Google Git
Sign in
android / platform / external / libhevc / 445e0d8d5ec10fcb8e43117c01ec6a671246ce69 / . / encoder / ihevce_decomp_pre_intra_pass.c
blob: 065fcbb5e77186c7724e415bbe5a57686bbcb3ab [file] [log] [blame]
/******************************************************************************
*
* Copyright (C) 2018 The Android Open Source Project
*
* Licensed under the Apache License, Version 2.0 (the "License");
* you may not use this file except in compliance with the License.
* You may obtain a copy of the License at:
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*
*****************************************************************************
* Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore
*/
/*!
******************************************************************************
* \file ihevce_decomp_pre_intra_pass.c
*
* \brief
* This file contains definitions related to frame decomposition done during
* pre intra processing
*
* \date
* 19/02/2013
*
* \author
* Ittiam
*
* List of Functions
* ihevce_intra_populate_mode_bits_cost()
* ihevce_8x8_sad_computer()
* ihevce_4x4_sad_computer()
* ihevce_ed_4x4_find_best_modes()
* ihevce_ed_calc_4x4_blk()
* ihevce_ed_calc_8x8_blk()
* ihevce_ed_calc_incomplete_ctb()
* ihevce_cu_level_qp_mod()
* ihevce_ed_calc_ctb()
* ihevce_ed_frame_init()
* ihevce_scale_by_2()
* ihevce_decomp_pre_intra_process_row()
* ihevce_decomp_pre_intra_process()
* ihevce_decomp_pre_intra_get_num_mem_recs()
* ihevce_decomp_pre_intra_get_mem_recs()
* ihevce_decomp_pre_intra_init()
* ihevce_decomp_pre_intra_frame_init()
* ihevce_merge_sort()
* ihevce_decomp_pre_intra_curr_frame_pre_intra_deinit()
*
******************************************************************************
*/
/*****************************************************************************/
/* File Includes */
/*****************************************************************************/
/* System include files */
#include <stdio.h>
#include <string.h>
#include <stdlib.h>
#include <assert.h>
#include <stdarg.h>
#include <stdint.h>
#include <math.h>
#include <limits.h>
/* User include files */
#include "ihevc_typedefs.h"
#include "itt_video_api.h"
#include "ihevce_api.h"
#include "rc_cntrl_param.h"
#include "rc_frame_info_collector.h"
#include "rc_look_ahead_params.h"
#include "ihevc_defs.h"
#include "ihevc_debug.h"
#include "ihevc_structs.h"
#include "ihevc_platform_macros.h"
#include "ihevc_deblk.h"
#include "ihevc_itrans_recon.h"
#include "ihevc_chroma_itrans_recon.h"
#include "ihevc_chroma_intra_pred.h"
#include "ihevc_intra_pred.h"
#include "ihevc_inter_pred.h"
#include "ihevc_mem_fns.h"
#include "ihevc_padding.h"
#include "ihevc_weighted_pred.h"
#include "ihevc_sao.h"
#include "ihevc_resi_trans.h"
#include "ihevc_quant_iquant_ssd.h"
#include "ihevc_cabac_tables.h"
#include "ihevce_defs.h"
#include "ihevce_hle_interface.h"
#include "ihevce_lap_enc_structs.h"
#include "ihevce_multi_thrd_structs.h"
#include "ihevce_multi_thrd_funcs.h"
#include "ihevce_me_common_defs.h"
#include "ihevce_had_satd.h"
#include "ihevce_error_codes.h"
#include "ihevce_bitstream.h"
#include "ihevce_cabac.h"
#include "ihevce_rdoq_macros.h"
#include "ihevce_function_selector.h"
#include "ihevce_enc_structs.h"
#include "ihevce_entropy_structs.h"
#include "ihevce_cmn_utils_instr_set_router.h"
#include "ihevce_ipe_instr_set_router.h"
#include "ihevce_decomp_pre_intra_structs.h"
#include "ihevce_decomp_pre_intra_pass.h"
#include "ihevce_enc_loop_structs.h"
#include "hme_datatype.h"
#include "hme_interface.h"
#include "hme_common_defs.h"
#include "ihevce_global_tables.h"
/*****************************************************************************/
/* Global variables */
/*****************************************************************************/
/**
*****************************************************************************
* @brief subset of intra modes to be evaluated during pre enc intra process
*****************************************************************************
*/
static const UWORD8 gau1_modes_to_eval[11] = { 0, 1, 26, 2, 6, 10, 14, 18, 22, 30, 34 };
/**
*****************************************************************************
* @brief list of pointers to luma intra pred functions
*****************************************************************************
*/
pf_intra_pred g_apf_lum_ip[NUM_IP_FUNCS];
/*****************************************************************************/
/* Function Definitions */
/*****************************************************************************/
/*!
******************************************************************************
* \if Function name : ihevce_intra_populate_mode_bits_cost \endif
*
* \brief: look-up table of cost of signalling an intra mode in the
* bitstream
*
*****************************************************************************
*/
static void ihevce_intra_populate_mode_bits_cost(UWORD16 *mode_bits_cost, WORD32 lambda)
{
WORD32 i;
// 5.5 * lambda
UWORD16 five_bits_cost = COMPUTE_RATE_COST_CLIP30(11, lambda, (LAMBDA_Q_SHIFT + 1));
for(i = 0; i < NUM_MODES; i++)
{
mode_bits_cost[i] = five_bits_cost;
}
}
/*!
******************************************************************************
* \if Function name : ihevce_8x8_sad_computer \endif
*
* \brief: compute sad between 2 8x8 blocks
*
*****************************************************************************
*/
UWORD16 ihevce_8x8_sad_computer(UWORD8 *src, UWORD8 *pred, WORD32 src_strd, WORD32 pred_strd)
{
UWORD16 sad = 0;
WORD32 i, j;
for(i = 0; i < 8; i++)
{
for(j = 0; j < 8; j++)
{
sad += ABS(src[j] - pred[j]);
}
src += src_strd;
pred += pred_strd;
}
return sad;
}
/*!
******************************************************************************
* \if Function name : ihevce_4x4_sad_computer \endif
*
* \brief: compute sad between 2 4x4 blocks
*
*****************************************************************************
*/
UWORD16 ihevce_4x4_sad_computer(UWORD8 *src, UWORD8 *pred, WORD32 src_strd, WORD32 pred_strd)
{
UWORD16 sad = 0;
WORD32 i, j;
for(i = 0; i < 4; i++)
{
for(j = 0; j < 4; j++)
{
sad += ABS(src[j] - pred[j]);
}
src += src_strd;
pred += pred_strd;
}
return sad;
}
/*!
******************************************************************************
* \if Function name : ihevce_ed_4x4_find_best_modes \endif
*
* \brief: evaluate input 4x4 block for pre-selected list intra modes and
* return best sad, cost
*
*****************************************************************************
*/
void ihevce_ed_4x4_find_best_modes(
UWORD8 *pu1_src,
WORD32 src_stride,
UWORD8 *ref,
UWORD16 *mode_bits_cost,
UWORD8 *pu1_best_modes,
WORD32 *pu1_best_sad_costs,
WORD32 u1_low_resol,
FT_SAD_COMPUTER *pf_4x4_sad_computer)
{
WORD32 i;
UWORD8 mode = 0, best_amode = 0, best_nmode = 0;
UWORD8 pred[16];
WORD32 sad = 0;
WORD32 sad_cost = 0;
WORD32 best_asad_cost = 0xFFFFF;
WORD32 best_nsad_cost = 0xFFFFF;
/* If lower layers, l1 or l2, all the 11 modes are evaluated */
/* If L0 layer, all modes excluding DC and Planar are evaluated */
if(1 == u1_low_resol)
i = 0;
else
i = 2;
/* Find the best non-angular and angular mode till level 4 */
for(; i < 11; i++)
{
mode = gau1_modes_to_eval[i];
g_apf_lum_ip[g_i4_ip_funcs[mode]](&ref[0], 0, &pred[0], 4, 4, mode);
sad = pf_4x4_sad_computer(pu1_src, pred, src_stride, 4);
sad_cost = sad + mode_bits_cost[mode];
if(mode < 2)
{
if(sad_cost < best_nsad_cost)
{
best_nmode = mode;
best_nsad_cost = sad_cost;
}
}
else
{
if(sad_cost < best_asad_cost)
{
best_amode = mode;
best_asad_cost = sad_cost;
}
}
}
pu1_best_modes[0] = best_amode;
pu1_best_sad_costs[0] = best_asad_cost;
if(1 == u1_low_resol)
{
pu1_best_modes[1] = best_nmode;
pu1_best_sad_costs[1] = best_nsad_cost;
}
}
/*!
******************************************************************************
* \if Function name : ihevce_ed_calc_4x4_blk \endif
*
* \brief: evaluate input 4x4 block for all intra modes and return best sad &
* cost
*
*****************************************************************************
*/
static void ihevce_ed_calc_4x4_blk(
ihevce_ed_blk_t *ps_ed,
UWORD8 *pu1_src,
WORD32 src_stride,
UWORD8 *ref,
UWORD16 *mode_bits_cost,
WORD32 *pi4_best_satd,
WORD32 i4_quality_preset,
WORD32 *pi4_best_sad_cost,
ihevce_ipe_optimised_function_list_t *ps_ipe_optimised_function_list)
{
WORD32 i, i_end;
UWORD8 mode, best_amode, best_nmode;
UWORD8 pred[16];
UWORD16 sad;
WORD32 sad_cost = 0;
WORD32 best_asad_cost = 0xFFFFF;
WORD32 best_nsad_cost = 0xFFFFF;
UWORD8 au1_best_modes[2];
WORD32 ai4_best_sad_costs[2];
/* L1/L2 resolution hence low resolution enable */
const WORD32 u1_low_resol = 1;
UWORD8 modes_to_eval[2];
ps_ipe_optimised_function_list->pf_ed_4x4_find_best_modes(
pu1_src,
src_stride,
ref,
mode_bits_cost,
au1_best_modes,
ai4_best_sad_costs,
u1_low_resol,
ps_ipe_optimised_function_list->pf_4x4_sad_computer);
best_nmode = au1_best_modes[1];
best_amode = au1_best_modes[0];
best_nsad_cost = ai4_best_sad_costs[1];
best_asad_cost = ai4_best_sad_costs[0];
*pi4_best_satd = best_asad_cost - mode_bits_cost[best_amode];
/* Around best level 4 angular mode, search for best level 2 mode */
modes_to_eval[0] = best_amode - 2;
modes_to_eval[1] = best_amode + 2;
i = 0;
i_end = 2;
if(best_amode == 2)
i = 1;
else if(best_amode == 34)
i_end = 1;
for(; i < i_end; i++)
{
mode = modes_to_eval[i];
g_apf_lum_ip[g_i4_ip_funcs[mode]](&ref[0], 0, &pred[0], 4, 4, mode);
sad = ps_ipe_optimised_function_list->pf_4x4_sad_computer(pu1_src, pred, src_stride, 4);
sad_cost = sad + mode_bits_cost[mode];
if(sad_cost < best_asad_cost)
{
best_amode = mode;
best_asad_cost = sad_cost;
*pi4_best_satd = sad;
}
}
if(i4_quality_preset < IHEVCE_QUALITY_P4)
{
/* Around best level 2 angular mode, search for best level 1 mode */
modes_to_eval[0] = best_amode - 1;
modes_to_eval[1] = best_amode + 1;
i = 0;
i_end = 2;
if(best_amode == 2)
i = 1;
else if(best_amode == 34)
i_end = 1;
for(; i < i_end; i++)
{
mode = modes_to_eval[i];
g_apf_lum_ip[g_i4_ip_funcs[mode]](&ref[0], 0, &pred[0], 4, 4, mode);
sad = ps_ipe_optimised_function_list->pf_4x4_sad_computer(pu1_src, pred, src_stride, 4);
sad_cost = sad + mode_bits_cost[mode];
if(sad_cost < best_asad_cost)
{
best_amode = mode;
best_asad_cost = sad_cost;
*pi4_best_satd = sad;
}
}
}
if(best_asad_cost < best_nsad_cost)
{
ps_ed->best_mode = best_amode;
*pi4_best_sad_cost = best_asad_cost;
}
else
{
ps_ed->best_mode = best_nmode;
*pi4_best_sad_cost = best_nsad_cost;
}
ps_ed->intra_or_inter = 0;
ps_ed->merge_success = 0;
}
/*!
******************************************************************************
* \if Function name : ihevce_ed_calc_8x8_blk \endif
*
* \brief: evaluate input 8x8 block for intra modes basing on the intra mode
* decisions made at 4x4 level. This function also makes a decision whether
* to split blk in to 4x4 partitions or not.
*
*****************************************************************************
*/
static void ihevce_ed_calc_8x8_blk(
ihevce_ed_ctxt_t *ps_ed_ctxt,
ihevce_ed_blk_t *ps_ed_8x8,
UWORD8 *pu1_src,
WORD32 src_stride,
WORD32 *nbr_flags_ptr,
WORD32 lambda,
WORD32 *pi4_best_satd,
WORD32 i4_layer_id,
WORD32 i4_quality_preset,
WORD32 *pi4_best_sad_cost_8x8_l1_ipe,
WORD32 *pi4_best_sad_8x8_l1_ipe,
ihevce_ipe_optimised_function_list_t *ps_ipe_optimised_function_list,
ihevce_cmn_opt_func_t *ps_cmn_utils_optimised_function_list)
{
ihevce_ed_blk_t *ps_ed_4x4 = ps_ed_8x8;
UWORD8 *pu1_src_arr[4];
WORD32 ai4_4x4_best_sad_cost[4];
WORD32 nbr_flags_c, nbr_flags_r;
UWORD8 *pu1_src_4x4;
WORD32 i, j;
func_selector_t *ps_func_selector = ps_ed_ctxt->ps_func_selector;
ihevc_intra_pred_luma_ref_substitution_ft *pf_intra_pred_luma_ref_substitution =
ps_func_selector->ihevc_intra_pred_luma_ref_substitution_fptr;
/* linearize ref samples for ipe of 8x8 block */
nbr_flags_c = nbr_flags_ptr[0];
nbr_flags_r = nbr_flags_ptr[1];
if(CHECK_TR_AVAILABLE(nbr_flags_r))
{
SET_TR_AVAILABLE(nbr_flags_c);
}
else
{
SET_TR_UNAVAILABLE(nbr_flags_c);
}
pf_intra_pred_luma_ref_substitution(
pu1_src - src_stride - 1,
pu1_src - src_stride,
pu1_src - 1,
src_stride,
8,
nbr_flags_c,
&ps_ed_ctxt->au1_ref_8x8[0][0],
0);
for(i = 0; i < 2; i++)
{
pu1_src_4x4 = pu1_src + i * 4 * src_stride;
for(j = 0; j < 2; j++)
{
WORD32 i4_best_satd;
pu1_src_arr[i * 2 + j] = pu1_src_4x4;
nbr_flags_c = nbr_flags_ptr[i * 8 + j];
/* linearize ref samples for ipe of 4x4 block */
pf_intra_pred_luma_ref_substitution(
pu1_src_4x4 - src_stride - 1,
pu1_src_4x4 - src_stride,
pu1_src_4x4 - 1,
src_stride,
4,
nbr_flags_c,
&ps_ed_ctxt->au1_ref_full_ctb[i * 2 + j][0],
0);
/* populates mode bits cost */
ihevce_intra_populate_mode_bits_cost(
&ps_ed_ctxt->au2_mode_bits_cost_full_ctb[i * 2 + j][0], lambda);
ihevce_ed_calc_4x4_blk(
ps_ed_4x4,
pu1_src_4x4,
src_stride,
&ps_ed_ctxt->au1_ref_full_ctb[i * 2 + j][0],
&ps_ed_ctxt->au2_mode_bits_cost_full_ctb[i * 2 + j][0],
&i4_best_satd,
i4_quality_preset,
&ai4_4x4_best_sad_cost[i * 2 + j],
ps_ipe_optimised_function_list);
pu1_src_4x4 += 4;
ps_ed_4x4 += 1;
}
}
/* 8x8 merge */
{
UWORD8 pred[64];
WORD32 merge_success;
WORD32 sad, satd, cost;
UWORD16 u2_sum_best_4x4_sad_cost = 0;
UWORD16 u2_sum_best_4x4_satd_cost = 0;
WORD32 i4_best_8x8_sad, i4_best_8x8_satd = 0;
UWORD16 u2_best_8x8_cost = (UWORD16)(-1);
UWORD8 u1_best_8x8_mode;
UWORD8 modes_to_eval[6];
UWORD8 u1_cond_4x4_satd;
UWORD8 mode;
/* init */
ps_ed_4x4 = ps_ed_8x8;
u1_best_8x8_mode = mode = ps_ed_4x4[0].best_mode;
merge_success =
(((ps_ed_4x4[0].best_mode == ps_ed_4x4[1].best_mode) +
(ps_ed_4x4[0].best_mode == ps_ed_4x4[2].best_mode) +
(ps_ed_4x4[0].best_mode == ps_ed_4x4[3].best_mode)) == 3);
*pi4_best_satd = 0;
for(i = 0; i < 4; i++)
{
u2_sum_best_4x4_sad_cost += ai4_4x4_best_sad_cost[i];
modes_to_eval[i] = ps_ed_4x4[i].best_mode;
}
u1_cond_4x4_satd = ((1 == i4_layer_id) || (!merge_success && i4_quality_preset < IHEVCE_QUALITY_P4));
if(u1_cond_4x4_satd)
{
/* Get SATD for 4x4 blocks */
for(i = 0; i < 4; i++)
{
mode = modes_to_eval[i];
g_apf_lum_ip[g_i4_ip_funcs[mode]](
&ps_ed_ctxt->au1_ref_full_ctb[i][0], 0, &pred[0], 4, 4, mode);
satd = ps_cmn_utils_optimised_function_list->pf_HAD_4x4_8bit(
pu1_src_arr[i], src_stride, &pred[0], 4, NULL, 0);
(ps_ed_4x4 + i)->i4_4x4_satd = satd;
u2_sum_best_4x4_satd_cost +=
(satd + ps_ed_ctxt->au2_mode_bits_cost_full_ctb[i][mode]);
*pi4_best_satd += satd;
}
}
if(!merge_success)
{
UWORD8 i1_start; /* no of modes to evaluate */
UWORD8 ai1_modes[6];
WORD32 i4_merge_success_stage2 = 0;
/* Prepare 6 candidates for 8x8 block. Two are DC and planar */
ai1_modes[4] = 0;
ai1_modes[5] = 1;
i1_start = 4;
/* Assign along with removing duplicates rest 4 candidates. */
for(i = 3; i >= 0; i--)
{
WORD8 i1_fresh_mode_flag = 1;
mode = modes_to_eval[i];
/* Check if duplicate already exists in ai1_modes */
for(j = i1_start; j < 6; j++)
{
if(mode == ai1_modes[j])
i1_fresh_mode_flag = 0;
}
if(i1_fresh_mode_flag)
{
i1_start--;
ai1_modes[i1_start] = mode;
}
}
if(i4_quality_preset < IHEVCE_QUALITY_P4)
{
// 7.5 * lambda to incorporate transform flags
u2_sum_best_4x4_satd_cost +=
(COMPUTE_RATE_COST_CLIP30(12, lambda, (LAMBDA_Q_SHIFT + 1)));
/* loop over all modes for calculating SATD */
for(i = i1_start; i < 6; i++)
{
mode = ai1_modes[i];
g_apf_lum_ip[g_i4_ip_funcs[mode]](
&ps_ed_ctxt->au1_ref_8x8[0][0], 0, &pred[0], 8, 8, mode);
satd = ps_cmn_utils_optimised_function_list->pf_HAD_8x8_8bit(
pu1_src_arr[0], src_stride, &pred[0], 8, NULL, 0);
cost = satd + ps_ed_ctxt->au2_mode_bits_cost_full_ctb[0][mode];
/* Update data corresponding to least 8x8 cost */
if(cost <= u2_best_8x8_cost)
{
u2_best_8x8_cost = cost;
i4_best_8x8_satd = satd;
u1_best_8x8_mode = mode;
}
}
/* 8x8 vs 4x4 decision based on SATD values */
if((u2_best_8x8_cost <= u2_sum_best_4x4_satd_cost) || (u2_best_8x8_cost <= 300))
{
i4_merge_success_stage2 = 1;
}
/* Find the SAD based cost for 8x8 block for best mode */
if(1 == i4_layer_id)
{
UWORD8 i4_best_8x8_mode = u1_best_8x8_mode;
WORD32 i4_best_8x8_sad_curr;
g_apf_lum_ip[g_i4_ip_funcs[i4_best_8x8_mode]](
&ps_ed_ctxt->au1_ref_8x8[0][0], 0, &pred[0], 8, 8, i4_best_8x8_mode);
i4_best_8x8_sad_curr = ps_ipe_optimised_function_list->pf_8x8_sad_computer(
pu1_src_arr[0], &pred[0], src_stride, 8);
*pi4_best_sad_cost_8x8_l1_ipe =
i4_best_8x8_sad_curr +
ps_ed_ctxt->au2_mode_bits_cost_full_ctb[0][i4_best_8x8_mode];
*pi4_best_sad_8x8_l1_ipe = i4_best_8x8_sad_curr;
}
}
else /*If high_speed or extreme speed*/
{
// 7.5 * lambda to incorporate transform flags
u2_sum_best_4x4_sad_cost +=
(COMPUTE_RATE_COST_CLIP30(12, lambda, (LAMBDA_Q_SHIFT + 1)));
/*Loop over all modes for calculating SAD*/
for(i = i1_start; i < 6; i++)
{
mode = ai1_modes[i];
g_apf_lum_ip[g_i4_ip_funcs[mode]](
&ps_ed_ctxt->au1_ref_8x8[0][0], 0, &pred[0], 8, 8, mode);
sad = ps_ipe_optimised_function_list->pf_8x8_sad_computer(
pu1_src_arr[0], &pred[0], src_stride, 8);
cost = sad + ps_ed_ctxt->au2_mode_bits_cost_full_ctb[0][mode];
/*Find the data correspoinding to least cost */
if(cost <= u2_best_8x8_cost)
{
u2_best_8x8_cost = cost;
i4_best_8x8_sad = sad;
u1_best_8x8_mode = mode;
}
}
/* 8x8 vs 4x4 decision based on SAD values */
if((u2_best_8x8_cost <= u2_sum_best_4x4_sad_cost) || (u2_best_8x8_cost <= 300))
{
i4_merge_success_stage2 = 1;
if(1 == i4_layer_id)
{
g_apf_lum_ip[g_i4_ip_funcs[u1_best_8x8_mode]](
&ps_ed_ctxt->au1_ref_8x8[0][0], 0, &pred[0], 8, 8, u1_best_8x8_mode);
i4_best_8x8_satd = ps_cmn_utils_optimised_function_list->pf_HAD_8x8_8bit(
pu1_src_arr[0], src_stride, &pred[0], 8, NULL, 0);
}
}
if(1 == i4_layer_id)
{
*pi4_best_sad_cost_8x8_l1_ipe = u2_best_8x8_cost;
*pi4_best_sad_8x8_l1_ipe = i4_best_8x8_sad;
}
}
if(i4_merge_success_stage2)
{
ps_ed_4x4->merge_success = 1;
ps_ed_4x4->best_merge_mode = u1_best_8x8_mode;
*pi4_best_satd = i4_best_8x8_satd;
}
}
else
{
ps_ed_4x4->merge_success = 1;
ps_ed_4x4->best_merge_mode = u1_best_8x8_mode;
if(1 == i4_layer_id)
{
mode = u1_best_8x8_mode;
g_apf_lum_ip[g_i4_ip_funcs[mode]](
&ps_ed_ctxt->au1_ref_8x8[0][0], 0, &pred[0], 8, 8, mode);
i4_best_8x8_sad = ps_ipe_optimised_function_list->pf_8x8_sad_computer(
pu1_src_arr[0], &pred[0], src_stride, 8);
*pi4_best_sad_cost_8x8_l1_ipe =
i4_best_8x8_sad + ps_ed_ctxt->au2_mode_bits_cost_full_ctb[0][mode];
*pi4_best_sad_8x8_l1_ipe = i4_best_8x8_sad;
i4_best_8x8_satd = ps_cmn_utils_optimised_function_list->pf_HAD_8x8_8bit(
pu1_src_arr[0], src_stride, &pred[0], 8, NULL, 0);
}
*pi4_best_satd = i4_best_8x8_satd;
}
}
}
/*!
******************************************************************************
* \if Function name : ihevce_ed_calc_ctb \endif
*
* \brief: performs L1/L2 8x8 and 4x4 intra mode analysis
*
*****************************************************************************
*/
void ihevce_ed_calc_ctb(
ihevce_ed_ctxt_t *ps_ed_ctxt,
ihevce_ed_blk_t *ps_ed_ctb,
ihevce_ed_ctb_l1_t *ps_ed_ctb_l1,
UWORD8 *pu1_src,
WORD32 src_stride,
WORD32 num_4x4_blks_x,
WORD32 num_4x4_blks_y,
WORD32 *nbr_flags,
WORD32 i4_layer_id,
ihevce_ipe_optimised_function_list_t *ps_ipe_optimised_function_list,
ihevce_cmn_opt_func_t *ps_cmn_utils_optimised_function_list)
{
ihevce_ed_blk_t *ps_ed_8x8;
UWORD8 *pu1_src_8x8;
WORD32 *nbr_flags_ptr;
WORD32 lambda = ps_ed_ctxt->lambda;
WORD32 i, j;
WORD32 z_scan_idx = 0;
WORD32 z_scan_act_idx = 0;
if(i4_layer_id == 1)
{
WORD32 i4_i;
for(i4_i = 0; i4_i < 64; i4_i++)
{
(ps_ed_ctb + i4_i)->i4_4x4_satd = -1;
}
for(i4_i = 0; i4_i < 16; i4_i++)
{
ps_ed_ctb_l1->i4_sum_4x4_satd[i4_i] = -2;
ps_ed_ctb_l1->i4_min_4x4_satd[i4_i] = 0x7FFFFFFF;
ps_ed_ctb_l1->i4_8x8_satd[i4_i][0] = -2;
ps_ed_ctb_l1->i4_8x8_satd[i4_i][1] = -2;
}
for(i4_i = 0; i4_i < 4; i4_i++)
{
ps_ed_ctb_l1->i4_16x16_satd[i4_i][0] = -2;
ps_ed_ctb_l1->i4_16x16_satd[i4_i][1] = -2;
ps_ed_ctb_l1->i4_16x16_satd[i4_i][2] = -2;
}
ps_ed_ctb_l1->i4_32x32_satd[0][0] = -2;
ps_ed_ctb_l1->i4_32x32_satd[0][1] = -2;
ps_ed_ctb_l1->i4_32x32_satd[0][2] = -2;
ps_ed_ctb_l1->i4_32x32_satd[0][3] = -2;
for(i4_i = 0; i4_i < 16; i4_i++)
{
ps_ed_ctb_l1->i4_best_sad_cost_8x8_l1_me[i4_i] = -1;
ps_ed_ctb_l1->i4_sad_cost_me_for_ref[i4_i] = -1;
ps_ed_ctb_l1->i4_sad_me_for_ref[i4_i] = -1;
ps_ed_ctb_l1->i4_best_sad_8x8_l1_me[i4_i] = -1;
ps_ed_ctb_l1->i4_best_sad_8x8_l1_me_for_decide[i4_i] = -1;
ps_ed_ctb_l1->i4_best_satd_8x8[i4_i] = -1;
ps_ed_ctb_l1->i4_best_sad_cost_8x8_l1_ipe[i4_i] = -1;
ps_ed_ctb_l1->i4_best_sad_8x8_l1_ipe[i4_i] = -1;
}
}
ASSERT((num_4x4_blks_x & 1) == 0);
ASSERT((num_4x4_blks_y & 1) == 0);
for(i = 0; i < num_4x4_blks_y / 2; i++)
{
pu1_src_8x8 = pu1_src + i * 2 * 4 * src_stride;
nbr_flags_ptr = &nbr_flags[0] + 2 * 8 * i;
for(j = 0; j < num_4x4_blks_x / 2; j++)
{
WORD32 i4_best_satd;
WORD32 i4_best_sad_cost_8x8_l1_ipe;
WORD32 i4_best_sad_8x8_l1_ipe;
z_scan_idx = gau1_ctb_raster_to_zscan[i * 2 * 16 + j * 2];
z_scan_act_idx = gau1_ctb_raster_to_zscan[i * 16 + j];
ASSERT(z_scan_act_idx <= 15);
ps_ed_8x8 = ps_ed_ctb + z_scan_idx;
ihevce_ed_calc_8x8_blk(
ps_ed_ctxt,
ps_ed_8x8,
pu1_src_8x8,
src_stride,
nbr_flags_ptr,
lambda,
&i4_best_satd,
i4_layer_id,
ps_ed_ctxt->i4_quality_preset,
&i4_best_sad_cost_8x8_l1_ipe,
&i4_best_sad_8x8_l1_ipe,
ps_ipe_optimised_function_list,
ps_cmn_utils_optimised_function_list);
ASSERT(i4_best_satd >= 0);
if(i4_layer_id == 1)
{
ps_ed_ctb_l1->i4_best_sad_cost_8x8_l1_ipe[z_scan_act_idx] =
i4_best_sad_cost_8x8_l1_ipe;
ps_ed_ctb_l1->i4_best_sad_8x8_l1_ipe[z_scan_act_idx] = i4_best_sad_8x8_l1_ipe;
ps_ed_ctb_l1->i4_best_satd_8x8[z_scan_act_idx] = i4_best_satd;
ps_ed_ctxt->i8_sum_best_satd += i4_best_satd;
ps_ed_ctxt->i8_sum_sq_best_satd += (i4_best_satd * i4_best_satd);
}
pu1_src_8x8 += 8;
nbr_flags_ptr += 2;
}
}
}
float fast_log2(float val)
{
union { float val; int32_t x; } u = { val };
float log_2 = (float)(((u.x >> 23) & 255) - 128);
u.x &= ~(255 << 23);
u.x += 127 << 23;
log_2 += ((-1.0f / 3) * u.val + 2) * u.val - 2.0f / 3;
return log_2;
}
/*!
******************************************************************************
* \if Function name : ihevce_cu_level_qp_mod \endif
*
* \brief: Performs CU level QP modulation
*
*****************************************************************************
*/
WORD32 ihevce_cu_level_qp_mod(
WORD32 frm_qscale,
WORD32 cu_satd,
long double frm_avg_activity,
float f_mod_strength,
WORD32 *pi4_act_factor,
WORD32 *pi4_q_scale_mod,
rc_quant_t *rc_quant_ctxt)
{
WORD32 cu_qscale;
WORD32 cu_qp;
*pi4_act_factor = (1 << QP_LEVEL_MOD_ACT_FACTOR);
if(cu_satd != -1 && (WORD32)frm_avg_activity != 0)
{
ULWORD64 sq_cur_satd = (cu_satd * cu_satd);
float log2_sq_cur_satd = fast_log2(1 + sq_cur_satd);
WORD32 qp_offset = f_mod_strength * (log2_sq_cur_satd - frm_avg_activity);
ASSERT(USE_SQRT_AVG_OF_SATD_SQR);
qp_offset = CLIP3(qp_offset, MIN_QP_MOD_OFFSET, MAX_QP_MOD_OFFSET);
*pi4_act_factor *= gad_look_up_activity[qp_offset + ABS(MIN_QP_MOD_OFFSET)];
ASSERT(*pi4_act_factor > 0);
cu_qscale = ((frm_qscale * (*pi4_act_factor)) + (1 << (QP_LEVEL_MOD_ACT_FACTOR - 1)));
cu_qscale >>= QP_LEVEL_MOD_ACT_FACTOR;
}
else
{
cu_qscale = frm_qscale;
}
cu_qscale = CLIP3(cu_qscale, rc_quant_ctxt->i2_min_qscale, rc_quant_ctxt->i2_max_qscale);
cu_qp = rc_quant_ctxt->pi4_qscale_to_qp[cu_qscale];
cu_qp = CLIP3(cu_qp, rc_quant_ctxt->i2_min_qp, rc_quant_ctxt->i2_max_qp);
*pi4_q_scale_mod = cu_qscale;
return (cu_qp);
}
/*!
******************************************************************************
* \if Function name : ihevce_ed_frame_init \endif
*
* \brief: Initialize frame context for early decision
*
*****************************************************************************
*/
void ihevce_ed_frame_init(void *pv_ed_ctxt, WORD32 i4_layer_no)
{
ihevce_ed_ctxt_t *ps_ed_ctxt = (ihevce_ed_ctxt_t *)pv_ed_ctxt;
g_apf_lum_ip[IP_FUNC_MODE_0] = ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_planar_fptr;
g_apf_lum_ip[IP_FUNC_MODE_1] = ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_dc_fptr;
g_apf_lum_ip[IP_FUNC_MODE_2] = ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_mode2_fptr;
g_apf_lum_ip[IP_FUNC_MODE_3TO9] =
ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_mode_3_to_9_fptr;
g_apf_lum_ip[IP_FUNC_MODE_10] = ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_horz_fptr;
g_apf_lum_ip[IP_FUNC_MODE_11TO17] =
ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_mode_11_to_17_fptr;
g_apf_lum_ip[IP_FUNC_MODE_18_34] =
ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_mode_18_34_fptr;
g_apf_lum_ip[IP_FUNC_MODE_19TO25] =
ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_mode_19_to_25_fptr;
g_apf_lum_ip[IP_FUNC_MODE_26] = ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_ver_fptr;
g_apf_lum_ip[IP_FUNC_MODE_27TO33] =
ps_ed_ctxt->ps_func_selector->ihevc_intra_pred_luma_mode_27_to_33_fptr;
if(i4_layer_no == 1)
{
ps_ed_ctxt->i8_sum_best_satd = 0;
ps_ed_ctxt->i8_sum_sq_best_satd = 0;
}
}
/**
********************************************************************************
*
* @brief downscales by 2 in horz and vertical direction, creates output of
* size wd/2 * ht/2
*
* @param[in] pu1_src : source pointer
* @param[in] src_stride : source stride
* @param[out] pu1_dst : destination pointer. Starting of a row.
* @param[in] dst_stride : destination stride
* @param[in] wd : width
* @param[in] ht : height
* @param[in] pu1_wkg_mem : working memory (atleast of size CEIL16(wd) * ht))
* @param[in] ht_offset : height offset of the block to be scaled
* @param[in] block_ht : height of the block to be scaled
* @param[in] wd_offset : width offset of the block to be scaled
* @param[in] block_wd : width of the block to be scaled
*
* @return void
*
* @remarks Assumption made block_ht should me multiple of 2. LANCZOS_SCALER
*
********************************************************************************
*/
void ihevce_scaling_filter_mxn(
UWORD8 *pu1_src,
WORD32 src_strd,
UWORD8 *pu1_scrtch,
WORD32 scrtch_strd,
UWORD8 *pu1_dst,
WORD32 dst_strd,
WORD32 ht,
WORD32 wd)
{
#define FILT_TAP_Q 8
#define N_TAPS 7
const WORD16 i4_ftaps[N_TAPS] = { -18, 0, 80, 132, 80, 0, -18 };
WORD32 i, j;
WORD32 tmp;
UWORD8 *pu1_src_tmp = pu1_src - 3 * src_strd;
UWORD8 *pu1_scrtch_tmp = pu1_scrtch;
/* horizontal filtering */
for(i = -3; i < ht + 2; i++)
{
for(j = 0; j < wd; j += 2)
{
tmp = (i4_ftaps[3] * pu1_src_tmp[j] +
i4_ftaps[2] * (pu1_src_tmp[j - 1] + pu1_src_tmp[j + 1]) +
i4_ftaps[1] * (pu1_src_tmp[j + 2] + pu1_src_tmp[j - 2]) +
i4_ftaps[0] * (pu1_src_tmp[j + 3] + pu1_src_tmp[j - 3]) +
(1 << (FILT_TAP_Q - 1))) >>
FILT_TAP_Q;
pu1_scrtch_tmp[j >> 1] = CLIP_U8(tmp);
}
pu1_scrtch_tmp += scrtch_strd;
pu1_src_tmp += src_strd;
}
/* vertical filtering */
pu1_scrtch_tmp = pu1_scrtch + 3 * scrtch_strd;
for(i = 0; i < ht; i += 2)
{
for(j = 0; j < (wd >> 1); j++)
{
tmp =
(i4_ftaps[3] * pu1_scrtch_tmp[j] +
i4_ftaps[2] * (pu1_scrtch_tmp[j + scrtch_strd] + pu1_scrtch_tmp[j - scrtch_strd]) +
i4_ftaps[1] *
(pu1_scrtch_tmp[j + 2 * scrtch_strd] + pu1_scrtch_tmp[j - 2 * scrtch_strd]) +
i4_ftaps[0] *
(pu1_scrtch_tmp[j + 3 * scrtch_strd] + pu1_scrtch_tmp[j - 3 * scrtch_strd]) +
(1 << (FILT_TAP_Q - 1))) >>
FILT_TAP_Q;
pu1_dst[j] = CLIP_U8(tmp);
}
pu1_dst += dst_strd;
pu1_scrtch_tmp += (scrtch_strd << 1);
}
}
void ihevce_scale_by_2(
UWORD8 *pu1_src,
WORD32 src_strd,
UWORD8 *pu1_dst,
WORD32 dst_strd,
WORD32 wd,
WORD32 ht,
UWORD8 *pu1_wkg_mem,
WORD32 ht_offset,
WORD32 block_ht,
WORD32 wd_offset,
WORD32 block_wd,
FT_COPY_2D *pf_copy_2d,
FT_SCALING_FILTER_BY_2 *pf_scaling_filter_mxn)
{
#define N_TAPS 7
#define MAX_BLK_SZ (MAX_CTB_SIZE + ((N_TAPS >> 1) << 1))
UWORD8 au1_cpy[MAX_BLK_SZ * MAX_BLK_SZ];
UWORD32 cpy_strd = MAX_BLK_SZ;
UWORD8 *pu1_cpy = au1_cpy + cpy_strd * (N_TAPS >> 1) + (N_TAPS >> 1);
UWORD8 *pu1_in, *pu1_out;
WORD32 in_strd, wkg_mem_strd;
WORD32 row_start, row_end;
WORD32 col_start, col_end;
WORD32 i, fun_select;
WORD32 ht_tmp, wd_tmp;
FT_SCALING_FILTER_BY_2 *ihevce_scaling_filters[2];
assert((wd & 1) == 0);
assert((ht & 1) == 0);
assert(block_wd <= MAX_CTB_SIZE);
assert(block_ht <= MAX_CTB_SIZE);
/* function pointers for filtering different dimensions */
ihevce_scaling_filters[0] = ihevce_scaling_filter_mxn;
ihevce_scaling_filters[1] = pf_scaling_filter_mxn;
/* handle boundary blks */
col_start = (wd_offset < (N_TAPS >> 1)) ? 1 : 0;
row_start = (ht_offset < (N_TAPS >> 1)) ? 1 : 0;
col_end = ((wd_offset + block_wd) > (wd - (N_TAPS >> 1))) ? 1 : 0;
row_end = ((ht_offset + block_ht) > (ht - (N_TAPS >> 1))) ? 1 : 0;
if(col_end && (wd % block_wd != 0))
{
block_wd = (wd % block_wd);
}
if(row_end && (ht % block_ht != 0))
{
block_ht = (ht % block_ht);
}
/* boundary blks needs to be padded, copy src to tmp buffer */
if(col_start || col_end || row_end || row_start)
{
UWORD8 *pu1_src_tmp = pu1_src + wd_offset + ht_offset * src_strd;
pu1_cpy -= (3 * (1 - col_start) + cpy_strd * 3 * (1 - row_start));
pu1_src_tmp -= (3 * (1 - col_start) + src_strd * 3 * (1 - row_start));
ht_tmp = block_ht + 3 * (1 - row_start) + 3 * (1 - row_end);
wd_tmp = block_wd + 3 * (1 - col_start) + 3 * (1 - col_end);
pf_copy_2d(pu1_cpy, cpy_strd, pu1_src_tmp, src_strd, wd_tmp, ht_tmp);
pu1_in = au1_cpy + cpy_strd * 3 + 3;
in_strd = cpy_strd;
}
else
{
pu1_in = pu1_src + wd_offset + ht_offset * src_strd;
in_strd = src_strd;
}
/*top padding*/
if(row_start)
{
UWORD8 *pu1_cpy_tmp = au1_cpy + cpy_strd * 3;
pu1_cpy = au1_cpy + cpy_strd * (3 - 1);
memcpy(pu1_cpy, pu1_cpy_tmp, block_wd + 6);
pu1_cpy -= cpy_strd;
memcpy(pu1_cpy, pu1_cpy_tmp, block_wd + 6);
pu1_cpy -= cpy_strd;
memcpy(pu1_cpy, pu1_cpy_tmp, block_wd + 6);
}
/*bottom padding*/
if(row_end)
{
UWORD8 *pu1_cpy_tmp = au1_cpy + cpy_strd * 3 + (block_ht - 1) * cpy_strd;
pu1_cpy = pu1_cpy_tmp + cpy_strd;
memcpy(pu1_cpy, pu1_cpy_tmp, block_wd + 6);
pu1_cpy += cpy_strd;
memcpy(pu1_cpy, pu1_cpy_tmp, block_wd + 6);
pu1_cpy += cpy_strd;
memcpy(pu1_cpy, pu1_cpy_tmp, block_wd + 6);
}
/*left padding*/
if(col_start)
{
UWORD8 *pu1_cpy_tmp = au1_cpy + 3;
pu1_cpy = au1_cpy;
for(i = 0; i < block_ht + 6; i++)
{
pu1_cpy[0] = pu1_cpy[1] = pu1_cpy[2] = pu1_cpy_tmp[0];
pu1_cpy += cpy_strd;
pu1_cpy_tmp += cpy_strd;
}
}
/*right padding*/
if(col_end)
{
UWORD8 *pu1_cpy_tmp = au1_cpy + 3 + block_wd - 1;
pu1_cpy = au1_cpy + 3 + block_wd;
for(i = 0; i < block_ht + 6; i++)
{
pu1_cpy[0] = pu1_cpy[1] = pu1_cpy[2] = pu1_cpy_tmp[0];
pu1_cpy += cpy_strd;
pu1_cpy_tmp += cpy_strd;
}
}
wkg_mem_strd = block_wd >> 1;
pu1_out = pu1_dst + (wd_offset >> 1);
fun_select = (block_wd % 16 == 0);
ihevce_scaling_filters[fun_select](
pu1_in, in_strd, pu1_wkg_mem, wkg_mem_strd, pu1_out, dst_strd, block_ht, block_wd);
/* Left padding of 16 for 1st block of every row */
if(wd_offset == 0)
{
UWORD8 u1_val;
WORD32 pad_wd = 16;
WORD32 pad_ht = block_ht >> 1;
UWORD8 *dst = pu1_dst;
for(i = 0; i < pad_ht; i++)
{
u1_val = dst[0];
memset(&dst[-pad_wd], u1_val, pad_wd);
dst += dst_strd;
}
}
if(wd == wd_offset + block_wd)
{
/* Right padding of (16 + (CEIL16(wd/2))-wd/2) for last block of every row */
/* Right padding is done only after processing of last block of that row is done*/
UWORD8 u1_val;
WORD32 pad_wd = 16 + CEIL16((wd >> 1)) - (wd >> 1) + 4;
WORD32 pad_ht = block_ht >> 1;
UWORD8 *dst = pu1_dst + (wd >> 1) - 1;
for(i = 0; i < pad_ht; i++)
{
u1_val = dst[0];
memset(&dst[1], u1_val, pad_wd);
dst += dst_strd;
}
if(ht_offset == 0)
{
/* Top padding of 16 is done for 1st row only after we reach end of that row */
pad_wd = dst_strd;
pad_ht = 16;
dst = pu1_dst - 16;
for(i = 1; i <= pad_ht; i++)
{
memcpy(dst - (i * dst_strd), dst, pad_wd);
}
}
/* Bottom padding of (16 + (CEIL16(ht/2)) - ht/2) is done only if we have
reached end of frame */
if(ht - ht_offset - block_ht == 0)
{
pad_wd = dst_strd;
pad_ht = 16 + CEIL16((ht >> 1)) - (ht >> 1) + 4;
dst = pu1_dst + (((block_ht >> 1) - 1) * dst_strd) - 16;
for(i = 1; i <= pad_ht; i++)
memcpy(dst + (i * dst_strd), dst, pad_wd);
}
}
}
/*!
******************************************************************************
* \if Function name : ihevce_decomp_pre_intra_process_row \endif
*
* \brief
* Row level function which down scales a given row by 2 in horz and vertical
* direction creates output of size wd/2 * ht/2. When decomposition is done
* from L1 to L2 pre intra analysis is done on L1
*
*****************************************************************************
*/
void ihevce_decomp_pre_intra_process_row(
UWORD8 *pu1_src,
WORD32 src_stride,
UWORD8 *pu1_dst_decomp,
WORD32 dst_stride,
WORD32 layer_wd,
WORD32 layer_ht,
UWORD8 *pu1_wkg_mem,
WORD32 ht_offset,
WORD32 block_ht,
WORD32 block_wd,
WORD32 num_col_blks,
WORD32 layer_no,
ihevce_ed_ctxt_t *ps_ed_ctxt,
ihevce_ed_blk_t *ps_ed_row,
ihevce_ed_ctb_l1_t *ps_ed_ctb_l1_row,
WORD32 num_4x4_blks_ctb_y,
WORD32 num_4x4_blks_last_ctb_x,
WORD32 skip_decomp,
WORD32 skip_pre_intra,
WORD32 row_block_no,
ctb_analyse_t *ps_ctb_analyse,
ihevce_ipe_optimised_function_list_t *ps_ipe_optimised_function_list,
ihevce_cmn_opt_func_t *ps_cmn_utils_optimised_function_list)
{
WORD32 do_pre_intra_analysis = ((layer_no == 1) || (layer_no == 2)) && (!skip_pre_intra);
WORD32 col_block_no;
WORD32 i, j;
if(!skip_decomp)
{
ctb_analyse_t *ps_ctb_analyse_curr = ps_ctb_analyse + row_block_no * num_col_blks;
for(col_block_no = 0; col_block_no < num_col_blks; col_block_no++)
{
ihevce_scale_by_2(
pu1_src,
src_stride,
pu1_dst_decomp,
dst_stride,
layer_wd,
layer_ht,
pu1_wkg_mem,
ht_offset,
block_ht,
block_wd * col_block_no,
block_wd,
ps_cmn_utils_optimised_function_list->pf_copy_2d,
ps_ipe_optimised_function_list->pf_scaling_filter_mxn);
/* Disable noise detection */
memset(
ps_ctb_analyse_curr->s_ctb_noise_params.au1_is_8x8Blk_noisy,
0,
sizeof(ps_ctb_analyse_curr->s_ctb_noise_params.au1_is_8x8Blk_noisy));
ps_ctb_analyse_curr->s_ctb_noise_params.i4_noise_present = 0;
ps_ctb_analyse_curr++;
}
}
if(do_pre_intra_analysis)
{
ihevce_ed_blk_t *ps_ed_ctb = ps_ed_row;
ihevce_ed_ctb_l1_t *ps_ed_ctb_l1 = ps_ed_ctb_l1_row;
WORD32 *nbr_flags_ptr = &ps_ed_ctxt->ai4_nbr_flags[0];
UWORD8 *pu1_src_pre_intra = pu1_src + (ht_offset * src_stride);
WORD32 num_4x4_blks_in_ctb = block_wd >> 2;
WORD32 src_inc_pre_intra = num_4x4_blks_in_ctb * 4;
WORD32 inc_ctb = num_4x4_blks_in_ctb * num_4x4_blks_in_ctb;
/* To analyse any given CTB we need to set the availability flags of the
* following neighbouring CTB: BL,L,TL,T,TR */
/* copy the neighbor flags for a general ctb (ctb inside the frame); not any corners */
memcpy(
ps_ed_ctxt->ai4_nbr_flags,
gau4_nbr_flags_8x8_4x4blks,
sizeof(gau4_nbr_flags_8x8_4x4blks));
/* set top flags unavailable for first ctb row */
if(ht_offset == 0)
{
for(j = 0; j < num_4x4_blks_in_ctb; j++)
{
SET_T_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j]);
SET_TR_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j]);
SET_TL_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j]);
}
}
/* set bottom left flags as not available for last row */
if(ht_offset + block_ht >= layer_ht)
{
for(j = 0; j < num_4x4_blks_in_ctb; j++)
{
SET_BL_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[(num_4x4_blks_ctb_y - 1) * 8 + j]);
}
}
/* set left flags unavailable for 1st ctb col */
for(j = 0; j < num_4x4_blks_ctb_y; j++)
{
SET_L_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j * 8]);
SET_BL_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j * 8]);
SET_TL_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j * 8]);
}
for(col_block_no = 0; col_block_no < num_col_blks; col_block_no++)
{
if(col_block_no == 1)
{
/* For the rest of the ctbs, set left flags available */
for(j = 0; j < num_4x4_blks_ctb_y; j++)
{
SET_L_AVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j * 8]);
}
for(j = 0; j < num_4x4_blks_ctb_y - 1; j++)
{
SET_BL_AVAILABLE(ps_ed_ctxt->ai4_nbr_flags[j * 8]);
SET_TL_AVAILABLE(ps_ed_ctxt->ai4_nbr_flags[(j + 1) * 8]);
}
if(ht_offset != 0)
{
SET_TL_AVAILABLE(ps_ed_ctxt->ai4_nbr_flags[0]);
}
}
if(col_block_no == num_col_blks - 1)
{
/* set top right flags unavailable for last ctb col */
for(i = 0; i < num_4x4_blks_ctb_y; i++)
{
SET_TR_UNAVAILABLE(ps_ed_ctxt->ai4_nbr_flags[i * 8 + num_4x4_blks_last_ctb_x - 1]);
}
}
/* Call intra analysis for the ctb */
ihevce_ed_calc_ctb(
ps_ed_ctxt,
ps_ed_ctb,
ps_ed_ctb_l1,
pu1_src_pre_intra,
src_stride,
(col_block_no == num_col_blks - 1) ? num_4x4_blks_last_ctb_x : num_4x4_blks_in_ctb,
num_4x4_blks_ctb_y,
nbr_flags_ptr,
layer_no,
ps_ipe_optimised_function_list,
ps_cmn_utils_optimised_function_list);
pu1_src_pre_intra += src_inc_pre_intra;
ps_ed_ctb += inc_ctb;
ps_ed_ctb_l1 += 1;
}
}
}
/*!
******************************************************************************
* \if Function name : ihevce_decomp_pre_intra_process \endif
*
* \brief
* Frame level function to decompose given layer L0 into coarser layers and
* perform intra analysis on layers below L0
*
*****************************************************************************
*/
void ihevce_decomp_pre_intra_process(
void *pv_ctxt,
ihevce_lap_output_params_t *ps_lap_out_prms,
frm_ctb_ctxt_t *ps_frm_ctb_prms,
void *pv_multi_thrd_ctxt,
WORD32 thrd_id,
WORD32 i4_ping_pong)
{
ihevce_decomp_pre_intra_master_ctxt_t *ps_master_ctxt = pv_ctxt;
ihevce_decomp_pre_intra_ctxt_t *ps_ctxt = ps_master_ctxt->aps_decomp_pre_intra_thrd_ctxt[thrd_id];
multi_thrd_ctxt_t *ps_multi_thrd = (multi_thrd_ctxt_t *)pv_multi_thrd_ctxt;
WORD32 i4_num_layers = ps_ctxt->i4_num_layers;
UWORD8 *pu1_wkg_mem = ps_ctxt->au1_wkg_mem;
ihevce_ed_ctxt_t *ps_ed_ctxt = ps_ctxt->ps_ed_ctxt;
ihevce_ed_ctb_l1_t *ps_ed_ctb_l1 = ps_ed_ctxt->ps_ed_ctb_l1;
ihevce_ed_blk_t *ps_ed;
WORD32 i4_layer_no;
WORD32 end_of_layer;
UWORD8 *pu1_src, *pu1_dst;
WORD32 src_stride, dst_stride;
WORD32 i4_layer_wd, i4_layer_ht;
WORD32 ht_offset, block_ht, row_block_no, num_row_blocks;
WORD32 block_wd, num_col_blks;
WORD32 skip_decomp, skip_pre_intra;
WORD32 inc_ctb;
ASSERT(i4_num_layers >= 3);
ps_ctxt->as_layers[0].pu1_inp = (UWORD8 *)ps_lap_out_prms->s_input_buf.pv_y_buf;
ps_ctxt->as_layers[0].i4_inp_stride = ps_lap_out_prms->s_input_buf.i4_y_strd;
ps_ctxt->as_layers[0].i4_actual_wd = ps_lap_out_prms->s_input_buf.i4_y_wd;
ps_ctxt->as_layers[0].i4_actual_ht = ps_lap_out_prms->s_input_buf.i4_y_ht;
/* This loop does decomp & intra by picking jobs from job queue */
for(i4_layer_no = 0; i4_layer_no < i4_num_layers; i4_layer_no++)
{
WORD32 idx = 0;
src_stride = ps_ctxt->as_layers[i4_layer_no].i4_inp_stride;
pu1_src = ps_ctxt->as_layers[i4_layer_no].pu1_inp;
i4_layer_wd = ps_ctxt->as_layers[i4_layer_no].i4_actual_wd;
i4_layer_ht = ps_ctxt->as_layers[i4_layer_no].i4_actual_ht;
pu1_dst = ps_ctxt->as_layers[i4_layer_no + 1].pu1_inp;
dst_stride = ps_ctxt->as_layers[i4_layer_no + 1].i4_inp_stride;
block_wd = ps_ctxt->as_layers[i4_layer_no].i4_decomp_blk_wd;
block_ht = ps_ctxt->as_layers[i4_layer_no].i4_decomp_blk_ht;
num_col_blks = ps_ctxt->as_layers[i4_layer_no].i4_num_col_blks;
num_row_blocks = ps_ctxt->as_layers[i4_layer_no].i4_num_row_blks;
inc_ctb = (block_wd >> 2) * (block_wd >> 2);
end_of_layer = 0;
skip_pre_intra = 1;
skip_decomp = 0;
if(i4_layer_no >= (ps_ctxt->i4_num_layers - 1))
{
skip_decomp = 1;
}
/* ------------ Loop over all the CTB rows & perform Decomp --------------- */
while(0 == end_of_layer)
{
job_queue_t *ps_pre_enc_job;
WORD32 num_4x4_blks_ctb_y = 0, num_4x4_blks_last_ctb_x = 0;
/* Get the current row from the job queue */
ps_pre_enc_job = (job_queue_t *)ihevce_pre_enc_grp_get_next_job(
pv_multi_thrd_ctxt, (DECOMP_JOB_LYR0 + i4_layer_no), 1, i4_ping_pong);
/* If all rows are done, set the end of layer flag to 1, */
if(NULL == ps_pre_enc_job)
{
end_of_layer = 1;
}
else
{
/* Obtain the current row's details from the job */
row_block_no = ps_pre_enc_job->s_job_info.s_decomp_job_info.i4_vert_unit_row_no;
ps_ctxt->as_layers[i4_layer_no].ai4_curr_row_no[idx] = row_block_no;
ht_offset = row_block_no * block_ht;
if(row_block_no < (num_row_blocks))
{
pu1_dst = ps_ctxt->as_layers[i4_layer_no + 1].pu1_inp +
((block_ht >> 1) * dst_stride * row_block_no);
/* call the row level processing function */
ihevce_decomp_pre_intra_process_row(
pu1_src,
src_stride,
pu1_dst,
dst_stride,
i4_layer_wd,
i4_layer_ht,
pu1_wkg_mem,
ht_offset,
block_ht,
block_wd,
num_col_blks,
i4_layer_no,
ps_ed_ctxt,
ps_ed,
ps_ed_ctb_l1,
num_4x4_blks_ctb_y,
num_4x4_blks_last_ctb_x,
skip_decomp,
skip_pre_intra,
row_block_no,
ps_ctxt->ps_ctb_analyse,
&ps_ctxt->s_ipe_optimised_function_list,
&ps_ctxt->s_cmn_opt_func);
}
idx++;
/* set the output dependency */
ihevce_pre_enc_grp_job_set_out_dep(
pv_multi_thrd_ctxt, ps_pre_enc_job, i4_ping_pong);
}
}
ps_ctxt->as_layers[i4_layer_no].i4_num_rows_processed = idx;
/* ------------ For the same rows perform preintra if required --------------- */
ihevce_ed_frame_init(ps_ed_ctxt, i4_layer_no);
if((1 == i4_layer_no) && (IHEVCE_QUALITY_P6 == ps_ctxt->i4_quality_preset))
{
WORD32 vert_ctr, ctb_ctr, i;
WORD32 ctb_ctr_blks = ps_ctxt->as_layers[1].i4_num_col_blks;
WORD32 vert_ctr_blks = ps_ctxt->as_layers[1].i4_num_row_blks;
if((ps_ctxt->i4_quality_preset == IHEVCE_QUALITY_P6) &&
(ps_lap_out_prms->i4_temporal_lyr_id > TEMPORAL_LAYER_DISABLE))
{
for(vert_ctr = 0; vert_ctr < vert_ctr_blks; vert_ctr++)
{
ihevce_ed_ctb_l1_t *ps_ed_ctb_row_l1 =
ps_ctxt->ps_ed_ctb_l1 + vert_ctr * ps_frm_ctb_prms->i4_num_ctbs_horz;
for(ctb_ctr = 0; ctb_ctr < ctb_ctr_blks; ctb_ctr++)
{
ihevce_ed_ctb_l1_t *ps_ed_ctb_curr_l1 = ps_ed_ctb_row_l1 + ctb_ctr;
for(i = 0; i < 16; i++)
{
ps_ed_ctb_curr_l1->i4_best_sad_cost_8x8_l1_ipe[i] = 0x7fffffff;
ps_ed_ctb_curr_l1->i4_best_sad_8x8_l1_ipe[i] = 0x7fffffff;
}
}
}
}
}
#if DISABLE_L2_IPE_IN_PB_L1_IN_B
if(((2 == i4_layer_no) && (ps_lap_out_prms->i4_pic_type == IV_I_FRAME ||
ps_lap_out_prms->i4_pic_type == IV_IDR_FRAME)) ||
((1 == i4_layer_no) &&
(ps_lap_out_prms->i4_temporal_lyr_id <= TEMPORAL_LAYER_DISABLE)) ||
((IHEVCE_QUALITY_P6 != ps_ctxt->i4_quality_preset) && (0 != i4_layer_no)))
#else
if((0 != i4_layer_no) &&
(1 != ((IHEVCE_QUALITY_P6 == ps_ctxt->i4_quality_preset) &&
(ps_lap_out_prms->i4_temporal_lyr_id > TEMPORAL_LAYER_DISABLE))))
#endif
{
WORD32 i4_num_rows = ps_ctxt->as_layers[i4_layer_no].i4_num_rows_processed;
ps_ed_ctxt->lambda = ps_ctxt->ai4_lambda[i4_layer_no];
if(0 == i4_layer_no)
{
ps_ed_ctxt->ps_ed_pic = NULL;
ps_ed_ctxt->ps_ed = NULL;
ps_ed_ctxt->ps_ed_ctb_l1_pic = NULL;
ps_ed_ctxt->ps_ed_ctb_l1 = NULL;
}
else if(1 == i4_layer_no)
{
ps_ed_ctxt->ps_ed_pic = ps_ctxt->ps_layer1_buf;
ps_ed_ctxt->ps_ed = ps_ctxt->ps_layer1_buf;
ps_ed_ctxt->ps_ed_ctb_l1_pic = ps_ctxt->ps_ed_ctb_l1;
ps_ed_ctxt->ps_ed_ctb_l1 = ps_ctxt->ps_ed_ctb_l1;
}
else if(2 == i4_layer_no)
{
ps_ed_ctxt->ps_ed_pic = ps_ctxt->ps_layer2_buf;
ps_ed_ctxt->ps_ed = ps_ctxt->ps_layer2_buf;
ps_ed_ctxt->ps_ed_ctb_l1_pic = NULL;
ps_ed_ctxt->ps_ed_ctb_l1 = NULL;
}
skip_decomp = 1;
skip_pre_intra = 0;
for(idx = 0; idx < i4_num_rows; idx++)
{
WORD32 num_4x4_blks_ctb_y = 0, num_4x4_blks_last_ctb_x = 0;
/* Obtain the current row's details from the job */
row_block_no = ps_ctxt->as_layers[i4_layer_no].ai4_curr_row_no[idx];
ht_offset = row_block_no * block_ht;
if(row_block_no < (num_row_blocks))
{
pu1_dst = ps_ctxt->as_layers[i4_layer_no + 1].pu1_inp +
((block_ht >> 1) * dst_stride * row_block_no);
if(i4_layer_no == 1 || i4_layer_no == 2)
{
ps_ed = ps_ed_ctxt->ps_ed + (row_block_no * inc_ctb * (num_col_blks));
ps_ed_ctb_l1 = ps_ed_ctxt->ps_ed_ctb_l1 + (row_block_no * num_col_blks);
ps_ed_ctxt->i4_quality_preset = ps_ctxt->i4_quality_preset;
num_4x4_blks_last_ctb_x = block_wd >> 2;
num_4x4_blks_ctb_y = block_ht >> 2;
if(row_block_no == num_row_blocks - 1)
{
if(i4_layer_ht % block_ht)
{
num_4x4_blks_ctb_y = ((i4_layer_ht % block_ht) + 3) >> 2;
}
}
if(i4_layer_wd % block_wd)
{
num_4x4_blks_last_ctb_x = ((i4_layer_wd % block_wd) + 3) >> 2;
}
}
/* call the row level processing function */
ihevce_decomp_pre_intra_process_row(
pu1_src,
src_stride,
pu1_dst,
dst_stride,
i4_layer_wd,
i4_layer_ht,
pu1_wkg_mem,
ht_offset,
block_ht,
block_wd,
num_col_blks,
i4_layer_no,
ps_ed_ctxt,
ps_ed,
ps_ed_ctb_l1,
num_4x4_blks_ctb_y,
num_4x4_blks_last_ctb_x,
skip_decomp,
skip_pre_intra,
row_block_no,
NULL,
&ps_ctxt->s_ipe_optimised_function_list,
&ps_ctxt->s_cmn_opt_func);
}
if(1 == i4_layer_no)
{
ps_multi_thrd->aai4_l1_pre_intra_done[i4_ping_pong][row_block_no] = 1;
}
}
for(idx = 0; idx < MAX_NUM_CTB_ROWS_FRM; idx++)
{
ps_ctxt->as_layers[i4_layer_no].ai4_curr_row_no[idx] = -1;
}
ps_ctxt->as_layers[i4_layer_no].i4_num_rows_processed = 0;
}
#if DISABLE_L2_IPE_IN_PB_L1_IN_B
if((IHEVCE_QUALITY_P6 == ps_ctxt->i4_quality_preset) &&
(((i4_layer_no == 2) && (ps_lap_out_prms->i4_pic_type == ISLICE)) ||
((i4_layer_no == 1) && (ps_lap_out_prms->i4_temporal_lyr_id > TEMPORAL_LAYER_DISABLE))))
{
WORD32 i4_num_rows = ps_ctxt->as_layers[i4_layer_no].i4_num_rows_processed;
if(1 == i4_layer_no)
{
for(idx = 0; idx < i4_num_rows; idx++)
{
row_block_no = ps_ctxt->as_layers[i4_layer_no].ai4_curr_row_no[idx];
{
ps_multi_thrd->aai4_l1_pre_intra_done[i4_ping_pong][row_block_no] = 1;
}
}
}
for(idx = 0; idx < MAX_NUM_CTB_ROWS_FRM; idx++)
{
ps_ctxt->as_layers[i4_layer_no].ai4_curr_row_no[idx] = -1;
}
ps_ctxt->as_layers[i4_layer_no].i4_num_rows_processed = 0;
}
#else
if((i4_layer_no != 0) && ((IHEVCE_QUALITY_P6 == ps_ctxt->i4_quality_preset) &&
(ps_lap_out_prms->i4_temporal_lyr_id > TEMPORAL_LAYER_DISABLE)))
{
WORD32 i4_num_rows = ps_ctxt->as_layers[i4_layer_no].i4_num_rows_processed;
for(idx = 0; idx < i4_num_rows; idx++)
{
row_block_no = ps_ctxt->as_layers[i4_layer_no].ai4_curr_row_no[idx];
if(1 == i4_layer_no)
{
ps_multi_thrd->aai4_l1_pre_intra_done[i4_ping_pong][row_block_no] = 1;
}
}
for(idx = 0; idx < MAX_NUM_CTB_ROWS_FRM; idx++)
{
ps_ctxt->as_layers[i4_layer_no].ai4_curr_row_no[idx] = -1;
}
ps_ctxt->as_layers[i4_layer_no].i4_num_rows_processed = 0;
}
#endif
}
}
/*!
************************************************************************
* \brief
* return number of records used by decomp pre intra
*
************************************************************************
*/
WORD32 ihevce_decomp_pre_intra_get_num_mem_recs(void)
{
return (NUM_DECOMP_PRE_INTRA_MEM_RECS);
}
/*!
************************************************************************
* @brief
* return each record attributes of decomp pre intra
************************************************************************
*/
WORD32 ihevce_decomp_pre_intra_get_mem_recs(
iv_mem_rec_t *ps_mem_tab, WORD32 i4_num_proc_thrds, WORD32 i4_mem_space)
{
/* memories should be requested assuming worst case requirememnts */
/* Module context structure */
ps_mem_tab[DECOMP_PRE_INTRA_CTXT].i4_mem_size = sizeof(ihevce_decomp_pre_intra_master_ctxt_t);
ps_mem_tab[DECOMP_PRE_INTRA_CTXT].e_mem_type = (IV_MEM_TYPE_T)i4_mem_space;
ps_mem_tab[DECOMP_PRE_INTRA_CTXT].i4_mem_alignment = 8;
/* Thread context structure */
ps_mem_tab[DECOMP_PRE_INTRA_THRDS_CTXT].i4_mem_size =
i4_num_proc_thrds * sizeof(ihevce_decomp_pre_intra_ctxt_t);
ps_mem_tab[DECOMP_PRE_INTRA_THRDS_CTXT].e_mem_type = (IV_MEM_TYPE_T)i4_mem_space;
ps_mem_tab[DECOMP_PRE_INTRA_THRDS_CTXT].i4_mem_alignment = 8;
/* early decision context structure */
ps_mem_tab[DECOMP_PRE_INTRA_ED_CTXT].i4_mem_size = i4_num_proc_thrds * sizeof(ihevce_ed_ctxt_t);
ps_mem_tab[DECOMP_PRE_INTRA_ED_CTXT].e_mem_type = (IV_MEM_TYPE_T)i4_mem_space;
ps_mem_tab[DECOMP_PRE_INTRA_ED_CTXT].i4_mem_alignment = 8;
return (NUM_DECOMP_PRE_INTRA_MEM_RECS);
}
/*!
************************************************************************
* @brief
* Init decomp pre intra context
************************************************************************
*/
void *ihevce_decomp_pre_intra_init(
iv_mem_rec_t *ps_mem_tab,
ihevce_static_cfg_params_t *ps_init_prms,
WORD32 i4_num_proc_thrds,
func_selector_t *ps_func_selector,
WORD32 i4_resolution_id,
UWORD8 u1_is_popcnt_available)
{
ihevce_decomp_pre_intra_master_ctxt_t *ps_mstr_ctxt = ps_mem_tab[DECOMP_PRE_INTRA_CTXT].pv_base;
ihevce_decomp_pre_intra_ctxt_t *ps_ctxt = ps_mem_tab[DECOMP_PRE_INTRA_THRDS_CTXT].pv_base;
ihevce_ed_ctxt_t *ps_ed_ctxt = ps_mem_tab[DECOMP_PRE_INTRA_ED_CTXT].pv_base;
ihevce_tgt_params_t *ps_tgt_prms = &ps_init_prms->s_tgt_lyr_prms.as_tgt_params[i4_resolution_id];
WORD32 min_cu_size = 1 << ps_init_prms->s_config_prms.i4_min_log2_cu_size;
WORD32 a_wd[MAX_NUM_HME_LAYERS], a_ht[MAX_NUM_HME_LAYERS];
WORD32 a_disp_wd[MAX_NUM_LAYERS], a_disp_ht[MAX_NUM_LAYERS];
WORD32 n_tot_layers;
WORD32 i, j, k;
/* Get the height and width of each layer */
*a_wd = ps_tgt_prms->i4_width + SET_CTB_ALIGN(ps_tgt_prms->i4_width, min_cu_size);
*a_ht = ps_tgt_prms->i4_height + SET_CTB_ALIGN(ps_tgt_prms->i4_height, min_cu_size);
n_tot_layers = hme_derive_num_layers(1, a_wd, a_ht, a_disp_wd, a_disp_ht);
ps_mstr_ctxt->i4_num_proc_thrds = i4_num_proc_thrds;
for(i = 0; i < ps_mstr_ctxt->i4_num_proc_thrds; i++)
{
ps_mstr_ctxt->aps_decomp_pre_intra_thrd_ctxt[i] = ps_ctxt;
ps_ctxt->i4_num_layers = n_tot_layers;
ps_ctxt->ps_ed_ctxt = ps_ed_ctxt;
for(j = 0; j < n_tot_layers; j++)
{
/** If CTB size= 64, decomp_blk_wd = 64 for L0, 32 for L1 , 16 for L2, 8 for L3 */
WORD32 max_ctb_size = 1 << ps_init_prms->s_config_prms.i4_max_log2_cu_size;
WORD32 decomp_blk_wd = max_ctb_size >> j;
WORD32 decomp_blk_ht = max_ctb_size >> j;
ps_ctxt->as_layers[j].i4_actual_wd = a_wd[j];
ps_ctxt->as_layers[j].i4_actual_ht = a_ht[j];
if(0 == j)
{
ps_ctxt->as_layers[j].i4_padded_ht = a_ht[j];
ps_ctxt->as_layers[j].i4_padded_wd = a_wd[j];
}
else
{
ps_ctxt->as_layers[j].i4_padded_ht = a_ht[j] + 32 + 4;
ps_ctxt->as_layers[j].i4_padded_wd = a_wd[j] + 32 + 4;
}
ps_ctxt->as_layers[j].pu1_inp = NULL;
ps_ctxt->as_layers[j].i4_inp_stride = 0;
ps_ctxt->as_layers[j].i4_decomp_blk_ht = decomp_blk_ht;
ps_ctxt->as_layers[j].i4_decomp_blk_wd = decomp_blk_wd;
ps_ctxt->as_layers[j].i4_num_row_blks = ((a_ht[j] + (decomp_blk_ht - 1)) / decomp_blk_ht);
ps_ctxt->as_layers[j].i4_num_col_blks = ((a_wd[j] + (decomp_blk_wd - 1)) / decomp_blk_wd);
for(k = 0; k < MAX_NUM_CTB_ROWS_FRM; k++)
{
ps_ctxt->as_layers[j].ai4_curr_row_no[k] = -1;
}
ps_ctxt->as_layers[j].i4_num_rows_processed = 0;
}
ps_ctxt->i4_quality_preset = ps_tgt_prms->i4_quality_preset;
if(ps_ctxt->i4_quality_preset == IHEVCE_QUALITY_P7)
{
ps_ctxt->i4_quality_preset = IHEVCE_QUALITY_P6;
}
if(ps_init_prms->s_coding_tools_prms.i4_vqet &
(1 << BITPOS_IN_VQ_TOGGLE_FOR_CONTROL_TOGGLER))
{
if(ps_init_prms->s_coding_tools_prms.i4_vqet &
(1 << BITPOS_IN_VQ_TOGGLE_FOR_ENABLING_NOISE_PRESERVATION))
{
ps_ctxt->i4_enable_noise_detection = 1;
}
else
{
ps_ctxt->i4_enable_noise_detection = 0;
}
}
else
{
ps_ctxt->i4_enable_noise_detection = 0;
}
ihevce_cmn_utils_instr_set_router(
&ps_ctxt->s_cmn_opt_func, u1_is_popcnt_available, ps_init_prms->e_arch_type);
ihevce_ipe_instr_set_router(
&ps_ctxt->s_ipe_optimised_function_list, ps_init_prms->e_arch_type);
ps_ed_ctxt->ps_func_selector = ps_func_selector;
ps_ctxt++;
ps_ed_ctxt++;
}
/* return the handle to caller */
return ((void *)ps_mstr_ctxt);
}
/*!
************************************************************************
* @brief
* Init decomp pre intra layer buffers
************************************************************************
*/
void ihevce_decomp_pre_intra_frame_init(
void *pv_ctxt,
UWORD8 **ppu1_decomp_lyr_bufs,
WORD32 *pi4_lyr_buf_stride,
ihevce_ed_blk_t *ps_layer1_buf,
ihevce_ed_blk_t *ps_layer2_buf,
ihevce_ed_ctb_l1_t *ps_ed_ctb_l1,
WORD32 i4_ol_sad_lambda_qf,
ctb_analyse_t *ps_ctb_analyse)
{
ihevce_decomp_pre_intra_master_ctxt_t *ps_master_ctxt = pv_ctxt;
ihevce_decomp_pre_intra_ctxt_t *ps_ctxt;
WORD32 i, j;
for(i = 0; i < ps_master_ctxt->i4_num_proc_thrds; i++)
{
ps_ctxt = ps_master_ctxt->aps_decomp_pre_intra_thrd_ctxt[i];
/* L0 layer (actual input) is registered in process call */
for(j = 1; j < ps_ctxt->i4_num_layers; j++)
{
ps_ctxt->as_layers[j].i4_inp_stride = pi4_lyr_buf_stride[j - 1];
ps_ctxt->as_layers[j].pu1_inp = ppu1_decomp_lyr_bufs[j - 1];
/* Populating the buffer pointers for layer1 and layer2 buffers to store the
structure for each 4x4 block after pre intra analysis on their respective layers */
if(j == 1)
{
WORD32 sad_lambda_l1 = (3 * i4_ol_sad_lambda_qf >> 2);
WORD32 temp = 1 << LAMBDA_Q_SHIFT;
WORD32 lambda = ((temp) > sad_lambda_l1) ? temp : sad_lambda_l1;
ps_ctxt->ps_layer1_buf = ps_layer1_buf;
ps_ctxt->ps_ed_ctb_l1 = ps_ed_ctb_l1;
ps_ctxt->ai4_lambda[j] = lambda;
}
else if(j == 2)
{
WORD32 sad_lambda_l2 = i4_ol_sad_lambda_qf >> 1;
WORD32 temp = 1 << LAMBDA_Q_SHIFT;
WORD32 lambda = ((temp) > sad_lambda_l2) ? temp : sad_lambda_l2;
ps_ctxt->ps_layer2_buf = ps_layer2_buf;
ps_ctxt->ai4_lambda[j] = lambda;
}
else
{
ps_ctxt->ai4_lambda[j] = -1;
}
}
/* make the ps_ctb_analyse refernce as a part of the private context */
ps_ctxt->ps_ctb_analyse = ps_ctb_analyse;
}
}
/**
*******************************************************************************
*
* @brief Merge Sort function.
*
* @par Description:
* This function sorts the data in the input array in ascending
* order using merge sort algorithm. Intermediate data obtained in
* merge sort are stored in output 2-D array.
*
* @param[in]
* pi4_input_val : Input 1-D array
* aai4_output_val: Output 2-D array containing elements sorted in sets of
* 4,16,64 etc.
* i4_length : length of the array
* i4_ip_sort_level: Input sort level. Specifies the level upto which array is sorted.
* It should be 1 if the array is unsorted. Should be 4 if array is sorted
* in sets of 4.
* i4_op_sort_level: Output sort level. Specify the level upto which sorting is required.
* If it is given as length of array it sorts for whole array.
*
*******************************************************************************
*/
void ihevce_merge_sort(
WORD32 *pi4_input_val,
WORD32 aai4_output_val[][64],
WORD32 i4_length,
WORD32 i4_ip_sort_level,
WORD32 i4_op_sort_level)
{
WORD32 i, j, k;
WORD32 count, level;
WORD32 temp[64];
WORD32 *pi4_temp_buf_cpy;
WORD32 *pi4_temp = &temp[0];
WORD32 calc_level;
pi4_temp_buf_cpy = pi4_temp;
GETRANGE(calc_level, i4_op_sort_level / i4_ip_sort_level);
calc_level = calc_level - 1;
/*** This function is written under the assumption that we need only intermediate values of
sort in the range of 4,16,64 etc. ***/
ASSERT((calc_level % 2) == 0);
/** One iteration of this for loop does 1 sets of sort and produces one intermediate value in 2 iterations **/
for(level = 0; level < calc_level; level++)
{
/** Merges adjacent sets of elements based on current sort level **/
for(count = 0; count < i4_length; (count = count + (i4_ip_sort_level * 2)))
{
i = 0;
j = 0;
if(pi4_input_val[i4_ip_sort_level - 1] < pi4_input_val[i4_ip_sort_level])
{
/*** Condition for early exit ***/
memcpy(&pi4_temp[0], pi4_input_val, sizeof(WORD32) * i4_ip_sort_level * 2);
}
else
{
for(k = 0; k < (i4_ip_sort_level * 2); k++)
{
if((i < i4_ip_sort_level) && (j < i4_ip_sort_level))
{
if(pi4_input_val[i] > pi4_input_val[j + i4_ip_sort_level])
{
/** copy to output array **/
pi4_temp[k] = pi4_input_val[j + i4_ip_sort_level];
j++;
}
else
{
/** copy to output array **/
pi4_temp[k] = pi4_input_val[i];
i++;
}
}
else if(i == i4_ip_sort_level)
{
/** copy the remaining data to output array **/
pi4_temp[k] = pi4_input_val[j + i4_ip_sort_level];
j++;
}
else
{
/** copy the remaining data to output array **/
pi4_temp[k] = pi4_input_val[i];
i++;
}
}
}
pi4_input_val += (i4_ip_sort_level * 2);
pi4_temp += (i4_ip_sort_level * 2);
}
pi4_input_val = pi4_temp - i4_length;
if(level % 2)
{
/** Assign a temp address for storing next sort level output as we will not need this data as output **/
pi4_temp = pi4_temp_buf_cpy;
}
else
{
/** Assign address for storing the intermediate data into output 2-D array **/
pi4_temp = aai4_output_val[level / 2];
}
i4_ip_sort_level *= 2;
}
}
/*!
************************************************************************
* @brief
* Calculate the average activities at 16*16 (8*8 in L1) and 32*32
* (8*8 in L2) block sizes. As this function accumulates activities
* across blocks of a frame, this needs to be called by only one thread
* and only after ensuring the processing of entire frame is done
************************************************************************
*/
void ihevce_decomp_pre_intra_curr_frame_pre_intra_deinit(
void *pv_pre_intra_ctxt,
pre_enc_me_ctxt_t *ps_curr_out,
frm_ctb_ctxt_t *ps_frm_ctb_prms)
{
ihevce_decomp_pre_intra_master_ctxt_t *ps_master_ctxt = pv_pre_intra_ctxt;
ihevce_decomp_pre_intra_ctxt_t *ps_ctxt = ps_master_ctxt->aps_decomp_pre_intra_thrd_ctxt[0];
ULWORD64 u8_frame_8x8_sum_act_sqr = 0;
LWORD64 ai8_frame_8x8_sum_act_sqr[2] = { 0, 0 };
WORD32 ai4_frame_8x8_sum_act[2] = { 0, 0 };
WORD32 ai4_frame_8x8_sum_blks[2] = { 0, 0 };
LWORD64 ai8_frame_16x16_sum_act_sqr[3] = { 0, 0, 0 };
WORD32 ai4_frame_16x16_sum_act[3] = { 0, 0, 0 };
WORD32 ai4_frame_16x16_sum_blks[3] = { 0, 0, 0 };
LWORD64 ai8_frame_32x32_sum_act_sqr[3] = { 0, 0, 0 };
WORD32 ai4_frame_32x32_sum_act[3] = { 0, 0, 0 };
WORD32 ai4_frame_32x32_sum_blks[3] = { 0, 0, 0 };
ihevce_ed_ctb_l1_t *ps_ed_ctb_pic_l1 = ps_curr_out->ps_ed_ctb_l1;
ihevce_ed_blk_t *ps_ed_blk_l1 = ps_curr_out->ps_layer1_buf;
WORD32 ctb_wd = ps_ctxt->as_layers[1].i4_decomp_blk_wd;
WORD32 h_ctb_cnt = ps_ctxt->as_layers[1].i4_num_col_blks;
WORD32 v_ctb_cnt = ps_ctxt->as_layers[1].i4_num_row_blks;
WORD32 sub_blk_cnt = ((ctb_wd >> 2) * (ctb_wd >> 2));
WORD32 i4_avg_noise_satd;
WORD32 ctb_ctr, vert_ctr;
WORD32 i, j, k;
{
/* Calculate min noise threshold */
/* Min noise threshold is calculated by taking average of lowest 1% satd val in
* the complete 4x4 frame satds */
#define MAX_SATD 64
#define SATD_NOISE_FLOOR_THRESHOLD 16
#define MIN_BLKS 2
WORD32 i4_layer_wd = ps_ctxt->as_layers[1].i4_actual_wd;
WORD32 i4_layer_ht = ps_ctxt->as_layers[1].i4_actual_ht;
WORD32 i4_min_blk = ((MIN_BLKS * (i4_layer_wd >> 1) * (i4_layer_ht >> 1)) / 100);
WORD32 i4_total_blks = 0;
WORD32 satd_hist[MAX_SATD];
LWORD64 i8_acc_satd = 0;
memset(satd_hist, 0, sizeof(satd_hist));
for(i = 0; i < sub_blk_cnt * h_ctb_cnt * v_ctb_cnt; i++)
{
if(ps_ed_blk_l1[i].i4_4x4_satd >= 0 && ps_ed_blk_l1[i].i4_4x4_satd < MAX_SATD)
{
satd_hist[ps_ed_blk_l1[i].i4_4x4_satd]++;
}
}
for(i = 0; i < MAX_SATD && i4_total_blks <= i4_min_blk; i++)
{
i4_total_blks += satd_hist[i];
i8_acc_satd += (i * satd_hist[i]);
}
if(i4_total_blks < i4_min_blk)
{
i4_avg_noise_satd = SATD_NOISE_FLOOR_THRESHOLD;
}
else
{
i4_avg_noise_satd = (WORD32)(i8_acc_satd + (i4_total_blks >> 1)) / i4_total_blks;
}
ps_curr_out->i4_avg_noise_thrshld_4x4 = i4_avg_noise_satd;
}
for(vert_ctr = 0; vert_ctr < v_ctb_cnt; vert_ctr++)
{
ihevce_ed_ctb_l1_t *ps_ed_ctb_row_l1 =
ps_ed_ctb_pic_l1 + vert_ctr * ps_frm_ctb_prms->i4_num_ctbs_horz;
ihevce_ed_blk_t *ps_ed = ps_ed_blk_l1 + (vert_ctr * sub_blk_cnt * h_ctb_cnt);
for(ctb_ctr = 0; ctb_ctr < h_ctb_cnt; ctb_ctr++, ps_ed += sub_blk_cnt)
{
ihevce_ed_ctb_l1_t *ps_ed_ctb_curr_l1 = ps_ed_ctb_row_l1 + ctb_ctr;
WORD8 b8_satd_eval[4];
WORD32 ai4_satd_4x4[64];
WORD32 ai4_satd_8x8[16]; // derived from accumulating 4x4 satds
WORD32 ai4_satd_16x16[4] = { 0 }; // derived from accumulating 8x8 satds
WORD32 i4_satd_32x32 = 0; // derived from accumulating 8x8 satds
/* This 2-D array will contain 4x4 satds sorted in ascending order in sets
* of 4, 16, 64 For example : '5 10 2 7 6 12 3 1' array input will return
* '2 5 7 10 1 3 6 12' if sorted in sets of 4 */
WORD32 aai4_sort_4_16_64_satd[3][64];
/* This 2-D array will contain 8x8 satds sorted in ascending order in sets of
* 4, 16***/
WORD32 aai4_sort_4_16_satd[2][64];
memset(b8_satd_eval, 1, sizeof(b8_satd_eval));
for(i = 0; i < 4; i++)
{
ihevce_ed_blk_t *ps_ed_b32 = &ps_ed[i * 16];
for(j = 0; j < 4; j++)
{
ihevce_ed_blk_t *ps_ed_b16 = &ps_ed_b32[j * 4];
WORD32 satd_sum = 0;
WORD32 blk_cnt = 0;
for(k = 0; k < 4; k++)
{
ihevce_ed_blk_t *ps_ed_b4 = &ps_ed_b16[k];
if(-1 != ps_ed_b4->i4_4x4_satd)
{
#define SUB_NOISE_THRSHLD 0
#if SUB_NOISE_THRSHLD
ps_ed_b4->i4_4x4_satd = ps_ed_b4->i4_4x4_satd - i4_avg_noise_satd;
if(ps_ed_b4->i4_4x4_satd < 0)
{
ps_ed_b4->i4_4x4_satd = 0;
}
#else
if(ps_ed_b4->i4_4x4_satd < i4_avg_noise_satd)
{
ps_ed_b4->i4_4x4_satd = i4_avg_noise_satd;
}
#endif
blk_cnt++;
satd_sum += ps_ed_b4->i4_4x4_satd;
}
ai4_satd_4x4[i * 16 + j * 4 + k] = ps_ed_b4->i4_4x4_satd;
}
ASSERT(blk_cnt == 0 || blk_cnt == 4);
if(blk_cnt == 0)
{
satd_sum = -1;
}
ai4_satd_8x8[i * 4 + j] = satd_sum;
ai4_satd_16x16[i] += satd_sum;
i4_satd_32x32 += satd_sum;
ps_ed_ctb_curr_l1->i4_sum_4x4_satd[i * 4 + j] = satd_sum;
}
}
{
/* This function will sort 64 elements in array ai4_satd_4x4 in ascending order
* to 3 arrays in sets of 4, 16, 64 into the 2-D array aai4_min_4_16_64_satd */
WORD32 array_length = sizeof(ai4_satd_4x4) / sizeof(WORD32);
ihevce_merge_sort(
&ai4_satd_4x4[0], aai4_sort_4_16_64_satd, array_length, 1, 64);
/* This function will sort 64 elements in array ai4_satd_8x8 in ascending order
* to 2 arrays in sets of 4, 16 into the 2-D array aai4_sum_4_16_satd_ctb */
array_length = sizeof(ai4_satd_8x8) / sizeof(WORD32);
ihevce_merge_sort(
&ai4_satd_8x8[0], aai4_sort_4_16_satd, array_length, 1, 16);
}
/* Populate avg satd to calculate modulation index and activity factors */
/* 16x16 */
for(i = 0; i < 4; i++)
{
for(j = 0; j < 4; j++)
{
WORD32 satd_sum = ps_ed_ctb_curr_l1->i4_sum_4x4_satd[i * 4 + j];
WORD32 satd_min = aai4_sort_4_16_64_satd[0][i * 16 + j * 4 + MEDIAN_CU_TU];
ASSERT(-2 != satd_sum);
ps_ed_ctb_curr_l1->i4_min_4x4_satd[i * 4 + j] = satd_min;
if(-1 != satd_sum)
{
ps_ed_ctb_curr_l1->i4_8x8_satd[i * 4 + j][0] = satd_sum;
ps_ed_ctb_curr_l1->i4_8x8_satd[i * 4 + j][1] = satd_min;
u8_frame_8x8_sum_act_sqr += (satd_sum * satd_sum);
ai4_frame_8x8_sum_act[0] += satd_sum;
ai8_frame_8x8_sum_act_sqr[0] += (satd_sum * satd_sum);
ai4_frame_8x8_sum_blks[0] += 1;
ai4_frame_8x8_sum_act[1] += satd_min;
ai8_frame_8x8_sum_act_sqr[1] += (satd_min * satd_min);
ai4_frame_8x8_sum_blks[1] += 1;
}
else
{
ps_ed_ctb_curr_l1->i4_8x8_satd[i * 4 + j][0] = -1;
ps_ed_ctb_curr_l1->i4_8x8_satd[i * 4 + j][1] = -1;
b8_satd_eval[i] = 0;
}
}
if(b8_satd_eval[i])
{
ps_ed_ctb_curr_l1->i4_16x16_satd[i][0] = ai4_satd_16x16[i];
ps_ed_ctb_curr_l1->i4_16x16_satd[i][1] = aai4_sort_4_16_satd[0][i * 4 + MEDIAN_CU_TU];
ps_ed_ctb_curr_l1->i4_16x16_satd[i][2] = aai4_sort_4_16_64_satd[1][i * 16 + MEDIAN_CU_TU_BY_2];
for(k = 0; k < 3; k++)
{
WORD32 satd = ps_ed_ctb_curr_l1->i4_16x16_satd[i][k];
ai4_frame_16x16_sum_act[k] += satd;
ai8_frame_16x16_sum_act_sqr[k] += (satd * satd);
ai4_frame_16x16_sum_blks[k] += 1;
}
}
else
{
ps_ed_ctb_curr_l1->i4_16x16_satd[i][0] = -1;
ps_ed_ctb_curr_l1->i4_16x16_satd[i][1] = -1;
ps_ed_ctb_curr_l1->i4_16x16_satd[i][2] = -1;
}
}
/*32x32*/
if(b8_satd_eval[0] && b8_satd_eval[1] && b8_satd_eval[2] && b8_satd_eval[3])
{
WORD32 aai4_sort_4_satd[1][64];
WORD32 array_length = sizeof(ai4_satd_16x16) / sizeof(WORD32);
WORD32 satd;
/* Sort 4 elements in ascending order */
ihevce_merge_sort(ai4_satd_16x16, aai4_sort_4_satd, array_length, 1, 4);
ps_ed_ctb_curr_l1->i4_32x32_satd[0][0] = aai4_sort_4_satd[0][MEDIAN_CU_TU];
ps_ed_ctb_curr_l1->i4_32x32_satd[0][1] = aai4_sort_4_16_satd[1][MEDIAN_CU_TU_BY_2];
ps_ed_ctb_curr_l1->i4_32x32_satd[0][2] = aai4_sort_4_16_64_satd[2][MEDIAN_CU_TU_BY_4];
ps_ed_ctb_curr_l1->i4_32x32_satd[0][3] = i4_satd_32x32;
for(k = 0; k < 3; k++)
{
WORD32 satd = ps_ed_ctb_curr_l1->i4_32x32_satd[0][k];
ai4_frame_32x32_sum_act[k] += satd;
ai8_frame_32x32_sum_act_sqr[k] += (satd * satd);
ai4_frame_32x32_sum_blks[k] += 1;
}
}
else
{
ps_ed_ctb_curr_l1->i4_32x32_satd[0][0] = -1;
ps_ed_ctb_curr_l1->i4_32x32_satd[0][1] = -1;
ps_ed_ctb_curr_l1->i4_32x32_satd[0][2] = -1;
ps_ed_ctb_curr_l1->i4_32x32_satd[0][3] = -1;
}
}
}
for(i = 0; i < 2; i++)
{
/*8x8*/
#if USE_SQRT_AVG_OF_SATD_SQR
ps_curr_out->i8_curr_frame_8x8_sum_act[i] = ai8_frame_8x8_sum_act_sqr[i];
#else
ps_curr_out->i8_curr_frame_8x8_sum_act[i] = ai4_frame_8x8_sum_act[i];
#endif
ps_curr_out->i4_curr_frame_8x8_sum_act_for_strength[i] = ai4_frame_8x8_sum_act[i];
ps_curr_out->i4_curr_frame_8x8_num_blks[i] = ai4_frame_8x8_sum_blks[i];
ps_curr_out->u8_curr_frame_8x8_sum_act_sqr = u8_frame_8x8_sum_act_sqr;
/*16x16*/
#if USE_SQRT_AVG_OF_SATD_SQR
ps_curr_out->i8_curr_frame_16x16_sum_act[i] = ai8_frame_16x16_sum_act_sqr[i];
#else
ps_curr_out->i8_curr_frame_16x16_sum_act[i] = ai4_frame_16x16_sum_act[i];
#endif
ps_curr_out->i4_curr_frame_16x16_num_blks[i] = ai4_frame_16x16_sum_blks[i];
/*32x32*/
#if USE_SQRT_AVG_OF_SATD_SQR
ps_curr_out->i8_curr_frame_32x32_sum_act[i] = ai8_frame_32x32_sum_act_sqr[i];
#else
ps_curr_out->i8_curr_frame_32x32_sum_act[i] = ai4_frame_32x32_sum_act[i];
#endif
ps_curr_out->i4_curr_frame_32x32_num_blks[i] = ai4_frame_32x32_sum_blks[i];
}
/*16x16*/
#if USE_SQRT_AVG_OF_SATD_SQR
ps_curr_out->i8_curr_frame_16x16_sum_act[2] = ai8_frame_16x16_sum_act_sqr[2];
#else
ps_curr_out->i8_curr_frame_16x16_sum_act[2] = ai4_frame_16x16_sum_act[2];
#endif
ps_curr_out->i4_curr_frame_16x16_num_blks[2] = ai4_frame_16x16_sum_blks[2];
/*32x32*/
#if USE_SQRT_AVG_OF_SATD_SQR
ps_curr_out->i8_curr_frame_32x32_sum_act[2] = ai8_frame_32x32_sum_act_sqr[2];
#else
ps_curr_out->i8_curr_frame_32x32_sum_act[2] = ai4_frame_32x32_sum_act[2];
#endif
ps_curr_out->i4_curr_frame_32x32_num_blks[2] = ai4_frame_32x32_sum_blks[2];
}
/*!
************************************************************************
* @brief
* accumulate L1 intra satd across all threads.
* Note: call to this function has to be made after all threads have
* finished preintra processing
*
************************************************************************
*/
LWORD64 ihevce_decomp_pre_intra_get_frame_satd(void *pv_ctxt, WORD32 *wd, WORD32 *ht)
{
ihevce_decomp_pre_intra_master_ctxt_t *ps_master_ctxt = pv_ctxt;
ihevce_decomp_pre_intra_ctxt_t *ps_ctxt = ps_master_ctxt->aps_decomp_pre_intra_thrd_ctxt[0];
LWORD64 satd_sum = ps_ctxt->ps_ed_ctxt->i8_sum_best_satd;
WORD32 i;
*wd = ps_ctxt->as_layers[1].i4_actual_wd;
*ht = ps_ctxt->as_layers[1].i4_actual_ht;
for(i = 1; i < ps_master_ctxt->i4_num_proc_thrds; i++)
{
ps_ctxt = ps_master_ctxt->aps_decomp_pre_intra_thrd_ctxt[i];
satd_sum += ps_ctxt->ps_ed_ctxt->i8_sum_best_satd;
}
return satd_sum;
}
LWORD64 ihevce_decomp_pre_intra_get_frame_satd_squared(void *pv_ctxt, WORD32 *wd, WORD32 *ht)
{
ihevce_decomp_pre_intra_master_ctxt_t *ps_master_ctxt = pv_ctxt;
ihevce_decomp_pre_intra_ctxt_t *ps_ctxt = ps_master_ctxt->aps_decomp_pre_intra_thrd_ctxt[0];
LWORD64 satd_sum = ps_ctxt->ps_ed_ctxt->i8_sum_sq_best_satd;
WORD32 i;
*wd = ps_ctxt->as_layers[1].i4_actual_wd;
*ht = ps_ctxt->as_layers[1].i4_actual_ht;
for(i = 1; i < ps_master_ctxt->i4_num_proc_thrds; i++)
{
ps_ctxt = ps_master_ctxt->aps_decomp_pre_intra_thrd_ctxt[i];
satd_sum += ps_ctxt->ps_ed_ctxt->i8_sum_sq_best_satd;
}
return satd_sum;
}