diff options
Diffstat (limited to 'src/video_dec/libvdpau/vdpau_h264.c')
| -rw-r--r-- | src/video_dec/libvdpau/vdpau_h264.c | 1014 |
1 files changed, 1014 insertions, 0 deletions
diff --git a/src/video_dec/libvdpau/vdpau_h264.c b/src/video_dec/libvdpau/vdpau_h264.c new file mode 100644 index 000000000..25ed62295 --- /dev/null +++ b/src/video_dec/libvdpau/vdpau_h264.c @@ -0,0 +1,1014 @@ +/* + * Copyright (C) 2008 Julian Scheel + * + * kate: space-indent on; indent-width 2; mixedindent off; indent-mode cstyle; remove-trailing-space on; + * + * This file is part of xine, a free video player. + * + * xine is free software; you can redistribute it and/or modify + * it under the terms of the GNU General Public License as published by + * the Free Software Foundation; either version 2 of the License, or + * (at your option) any later version. + * + * xine 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 General Public License for more details. + * + * You should have received a copy of the GNU General Public License + * along with this program; if not, write to the Free Software + * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110, USA + * + * vdpau_h264.c: H264 Video Decoder utilizing nvidia VDPAU engine + */ + +#define LOG_MODULE "vdpau_h264" + +#ifdef HAVE_CONFIG_H +#include "config.h" +#endif + +#include <stdio.h> +#include <stdlib.h> +#include <string.h> +#include <sys/types.h> +#include <unistd.h> +#include <vdpau/vdpau.h> + +#include <xine/xine_internal.h> +#include <xine/video_out.h> +#include <xine/buffer.h> +#include <xine/xineutils.h> +#include "bswap.h" +#include "accel_vdpau.h" +#include "h264_parser.h" +#include "dpb.h" +#include "cpb.h" + +//#define DEBUG_H264 + +#define VIDEOBUFSIZE 128*1024 + +typedef struct { + video_decoder_class_t decoder_class; +} vdpau_h264_class_t; + +typedef struct vdpau_h264_decoder_s { + video_decoder_t video_decoder; /* parent video decoder structure */ + + vdpau_h264_class_t *class; + xine_stream_t *stream; + + /* these are traditional variables in a video decoder object */ + uint64_t video_step; /* frame duration in pts units */ + uint64_t reported_video_step; /* frame duration in pts units */ + + int width; /* the width of a video frame */ + int height; /* the height of a video frame */ + double ratio; /* the width to height ratio */ + + + struct h264_parser *nal_parser; /* h264 nal parser. extracts stream data for vdpau */ + + struct decoded_picture *incomplete_pic; + uint32_t last_top_field_order_cnt; + + int have_frame_boundary_marks; + int wait_for_frame_start; + + VdpDecoder decoder; + int decoder_started; + int progressive_cnt; /* count of progressive marked frames in line */ + + VdpColorStandard color_standard; + VdpDecoderProfile profile; + vdpau_accel_t *vdpau_accel; + + xine_t *xine; + + struct coded_picture *completed_pic; + vo_frame_t *dangling_img; + + uint8_t *codec_private; + uint32_t codec_private_len; + + int vdp_runtime_nr; + + int reset; + +} vdpau_h264_decoder_t; + +static void vdpau_h264_reset (video_decoder_t *this_gen); +static void vdpau_h264_flush (video_decoder_t *this_gen); + +/************************************************************************** + * vdpau_h264 specific decode functions + *************************************************************************/ + +/************************************************************************** + * xine video plugin functions + *************************************************************************/ + +#ifdef DEBUG_H264 +static inline void dump_pictureinfo_h264(VdpPictureInfoH264 *pic) +{ + printf("C: slice_count: %d\n", pic->slice_count); + printf("C: field_order_cnt[0]: %d\n", pic->field_order_cnt[0]); + printf("C: field_order_cnt[1]: %d\n", pic->field_order_cnt[1]); + printf("C: is_reference: %d\n", pic->is_reference); + printf("C: frame_num: %d\n", pic->frame_num); + printf("C: field_pic_flag: %d\n", pic->field_pic_flag); + printf("C: bottom_field_flag: %d\n", pic->bottom_field_flag); + printf("C: num_ref_frames: %d\n", pic->num_ref_frames); + printf("C: mb_adaptive_frame_field_flag: %d\n", pic->mb_adaptive_frame_field_flag); + printf("C: constrained_intra_pred_flag: %d\n", pic->constrained_intra_pred_flag); + printf("C: weighted_pred_flag: %d\n", pic->weighted_pred_flag); + printf("C: weighted_bipred_idc: %d\n", pic->weighted_bipred_idc); + printf("C: frame_mbs_only_flag: %d\n", pic->frame_mbs_only_flag); + printf("C: transform_8x8_mode_flag: %d\n", pic->transform_8x8_mode_flag); + printf("C: chroma_qp_index_offset: %d\n", pic->chroma_qp_index_offset); + printf("C: second_chroma_qp_index_offset: %d\n", pic->second_chroma_qp_index_offset); + printf("C: pic_init_qp_minus26: %d\n", pic->pic_init_qp_minus26); + printf("C: num_ref_idx_l0_active_minus1: %d\n", pic->num_ref_idx_l0_active_minus1); + printf("C: num_ref_idx_l1_active_minus1: %d\n", pic->num_ref_idx_l1_active_minus1); + printf("C: log2_max_frame_num_minus4: %d\n", pic->log2_max_frame_num_minus4); + printf("C: pic_order_cnt_type: %d\n", pic->pic_order_cnt_type); + printf("C: log2_max_pic_order_cnt_lsb_minus4: %d\n", pic->log2_max_pic_order_cnt_lsb_minus4); + printf("C: delta_pic_order_always_zero_flag: %d\n", pic->delta_pic_order_always_zero_flag); + printf("C: direct_8x8_inference_flag: %d\n", pic->direct_8x8_inference_flag); + printf("C: entropy_coding_mode_flag: %d\n", pic->entropy_coding_mode_flag); + printf("C: pic_order_present_flag: %d\n", pic->pic_order_present_flag); + printf("C: deblocking_filter_control_present_flag: %d\n", pic->deblocking_filter_control_present_flag); + printf("C: redundant_pic_cnt_present_flag: %d\n", pic->redundant_pic_cnt_present_flag); + + int i, j; + for(i = 0; i < 6; i++) { + printf("C: scalint_list4x4[%d]:\nC:", i); + for(j = 0; j < 16; j++) { + printf(" [%d]", pic->scaling_lists_4x4[i][j]); + if(j%8 == 0) + printf("\nC:"); + } + printf("C: \n"); + } + for(i = 0; i < 2; i++) { + printf("C: scalint_list8x8[%d]:\nC:", i); + for(j = 0; j < 64; j++) { + printf(" [%d] ", pic->scaling_lists_8x8[i][j]); + if(j%8 == 0) + printf("\nC:"); + } + printf("C: \n"); + } + + //int i; + for(i = 0; i < 16; i++) { + if(pic->referenceFrames[i].surface != VDP_INVALID_HANDLE) { + printf("C: -------------------\n"); + printf("C: Reference Frame %d:\n", i); + printf("C: frame_idx: %d\n", pic->referenceFrames[i].frame_idx); + printf("C: field_order_cnt[0]: %d\n", pic->referenceFrames[i].field_order_cnt[0]); + printf("C: field_order_cnt[1]: %d\n", pic->referenceFrames[i].field_order_cnt[0]); + printf("C: is_long_term: %d\n", pic->referenceFrames[i].is_long_term); + printf("C: top_is_reference: %d\n", pic->referenceFrames[i].top_is_reference); + printf("C: bottom_is_reference: %d\n", pic->referenceFrames[i].bottom_is_reference); + } + } + printf("C: ---------------------------------------------------------------\n"); + /*memcpy(pic.scaling_lists_4x4, pps->scaling_lists_4x4, 6*16); + memcpy(pic.scaling_lists_8x8, pps->scaling_lists_8x8, 2*64); + memcpy(pic.referenceFrames, this->reference_frames, sizeof(this->reference_frames));*/ + +} +#endif + +static void set_ratio(video_decoder_t *this_gen) +{ + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *)this_gen; + + this->ratio = (double)this->width / (double)this->height; + if(this->completed_pic->sps_nal->sps.vui_parameters.aspect_ration_info_present_flag) { + switch(this->completed_pic->sps_nal->sps.vui_parameters.aspect_ratio_idc) { + case ASPECT_1_1: + this->ratio = 1 * this->ratio; + break; + case ASPECT_12_11: + this->ratio *= 12.0/11.0; + break; + case ASPECT_10_11: + this->ratio *= 10.0/11.0; + break; + case ASPECT_16_11: + this->ratio *= 16.0/11.0; + break; + case ASPECT_40_33: + this->ratio *= 40.0/33.0; + break; + case ASPECT_24_11: + this->ratio *= 24.0/11.0; + break; + case ASPECT_20_11: + this->ratio *= 20.0/11.0; + break; + case ASPECT_32_11: + this->ratio *= 32.0/11.0; + break; + case ASPECT_80_33: + this->ratio *= 80.0/33.0; + break; + case ASPECT_18_11: + this->ratio *= 18.0/11.0; + break; + case ASPECT_15_11: + this->ratio *= 15.0/11.0; + break; + case ASPECT_64_33: + this->ratio *= 64.0/33.0; + break; + case ASPECT_160_99: + this->ratio *= 160.0/99.0; + break; + case ASPECT_4_3: + this->ratio *= 4.0/3.0; + break; + case ASPECT_3_2: + this->ratio *= 3.0/2.0; + break; + case ASPECT_2_1: + this->ratio *= 2.0/1.0; + break; + case ASPECT_EXTENDED_SAR: + this->ratio *= + (double)this->completed_pic->sps_nal->sps.vui_parameters.sar_width/ + (double)this->completed_pic->sps_nal->sps.vui_parameters.sar_height; + break; + } + } +} + +static void fill_vdpau_pictureinfo_h264(video_decoder_t *this_gen, uint32_t slice_count, VdpPictureInfoH264 *pic) +{ + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *)this_gen; + + struct pic_parameter_set_rbsp *pps = &this->completed_pic->pps_nal->pps; + struct seq_parameter_set_rbsp *sps = &this->completed_pic->sps_nal->sps; + struct slice_header *slc = &this->completed_pic->slc_nal->slc; + + pic->slice_count = slice_count; + pic->field_order_cnt[0] = this->completed_pic->top_field_order_cnt; + pic->field_order_cnt[1] = this->completed_pic->bottom_field_order_cnt; + pic->is_reference = + (this->completed_pic->flag_mask & REFERENCE) ? VDP_TRUE : VDP_FALSE; + pic->frame_num = slc->frame_num; + pic->field_pic_flag = slc->field_pic_flag; + pic->bottom_field_flag = slc->bottom_field_flag; + pic->num_ref_frames = sps->num_ref_frames; + pic->mb_adaptive_frame_field_flag = sps->mb_adaptive_frame_field_flag && !slc->field_pic_flag; + pic->constrained_intra_pred_flag = pps->constrained_intra_pred_flag; + pic->weighted_pred_flag = pps->weighted_pred_flag; + pic->weighted_bipred_idc = pps->weighted_bipred_idc; + pic->frame_mbs_only_flag = sps->frame_mbs_only_flag; + pic->transform_8x8_mode_flag = pps->transform_8x8_mode_flag; + pic->chroma_qp_index_offset = pps->chroma_qp_index_offset; + pic->second_chroma_qp_index_offset = pps->second_chroma_qp_index_offset; + pic->pic_init_qp_minus26 = pps->pic_init_qp_minus26; + pic->num_ref_idx_l0_active_minus1 = pps->num_ref_idx_l0_active_minus1; + pic->num_ref_idx_l1_active_minus1 = pps->num_ref_idx_l1_active_minus1; + pic->log2_max_frame_num_minus4 = sps->log2_max_frame_num_minus4; + pic->pic_order_cnt_type = sps->pic_order_cnt_type; + pic->log2_max_pic_order_cnt_lsb_minus4 = sps->log2_max_pic_order_cnt_lsb_minus4; + pic->delta_pic_order_always_zero_flag = sps->delta_pic_order_always_zero_flag; + pic->direct_8x8_inference_flag = sps->direct_8x8_inference_flag; + pic->entropy_coding_mode_flag = pps->entropy_coding_mode_flag; + pic->pic_order_present_flag = pps->pic_order_present_flag; + pic->deblocking_filter_control_present_flag = pps->deblocking_filter_control_present_flag; + pic->redundant_pic_cnt_present_flag = pps->redundant_pic_cnt_present_flag; + + memcpy(pic->scaling_lists_4x4, pps->scaling_lists_4x4, sizeof(pic->scaling_lists_4x4)); + memcpy(pic->scaling_lists_8x8, pps->scaling_lists_8x8, sizeof(pic->scaling_lists_8x8)); + + /* set num_ref_frames to the number of actually available reference frames, + * if this is not set generation 3 decoders will fail. */ + /*pic->num_ref_frames =*/ + fill_vdpau_reference_list(this->nal_parser->dpb, pic->referenceFrames); + +} + +static int check_progressive(video_decoder_t *this_gen, struct decoded_picture *dpic) +{ + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *)this_gen; + int progressive = 0; + int i; + + for(i = 0; i < 2; i++) { + struct coded_picture *pic = dpic->coded_pic[i]; + if (!pic) { + continue; + } + + if (pic->flag_mask & PIC_STRUCT_PRESENT && pic->sei_nal != NULL) { + uint8_t pic_struct = pic->sei_nal->sei.pic_timing.pic_struct; + + if (pic_struct == DISP_FRAME) { + progressive = 1; + continue; + } else if (pic_struct == DISP_TOP_BOTTOM || + pic_struct == DISP_BOTTOM_TOP) { + progressive = 0; + break; + } + + /* FIXME: seems unreliable, maybe it's has to be interpreted more complex */ + /*if (pic->sei_nal->sei.pic_timing.ct_type == CT_INTERLACED) { + return 0; + } else if (pic->sei_nal->sei.pic_timing.ct_type == CT_PROGRESSIVE) { + return 1; + } */ + } + + if (pic->slc_nal->slc.field_pic_flag && pic->pps_nal->pps.pic_order_present_flag) { + if(pic->slc_nal->slc.delta_pic_order_cnt_bottom == 1 || + pic->slc_nal->slc.delta_pic_order_cnt_bottom == -1) { + progressive = 0; + break; + } else { + progressive = 1; + continue; + } + } + if (!pic->slc_nal->slc.field_pic_flag && pic->sps_nal->sps.frame_mbs_only_flag) { + progressive = 1; + continue; + } + } + + if (progressive) { + this->progressive_cnt++; + } else { + this->progressive_cnt = 0; + } + + /* only switch to progressive mode if at least 5 + * frames in order were marked as progressive */ + return (this->progressive_cnt >= 5); +} + +static int vdpau_decoder_init(video_decoder_t *this_gen) +{ + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *)this_gen; + vo_frame_t *img; + + if(this->width == 0) { + this->width = this->completed_pic->sps_nal->sps.pic_width; + this->height = this->completed_pic->sps_nal->sps.pic_height; + } + + set_ratio(this_gen); + + _x_stream_info_set( this->stream, XINE_STREAM_INFO_VIDEO_WIDTH, this->width ); + _x_stream_info_set( this->stream, XINE_STREAM_INFO_VIDEO_HEIGHT, this->height ); + _x_stream_info_set( this->stream, XINE_STREAM_INFO_VIDEO_RATIO, ((double)10000*this->ratio) ); + _x_stream_info_set( this->stream, XINE_STREAM_INFO_FRAME_DURATION, (this->reported_video_step = this->video_step) ); + _x_meta_info_set_utf8( this->stream, XINE_META_INFO_VIDEOCODEC, "H264/AVC (vdpau)" ); + xine_event_t event; + xine_format_change_data_t data; + event.type = XINE_EVENT_FRAME_FORMAT_CHANGE; + event.stream = this->stream; + event.data = &data; + event.data_length = sizeof(data); + data.width = this->width; + data.height = this->height; + data.aspect = this->ratio; + xine_event_send( this->stream, &event ); + + switch(this->completed_pic->sps_nal->sps.profile_idc) { + case 100: + this->profile = VDP_DECODER_PROFILE_H264_HIGH; + break; + case 77: + this->profile = VDP_DECODER_PROFILE_H264_MAIN; + break; + case 66: + default: + // nvidia's VDPAU doesn't support BASELINE. But most (every?) streams marked BASELINE do not use BASELINE specifics, + // so, just force MAIN. + //this->profile = VDP_DECODER_PROFILE_H264_BASELINE; + this->profile = VDP_DECODER_PROFILE_H264_MAIN; + break; + } + + // Level 4.1 limits: + int ref_frames = 0; + if(this->completed_pic->sps_nal->sps.num_ref_frames) { + ref_frames = this->completed_pic->sps_nal->sps.num_ref_frames; + } else { + uint32_t round_width = (this->width + 15) & ~15; + uint32_t round_height = (this->height + 15) & ~15; + uint32_t surf_size = (round_width * round_height * 3) / 2; + ref_frames = (12 * 1024 * 1024) / surf_size; + } + + if (ref_frames > 16) { + ref_frames = 16; + } + + xprintf(this->xine, XINE_VERBOSITY_LOG, "Allocate %d reference frames\n", + ref_frames); + /* get the vdpau context from vo */ + //(this->stream->video_out->open) (this->stream->video_out, this->stream); + img = this->stream->video_out->get_frame (this->stream->video_out, + this->width, this->height, + this->ratio, + XINE_IMGFMT_VDPAU, VO_BOTH_FIELDS | this->reset); + this->reset = 0; + + this->vdpau_accel = (vdpau_accel_t*)img->accel_data; + + img->free(img); + img = NULL; + + /*VdpBool is_supported; + uint32_t max_level, max_references, max_width, max_height;*/ + if(this->vdpau_accel->vdp_runtime_nr > 0) { + xprintf(this->xine, XINE_VERBOSITY_LOG, + "Create decoder: vdp_device: %d, profile: %d, res: %dx%d\n", + this->vdpau_accel->vdp_device, this->profile, this->width, this->height); + + VdpStatus status = this->vdpau_accel->vdp_decoder_create(this->vdpau_accel->vdp_device, + this->profile, this->width, this->height, 16, &this->decoder); + + if(status != VDP_STATUS_OK) { + xprintf(this->xine, XINE_VERBOSITY_LOG, "vdpau_h264: ERROR: VdpDecoderCreate returned status != OK (%s)\n", this->vdpau_accel->vdp_get_error_string(status)); + return 0; + } + } + return 1; +} + +static void draw_frames(video_decoder_t *this_gen, int flush) +{ + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *)this_gen; + + struct decoded_picture *decoded_pic = NULL; + while ((decoded_pic = dpb_get_next_out_picture(this->nal_parser->dpb, flush)) != NULL) { + decoded_pic->img->top_field_first = dp_top_field_first(decoded_pic); + decoded_pic->img->progressive_frame = check_progressive(this_gen, decoded_pic); +#ifdef DEBUG_H264 + printf("progressive: %d\n", decoded_pic->img->progressive_frame); +#endif + if (flush) { + xprintf(this->xine, XINE_VERBOSITY_DEBUG, + "h264 flush, draw pts: %"PRId64"\n", decoded_pic->img->pts); + } + + decoded_pic->img->draw(decoded_pic->img, this->stream); + dpb_unmark_picture_delayed(this->nal_parser->dpb, decoded_pic); + decoded_pic = NULL; + } +} + +static int vdpau_decoder_render(video_decoder_t *this_gen, VdpBitstreamBuffer *vdp_buffer, uint32_t slice_count) +{ + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *)this_gen; + vo_frame_t *img = NULL; + + /* if we wait for a second field for this frame, we + * have to render to the same surface again. + */ + if (this->incomplete_pic) { + img = this->incomplete_pic->img; + } + + // FIXME: what is if this is the second field of a field coded + // picture? - should we keep the first field in dpb? + if(this->completed_pic->flag_mask & IDR_PIC) { + dpb_flush(this->nal_parser->dpb); + if(this->incomplete_pic) { + release_decoded_picture(this->incomplete_pic); + this->incomplete_pic = NULL; + } + } + + struct seq_parameter_set_rbsp *sps = &this->completed_pic->sps_nal->sps; + struct slice_header *slc = &this->completed_pic->slc_nal->slc; + + if(sps->vui_parameters_present_flag && + sps->vui_parameters.timing_info_present_flag && + this->video_step == 0) { + this->video_step = 2*90000/(1/((double)sps->vui_parameters.num_units_in_tick/(double)sps->vui_parameters.time_scale)); + } + + /* go and decode a frame */ + + /* check if we expect a second field, but got a frame */ + if (this->incomplete_pic && img) { + if ((this->completed_pic->slc_nal->slc.frame_num != + this->incomplete_pic->coded_pic[0]->slc_nal->slc.frame_num) || + !slc->field_pic_flag) { + xprintf(this->xine, XINE_VERBOSITY_DEBUG, "H264 warning: Expected a second field, stream might be broken\n"); + + /* remove this pic from dpb, as it is not complete */ + dpb_unmark_picture_delayed(this->nal_parser->dpb, this->incomplete_pic); + dpb_unmark_reference_picture(this->nal_parser->dpb, this->incomplete_pic); + + release_decoded_picture(this->incomplete_pic); + this->incomplete_pic = NULL; + img = NULL; + } + } + + + VdpPictureInfoH264 pic; + + fill_vdpau_pictureinfo_h264(this_gen, slice_count, &pic); + +#ifdef DEBUG_H264 + dump_pictureinfo_h264(&pic); + + int i; + printf("E: Bytes used: %d\n", vdp_buffer->bitstream_bytes); + printf("E: Decode data: \nE:"); + for(i = 0; i < ((vdp_buffer->bitstream_bytes < 20) ? vdp_buffer->bitstream_bytes : 20); i++) { + printf("%02x ", ((uint8_t*)vdp_buffer->bitstream)[i]); + if((i+1) % 10 == 0) + printf("\nE:"); + } + printf("\n...\n"); + for(i = vdp_buffer->bitstream_bytes - 20; i < vdp_buffer->bitstream_bytes; i++) { + printf("%02x ", ((uint8_t*)vdp_buffer->bitstream)[i]); + if((i+1) % 10 == 0) + printf("\nE:"); + } + printf("\nE: ---------------------------------------------------------------\n"); +#endif + + if(!this->decoder_started && !pic.is_reference) + return 0; + + this->decoder_started = 1; + + if(img == NULL) { + img = this->stream->video_out->get_frame (this->stream->video_out, + this->width, this->height, + this->ratio, + XINE_IMGFMT_VDPAU, VO_BOTH_FIELDS); + this->vdpau_accel = (vdpau_accel_t*)img->accel_data; + + img->duration = this->video_step; + img->pts = this->completed_pic->pts; + + if (this->dangling_img) { + xprintf(this->xine, XINE_VERBOSITY_LOG, + "broken stream: current img wasn't processed -- freeing it!\n"); + this->dangling_img->free(this->dangling_img); + } + this->dangling_img = img; + } else { + if (img->pts == 0) { + img->pts = this->completed_pic->pts; + } + } + + if(this->vdp_runtime_nr != *(this->vdpau_accel->current_vdp_runtime_nr)) { + xprintf(this->xine, XINE_VERBOSITY_LOG, + "VDPAU was preempted. Reinitialise the decoder.\n"); + this->decoder = VDP_INVALID_HANDLE; + vdpau_h264_reset(this_gen); + this->vdp_runtime_nr = this->vdpau_accel->vdp_runtime_nr; + return 0; + } + + VdpVideoSurface surface = this->vdpau_accel->surface; + + /*xprintf(this->xine, XINE_VERBOSITY_DEBUG, + "Decode: NUM: %d, REF: %d, BYTES: %d, PTS: %lld\n", pic.frame_num, pic.is_reference, vdp_buffer->bitstream_bytes, this->completed_pic->pts);*/ + VdpStatus status = this->vdpau_accel->vdp_decoder_render(this->decoder, + surface, (VdpPictureInfo*)&pic, 1, vdp_buffer); + + /* free the image data */ + if(((uint8_t*)vdp_buffer->bitstream) != NULL) { + free((uint8_t*)vdp_buffer->bitstream); + } + + process_mmc_operations(this->nal_parser, this->completed_pic); + + if(status != VDP_STATUS_OK) + { + xprintf(this->xine, XINE_VERBOSITY_LOG, "vdpau_h264: Decoder failure: %s\n", this->vdpau_accel->vdp_get_error_string(status)); + if (this->dangling_img) + this->dangling_img->free(this->dangling_img); + img = NULL; + this->dangling_img = NULL; + free_coded_picture(this->completed_pic); + this->completed_pic = NULL; + } + else { + img->bad_frame = 0; + + if(!img->progressive_frame && this->completed_pic->repeat_pic) + img->repeat_first_field = 1; + //else if(img->progressive_frame && this->nal_parser->current_nal->repeat_pic) + // img->duration *= this->nal_parser->current_nal->repeat_pic; + + /* only bt601 and bt701 handled so far. others seem to be rarely used */ + if(sps->vui_parameters.colour_description_present) { + switch (sps->vui_parameters.colour_primaries) { + case 1: + this->color_standard = VDP_COLOR_STANDARD_ITUR_BT_709; + break; + case 5: + case 6: + default: + this->color_standard = VDP_COLOR_STANDARD_ITUR_BT_601; + break; + } + } + + this->vdpau_accel->color_standard = this->color_standard; + + struct decoded_picture *decoded_pic = NULL; + + + uint8_t draw_frame = 0; + if (!slc->field_pic_flag) { /* frame coded: simply add to dpb */ + decoded_pic = init_decoded_picture(this->completed_pic, img); + this->completed_pic = NULL; + this->dangling_img = NULL; + + dpb_add_picture(this->nal_parser->dpb, decoded_pic, sps->num_ref_frames); + + draw_frame = 1; + } else { /* field coded: check for second field */ + if (!this->incomplete_pic) { + decoded_pic = init_decoded_picture(this->completed_pic, img); + this->completed_pic = NULL; + this->dangling_img = NULL; + this->incomplete_pic = decoded_pic; + lock_decoded_picture(this->incomplete_pic); + + dpb_add_picture(this->nal_parser->dpb, decoded_pic, sps->num_ref_frames); + + /* don't do a draw yet as the field was incomplete */ + draw_frame = 0; + } else { + decoded_pic = this->incomplete_pic; + lock_decoded_picture(decoded_pic); + + /* picture is complete now */ + release_decoded_picture(this->incomplete_pic); + this->incomplete_pic = NULL; + this->dangling_img = NULL; + + decoded_pic_add_field(decoded_pic, this->completed_pic); + this->completed_pic = NULL; + + draw_frame = 1; + } + } + + release_decoded_picture(decoded_pic); + + /* draw the next frame in display order */ + if (draw_frame) { + draw_frames(this_gen, 0); + } + } + + return 1; +} + +/* + * This function receives a buffer of data from the demuxer layer and + * figures out how to handle it based on its header flags. + */ +static void vdpau_h264_decode_data (video_decoder_t *this_gen, + buf_element_t *buf) { + + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *) this_gen; + + VdpBitstreamBuffer vdp_buffer; + vdp_buffer.struct_version = VDP_BITSTREAM_BUFFER_VERSION; + + /* a video decoder does not care about this flag (?) */ + if (buf->decoder_flags & BUF_FLAG_PREVIEW) + return; + + if(buf->decoder_flags & BUF_FLAG_FRAME_START || buf->decoder_flags & BUF_FLAG_FRAME_END) + this->have_frame_boundary_marks = 1; + + if (buf->decoder_flags & BUF_FLAG_FRAMERATE) { + this->video_step = buf->decoder_info[0]; + _x_stream_info_set(this->stream, XINE_STREAM_INFO_FRAME_DURATION, this->video_step); + } + + if (this->video_step != this->reported_video_step){ + _x_stream_info_set(this->stream, XINE_STREAM_INFO_FRAME_DURATION, (this->reported_video_step = this->video_step)); + } + + if (buf->decoder_flags & BUF_FLAG_STDHEADER) { /* need to initialize */ + this->have_frame_boundary_marks = 0; + + xine_bmiheader *bih = (xine_bmiheader*)buf->content; + this->width = bih->biWidth; + this->height = bih->biHeight; + + uint8_t *codec_private = buf->content + sizeof(xine_bmiheader); + uint32_t codec_private_len = bih->biSize - sizeof(xine_bmiheader); + this->codec_private_len = codec_private_len; + this->codec_private = malloc(codec_private_len); + memcpy(this->codec_private, codec_private, codec_private_len); + + if(codec_private_len > 0) { + parse_codec_private(this->nal_parser, codec_private, codec_private_len); + } + } else if (buf->decoder_flags & BUF_FLAG_SPECIAL) { + this->have_frame_boundary_marks = 0; + + if(buf->decoder_info[1] == BUF_SPECIAL_DECODER_CONFIG) { + uint8_t *codec_private = buf->decoder_info_ptr[2]; + uint32_t codec_private_len = buf->decoder_info[2]; + this->codec_private_len = codec_private_len; + this->codec_private = malloc(codec_private_len); + memcpy(this->codec_private, codec_private, codec_private_len); + + if(codec_private_len > 0) { + parse_codec_private(this->nal_parser, codec_private, codec_private_len); + } + } else if (buf->decoder_info[1] == BUF_SPECIAL_PALETTE) { + xprintf(this->xine, XINE_VERBOSITY_LOG, + "SPECIAL PALETTE is not yet handled\n"); + } else + xprintf(this->xine, XINE_VERBOSITY_LOG, + "UNKNOWN SPECIAL HEADER\n"); + + } else { + /* parse the first nal packages to retrieve profile type */ + int len = 0; + + while(len < buf->size && !(this->wait_for_frame_start && !(buf->decoder_flags & BUF_FLAG_FRAME_START))) { + this->wait_for_frame_start = 0; + len += parse_frame(this->nal_parser, buf->content + len, buf->size - len, + buf->pts, + (uint8_t**)&vdp_buffer.bitstream, &vdp_buffer.bitstream_bytes, &this->completed_pic); + + if(this->decoder == VDP_INVALID_HANDLE && + this->completed_pic && + this->completed_pic->sps_nal != NULL && + this->completed_pic->sps_nal->sps.pic_width > 0 && + this->completed_pic->sps_nal->sps.pic_height > 0) { + + vdpau_decoder_init(this_gen); + } + + if(this->completed_pic && + this->completed_pic->sps_nal != NULL && + this->completed_pic->sps_nal->sps.vui_parameters_present_flag && + this->completed_pic->sps_nal->sps.vui_parameters.bitstream_restriction_flag) { + + this->nal_parser->dpb->max_reorder_frames = + this->completed_pic->sps_nal->sps.vui_parameters.num_reorder_frames + 1; + this->nal_parser->dpb->max_dpb_frames = this->completed_pic->sps_nal->sps.vui_parameters.max_dec_frame_buffering + 1; + + xprintf(this->xine, XINE_VERBOSITY_DEBUG, + "max reorder count: %d, max dpb count %d\n", + this->nal_parser->dpb->max_reorder_frames, + this->nal_parser->dpb->max_dpb_frames); + } + + if(this->decoder != VDP_INVALID_HANDLE && + vdp_buffer.bitstream_bytes > 0 && + this->completed_pic->slc_nal != NULL && + this->completed_pic->pps_nal != NULL) { + vdpau_decoder_render(this_gen, &vdp_buffer, this->completed_pic->slice_cnt); + } else if (this->completed_pic != NULL) { + free_coded_picture(this->completed_pic); + } + + /* in case the last nal was detected as END_OF_SEQUENCE + * we will flush the dpb, so that all pictures get drawn + */ + if(this->nal_parser->last_nal_res == 3) { + xprintf(this->xine, XINE_VERBOSITY_DEBUG, + "END_OF_SEQUENCE, flush buffers\n"); + vdpau_h264_flush(this_gen); + } + } + } + + if(buf->decoder_flags & BUF_FLAG_FRAME_END) + this->wait_for_frame_start = 0; +} + +/* + * This function is called when xine needs to flush the system. + */ +static void vdpau_h264_flush (video_decoder_t *this_gen) { + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t*) this_gen; + struct decoded_picture *decoded_pic = NULL; + + if(this->dangling_img){ + this->dangling_img->free(this->dangling_img); + this->dangling_img = NULL; + } + + if (this->incomplete_pic) { + release_decoded_picture(this->incomplete_pic); + this->incomplete_pic = NULL; + } + + draw_frames(this_gen, 1); + dpb_free_all(this->nal_parser->dpb); + this->reset = VO_NEW_SEQUENCE_FLAG; +} + +/* + * This function resets the video decoder. + */ +static void vdpau_h264_reset (video_decoder_t *this_gen) { + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *) this_gen; + + dpb_free_all(this->nal_parser->dpb); + + if (this->decoder != VDP_INVALID_HANDLE) { + this->vdpau_accel->vdp_decoder_destroy( this->decoder ); + this->decoder = VDP_INVALID_HANDLE; + } + + // Doing a full parser reinit here works more reliable than + // resetting + + //reset_parser(this->nal_parser); + free_parser(this->nal_parser); + this->nal_parser = init_parser(this->xine); + + this->color_standard = VDP_COLOR_STANDARD_ITUR_BT_601; + this->video_step = 0; + + if(this->codec_private_len > 0) { + parse_codec_private(this->nal_parser, this->codec_private, this->codec_private_len); + + /* if the stream does not contain frame boundary marks we + * have to hope that the next nal will start with the next + * incoming buf... seems to work, though... + */ + this->wait_for_frame_start = this->have_frame_boundary_marks; + } + + if (this->incomplete_pic) { + release_decoded_picture(this->incomplete_pic); + this->incomplete_pic = NULL; + } + + if (this->dangling_img) { + this->dangling_img->free(this->dangling_img); + this->dangling_img = NULL; + } + + this->progressive_cnt = 0; + this->reset = VO_NEW_SEQUENCE_FLAG; +} + +/* + * The decoder should forget any stored pts values here. + */ +static void vdpau_h264_discontinuity (video_decoder_t *this_gen) { + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *) this_gen; + + dpb_clear_all_pts(this->nal_parser->dpb); + this->reset = VO_NEW_SEQUENCE_FLAG; +} + +/* + * This function frees the video decoder instance allocated to the decoder. + */ +static void vdpau_h264_dispose (video_decoder_t *this_gen) { + + vdpau_h264_decoder_t *this = (vdpau_h264_decoder_t *) this_gen; + + if (this->incomplete_pic) { + release_decoded_picture(this->incomplete_pic); + this->incomplete_pic = NULL; + } + + if (this->dangling_img) { + this->dangling_img->free(this->dangling_img); + this->dangling_img = NULL; + } + + dpb_free_all(this->nal_parser->dpb); + + if (this->decoder != VDP_INVALID_HANDLE) { + this->vdpau_accel->vdp_decoder_destroy( this->decoder ); + this->decoder = VDP_INVALID_HANDLE; + } + + this->stream->video_out->close( this->stream->video_out, this->stream ); + + free_parser (this->nal_parser); + free (this_gen); +} + +/* + * This function allocates, initializes, and returns a private video + * decoder structure. + */ +static video_decoder_t *open_plugin (video_decoder_class_t *class_gen, xine_stream_t *stream) { + + vdpau_h264_decoder_t *this ; + + /* the videoout must be vdpau-capable to support this decoder */ + if ( !(stream->video_driver->get_capabilities(stream->video_driver) & VO_CAP_VDPAU_H264) ) + return NULL; + + /* now check if vdpau has free decoder resource */ + vo_frame_t *img = stream->video_out->get_frame( stream->video_out, 1920, 1080, 1, XINE_IMGFMT_VDPAU, VO_BOTH_FIELDS ); + vdpau_accel_t *accel = (vdpau_accel_t*)img->accel_data; + int runtime_nr = accel->vdp_runtime_nr; + img->free(img); + VdpDecoder decoder; + VdpStatus st = accel->vdp_decoder_create( accel->vdp_device, VDP_DECODER_PROFILE_H264_MAIN, 1920, 1080, 16, &decoder ); + if ( st!=VDP_STATUS_OK ) { + lprintf( "can't create vdpau decoder.\n" ); + return NULL; + } + + accel->vdp_decoder_destroy( decoder ); + + this = (vdpau_h264_decoder_t *) calloc(1, sizeof(vdpau_h264_decoder_t)); + + this->nal_parser = init_parser(stream->xine); + + this->video_decoder.decode_data = vdpau_h264_decode_data; + this->video_decoder.flush = vdpau_h264_flush; + this->video_decoder.reset = vdpau_h264_reset; + this->video_decoder.discontinuity = vdpau_h264_discontinuity; + this->video_decoder.dispose = vdpau_h264_dispose; + + this->stream = stream; + this->xine = stream->xine; + this->class = (vdpau_h264_class_t *) class_gen; + + this->decoder = VDP_INVALID_HANDLE; + this->vdp_runtime_nr = runtime_nr; + this->color_standard = VDP_COLOR_STANDARD_ITUR_BT_601; + this->progressive_cnt = 0; + + this->reset = VO_NEW_SEQUENCE_FLAG; + + (this->stream->video_out->open) (this->stream->video_out, this->stream); + + return &this->video_decoder; +} + +/* + * This function allocates a private video decoder class and initializes + * the class's member functions. + */ +static void *init_plugin (xine_t *xine, void *data) { + + vdpau_h264_class_t *this; + + this = (vdpau_h264_class_t *) calloc(1, sizeof(vdpau_h264_class_t)); + + this->decoder_class.open_plugin = open_plugin; + this->decoder_class.identifier = "vdpau_h264"; + this->decoder_class.description = + N_("vdpau_h264: h264 decoder plugin using VDPAU hardware decoding.\n" + "Must be used along with video_out_vdpau."); + this->decoder_class.dispose = default_video_decoder_class_dispose; + + return this; +} + +/* + * This is a list of all of the internal xine video buffer types that + * this decoder is able to handle. Check src/xine-engine/buffer.h for a + * list of valid buffer types (and add a new one if the one you need does + * not exist). Terminate the list with a 0. + */ +static const uint32_t video_types[] = { + /* BUF_VIDEO_FOOVIDEO, */ + BUF_VIDEO_H264, + 0 +}; + +/* + * This data structure combines the list of supported xine buffer types and + * the priority that the plugin should be given with respect to other + * plugins that handle the same buffer type. A plugin with priority (n+1) + * will be used instead of a plugin with priority (n). + */ +static const decoder_info_t dec_info_video = { + video_types, /* supported types */ + 7 /* priority */ +}; + +/* + * The plugin catalog entry. This is the only information that this plugin + * will export to the public. + */ +const plugin_info_t xine_plugin_info[] EXPORTED = { + /* { type, API, "name", version, special_info, init_function } */ + { PLUGIN_VIDEO_DECODER | PLUGIN_MUST_PRELOAD, 19, "vdpau_h264", XINE_VERSION_CODE, &dec_info_video, init_plugin }, + { PLUGIN_NONE, 0, "", 0, NULL, NULL } +}; |