vulkan-playground/vaapi_encoder_hevc.h

#ifndef VAAPI_ENCODER_HEVC_H
#define VAAPI_ENCODER_HEVC_H

#include "vaapi_encoder.h"
#include <cstring>
#include <va/va_enc_hevc.h>

#define USE_P010 1

#define NAL_REF_IDC_NONE        0
#define NAL_REF_IDC_LOW         1
#define NAL_REF_IDC_MEDIUM      2
#define NAL_REF_IDC_HIGH        3

// SLICE TYPE HEVC ENUM
enum {
	SLICE_B = 0,
	SLICE_P = 1,
	SLICE_I = 2,
};
#define IS_I_SLICE(type) (SLICE_I == (type))
#define IS_P_SLICE(type) (SLICE_P == (type))
#define IS_B_SLICE(type) (SLICE_B == (type))


#define ENTROPY_MODE_CAVLC      0
#define ENTROPY_MODE_CABAC      1

#define PROFILE_IDC_MAIN        1
#define PROFILE_IDC_MAIN10      2

enum NALUType {
	NALU_TRAIL_N        = 0x00, // Coded slice segment of a non-TSA, non-STSA trailing picture - slice_segment_layer_rbsp, VLC
	NALU_TRAIL_R        = 0x01, // Coded slice segment of a non-TSA, non-STSA trailing picture - slice_segment_layer_rbsp, VLC
	NALU_TSA_N          = 0x02, // Coded slice segment of a TSA picture - slice_segment_layer_rbsp, VLC
	NALU_TSA_R          = 0x03, // Coded slice segment of a TSA picture - slice_segment_layer_rbsp, VLC
	NALU_STSA_N         = 0x04, // Coded slice of an STSA picture - slice_layer_rbsp, VLC
	NALU_STSA_R         = 0x05, // Coded slice of an STSA picture - slice_layer_rbsp, VLC
	NALU_RADL_N         = 0x06, // Coded slice of an RADL picture - slice_layer_rbsp, VLC
	NALU_RADL_R         = 0x07, // Coded slice of an RADL picture - slice_layer_rbsp, VLC
	NALU_RASL_N         = 0x08, // Coded slice of an RASL picture - slice_layer_rbsp, VLC
	NALU_RASL_R         = 0x09, // Coded slice of an RASL picture - slice_layer_rbsp, VLC
	/* 0x0a..0x0f - Reserved */
	NALU_BLA_W_LP       = 0x10, // Coded slice segment of an BLA picture - slice_segment_layer_rbsp, VLC
	NALU_BLA_W_DLP      = 0x11, // Coded slice segment of an BLA picture - slice_segment_layer_rbsp, VLC
	NALU_BLA_N_LP       = 0x12, // Coded slice segment of an BLA picture - slice_segment_layer_rbsp, VLC
	NALU_IDR_W_DLP      = 0x13, // Coded slice segment of an IDR picture - slice_segment_layer_rbsp, VLC
	NALU_IDR_N_LP       = 0x14, // Coded slice segment of an IDR picture - slice_segment_layer_rbsp, VLC
	NALU_CRA            = 0x15, // Coded slice segment of an CRA picture - slice_segment_layer_rbsp, VLC
	/* 0x16..0x1f - Reserved */
	NALU_VPS            = 0x20, // Video parameter set - video_parameter_set_rbsp, non-VLC
	NALU_SPS            = 0x21, // Sequence parameter set - seq_parameter_set_rbsp, non-VLC
	NALU_PPS            = 0x22, // Picture parameter set - pic_parameter_set_rbsp, non-VLC
	NALU_AUD            = 0x23, // Access unit delimiter - access_unit_delimiter_rbsp, non-VLC
	NALU_EOS            = 0x24, // End of sequence - end_of_seq_rbsp, non-VLC
	NALU_EOB            = 0x25, // End of bitsteam - end_of_bitsteam_rbsp, non-VLC
	NALU_FD             = 0x26, // Filler data - filler_data_rbsp, non-VLC
	NALU_PREFIX_SEI     = 0x27, // Supplemental enhancement information (SEI) - sei_rbsp, non_VLC
	NALU_SUFFIX_SEI     = 0x28, // Supplemental enhancement information (SEI) - sei_rbsp, non_VLC
	/* 0x29..0x2f - Reserved */
	/* 0x30..0x3f - Unspecified */
	//this should be the last element of this enum
	//chagne this value if NAL unit type increased
	MAX_HEVC_NAL_TYPE   = 0x3f,
};

struct BaseBitstreamHEVC : BaseBitstream<16>
{
	inline void NalStartCodePrefix(int nal_unit_type)
	{
		if (nal_unit_type == NALU_VPS ||
			nal_unit_type == NALU_SPS ||
			nal_unit_type == NALU_PPS ||
			nal_unit_type == NALU_AUD)
			PutUI(0x00000001, 32);
		else
			PutUI(0x000001, 24);
	}
	inline void NalHeader(int nal_unit_type)
	{
		PutUI(0, 1);                /* forbidden_zero_bit: 0 */
		PutUI(nal_unit_type, 6);
		PutUI(0, 6);
		PutUI(1, 3);
	}
	void ProfileTier(const VAEncSequenceParameterBufferHEVC *seq)
	{
		uint32_t i = 0;
		PutUI(0, 2); // general_profile_space
		PutUI(seq->general_tier_flag, 1); // general_tier_flag
		PutUI(seq->general_profile_idc, 5); // general_profile_idc //  real_hevc_profile

		//for (i = 0; i < 32; i++)
		//PutUI(protier_param.general_profile_compatibility_flag[i], 1);
		//PutUI(1 << 30, 32);
		// ptps->general_profile_compatibility_flag[ptps->general_profile_idc] = 1;
		// todo: configurable flags?
		for (i = 0; i < 32; i++)
			PutUI(i == seq->general_profile_idc, 1);
		PutUI(1, 1); //general_progressive_source_flag
		PutUI(0, 1); // general_interlaced_source_flag
		PutUI(1, 1); // general_non_packed_constraint_flag
		PutUI(1, 1); // general_frame_only_constraint_flag
		// reserved?
		PutUI(0, 16);
		PutUI(0, 16);
		PutUI(0, 12);
		PutUI(seq->general_level_idc, 8); // general_level_idc
		//ptps->general_level_idc = 30;
		//ptps->general_level_idc = ptps->general_level_idc * 4;
	}
};
#define TEMPORAL_ID_NESTING 1
#define POC_BITS 16

#define ALIGN16(x)  ((x+15)&~15)

struct PackedVPSHEVC : BaseBitstreamHEVC
{
	PackedVPSHEVC(const VAEncSequenceParameterBufferHEVC *seq): BaseBitstreamHEVC()
	{
		NalStartCodePrefix(NALU_VPS);
		NalHeader(NALU_VPS);
		uint32_t i = 0;
		PutUI(0, 4); // vps.vps_video_parameter_set_id
		PutUI(3, 2);  //vps_reserved_three_2bits
		//vps_base_layer_internal_flag:1
		//vps_base_layer_available_flag:1
		PutUI(0, 6);  //vps_reserved_zero_6bits
		// vps_max_layers_minus1:0

		PutUI(0, 3); // vps_max_sub_layers_minus1
		PutUI(TEMPORAL_ID_NESTING, 1); // vps_temporal_id_nesting_flag
		PutUI(0xFFFF, 16); //vps_reserved_0xffff_16bits
		ProfileTier(seq);

		PutUI(0, 1); // vps.vps_sub_layer_ordering_info_present_flag

		//	for (i = (vps.vps_sub_layer_ordering_info_present_flag ? 0 : vps.vps_max_sub_layers_minus1); i <= vps.vps_max_sub_layers_minus1; i++) {
		// NOTE: In teddi and mv_encoder, the setting is max_dec_pic_buffering.
		// here just follow the spec 7.3.2.1
		// todo: check this. At least, breaks some hevc parsers when set to 0
		PutUE(1); //vps.vps_max_dec_pic_buffering_minus1[i]
		PutUE(0);//vps.vps_max_num_reorder_pics[i]
		PutUE(0);//vps.vps_max_latency_increase_plus1[i]
		//}
		//*/

		PutUI(0, 6); // vps.vps_max_nuh_reserved_zero_layer_id
		PutUE(0); // vps.vps_num_op_sets_minus1

		PutUI(0, 1); // vps.vps_num_op_sets_minus1
		/*&
	if (vps.vps_timing_info_present_flag) {
		PutUE(vps.vps_num_units_in_tick);
		PutUE(vps.vps_time_scale);
		PutUE(vps.vps_poc_proportional_to_timing_flag);
		if (vps.vps_poc_proportional_to_timing_flag) {
			PutUE(vps.vps_num_ticks_poc_diff_one_minus1);
		}
		PutUE(vps.vps_num_hrd_parameters);
		for (i = 0; i < vps.vps_num_hrd_parameters; i++) {
			PutUE(vps.hrd_layer_set_idx[i]);
			if (i > 0) {
				PutUI(vps.cprms_present_flag[i], 1);
			}
		}
	}
*/
		// todo: bitstream restrictions?
		// no extension flag
		PutUI(0, 1);
		RBSPTrailingBits();
		End();
	}
};

struct PackedSPSHEVC : BaseBitstreamHEVC
{
	PackedSPSHEVC(const VAEncSequenceParameterBufferHEVC *sps, int width, int height): BaseBitstreamHEVC()
	{
		NalStartCodePrefix(NALU_SPS);
		NalHeader(NALU_SPS);
		uint32_t  i = 0;
		PutUI(0, 4); // sps.sps_video_parameter_set_id
		PutUI(0, 3); // sps.sps_max_sub_layers_minus1
		PutUI(TEMPORAL_ID_NESTING, 1); // sps.sps_temporal_id_nesting_flag
		ProfileTier(sps);

		PutUE(0); // sps.sps_seq_parameter_set_id
		PutUE(sps->seq_fields.bits.chroma_format_idc); // sps.chroma_format_idc
		// 4:2:0

		// todo: check if we can use 4:4:4/4:2:2
		if (sps->seq_fields.bits.chroma_format_idc == 3) {
			PutUI(sps->seq_fields.bits.separate_colour_plane_flag, 1);

		}
		PutUE(ALIGN16(width) ); //sps.pic_width_in_luma_samples
		PutUE(ALIGN16(height)); //sps.pic_height_in_luma_samples
		bool conformance_window_flag = ALIGN16(width) != width || ALIGN16(height) != height;

		PutUI(1, 1); // sps.conformance_window_flag

		if (conformance_window_flag) { // sps.
			PutUE(0); // sps.conf_win_left_offset
			PutUE((ALIGN16(width) - width) >> 1); // sps.conf_win_right_offset
			PutUE(0); // sps.conf_win_top_offset
			PutUE((ALIGN16(height) - height) >> 1); // sps.conf_win_bottom_offset
		}

		PutUE(sps->seq_fields.bits.bit_depth_luma_minus8); //sps.bit_depth_luma_minus8
		PutUE(sps->seq_fields.bits.bit_depth_chroma_minus8); //sps.bit_depth_chroma_minus8

		PutUE(POC_BITS - 4); // sps.log2_max_pic_order_cnt_lsb_minus4

		PutUI(0 , 1); //sps.sps_sub_layer_ordering_info_present_flag

		//for (i = (sps.sps_sub_layer_ordering_info_present_flag ? 0 : sps.sps_max_sub_layers_minus1); i <= sps.sps_max_sub_layers_minus1; i++) {
		// NOTE: In teddi and mv_encoder, the setting is max_dec_pic_buffering.
		// here just follow the spec 7.3.2.2
		// todo: check. At least, breaks some hevc parsers when set to 0
		PutUE(0);//sps.sps_max_dec_pic_buffering_minus1[i]
		PutUE(0);//sps.sps_max_num_reorder_pics[i]
		PutUE(0); //sps.sps_max_latency_increase_plus1[i]
		//}

		PutUE(sps->log2_min_luma_coding_block_size_minus3); //  sps.log2_min_luma_coding_block_size_minus3
		PutUE(sps->log2_diff_max_min_luma_coding_block_size);//sps.log2_diff_max_min_luma_coding_block_size
		PutUE(sps->log2_min_transform_block_size_minus2);//sps.log2_min_luma_transform_block_size_minus2
		PutUE(sps->log2_diff_max_min_transform_block_size);//sps.log2_diff_max_min_luma_transform_block_size
		PutUE(sps->max_transform_hierarchy_depth_inter);//sps.max_transform_hierarchy_depth_inter
		PutUE(sps->max_transform_hierarchy_depth_inter);//sps.max_transform_hierarchy_depth_intra

		assert(!sps->seq_fields.bits.scaling_list_enabled_flag);
		// scaling_list_enabled_flag is set as 0 in fill_sps_header() for now
		PutUI(0, 1); // sps.scaling_list_enabled_flag
		/*if (sps.scaling_list_enabled_flag) {
		PutUI(sps.sps_scaling_list_data_present_flag, 1);
		if (sps.sps_scaling_list_data_present_flag) {
			//scaling_list_data();
		}
	}*/

		PutUI(sps->seq_fields.bits.amp_enabled_flag, 1); // sps.amp_enabled_flag
		PutUI(sps->seq_fields.bits.sample_adaptive_offset_enabled_flag, 1); // sps.sample_adaptive_offset_enabled_flag

		// pcm_enabled_flag is set as 0 in fill_sps_header() for now
		PutUI(sps->seq_fields.bits.pcm_enabled_flag, 1);//sps.pcm_enabled_flag
		assert(!sps->seq_fields.bits.pcm_enabled_flag);
		if (sps->seq_fields.bits.pcm_enabled_flag) {
			PutUI(sps->pcm_sample_bit_depth_luma_minus1, 4);
			PutUI(sps->pcm_sample_bit_depth_chroma_minus1, 4);
			PutUE(sps->log2_min_pcm_luma_coding_block_size_minus3);
			PutUE(sps->log2_max_pcm_luma_coding_block_size_minus3 - sps->log2_min_pcm_luma_coding_block_size_minus3); //log2_diff_max_min_pcm_luma_coding_block_size
			PutUI(sps->seq_fields.bits.pcm_loop_filter_disabled_flag, 1);
		}

		PutUE(1);// sps.num_short_term_ref_pic_sets
		PutUE(1); // num_negative_pics
		PutUE(0); // num_positive_pics
		PutUE(0); //delta_poc_s0_minus1[i]
		PutUI(1, 1);//used_by_curr_pic_s0_flag[i]
		/*for (i = 0; i < sps.num_short_term_ref_pic_sets; i++) {
		pack_short_term_ref_pic_setp(bs, &sps.strp[i], i == 0);
	}*/

		// long_term_ref_pics_present_flag is set as 0 in fill_sps_header() for now
		PutUI(0, 1); // sps.long_term_ref_pics_present_flag
		/*if (sps.long_term_ref_pics_present_flag) {
		PutUE(sps.num_long_term_ref_pics_sps);
		for (i = 0; i < sps.num_long_term_ref_pics_sps; i++) {
			PutUE(sps.lt_ref_pic_poc_lsb_sps[i]);
			PutUI(sps.used_by_curr_pic_lt_sps_flag[i], 1);
		}
	}*/

		PutUI(sps->seq_fields.bits.sps_temporal_mvp_enabled_flag, 1); // sps.sps_temporal_mvp_enabled_flag
		PutUI(sps->seq_fields.bits.strong_intra_smoothing_enabled_flag, 1); //sps.strong_intra_smoothing_enabled_flag

		assert(!sps->vui_parameters_present_flag);
		PutUI(sps->vui_parameters_present_flag, 1); // sps.vui_parameters_present_flag

		PutUI(0, 1); // sps.sps_extension_present_flag
		RBSPTrailingBits();
		End();
	}
};

#define PPS_CABAC_INIT_PRESENT_FLAG 1

struct PackedPPSHEVC : BaseBitstreamHEVC
{
	PackedPPSHEVC(const VAEncPictureParameterBufferHEVC *pic): BaseBitstreamHEVC()
	{
		NalStartCodePrefix(NALU_PPS);
		NalHeader(NALU_PPS);
		uint32_t  i = 0;
		bool deblocking_filter_control_present_flag = false;
		PutUE(0); // pps.pps_pic_parameter_set_id
		PutUE(0); // pps.pps_seq_parameter_set_id
		PutUI(pic->pic_fields.bits.dependent_slice_segments_enabled_flag, 1); // pps.dependent_slice_segments_enabled_flag // TODO: !!!
		PutUI(0, 1); //pps.output_flag_present_flag
		PutUI(0, 3); // pps.num_extra_slice_header_bits
		PutUI(pic->pic_fields.bits.sign_data_hiding_enabled_flag, 1); //pps.sign_data_hiding_enabled_flag
		PutUI(PPS_CABAC_INIT_PRESENT_FLAG, 1); // pps.cabac_init_present_flag

		PutUE(pic->num_ref_idx_l0_default_active_minus1); //pps.num_ref_idx_l0_default_active_minus1
		PutUE(pic->num_ref_idx_l1_default_active_minus1); //pps.num_ref_idx_l1_default_active_minus1
		PutSE(pic->pic_init_qp - 26); //pps.init_qp_minus26

		PutUI(pic->pic_fields.bits.constrained_intra_pred_flag, 1); //pps.constrained_intra_pred_flag
		PutUI(pic->pic_fields.bits.transform_skip_enabled_flag, 1); //pps.transform_skip_enabled_flag

		PutUI(pic->pic_fields.bits.cu_qp_delta_enabled_flag, 1); //pps.cu_qp_delta_enabled_flag
		if (pic->pic_fields.bits.cu_qp_delta_enabled_flag) {
			PutUE(pic->diff_cu_qp_delta_depth);//diff_cu_qp_delta_depth
		}

		PutSE(pic->pps_cb_qp_offset);//pps.pps_cb_qp_offset
		PutSE(pic->pps_cr_qp_offset);//pps.pps_cr_qp_offset)

		PutUI(0, 1);//pps.pps_slice_chroma_qp_offsets_present_flag
		PutUI(pic->pic_fields.bits.weighted_pred_flag, 1);//pps.weighted_pred_flag
		PutUI(pic->pic_fields.bits.weighted_bipred_flag, 1); //pps.weighted_bipred_flag
		PutUI(pic->pic_fields.bits.transquant_bypass_enabled_flag, 1);//pps.transquant_bypass_enabled_flag
		PutUI(pic->pic_fields.bits.tiles_enabled_flag, 1);//pps.tiles_enabled_flag
		PutUI(pic->pic_fields.bits.entropy_coding_sync_enabled_flag, 1);//pps.entropy_coding_sync_enabled_flag

		if (pic->pic_fields.bits.tiles_enabled_flag) {
			bool uniform_spacing_flag = false;
			PutUE(pic->num_tile_columns_minus1);
			PutUE(pic->num_tile_rows_minus1);
			PutUI(uniform_spacing_flag, 1);//uniform_spacing_flag
			if (!uniform_spacing_flag) {
				for (i = 0; i < pic->num_tile_columns_minus1; i++) {
					PutUE(pic->column_width_minus1[i]);
				}

				for (i = 0; i < pic->num_tile_rows_minus1; i++) {
					PutUE(pic->row_height_minus1[i]);
				}

			}
			PutUI(pic->pic_fields.bits.loop_filter_across_tiles_enabled_flag, 1);
		}

		PutUI(pic->pic_fields.bits.pps_loop_filter_across_slices_enabled_flag, 1); // pps.pps_loop_filter_across_slices_enabled_flag
		PutUI(deblocking_filter_control_present_flag, 1);//pps.deblocking_filter_control_present_flag
		if (deblocking_filter_control_present_flag) {
			bool deblocking_filter_override_enabled_flag = false;
			bool  pps_deblocking_filter_disabled_flag = true;
			PutUI(deblocking_filter_override_enabled_flag, 1);
			PutUI(pps_deblocking_filter_disabled_flag, 1);
			int beta_offset_div2 = 0;
			int tc_offset_div2 = 0;
			if (!pps_deblocking_filter_disabled_flag) {
				PutSE(beta_offset_div2);
				PutSE(tc_offset_div2);
			}
		}

		// pps_scaling_list_data_present_flag is set as 0 in fill_pps_header() for now
		assert(!pic->pic_fields.bits.scaling_list_data_present_flag);
		PutUI(pic->pic_fields.bits.scaling_list_data_present_flag, 1);//pps.pps_scaling_list_data_present_flag
		/*if (pps.pps_scaling_list_data_present_flag) {
		//scaling_list_data();
	}*/

		PutUI(0, 1);//pps.lists_modification_present_flag
		PutUE(0); //pps.log2_parallel_merge_level_minus2
		PutUI(0, 1);//pps.slice_segment_header_extension_present_flag

		PutUI(0, 1); //pps.pps_extension_present_flag
		/*if (pps.pps_extension_present_flag) {
		PutUI(pps.pps_range_extension_flag, 1);
		PutUI(pps.pps_multilayer_extension_flag, 1);
		PutUI(pps.pps_3d_extension_flag, 1);
		PutUI(pps.pps_extension_5bits, 1);

	}

	if (pps.pps_range_extension_flag) {
		if (pps.transform_skip_enabled_flag)
			PutUE(pps.log2_max_transform_skip_block_size_minus2);
		PutUI(pps.cross_component_prediction_enabled_flag, 1);
		PutUI(pps.chroma_qp_offset_list_enabled_flag, 1);

		if (pps.chroma_qp_offset_list_enabled_flag) {
			PutUE(pps.diff_cu_chroma_qp_offset_depth);
			PutUE(pps.chroma_qp_offset_list_len_minus1);
			for (i = 0; i <= pps.chroma_qp_offset_list_len_minus1; i++) {
				PutUE(pps.cb_qp_offset_list[i]);
				PutUE(pps.cr_qp_offset_list[i]);
			}
		}

		PutUE(pps.log2_sao_offset_scale_luma);
		PutUE(pps.log2_sao_offset_scale_chroma);
	}
*/
		RBSPTrailingBits();
		End();
	}
};
template <int slice_type>
struct PackedSliceHEVC : BaseBitstreamHEVC
{
	inline PackedSliceHEVC(int framenum, const VAEncSequenceParameterBufferHEVC *sps, const VAEncSliceParameterBufferHEVC *slice, const VAEncPictureParameterBufferHEVC *pic): BaseBitstreamHEVC()
	{
		uint8_t nal_unit_type = NALU_TRAIL_R;
		//int gop_ref_distance = ip_period;
		int i = 0;
		//int short_term_ref_pic_set_sps_flag = 1; // !is_idr;
		int slice_qp_delta = slice->slice_qp_delta;
		int pic_order_cnt_lsb = framenum;
		if constexpr (slice_type == SLICE_I)
			nal_unit_type = NALU_IDR_W_DLP;
		NalStartCodePrefix(nal_unit_type);
		NalHeader(nal_unit_type);
		PutUI(1, 1);// first_slice_segment_in_pic_flag


		// idr
		if (nal_unit_type >= 16 && nal_unit_type <= 23)
			PutUI(1, 1); //no_output_of_prior_pics_flag

		PutUE(0);//slice_pic_parameter_set_id

		/*if (!slice_header->first_slice_segment_in_pic_flag) {
		if (slice_header->dependent_slice_segment_flag) {
			PutUI(slice_header->dependent_slice_segment_flag, 1);
		}

		PutUI(slice_header->slice_segment_address,
			   (uint8_t)(ceil(log(slice_header->picture_height_in_ctus * slice_header->picture_width_in_ctus) / log(2.0))));
	}*/
		// !slice_header->dependent_slice_segment_flag
		if (!slice->slice_fields.bits.dependent_slice_segment_flag) {
			/*for (i = 0; i < pps->num_extra_slice_header_bits; i++) {
			PutUI(slice_header->slice_reserved_undetermined_flag[i], 1);
		}*/
			PutUE(slice_type);
			/*if (pps->output_flag_present_flag) {
			PutUI(slice_header->pic_output_flag, 1);
		}*/

			if(sps->seq_fields.bits.separate_colour_plane_flag)
				PutUI(slice->slice_fields.bits.colour_plane_id, 2);

			if (!(nal_unit_type == NALU_IDR_W_DLP || nal_unit_type == NALU_IDR_N_LP)) {
				// slice_header->pic_order_cnt_lsb
				PutUI(pic_order_cnt_lsb, POC_BITS );//(sps->log2_max_pic_order_cnt_lsb_minus4 + 4)
				PutUI(1, 1); // short_term_ref_pic_set_sps_flag
				// assume we are only pushing I-slices on IDR frames, SPS only references (n-1)th frame for now
				// this should be restored when p-slice references something different or using CRA frames
#if 0
			if (!short_term_ref_pic_set_sps_flag) {
				// refer to Teddi
				if (sps->num_short_term_ref_pic_sets > 0)
					PutUI(0, 1); // inter_ref_pic_set_prediction_flag, always 0 for now

				PutUE(slice_header->strp.num_negative_pics);
				PutUE(slice_header->strp.num_positive_pics);

				// below chunks of codes (majorly two big 'for' blocks) are refering both
				// Teddi and mv_encoder, they look kind of ugly, however, keep them as these
				// since it will be pretty easy to update if change/update in Teddi side.
				// According to Teddi, these are CModel Implementation.
				int prev = 0;
				int frame_cnt_in_gop = slice_header->pic_order_cnt_lsb / 2;
				// this is the first big 'for' block
				for (i = 0; i < slice_header->strp.num_negative_pics; i++) {
					// Low Delay B case
					if (1 == gop_ref_distance) {
						PutUE(0 /*delta_poc_s0_minus1*/);
					} else {
						// For Non-BPyramid GOP i.e B0 type
						if (num_active_ref_p > 1) {
							// DeltaPOC Equals NumB
							int DeltaPoc = -(int)(gop_ref_distance);
							PutUE(prev - DeltaPoc - 1 /*delta_poc_s0_minus1*/);
						} else {
							//  the big 'if' wraps here is -
							//     if (!slice_header->short_term_ref_pic_set_sps_flag)
							// From the Teddi logic, the short_term_ref_pic_set_sps_flag only can be '0'
							// either for B-Prymid or first several frames in a GOP in multi-ref cases
							// when there are not enough backward refs.
							// So though there are really some codes under this 'else'in Teddi, don't
							// want to introduce them in MEA to avoid confusion, and put an assert
							// here to guard that there is new case we need handle in the future.
							assert(0);
						}
					}
					PutUI(1 /*used_by_curr_pic_s0_flag*/, 1);
				}

				prev = 0;
				// this is the second big 'for' block
				for (i = 0; i < slice_header->strp.num_positive_pics; i++) {
					// Non-BPyramid GOP
					if (num_active_ref_p > 1) {
						// MultiRef Case
						if (frame_cnt_in_gop < gop_ref_distance) {
							int DeltaPoc = (int)(gop_ref_distance - frame_cnt_in_gop);
							PutUE(DeltaPoc - prev - 1 /*delta_poc_s1_minus1*/);
						} else if (frame_cnt_in_gop > gop_ref_distance) {
							int DeltaPoc = (int)(gop_ref_distance * slice_header->strp.num_negative_pics - frame_cnt_in_gop);
							PutUE(DeltaPoc - prev - 1 /*delta_poc_s1_minus1*/);
						}
					} else {
						//  the big 'if' wraps here is -
						//     if (!slice_header->short_term_ref_pic_set_sps_flag)
						// From the Teddi logic, the short_term_ref_pic_set_sps_flag only can be '0'
						// either for B-Prymid or first several frames in a GOP in multi-ref cases
						// when there are not enough backward refs.
						// So though there are really some codes under this 'else'in Teddi, don't
						// want to introduce them in MEA to avoid confusion, and put an assert
						// here to guard that there is new case we need handle in the future.
						assert(0);
					}
					PutUI(1 /*used_by_curr_pic_s1_flag*/, 1);
				}
			} else if (sps->num_short_term_ref_pic_sets > 1)
				PutUI(slice_header->short_term_ref_pic_set_idx,
					   (uint8_t)(ceil(log(sps->num_short_term_ref_pic_sets) / log(2.0))));
#endif
// no long term refs
#if 0
			if (sps->long_term_ref_pics_present_flag) {
				if (sps->num_long_term_ref_pics_sps > 0)
					PutUE(slice_header->num_long_term_sps);

				PutUE(slice_header->num_long_term_pics);
			}
#endif

				if (sps->seq_fields.bits.sps_temporal_mvp_enabled_flag)
					PutUI(slice->slice_fields.bits.slice_temporal_mvp_enabled_flag, 1);
			}

			if (sps->seq_fields.bits.sample_adaptive_offset_enabled_flag ) { // sample_adaptive_offset_enabled_flag
				PutUI(slice->slice_fields.bits.slice_sao_luma_flag, 1);// slice_sao_luma_flag
				PutUI(slice->slice_fields.bits.slice_sao_chroma_flag, 1);//slice_sao_chroma_flag
			}

			if constexpr(slice_type != SLICE_I) {
				PutUI(slice->slice_fields.bits.num_ref_idx_active_override_flag, 1); //num_ref_idx_active_override_flag

				if (slice->slice_fields.bits.num_ref_idx_active_override_flag) {
					PutUE(slice->num_ref_idx_l0_active_minus1);
					//if (slice->slice_type == SLICE_B)
					//PutUE(slice->num_ref_idx_l1_active_minus1);
				}
#if 0
			if (pps->lists_modification_present_flag &&  slice_header->num_poc_total_cur > 1) {
				/* ref_pic_list_modification */
				PutUI(slice_header->ref_pic_list_modification_flag_l0, 1);

				if (slice_header->ref_pic_list_modification_flag_l0) {
					for (i = 0; i <= slice_header->num_ref_idx_l0_active_minus1; i++) {
						PutUI(slice_header->list_entry_l0[i],
							   (uint8_t)(ceil(log(slice_header->num_poc_total_cur) / log(2.0))));
					}
				}

				PutUI(slice_header->ref_pic_list_modification_flag_l1, 1);

				if (slice_header->ref_pic_list_modification_flag_l1) {
					for (i = 0; i <= slice_header->num_ref_idx_l1_active_minus1; i++) {
						PutUI(slice_header->list_entry_l1[i],
							   (uint8_t)(ceil(log(slice_header->num_poc_total_cur) / log(2.0))));
					}
				}
			}
#endif
				/*if (slice_header->slice_type == SLICE_B) {
				PutUI(slice_header->mvd_l1_zero_flag, 1);
			}*/

				if (PPS_CABAC_INIT_PRESENT_FLAG) {
					PutUI(slice->slice_fields.bits.cabac_init_flag, 1); //slice_header->cabac_init_present_flag
				}

				if (slice->slice_fields.bits.slice_temporal_mvp_enabled_flag) {
					int collocated_from_l0_flag = 1;

					if (slice->slice_type == SLICE_B) {
						collocated_from_l0_flag = slice->slice_fields.bits.collocated_from_l0_flag;
						PutUI(collocated_from_l0_flag , 1);
					}

					if (((collocated_from_l0_flag && (slice->num_ref_idx_l0_active_minus1 > 0)) ||
						(!collocated_from_l0_flag && (slice->num_ref_idx_l1_active_minus1 > 0)))) {
						PutUE(pic->num_ref_idx_l0_default_active_minus1); // collocated_ref_idx
					}
				}

				PutUE(5 - slice->max_num_merge_cand);//slice_header->five_minus_max_num_merge_cand
			}

			PutSE(slice_qp_delta);

			/*if (pps->chroma_qp_offset_list_enabled_flag) {
			PutSE(slice_header->slice_qp_delta_cb);
			PutSE(slice_header->slice_qp_delta_cr);
		}

		if (pps->deblocking_filter_override_enabled_flag) {
			PutUI(slice_header->deblocking_filter_override_flag, 1);
		}
		if (slice_header->deblocking_filter_override_flag) {
			PutUI(slice_header->disable_deblocking_filter_flag, 1);

			if (!slice_header->disable_deblocking_filter_flag) {
				PutSE(slice_header->beta_offset_div2);
				PutSE(slice_header->tc_offset_div2);
			}
		}*/

			if (pic->pic_fields.bits.pps_loop_filter_across_slices_enabled_flag &&
				(slice->slice_fields.bits.slice_sao_luma_flag || slice->slice_fields.bits.slice_sao_chroma_flag ||
				 !slice->slice_fields.bits.slice_deblocking_filter_disabled_flag)) {
				PutUI(slice->slice_fields.bits.slice_loop_filter_across_slices_enabled_flag, 1);
			}

		}

		if ((pic->pic_fields.bits.tiles_enabled_flag) || (pic->pic_fields.bits.entropy_coding_sync_enabled_flag)) {
			int num_entry_point_offsets = 0, offset_len_minus1 = 0;
			PutUE(num_entry_point_offsets);

			if (num_entry_point_offsets > 0) {
				PutUE(offset_len_minus1);
			}
		}

		/*if (pps->slice_segment_header_extension_present_flag) {
		int slice_header_extension_length = 0;

		PutUE(slice_header_extension_length);
	}*/
		RBSPTrailingBits();
		End();
	}
};


struct VaapiEncoderHEVC: VaapiEncoder
{
	VAEncSequenceParameterBufferHEVC seq;
	VAEncPictureParameterBufferHEVC pic;
	VAEncSliceParameterBufferHEVC slice;
	void InitParameters()
	{
		int width_in_mbs, height_in_mbs;
		int frame_cropping_flag = 0;
		int frame_crop_bottom_offset = 0;

		width_in_mbs = (width + 15) / 16;
		height_in_mbs = (height + 15) / 16;

		// sps
		// seems work
		seq.seq_fields.bits.sps_temporal_mvp_enabled_flag = 1;
		// todo: check if we can use 4:4:4/4:2:2
		seq.seq_fields.bits.chroma_format_idc = 1; // 4:2:0
		seq.pic_width_in_luma_samples =  ALIGN16(width);
		seq.pic_height_in_luma_samples =  ALIGN16(height);
		// todo: tunable block size
		seq.log2_diff_max_min_luma_coding_block_size = 2;
		seq.log2_diff_max_min_transform_block_size = 3;
		seq.log2_min_transform_block_size_minus2 = 0;
		seq.max_transform_hierarchy_depth_inter = 2;
		seq.max_transform_hierarchy_depth_intra = 2;
		seq.seq_fields.bits.amp_enabled_flag = 1;
		// broken on intel? or broken bistream?
		seq.seq_fields.bits.sample_adaptive_offset_enabled_flag = 0;
		seq.intra_idr_period = 32767;
		seq.intra_period = 32767;
		seq.ip_period = 1;
		if(USE_P010)
		{
			seq.seq_fields.bits.bit_depth_chroma_minus8 = 2;
			seq.seq_fields.bits.bit_depth_luma_minus8 = 2;
		}
		seq.bits_per_second = 150*1024*1024;//(long long)r->width * r->height * 12 * 90 / 50;

		// vps
		// (none?)
		// profile
		seq.general_level_idc = 120;
		seq.general_profile_idc = USE_P010? 2: 1;

		// pps
		pic.pic_fields.bits.dependent_slice_segments_enabled_flag = 1; // seens work both
		pic.pic_fields.bits.transform_skip_enabled_flag = 1;
		// pic
		pic.collocated_ref_pic_index = 0;//255;
		pic.pic_init_qp = 26;
		pic.nal_unit_type = NALU_IDR_W_DLP;
		pic.pic_fields.bits.idr_pic_flag = 1;
		pic.pic_fields.bits.coding_type = 1;
		pic.pic_fields.bits.reference_pic_flag = 1;
		// seems work
		pic.pic_fields.bits.pps_loop_filter_across_slices_enabled_flag = 1;

		pic.pic_fields.bits.cu_qp_delta_enabled_flag = 1; // CBR
		if(pic.pic_fields.bits.cu_qp_delta_enabled_flag)
			pic.diff_cu_qp_delta_depth = 2;
		for(int i = 0; i < 15; i++)
		{
			pic.reference_frames[i].picture_id = VA_INVALID_SURFACE;
			pic.reference_frames[i].flags = VA_PICTURE_HEVC_INVALID;
			pic.reference_frames[i].pic_order_cnt = 0;
		}

		// slice
		//slice.slice_fields.bits.num_ref_idx_active_override_flag = 0;
		slice.slice_qp_delta = 0;
		int lcu_size = 32; // todo: block size settings?
		int picture_width_in_ctus = (width + lcu_size - 1) / lcu_size;
		int picture_height_in_ctus = (height + lcu_size - 1) / lcu_size;
		slice.num_ctu_in_slice = picture_width_in_ctus * picture_height_in_ctus;
		slice.max_num_merge_cand = 5;
		// seems works
		//slice.slice_fields.bits.collocated_from_l0_flag = 1;

		// broken on intel???
		//slice.slice_fields.bits.slice_sao_chroma_flag = 1;
		//slice.slice_fields.bits.slice_sao_luma_flag = 1;
		memset((void*)slice.ref_pic_list0, -1, sizeof(slice.ref_pic_list0));
		memset((void*)slice.ref_pic_list1, -1, sizeof(slice.ref_pic_list1));

		for(int i = 0; i < 15; i++)
		{
			slice.ref_pic_list0[i].flags = -1;//VA_PICTURE_HEVC_INVALID;
			slice.ref_pic_list0[i].picture_id = VA_INVALID_SURFACE;
			slice.ref_pic_list0[i].pic_order_cnt = -1;
			slice.ref_pic_list1[i].flags = -1;//VA_PICTURE_HEVC_INVALID;
			slice.ref_pic_list1[i].picture_id = VA_INVALID_SURFACE;
			slice.ref_pic_list1[i].pic_order_cnt = -1;
		}
	}
	bool Setup(int drm_fd, int width, int height, const char *filename, int *dmabuf_fd, uint64_t *mod, uint32_t *size, uint32_t *offset, uint32_t *pitch1, uint32_t *pitch2, uint64_t *modifiers, int modifierscount)
	{
		VAProfile profile = VAProfileHEVCMain10;
		if(!SetupVA(profile, drm_fd, width, height, filename, dmabuf_fd, mod, size, offset, pitch1, pitch2, modifiers, modifierscount))
		{
			VaapiEncoder::Destroy();
			return false;
		}
		if(!CreateContext(profile, VA_RT_FORMAT_YUV420_10, VA_FOURCC_P010, VA_RC_CBR))
		{
			VaapiEncoder::Destroy();
			return false;
		}
		InitParameters();

		return true;
	}
	inline void EncodeIDR(int idx)
	{
		VAStatus status = vaBeginPicture(dpy, ctx, inputFrames[idx]);

		VABufferID seqb = CreateParamererBuffer(VAEncSequenceParameterBufferType, seq);
		slice.slice_type = SLICE_I;
		slice.ref_pic_list0[0].pic_order_cnt = -1;
		slice.ref_pic_list0[0].picture_id = VA_INVALID_SURFACE;
		slice.ref_pic_list0[0].flags = -1;
		VABufferID sliceb = CreateParamererBuffer(VAEncSliceParameterBufferType, slice);
		VABufferID output = CreateOutputBuf(width * height);

		pic.decoded_curr_pic.picture_id = reference_picture[frame_count % 2];
		pic.decoded_curr_pic.pic_order_cnt = frame_count;
		pic.reference_frames[0].picture_id = VA_INVALID_SURFACE;
		pic.reference_frames[0].flags = VA_PICTURE_HEVC_INVALID;
		pic.reference_frames[0].pic_order_cnt = 0;
		pic.coded_buf = output;
		pic.pic_fields.bits.idr_pic_flag = 1;
		pic.nal_unit_type = NALU_IDR_W_DLP;
		pic.pic_fields.bits.coding_type = 1;
		VABufferID picb = CreateParamererBuffer(VAEncPictureParameterBufferType, pic);
		VABufferID fpsb = CreateMiscParameterBuffer(VAEncMiscParameterTypeFrameRate,VAEncMiscParameterFrameRate{.framerate = 90} );
		VABufferID hrdb = CreateMiscParameterBuffer(VAEncMiscParameterTypeHRD, VAEncMiscParameterHRD{});
		VABufferID rcb = CreateMiscParameterBuffer(VAEncMiscParameterTypeRateControl, VAEncMiscParameterRateControl{
																							.bits_per_second = 150*1024*1024,
																							.target_percentage = 66,
																							.window_size = 1000,
																							.initial_qp = 25
																							});
		VABufferID ppps[2];
		VABufferID pvps[2];
		VABufferID psps[2];
		CreatePackedBuffer(pvps[0], pvps[1], VAEncPackedHeaderSequence, PackedVPSHEVC(&seq));
		CreatePackedBuffer(psps[0], psps[1], VAEncPackedHeaderSequence, PackedSPSHEVC(&seq, width, height));
		CreatePackedBuffer(ppps[0], ppps[1], VAEncPackedHeaderPicture, PackedPPSHEVC(&pic));

		VABufferID pslice[2];
		CreatePackedBuffer(pslice[0], pslice[1], VAEncPackedHeaderSlice,
						   PackedSliceHEVC<SLICE_I>(frame_count, &seq, &slice, &pic));
		VABufferID buffers[] = {seqb, pvps[0],pvps[1],psps[0], psps[1], fpsb, hrdb, rcb, ppps[0], ppps[1], picb, pslice[0], pslice[1], sliceb };
		vaRenderPicture(dpy, ctx, buffers, sizeof(buffers) / sizeof(buffers[0]) );
		status = vaEndPicture(dpy, ctx);
		if(status != VA_STATUS_SUCCESS)
			abort();
		PushOutput(output);
		/*status = vaSyncSurface(dpy, inputFrames[idx]);
		status = vaSyncBuffer(dpy, output, 1000000000);
		WriteOutput(output);*/
		for(int i = 0; i < sizeof(buffers) / sizeof(buffers[0]); i++)
			vaDestroyBuffer(dpy, buffers[i]);
		frame_count++;
	}
	inline void EncodeP(int idx)
	{
		VAStatus status = vaBeginPicture(dpy, ctx, inputFrames[idx]);
		// todo: chain slice/output buffers, patch POC in slice buffers???
		slice.slice_type = SLICE_P;
		slice.ref_pic_list0[0].pic_order_cnt = frame_count - 1;
		slice.ref_pic_list0[0].picture_id = reference_picture[(frame_count - 1)% 2];
		slice.ref_pic_list0[0].flags = 0;//VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE;
		VABufferID sliceb = CreateParamererBuffer(VAEncSliceParameterBufferType, slice);
		VABufferID output = CreateOutputBuf(width * height);
		pic.decoded_curr_pic.picture_id = reference_picture[frame_count % 2];
		pic.decoded_curr_pic.pic_order_cnt = frame_count;
		pic.reference_frames[0].picture_id = reference_picture[(frame_count + 1) % 2];
		pic.reference_frames[0].flags = 0;//VA_PICTURE_HEVC_RPS_ST_CURR_BEFORE;
		pic.reference_frames[0].pic_order_cnt = frame_count - 1;
		pic.coded_buf = output;
		pic.pic_fields.bits.idr_pic_flag = 0;
		pic.nal_unit_type = NALU_TRAIL_R;
		pic.pic_fields.bits.coding_type = 2;
		VABufferID picb = CreateParamererBuffer(VAEncPictureParameterBufferType, pic);
		VABufferID pslice[2];
		CreatePackedBuffer(pslice[0], pslice[1], VAEncPackedHeaderSlice,
						   PackedSliceHEVC<SLICE_P>(frame_count, &seq, &slice, &pic));
		VABufferID buffers[] = {picb, pslice[0], pslice[1], sliceb };
		vaRenderPicture(dpy, ctx, buffers, sizeof(buffers) / sizeof(buffers[0]) );

		status = vaEndPicture(dpy, ctx);
		PushOutput(output);
		/*status = vaSyncSurface(dpy, inputFrames[idx]);
		status = vaSyncBuffer(dpy, output, 1000000000);
		WriteOutput(output);*/
		for(int i = 0; i < sizeof(buffers) / sizeof(buffers[0]); i++)
			vaDestroyBuffer(dpy, buffers[i]);
		frame_count++;
	}
};


#endif // VAAPI_ENCODER_HEVC_H