> To: ffmpeg-devel@ffmpeg.org
> Cc: jianfeng.zheng <jianfeng.zheng@mthreads.com>
> Subject: [FFmpeg-devel] [PATCH v1 1/2] vaapi: add vaapi_cavs support
>
> see https://github.com/intel/libva/pull/738
>
> [Moore Threads](https://www.mthreads.com) (short for Mthreads) is a
> Chinese GPU manufacturer. All our products, like MTTS70/MTTS80/.. ,
> support AVS/AVS+ HW decoding at max 2k resolution.
Please use description more objective and neutrality.
>
> Signed-off-by: jianfeng.zheng <jianfeng.zheng@mthreads.com>
> ---
> configure | 14 ++
> libavcodec/Makefile | 1 +
> libavcodec/cavs.c | 12 +
> libavcodec/cavs.h | 36 ++-
> libavcodec/cavs_parser.c | 16 ++
> libavcodec/cavsdec.c | 473 +++++++++++++++++++++++++++++++++-----
> libavcodec/defs.h | 3 +
> libavcodec/hwaccels.h | 1 +
> libavcodec/profiles.c | 6 +
> libavcodec/profiles.h | 1 +
> libavcodec/vaapi_cavs.c | 164 +++++++++++++
> libavcodec/vaapi_decode.c | 4 +
> 12 files changed, 669 insertions(+), 62 deletions(-)
> create mode 100644 libavcodec/vaapi_cavs.c
>
> diff --git a/configure b/configure
> index c8ae0a061d..89759eda5d 100755
> --- a/configure
> +++ b/configure
> @@ -2463,6 +2463,7 @@ HAVE_LIST="
> xmllint
> zlib_gzip
> openvino2
> + va_profile_avs
> "
>
> # options emitted with CONFIG_ prefix but not available on the command line
> @@ -3202,6 +3203,7 @@ wmv3_dxva2_hwaccel_select="vc1_dxva2_hwaccel"
> wmv3_nvdec_hwaccel_select="vc1_nvdec_hwaccel"
> wmv3_vaapi_hwaccel_select="vc1_vaapi_hwaccel"
> wmv3_vdpau_hwaccel_select="vc1_vdpau_hwaccel"
> +cavs_vaapi_hwaccel_deps="vaapi va_profile_avs VAPictureParameterBufferAVS"
>
> # hardware-accelerated codecs
> mediafoundation_deps="mftransform_h MFCreateAlignedMemoryBuffer"
> @@ -7175,6 +7177,18 @@ if enabled vaapi; then
> check_type "va/va.h va/va_enc_vp8.h" "VAEncPictureParameterBufferVP8"
> check_type "va/va.h va/va_enc_vp9.h" "VAEncPictureParameterBufferVP9"
> check_type "va/va.h va/va_enc_av1.h" "VAEncPictureParameterBufferAV1"
> +
> + #
> + # Using 'VA_CHECK_VERSION' in source codes make things easy. But we have to wait
> + # until newly added VAProfile being distributed by VAAPI released version.
> + #
> + # Before or after that, we can use auto-detection to keep version compatibility.
> + # It always works.
> + #
> + disable va_profile_avs &&
> + test_code cc va/va.h "VAProfile p1 = VAProfileAVSJizhun, p2 = VAProfileAVSGuangdian;" &&
> + enable va_profile_avs
> + enabled va_profile_avs && check_type "va/va.h va/va_dec_avs.h" "VAPictureParameterBufferAVS"
> fi
>
> if enabled_all opencl libdrm ; then
> diff --git a/libavcodec/Makefile b/libavcodec/Makefile
> index bb42095165..7d92375fed 100644
> --- a/libavcodec/Makefile
> +++ b/libavcodec/Makefile
> @@ -1055,6 +1055,7 @@ OBJS-$(CONFIG_VP9_VAAPI_HWACCEL) += vaapi_vp9.o
> OBJS-$(CONFIG_VP9_VDPAU_HWACCEL) += vdpau_vp9.o
> OBJS-$(CONFIG_VP9_VIDEOTOOLBOX_HWACCEL) += videotoolbox_vp9.o
> OBJS-$(CONFIG_VP8_QSV_HWACCEL) += qsvdec.o
> +OBJS-$(CONFIG_CAVS_VAAPI_HWACCEL) += vaapi_cavs.o
>
> # Objects duplicated from other libraries for shared builds
> SHLIBOBJS += log2_tab.o reverse.o
> diff --git a/libavcodec/cavs.c b/libavcodec/cavs.c
> index fdd577f7fb..ed7b278336 100644
> --- a/libavcodec/cavs.c
> +++ b/libavcodec/cavs.c
Please split the patch.
> @@ -810,6 +810,14 @@ av_cold int ff_cavs_init(AVCodecContext *avctx)
> if (!h->cur.f || !h->DPB[0].f || !h->DPB[1].f)
> return AVERROR(ENOMEM);
>
> + h->out[0].f = av_frame_alloc();
> + h->out[1].f = av_frame_alloc();
> + h->out[2].f = av_frame_alloc();
> + if (!h->out[0].f || !h->out[1].f || !h->out[2].f) {
> + ff_cavs_end(avctx);
> + return AVERROR(ENOMEM);
> + }
> +
> h->luma_scan[0] = 0;
> h->luma_scan[1] = 8;
> h->intra_pred_l[INTRA_L_VERT] = intra_pred_vert;
> @@ -840,6 +848,10 @@ av_cold int ff_cavs_end(AVCodecContext *avctx)
> av_frame_free(&h->DPB[0].f);
> av_frame_free(&h->DPB[1].f);
>
> + av_frame_free(&h->out[0].f);
> + av_frame_free(&h->out[1].f);
> + av_frame_free(&h->out[2].f);
> +
> av_freep(&h->top_qp);
> av_freep(&h->top_mv[0]);
> av_freep(&h->top_mv[1]);
> diff --git a/libavcodec/cavs.h b/libavcodec/cavs.h
> index 244c322b35..ef03c1a974 100644
> --- a/libavcodec/cavs.h
> +++ b/libavcodec/cavs.h
> @@ -39,8 +39,10 @@
> #define EXT_START_CODE 0x000001b5
> #define USER_START_CODE 0x000001b2
> #define CAVS_START_CODE 0x000001b0
> +#define VIDEO_SEQ_END_CODE 0x000001b1
> #define PIC_I_START_CODE 0x000001b3
> #define PIC_PB_START_CODE 0x000001b6
> +#define VIDEO_EDIT_CODE 0x000001b7
>
> #define A_AVAIL 1
> #define B_AVAIL 2
> @@ -164,10 +166,15 @@ struct dec_2dvlc {
> typedef struct AVSFrame {
> AVFrame *f;
> int poc;
> + int outputed;
> +
> + AVBufferRef *hwaccel_priv_buf;
> + void *hwaccel_picture_private;
> } AVSFrame;
>
> typedef struct AVSContext {
> AVCodecContext *avctx;
> + int got_pix_fmt;
> BlockDSPContext bdsp;
> H264ChromaContext h264chroma;
> VideoDSPContext vdsp;
> @@ -175,6 +182,7 @@ typedef struct AVSContext {
> GetBitContext gb;
> AVSFrame cur; ///< currently decoded frame
> AVSFrame DPB[2]; ///< reference frames
> + AVSFrame out[3]; ///< output queue, size 2 maybe enough
> int dist[2]; ///< temporal distances from current frame to ref frames
> int low_delay;
> int profile, level;
> @@ -182,12 +190,38 @@ typedef struct AVSContext {
> int mb_width, mb_height;
> int width, height;
> int stream_revision; ///<0 for samples from 2006, 1 for rm52j encoder
> - int progressive;
> + int progressive_seq;
> + int progressive_frame;
> int pic_structure;
> + int no_forward_ref_flag;
> + int pb_field_enhanced_flag; ///< only used in GUANGDIAN
> int skip_mode_flag; ///< select between skip_count or one skip_flag per MB
> int loop_filter_disable;
> int alpha_offset, beta_offset;
> int ref_flag;
> +
> + /** \defgroup guangdian profile
> + * @{
> + */
> + int aec_flag;
> + int weight_quant_flag;
> + int chroma_quant_param_delta_cb;
> + int chroma_quant_param_delta_cr;
> + uint8_t wqm_8x8[64];
> + /**@}*/
> +
> + /** \defgroup slice weighting
> + * FFmpeg don't support slice weighting natively, but maybe needed for HWaccel.
> + * @{
> + */
> + uint32_t slice_weight_pred_flag : 1;
> + uint32_t mb_weight_pred_flag : 1;
> + uint8_t luma_scale[4];
> + int8_t luma_shift[4];
> + uint8_t chroma_scale[4];
> + int8_t chroma_shift[4];
> + /**@}*/
> +
> int mbx, mby, mbidx; ///< macroblock coordinates
> int flags; ///< availability flags of neighbouring macroblocks
> int stc; ///< last start code
> diff --git a/libavcodec/cavs_parser.c b/libavcodec/cavs_parser.c
> index 4a03effd0f..a41a82d8d1 100644
> --- a/libavcodec/cavs_parser.c
> +++ b/libavcodec/cavs_parser.c
> @@ -65,6 +65,22 @@ static int cavs_find_frame_end(ParseContext *pc, const uint8_t *buf,
> pc->state=-1;
> return i-3;
> }
> + if((state&0xFFFFFF00) == 0x100){
> + if(state != EXT_START_CODE && state != USER_START_CODE){
> + state = state >> 8;
> + break;
> + }
> + }
> + }
> + for(; i<buf_size; i++){
> + state= (state<<8) | buf[i];
> + if((state&0xFFFFFF00) == 0x100){
> + if(state > SLICE_MAX_START_CODE){
> + pc->frame_start_found=0;
> + pc->state=-1;
> + return i-3;
> + }
> + }
Why? A parser should split data into frames, not slices.
> }
> }
> pc->frame_start_found= pic_found;
> diff --git a/libavcodec/cavsdec.c b/libavcodec/cavsdec.c
> index b356da0b04..18c38cd3ff 100644
> --- a/libavcodec/cavsdec.c
> +++ b/libavcodec/cavsdec.c
> @@ -25,11 +25,15 @@
> * @author Stefan Gehrer <stefan.gehrer@gmx.de>
> */
>
> +#include "config_components.h"
> #include "libavutil/avassert.h"
> #include "libavutil/emms.h"
> #include "avcodec.h"
> #include "get_bits.h"
> #include "golomb.h"
> +#include "hwaccel_internal.h"
> +#include "hwconfig.h"
> +#include "profiles.h"
> #include "cavs.h"
> #include "codec_internal.h"
> #include "decode.h"
> @@ -37,6 +41,43 @@
> #include "mpeg12data.h"
> #include "startcode.h"
>
> +static const uint8_t default_wq_param[4][6] = {
> + {128, 98, 106, 116, 116, 128},
> + {135, 143, 143, 160, 160, 213},
> + {128, 98, 106, 116, 116, 128},
> + {128, 128, 128, 128, 128, 128},
> +};
> +static const uint8_t wq_model_2_param[4][64] = {
> + {
> + 0, 0, 0, 4, 4, 4, 5, 5,
> + 0, 0, 3, 3, 3, 3, 5, 5,
> + 0, 3, 2, 2, 1, 1, 5, 5,
> + 4, 3, 2, 2, 1, 5, 5, 5,
> + 4, 3, 1, 1, 5, 5, 5, 5,
> + 4, 3, 1, 5, 5, 5, 5, 5,
> + 5, 5, 5, 5, 5, 5, 5, 5,
> + 5, 5, 5, 5, 5, 5, 5, 5,
> + }, {
> + 0, 0, 0, 4, 4, 4, 5, 5,
> + 0, 0, 4, 4, 4, 4, 5, 5,
> + 0, 3, 2, 2, 2, 1, 5, 5,
> + 3, 3, 2, 2, 1, 5, 5, 5,
> + 3, 3, 2, 1, 5, 5, 5, 5,
> + 3, 3, 1, 5, 5, 5, 5, 5,
> + 5, 5, 5, 5, 5, 5, 5, 5,
> + 5, 5, 5, 5, 5, 5, 5, 5,
> + }, {
> + 0, 0, 0, 4, 4, 3, 5, 5,
> + 0, 0, 4, 4, 3, 2, 5, 5,
> + 0, 4, 4, 3, 2, 1, 5, 5,
> + 4, 4, 3, 2, 1, 5, 5, 5,
> + 4, 3, 2, 1, 5, 5, 5, 5,
> + 3, 2, 1, 5, 5, 5, 5, 5,
> + 5, 5, 5, 5, 5, 5, 5, 5,
> + 5, 5, 5, 5, 5, 5, 5, 5,
> + }
> +};
> +
> static const uint8_t mv_scan[4] = {
> MV_FWD_X0, MV_FWD_X1,
> MV_FWD_X2, MV_FWD_X3
> @@ -927,7 +968,11 @@ static int decode_mb_b(AVSContext *h, enum cavs_mb mb_type)
>
> static inline int decode_slice_header(AVSContext *h, GetBitContext *gb)
> {
> - if (h->stc > 0xAF)
> + int i, nref;
> +
> + av_log(h->avctx, AV_LOG_TRACE, "slice start code 0x%02x\n", h->stc);
> +
> + if (h->stc > SLICE_MAX_START_CODE)
> av_log(h->avctx, AV_LOG_ERROR, "unexpected start code 0x%02x\n", h->stc);
>
> if (h->stc >= h->mb_height) {
> @@ -946,14 +991,119 @@ static inline int decode_slice_header(AVSContext *h, GetBitContext *gb)
> }
> /* inter frame or second slice can have weighting params */
> if ((h->cur.f->pict_type != AV_PICTURE_TYPE_I) ||
> - (!h->pic_structure && h->mby >= h->mb_width / 2))
> - if (get_bits1(gb)) { //slice_weighting_flag
> - av_log(h->avctx, AV_LOG_ERROR,
> - "weighted prediction not yet supported\n");
> + (!h->pic_structure && h->mby >= h->mb_height / 2)) {
> + h->slice_weight_pred_flag = get_bits1(gb);
> + if (h->slice_weight_pred_flag) {
> + nref = h->cur.f->pict_type == AV_PICTURE_TYPE_I ? 1 : (h->pic_structure ? 2 : 4);
> + for (i = 0; i < nref; i++) {
> + h->luma_scale[i] = get_bits(gb, 8);
> + h->luma_shift[i] = get_sbits(gb, 8);
> + skip_bits1(gb);
> + h->chroma_scale[i] = get_bits(gb, 8);
> + h->chroma_shift[i] = get_sbits(gb, 8);
> + skip_bits1(gb);
> + }
> + h->mb_weight_pred_flag = get_bits1(gb);
> + if (!h->avctx->hwaccel) {
> + av_log(h->avctx, AV_LOG_ERROR,
> + "weighted prediction not yet supported\n");
> + }
> }
> + }
> + if (h->aec_flag) {
> + align_get_bits(gb);
> + }
> +
> + return 0;
> +}
> +
> +/**
> + * skip stuffing bits before next start code "0x000001"
> + * @return '0' no stuffing bits placed at h->gb being skip, else '1'.
> + */
> +static inline int skip_stuffing_bits(AVSContext *h)
> +{
> + GetBitContext gb0 = h->gb;
> + GetBitContext *gb = &h->gb;
> + const uint8_t *start;
> + const uint8_t *ptr;
> + const uint8_t *end;
> + int align;
> + int stuffing_zeros;
> +
> +#if 0
> + /**
> + * skip 1 bit stuffing_bit '1' and 0~7 bit stuffing_bit '0'
> + */
> + if (!get_bits1(gb)) {
> + av_log(h->avctx, AV_LOG_WARNING, "NOT stuffing_bit '1'\n");
> + goto restore_get_bits;
> + }
> + align = (-get_bits_count(gb)) & 7;
> + if (show_bits_long(gb, align)) {
> + av_log(h->avctx, AV_LOG_WARNING, "NOT %d stuffing_bit '0..0'\n", align);
> + goto restore_get_bits;
> + }
> +#else
> + /**
> + * Seems like not all the stream follow "next_start_code()" strictly.
> + */
> + align = (-get_bits_count(gb)) & 7;
> + if (align == 0 && show_bits_long(gb, 8) == 0x80) {
> + skip_bits_long(gb, 8);
> + }
> +#endif
> +
> + /**
> + * skip leading zero bytes before 0x 00 00 01 stc
> + */
> + ptr = start = align_get_bits(gb);
> + end = gb->buffer_end;
> + while (ptr < end && *ptr == 0)
> + ptr++;
> +
> + if ((ptr >= end) || (*ptr == 1 && ptr - start >= 2)) {
> + stuffing_zeros = (ptr >= end ? end - start : ptr - start - 2);
> + if (stuffing_zeros > 0)
> + av_log(h->avctx, AV_LOG_DEBUG, "Skip 0x%x stuffing zeros @0x%x.\n",
> + stuffing_zeros, (int)(start - gb->buffer));
> + skip_bits_long(gb, stuffing_zeros * 8);
> + return 1;
> + } else {
> + av_log(h->avctx, AV_LOG_DEBUG, "No next_start_code() found @0x%x.\n",
> + (int)(start - gb->buffer));
> + goto restore_get_bits;
> + }
> +
> +restore_get_bits:
> + h->gb = gb0;
> return 0;
> }
>
> +static inline int skip_extension_and_user_data(AVSContext *h)
> +{
> + int stc = -1;
> + const uint8_t *start = align_get_bits(&h->gb);
> + const uint8_t *end = h->gb.buffer_end;
> + const uint8_t *ptr, *next;
> +
> + for (ptr = start; ptr + 4 < end; ptr = next) {
> + stc = show_bits_long(&h->gb, 32);
> + if (stc != EXT_START_CODE && stc != USER_START_CODE) {
> + break;
> + }
> + next = avpriv_find_start_code(ptr + 4, end, &stc);
> + if (next < end) {
> + next -= 4;
> + }
> + skip_bits(&h->gb, (next - ptr) * 8);
> + av_log(h->avctx, AV_LOG_DEBUG, "skip %d byte ext/user data\n",
> + (int)(next - ptr));
> + }
> +
> + return ptr > start;
> +}
> +
> static inline int check_for_slice(AVSContext *h)
> {
> GetBitContext *gb = &h->gb;
> @@ -981,44 +1131,133 @@ static inline int check_for_slice(AVSContext *h)
> * frame level
> *
> ****************************************************************************/
> +static int hwaccel_pic(AVSContext *h)
> +{
> + int ret = 0;
> + int stc = -1;
> + const uint8_t *frm_start = align_get_bits(&h->gb);
> + const uint8_t *frm_end = h->gb.buffer_end;
> + const uint8_t *slc_start = frm_start;
> + const uint8_t *slc_end = frm_end;
> + GetBitContext gb = h->gb;
> + const FFHWAccel *hwaccel = ffhwaccel(h->avctx->hwaccel);
> +
> + ret = hwaccel->start_frame(h->avctx, NULL, 0);
> + if (ret < 0)
> + return ret;
> +
> + for (slc_start = frm_start; slc_start + 4 < frm_end; slc_start = slc_end) {
> + slc_end = avpriv_find_start_code(slc_start + 4, frm_end, &stc);
> + if (slc_end < frm_end) {
> + slc_end -= 4;
> + }
> +
> + init_get_bits(&h->gb, slc_start, (slc_end - slc_start) * 8);
> + if (!check_for_slice(h)) {
> + break;
> + }
> +
> + ret = hwaccel->decode_slice(h->avctx, slc_start, slc_end - slc_start);
> + if (ret < 0) {
> + break;
> + }
> + }
> +
> + h->gb = gb;
> + skip_bits(&h->gb, (slc_start - frm_start) * 8);
> +
> + if (ret < 0)
> + return ret;
> +
> + return hwaccel->end_frame(h->avctx);
> +}
> +
> +/**
> + * @brief remove frame out of dpb
> + */
> +static void cavs_frame_unref(AVSFrame *frame)
> +{
> + /* frame->f can be NULL if context init failed */
> + if (!frame->f || !frame->f->buf[0])
> + return;
> +
> + av_buffer_unref(&frame->hwaccel_priv_buf);
> + frame->hwaccel_picture_private = NULL;
> +
> + av_frame_unref(frame->f);
> +}
> +
> +static int output_one_frame(AVSContext *h, AVFrame *data, int *got_frame)
> +{
> + if (h->out[0].f->buf[0]) {
> + av_log(h->avctx, AV_LOG_DEBUG, "output frame: poc=%d\n", h->out[0].poc);
> + av_frame_move_ref(data, h->out[0].f);
> + *got_frame = 1;
> +
> + // out[0] <- out[1] <- out[2] <- out[0]
> + cavs_frame_unref(&h->out[2]);
> + FFSWAP(AVSFrame, h->out[0], h->out[2]);
> + FFSWAP(AVSFrame, h->out[0], h->out[1]);
> +
> + return 1;
> + }
> +
> + return 0;
> +}
> +
> +static void queue_one_frame(AVSContext *h, AVSFrame *out)
> +{
> + int idx = !h->out[0].f->buf[0] ? 0 : (!h->out[1].f->buf[0] ? 1 : 2);
> + av_log(h->avctx, AV_LOG_DEBUG, "queue in out[%d]: poc=%d\n", idx, out->poc);
> + av_frame_ref(h->out[idx].f, out->f);
> + h->out[idx].poc = out->poc;
> +}
>
> static int decode_pic(AVSContext *h)
> {
> int ret;
> int skip_count = -1;
> enum cavs_mb mb_type;
> + char tc[4];
>
> if (!h->top_qp) {
> av_log(h->avctx, AV_LOG_ERROR, "No sequence header decoded yet\n");
> return AVERROR_INVALIDDATA;
> }
>
> - av_frame_unref(h->cur.f);
> + cavs_frame_unref(&h->cur);
> +
> + skip_bits(&h->gb, 16);//bbv_delay
> + if (h->profile == AV_PROFILE_CAVS_GUANGDIAN) {
> + skip_bits(&h->gb, 8);//bbv_dwlay_extension
> + }
>
> - skip_bits(&h->gb, 16);//bbv_dwlay
> if (h->stc == PIC_PB_START_CODE) {
> h->cur.f->pict_type = get_bits(&h->gb, 2) + AV_PICTURE_TYPE_I;
> if (h->cur.f->pict_type > AV_PICTURE_TYPE_B) {
> av_log(h->avctx, AV_LOG_ERROR, "illegal picture type\n");
> return AVERROR_INVALIDDATA;
> }
> +
> /* make sure we have the reference frames we need */
> - if (!h->DPB[0].f->data[0] ||
> - (!h->DPB[1].f->data[0] && h->cur.f->pict_type == AV_PICTURE_TYPE_B))
> + if (!h->DPB[0].f->buf[0] ||
> + (!h->DPB[1].f->buf[0] && h->cur.f->pict_type == AV_PICTURE_TYPE_B)) {
> + av_log(h->avctx, AV_LOG_ERROR, "Invalid reference frame\n");
> return AVERROR_INVALIDDATA;
> + }
> } else {
> h->cur.f->pict_type = AV_PICTURE_TYPE_I;
> - if (get_bits1(&h->gb))
> - skip_bits(&h->gb, 24);//time_code
> - /* old sample clips were all progressive and no low_delay,
> - bump stream revision if detected otherwise */
> - if (h->low_delay || !(show_bits(&h->gb, 9) & 1))
> - h->stream_revision = 1;
> - /* similarly test top_field_first and repeat_first_field */
> - else if (show_bits(&h->gb, 11) & 3)
> - h->stream_revision = 1;
> - if (h->stream_revision > 0)
> - skip_bits(&h->gb, 1); //marker_bit
> + if (get_bits1(&h->gb)) { //time_code
> + skip_bits(&h->gb, 1);
> + tc[0] = get_bits(&h->gb, 5);
> + tc[1] = get_bits(&h->gb, 6);
> + tc[2] = get_bits(&h->gb, 6);
> + tc[3] = get_bits(&h->gb, 6);
> + av_log(h->avctx, AV_LOG_DEBUG, "timecode: %d:%d:%d.%d\n",
> + tc[0], tc[1], tc[2], tc[3]);
> + }
> +
> + skip_bits(&h->gb, 1);
> }
>
> if (get_bits_left(&h->gb) < 23)
> @@ -1029,6 +1268,17 @@ static int decode_pic(AVSContext *h)
> if (ret < 0)
> return ret;
>
> + if (h->avctx->hwaccel) {
> + const FFHWAccel *hwaccel = ffhwaccel(h->avctx->hwaccel);
> + av_assert0(!h->cur.hwaccel_picture_private);
> + if (hwaccel->frame_priv_data_size) {
> + h->cur.hwaccel_priv_buf = av_buffer_allocz(hwaccel->frame_priv_data_size);
> + if (!h->cur.hwaccel_priv_buf)
> + return AVERROR(ENOMEM);
> + h->cur.hwaccel_picture_private = h->cur.hwaccel_priv_buf->data;
> + }
> + }
> +
> if (!h->edge_emu_buffer) {
> int alloc_size = FFALIGN(FFABS(h->cur.f->linesize[0]) + 32, 32);
> h->edge_emu_buffer = av_mallocz(alloc_size * 2 * 24);
> @@ -1039,6 +1289,8 @@ static int decode_pic(AVSContext *h)
> if ((ret = ff_cavs_init_pic(h)) < 0)
> return ret;
> h->cur.poc = get_bits(&h->gb, 8) * 2;
> + av_log(h->avctx, AV_LOG_DEBUG, "poc=%d, type=%d\n",
> + h->cur.poc, h->cur.f->pict_type);
>
> /* get temporal distances and MV scaling factors */
> if (h->cur.f->pict_type != AV_PICTURE_TYPE_B) {
> @@ -1052,8 +1304,12 @@ static int decode_pic(AVSContext *h)
> if (h->cur.f->pict_type == AV_PICTURE_TYPE_B) {
> h->sym_factor = h->dist[0] * h->scale_den[1];
> if (FFABS(h->sym_factor) > 32768) {
> + av_log(h->avctx, AV_LOG_ERROR, "poc=%d/%d/%d, dist=%d/%d\n",
> + h->DPB[1].poc, h->DPB[0].poc, h->cur.poc, h->dist[0], h->dist[1]);
> av_log(h->avctx, AV_LOG_ERROR, "sym_factor %d too large\n", h->sym_factor);
> - return AVERROR_INVALIDDATA;
> +
> + if (!h->avctx->hwaccel)
> + return AVERROR_INVALIDDATA;
> }
> } else {
> h->direct_den[0] = h->dist[0] ? 16384 / h->dist[0] : 0;
> @@ -1062,9 +1318,9 @@ static int decode_pic(AVSContext *h)
>
> if (h->low_delay)
> get_ue_golomb(&h->gb); //bbv_check_times
> - h->progressive = get_bits1(&h->gb);
> + h->progressive_frame = get_bits1(&h->gb);
> h->pic_structure = 1;
> - if (!h->progressive)
> + if (!h->progressive_frame)
> h->pic_structure = get_bits1(&h->gb);
> if (!h->pic_structure && h->stc == PIC_PB_START_CODE)
> skip_bits1(&h->gb); //advanced_pred_mode_disable
> @@ -1073,14 +1329,18 @@ static int decode_pic(AVSContext *h)
> h->pic_qp_fixed =
> h->qp_fixed = get_bits1(&h->gb);
> h->qp = get_bits(&h->gb, 6);
> + h->skip_mode_flag = 0;
> + h->ref_flag = 0;
> if (h->cur.f->pict_type == AV_PICTURE_TYPE_I) {
> - if (!h->progressive && !h->pic_structure)
> - skip_bits1(&h->gb);//what is this?
> + if (!h->progressive_frame && !h->pic_structure)
> + h->skip_mode_flag = get_bits1(&h->gb);
> skip_bits(&h->gb, 4); //reserved bits
> } else {
> if (!(h->cur.f->pict_type == AV_PICTURE_TYPE_B && h->pic_structure == 1))
> h->ref_flag = get_bits1(&h->gb);
> - skip_bits(&h->gb, 4); //reserved bits
> + h->no_forward_ref_flag = get_bits1(&h->gb);
> + h->pb_field_enhanced_flag = get_bits1(&h->gb);
> + skip_bits(&h->gb, 2); //reserved bits
> h->skip_mode_flag = get_bits1(&h->gb);
> }
> h->loop_filter_disable = get_bits1(&h->gb);
> @@ -1096,8 +1356,46 @@ static int decode_pic(AVSContext *h)
> h->alpha_offset = h->beta_offset = 0;
> }
>
> + h->weight_quant_flag = 0;
> + if (h->profile == AV_PROFILE_CAVS_GUANGDIAN) {
> + h->weight_quant_flag = get_bits1(&h->gb);
> + if (h->weight_quant_flag) {
> + int wq_param[6] = {128, 128, 128, 128, 128, 128};
> + int i, wqp_index, wq_model;
> + const uint8_t *m2p;
> +
> + skip_bits1(&h->gb);
> + if (!get_bits1(&h->gb)) {
> + h->chroma_quant_param_delta_cb = get_se_golomb(&h->gb);
> + h->chroma_quant_param_delta_cr = get_se_golomb(&h->gb);
> + }
> + wqp_index = get_bits(&h->gb, 2);
> + wq_model = get_bits(&h->gb, 2);
> + m2p = wq_model_2_param[wq_model];
> +
> + for (i = 0; i < 6; i++) {
> + int delta = (wqp_index == 1 || wqp_index == 2) ? get_se_golomb(&h->gb) : 0;
> + wq_param[i] = default_wq_param[wqp_index][i] + delta;
> + av_log(h->avctx, AV_LOG_DEBUG, "wqp[%d]=%d\n", i, wq_param[i]);
> + }
> + for (i = 0; i < 64; i++) {
> + h->wqm_8x8[i] = wq_param[ m2p[i] ];
> + }
> + } else {
> + memset(h->wqm_8x8, 128, sizeof(h->wqm_8x8));
> + }
> + h->aec_flag = get_bits1(&h->gb);
> + av_log(h->avctx, AV_LOG_DEBUG, "wq_flag=%d, aec_flag=%d\n",
> + h->weight_quant_flag, h->aec_flag);
> + }
> +
> + skip_stuffing_bits(h);
> + skip_extension_and_user_data(h);
> +
> ret = 0;
> - if (h->cur.f->pict_type == AV_PICTURE_TYPE_I) {
> + if (h->avctx->hwaccel) {
> + ret = hwaccel_pic(h);
> + } else if (h->cur.f->pict_type == AV_PICTURE_TYPE_I) {
> do {
> check_for_slice(h);
> ret = decode_mb_i(h, 0);
> @@ -1160,11 +1458,6 @@ static int decode_pic(AVSContext *h)
> } while (ff_cavs_next_mb(h));
> }
> emms_c();
> - if (ret >= 0 && h->cur.f->pict_type != AV_PICTURE_TYPE_B) {
> - av_frame_unref(h->DPB[1].f);
> - FFSWAP(AVSFrame, h->cur, h->DPB[1]);
> - FFSWAP(AVSFrame, h->DPB[0], h->DPB[1]);
> - }
> return ret;
> }
>
> @@ -1181,13 +1474,8 @@ static int decode_seq_header(AVSContext *h)
> int ret;
>
> h->profile = get_bits(&h->gb, 8);
> - if (h->profile != 0x20) {
> - avpriv_report_missing_feature(h->avctx,
> - "only supprt JiZhun profile");
> - return AVERROR_PATCHWELCOME;
> - }
> h->level = get_bits(&h->gb, 8);
> - skip_bits1(&h->gb); //progressive sequence
> + h->progressive_seq = get_bits1(&h->gb);
>
> width = get_bits(&h->gb, 14);
> height = get_bits(&h->gb, 14);
> @@ -1214,6 +1502,9 @@ static int decode_seq_header(AVSContext *h)
> skip_bits1(&h->gb); //marker_bit
> skip_bits(&h->gb, 12); //bit_rate_upper
> h->low_delay = get_bits1(&h->gb);
> + av_log(h->avctx, AV_LOG_DEBUG,
> + "seq: profile=0x%02x, level=0x%02x, size=%dx%d, low_delay=%d\n",
> + h->profile, h->level, width, height, h->low_delay);
>
> ret = ff_set_dimensions(h->avctx, width, height);
> if (ret < 0)
> @@ -1239,43 +1530,61 @@ static int cavs_decode_frame(AVCodecContext *avctx, AVFrame *rframe,
> int *got_frame, AVPacket *avpkt)
> {
> AVSContext *h = avctx->priv_data;
> - const uint8_t *buf = avpkt->data;
> - int buf_size = avpkt->size;
> uint32_t stc = -1;
> int input_size, ret;
> const uint8_t *buf_end;
> const uint8_t *buf_ptr;
> int frame_start = 0;
>
> - if (buf_size == 0) {
> - if (!h->low_delay && h->DPB[0].f->data[0]) {
> - *got_frame = 1;
> - av_frame_move_ref(rframe, h->DPB[0].f);
> + if (avpkt->size == 0) {
> + if (h->DPB[0].f->buf[0] && !h->DPB[0].outputed) {
> + queue_one_frame(h, &h->DPB[0]);
> + cavs_frame_unref(&h->DPB[0]);
> }
> + output_one_frame(h, rframe, got_frame);
> return 0;
> }
>
> h->stc = 0;
>
> - buf_ptr = buf;
> - buf_end = buf + buf_size;
> - for(;;) {
> + buf_ptr = avpkt->data;
> + buf_end = avpkt->data + avpkt->size;
> + for(; buf_ptr < buf_end;) {
> buf_ptr = avpriv_find_start_code(buf_ptr, buf_end, &stc);
> if ((stc & 0xFFFFFE00) || buf_ptr == buf_end) {
> if (!h->stc)
> av_log(h->avctx, AV_LOG_WARNING, "no frame decoded\n");
> - return FFMAX(0, buf_ptr - buf);
> + return FFMAX(0, buf_ptr - avpkt->data);
> }
> input_size = (buf_end - buf_ptr) * 8;
> + av_log(h->avctx, AV_LOG_TRACE, "Found start code 0x%04x, sz=%d\n",
> + stc, input_size / 8);
> switch (stc) {
> case CAVS_START_CODE:
> init_get_bits(&h->gb, buf_ptr, input_size);
> - decode_seq_header(h);
> + if ((ret = decode_seq_header(h)) < 0)
> + return ret;
> + avctx->profile = h->profile;
> + avctx->level = h->level;
> + if (!h->got_pix_fmt) {
> + h->got_pix_fmt = 1;
> + ret = ff_get_format(avctx, avctx->codec->pix_fmts);
> + if (ret < 0)
> + return ret;
> +
> + avctx->pix_fmt = ret;
> +
> + if (h->profile == AV_PROFILE_CAVS_GUANGDIAN && !avctx->hwaccel) {
> + av_log(avctx, AV_LOG_ERROR, "Your platform doesn't suppport hardware"
> + " accelerated for CAVS Guangdian Profile decoding.\n");
> + return AVERROR(ENOTSUP);
> + }
> + }
> break;
> case PIC_I_START_CODE:
> if (!h->got_keyframe) {
> - av_frame_unref(h->DPB[0].f);
> - av_frame_unref(h->DPB[1].f);
> + cavs_frame_unref(&h->DPB[0]);
> + cavs_frame_unref(&h->DPB[1]);
> h->got_keyframe = 1;
> }
> case PIC_PB_START_CODE:
> @@ -1285,23 +1594,39 @@ static int cavs_decode_frame(AVCodecContext *avctx, AVFrame *rframe,
> if (*got_frame)
> av_frame_unref(rframe);
> *got_frame = 0;
> - if (!h->got_keyframe)
> + if (!h->got_keyframe) {
> + av_log(avctx, AV_LOG_ERROR, "No keyframe decoded before P/B frame.\n");
> break;
> + }
> init_get_bits(&h->gb, buf_ptr, input_size);
> h->stc = stc;
> - if (decode_pic(h))
> - break;
> - *got_frame = 1;
> + if ((ret = decode_pic(h)) < 0)
> + return ret;
> + buf_ptr = align_get_bits(&h->gb);
> +
> + h->cur.outputed = 0;
> if (h->cur.f->pict_type != AV_PICTURE_TYPE_B) {
> - if (h->DPB[!h->low_delay].f->data[0]) {
> - if ((ret = av_frame_ref(rframe, h->DPB[!h->low_delay].f)) < 0)
> - return ret;
> - } else {
> - *got_frame = 0;
> + // at most one delay
> + if (h->DPB[0].f->buf[0] && !h->DPB[0].outputed) {
> + queue_one_frame(h, &h->DPB[0]);
> + h->DPB[0].outputed = 1;
> + }
> +
> + if (h->low_delay) {
> + queue_one_frame(h, &h->cur);
> + h->cur.outputed = 1;
> }
> +
> + // null -> curr -> DPB[0] -> DPB[1]
> + cavs_frame_unref(&h->DPB[1]);
> + FFSWAP(AVSFrame, h->cur, h->DPB[1]);
> + FFSWAP(AVSFrame, h->DPB[0], h->DPB[1]);
> } else {
> - av_frame_move_ref(rframe, h->cur.f);
> + queue_one_frame(h, &h->cur);
> + cavs_frame_unref(&h->cur);
> }
> +
> + output_one_frame(h, rframe, got_frame);
> break;
> case EXT_START_CODE:
> //mpeg_decode_extension(avctx, buf_ptr, input_size);
> @@ -1309,16 +1634,34 @@ static int cavs_decode_frame(AVCodecContext *avctx, AVFrame *rframe,
> case USER_START_CODE:
> //mpeg_decode_user_data(avctx, buf_ptr, input_size);
> break;
> + case VIDEO_EDIT_CODE:
> + av_log(h->avctx, AV_LOG_WARNING, "Skip video_edit_code\n");
> + break;
> + case VIDEO_SEQ_END_CODE:
> + av_log(h->avctx, AV_LOG_WARNING, "Skip video_sequence_end_code\n");
> + break;
> default:
> if (stc <= SLICE_MAX_START_CODE) {
> + h->stc = stc & 0xff;
> init_get_bits(&h->gb, buf_ptr, input_size);
> decode_slice_header(h, &h->gb);
> + } else {
> + av_log(h->avctx, AV_LOG_WARNING, "Skip unsupported start code 0x%04X\n", stc);
> }
> break;
> }
> }
> + return (buf_ptr - avpkt->data);
> }
>
> +static const enum AVPixelFormat cavs_hwaccel_pixfmt_list_420[] = {
> +#if CONFIG_CAVS_VAAPI_HWACCEL
> + AV_PIX_FMT_VAAPI,
> +#endif
> + AV_PIX_FMT_YUV420P,
> + AV_PIX_FMT_NONE
> +};
> +
> const FFCodec ff_cavs_decoder = {
> .p.name = "cavs",
> CODEC_LONG_NAME("Chinese AVS (Audio Video Standard) (AVS1-P2, JiZhun profile)"),
> @@ -1331,4 +1674,12 @@ const FFCodec ff_cavs_decoder = {
> .p.capabilities = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY,
> .flush = cavs_flush,
> .caps_internal = FF_CODEC_CAP_INIT_CLEANUP,
> + .p.pix_fmts = cavs_hwaccel_pixfmt_list_420,
> + .hw_configs = (const AVCodecHWConfigInternal *const []) {
> +#if CONFIG_CAVS_VAAPI_HWACCEL
> + HWACCEL_VAAPI(cavs),
> +#endif
> + NULL
> + },
> + .p.profiles = NULL_IF_CONFIG_SMALL(ff_cavs_profiles),
> };
> diff --git a/libavcodec/defs.h b/libavcodec/defs.h
> index 00d840ec19..d59816a70f 100644
> --- a/libavcodec/defs.h
> +++ b/libavcodec/defs.h
> @@ -192,6 +192,9 @@
> #define AV_PROFILE_EVC_BASELINE 0
> #define AV_PROFILE_EVC_MAIN 1
>
> +#define AV_PROFILE_CAVS_JIZHUN 0x20
> +#define AV_PROFILE_CAVS_GUANGDIAN 0x48
> +
>
> #define AV_LEVEL_UNKNOWN -99
>
> diff --git a/libavcodec/hwaccels.h b/libavcodec/hwaccels.h
> index 5171e4c7d7..a1a973b460 100644
> --- a/libavcodec/hwaccels.h
> +++ b/libavcodec/hwaccels.h
> @@ -89,5 +89,6 @@ extern const struct FFHWAccel ff_wmv3_dxva2_hwaccel;
> extern const struct FFHWAccel ff_wmv3_nvdec_hwaccel;
> extern const struct FFHWAccel ff_wmv3_vaapi_hwaccel;
> extern const struct FFHWAccel ff_wmv3_vdpau_hwaccel;
> +extern const struct FFHWAccel ff_cavs_vaapi_hwaccel;
>
> #endif /* AVCODEC_HWACCELS_H */
> diff --git a/libavcodec/profiles.c b/libavcodec/profiles.c
> index 5bb8f150e6..b312f12281 100644
> --- a/libavcodec/profiles.c
> +++ b/libavcodec/profiles.c
> @@ -200,4 +200,10 @@ const AVProfile ff_evc_profiles[] = {
> { AV_PROFILE_UNKNOWN },
> };
>
> +const AVProfile ff_cavs_profiles[] = {
> + { AV_PROFILE_CAVS_JIZHUN, "Jizhun" },
> + { AV_PROFILE_CAVS_GUANGDIAN, "Guangdian" },
> + { AV_PROFILE_UNKNOWN },
> +};
> +
> #endif /* !CONFIG_SMALL */
> diff --git a/libavcodec/profiles.h b/libavcodec/profiles.h
> index 270430a48b..9a2b348ad4 100644
> --- a/libavcodec/profiles.h
> +++ b/libavcodec/profiles.h
> @@ -75,5 +75,6 @@ extern const AVProfile ff_prores_profiles[];
> extern const AVProfile ff_mjpeg_profiles[];
> extern const AVProfile ff_arib_caption_profiles[];
> extern const AVProfile ff_evc_profiles[];
> +extern const AVProfile ff_cavs_profiles[];
>
> #endif /* AVCODEC_PROFILES_H */
> diff --git a/libavcodec/vaapi_cavs.c b/libavcodec/vaapi_cavs.c
> new file mode 100644
> index 0000000000..4a7a9b95ad
> --- /dev/null
> +++ b/libavcodec/vaapi_cavs.c
> @@ -0,0 +1,164 @@
> +/*
> + * AVS (Chinese GY/T 257.1b
_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".
prev parent reply other threads:[~2024-01-20 4:21 UTC|newest]
Thread overview: 2+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-01-19 15:49 jianfeng.zheng
2024-01-20 4:20 ` Zhao Zhili [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=tencent_7A2D279AA5401C9AEA9F16ABBE4E7570E708@qq.com \
--to=quinkblack@foxmail.com \
--cc=ffmpeg-devel@ffmpeg.org \
--cc=jianfeng.zheng@mthreads.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Git Inbox Mirror of the ffmpeg-devel mailing list - see https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
This inbox may be cloned and mirrored by anyone:
git clone --mirror https://master.gitmailbox.com/ffmpegdev/0 ffmpegdev/git/0.git
# If you have public-inbox 1.1+ installed, you may
# initialize and index your mirror using the following commands:
public-inbox-init -V2 ffmpegdev ffmpegdev/ https://master.gitmailbox.com/ffmpegdev \
ffmpegdev@gitmailbox.com
public-inbox-index ffmpegdev
Example config snippet for mirrors.
AGPL code for this site: git clone https://public-inbox.org/public-inbox.git