Home | History | Annotate | Download | only in common
      1 /******************************************************************************
      2 *
      3 * Copyright (C) 2012 Ittiam Systems Pvt Ltd, Bangalore
      4 *
      5 * Licensed under the Apache License, Version 2.0 (the "License");
      6 * you may not use this file except in compliance with the License.
      7 * You may obtain a copy of the License at:
      8 *
      9 * http://www.apache.org/licenses/LICENSE-2.0
     10 *
     11 * Unless required by applicable law or agreed to in writing, software
     12 * distributed under the License is distributed on an "AS IS" BASIS,
     13 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
     14 * See the License for the specific language governing permissions and
     15 * limitations under the License.
     16 *
     17 ******************************************************************************/
     18 /**
     19  *******************************************************************************
     20  * @file
     21  *  ihevc_itrans.c
     22  *
     23  * @brief
     24  *  Contains function definitions for single stage  inverse transform
     25  *
     26  * @author
     27  *  100470
     28  *
     29  * @par List of Functions:
     30  *  - ihevc_itrans_4x4_ttype1()
     31  *  - ihevc_itrans_4x4()
     32  *  - ihevc_itrans_8x8()
     33  *  - ihevc_itrans_16x16()
     34  *  - ihevc_itrans_32x32()
     35  *
     36  * @remarks
     37  *  None
     38  *
     39  *******************************************************************************
     40  */
     41 #include <stdio.h>
     42 #include <string.h>
     43 #include "ihevc_typedefs.h"
     44 #include "ihevc_macros.h"
     45 #include "ihevc_platform_macros.h"
     46 #include "ihevc_defs.h"
     47 #include "ihevc_trans_tables.h"
     48 #include "ihevc_func_selector.h"
     49 #include "ihevc_trans_macros.h"
     50 
     51 #define NON_OPTIMIZED 1
     52 
     53 /**
     54  *******************************************************************************
     55  *
     56  * @brief
     57  *  This function performs Single stage  Inverse transform type 1 (DST) for
     58  * 4x4 input block
     59  *
     60  * @par Description:
     61  *  Performs single stage 4x4 inverse transform type 1  by utilizing the
     62  * symmetry of transformation matrix  and reducing number of multiplications
     63  * wherever  possible but keeping the number of operations
     64  * (addition,multiplication and shift)same
     65  *
     66  * @param[in] pi2_src
     67  *  Input 4x4 coefficients
     68  *
     69  * @param[out] pi2_dst
     70  *  Output 4x4 block
     71  *
     72  * @param[in] src_strd
     73  *  Input stride
     74  *
     75  * @param[in] dst_strd
     76  *  Output Stride
     77  *
     78  * @param[in] i4_shift
     79  *  Output shift
     80  *
     81  * @param[in] zero_cols
     82  *  Zero columns in pi2_src
     83  *
     84  * @returns  Void
     85  *
     86  * @remarks
     87  *  None
     88  *
     89  *******************************************************************************
     90  */
     91 
     92 
     93 void ihevc_itrans_4x4_ttype1(WORD16 *pi2_src,
     94                              WORD16 *pi2_dst,
     95                              WORD32 src_strd,
     96                              WORD32 dst_strd,
     97                              WORD32 i4_shift,
     98                              WORD32 zero_cols)
     99 {
    100     WORD32 i, c[4];
    101     WORD32 add;
    102 
    103     add = 1 << (i4_shift - 1);
    104 
    105     for(i = 0; i < TRANS_SIZE_4; i++)
    106     {
    107         /* Checking for Zero Cols */
    108         if((zero_cols & 1) == 1)
    109         {
    110             memset(pi2_dst, 0, TRANS_SIZE_4 * sizeof(WORD16));
    111         }
    112         else
    113         {
    114             // Intermediate Variables
    115             c[0] = pi2_src[0] + pi2_src[2 * src_strd];
    116             c[1] = pi2_src[2 * src_strd] + pi2_src[3 * src_strd];
    117             c[2] = pi2_src[0] - pi2_src[3 * src_strd];
    118             c[3] = 74 * pi2_src[src_strd];
    119 
    120             pi2_dst[0] =
    121                             CLIP_S16((29 * c[0] + 55 * c[1] + c[3] + add) >> i4_shift);
    122             pi2_dst[1] =
    123                             CLIP_S16((55 * c[2] - 29 * c[1] + c[3] + add) >> i4_shift);
    124             pi2_dst[2] =
    125                             CLIP_S16((74 * (pi2_src[0] - pi2_src[2 * src_strd] + pi2_src[3 * src_strd]) + add) >> i4_shift);
    126             pi2_dst[3] =
    127                             CLIP_S16((55 * c[0] + 29 * c[2] - c[3] + add) >> i4_shift);
    128         }
    129         pi2_src++;
    130         pi2_dst += dst_strd;
    131         zero_cols = zero_cols >> 1;
    132     }
    133 }
    134 
    135 
    136 /**
    137  *******************************************************************************
    138  *
    139  * @brief
    140  *  This function performs Single stage  Inverse transform for 4x4 input
    141  * block
    142  *
    143  * @par Description:
    144  *  Performs single stage 4x4 inverse transform by utilizing  the symmetry of
    145  * transformation matrix and reducing number  of multiplications wherever
    146  * possible but keeping the  number of operations(addition,multiplication and
    147  * shift)  same
    148  *
    149  * @param[in] pi2_src
    150  *  Input 4x4 coefficients
    151  *
    152  * @param[out] pi2_dst
    153  *  Output 4x4 block
    154  *
    155  * @param[in] src_strd
    156  *  Input stride
    157  *
    158  * @param[in] dst_strd
    159  *  Output Stride
    160  *
    161  * @param[in] i4_shift
    162  *  Output shift
    163  *
    164  * @param[in] zero_cols
    165  *  Zero columns in pi2_src
    166  *
    167  * @returns  Void
    168  *
    169  * @remarks
    170  *  None
    171  *
    172  *******************************************************************************
    173  */
    174 
    175 #if NON_OPTIMIZED
    176 void ihevc_itrans_4x4(WORD16 *pi2_src,
    177                       WORD16 *pi2_dst,
    178                       WORD32 src_strd,
    179                       WORD32 dst_strd,
    180                       WORD32 i4_shift,
    181                       WORD32 zero_cols)
    182 {
    183     WORD32 j;
    184     WORD32 e[2], o[2];
    185     WORD32 add;
    186 
    187     add = 1 << (i4_shift - 1);
    188 
    189     for(j = 0; j < TRANS_SIZE_4; j++)
    190     {
    191         /* Checking for Zero Cols */
    192         if((zero_cols & 1) == 1)
    193         {
    194             memset(pi2_dst, 0, TRANS_SIZE_4 * sizeof(WORD16));
    195         }
    196         else
    197         {
    198 
    199             /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
    200             o[0] = g_ai2_ihevc_trans_4[1][0] * pi2_src[src_strd]
    201                             + g_ai2_ihevc_trans_4[3][0] * pi2_src[3 * src_strd];
    202             o[1] = g_ai2_ihevc_trans_4[1][1] * pi2_src[src_strd]
    203                             + g_ai2_ihevc_trans_4[3][1] * pi2_src[3 * src_strd];
    204             e[0] = g_ai2_ihevc_trans_4[0][0] * pi2_src[0]
    205                             + g_ai2_ihevc_trans_4[2][0] * pi2_src[2 * src_strd];
    206             e[1] = g_ai2_ihevc_trans_4[0][1] * pi2_src[0]
    207                             + g_ai2_ihevc_trans_4[2][1] * pi2_src[2 * src_strd];
    208 
    209             pi2_dst[0] =
    210                             CLIP_S16(((e[0] + o[0] + add) >> i4_shift));
    211             pi2_dst[1] =
    212                             CLIP_S16(((e[1] + o[1] + add) >> i4_shift));
    213             pi2_dst[2] =
    214                             CLIP_S16(((e[1] - o[1] + add) >> i4_shift));
    215             pi2_dst[3] =
    216                             CLIP_S16(((e[0] - o[0] + add) >> i4_shift));
    217 
    218         }
    219         pi2_src++;
    220         pi2_dst += dst_strd;
    221         zero_cols = zero_cols >> 1;
    222     }
    223 }
    224 #else
    225 void ihevc_itrans_4x4(WORD16 *pi2_src,
    226                       WORD16 *pi2_dst,
    227                       WORD32 src_strd,
    228                       WORD32 dst_strd,
    229                       WORD32 i4_shift,
    230                       WORD32 zero_cols)
    231 {
    232     WORD32 j;
    233     WORD32 e[2], o[2];
    234     WORD32 add;
    235 
    236     add = 1 << (i4_shift - 1);
    237 
    238     /***************************************************************************/
    239     /* Transform Matrix 4x4                                                    */
    240     /*      0   1   2   3                                                      */
    241     /* 0 { 64, 64, 64, 64},                                                    */
    242     /* 1 { 83, 36,-36,-83},                                                    */
    243     /* 2 { 64,-64,-64, 64},                                                    */
    244     /* 3 { 36,-83, 83,-36}                                                     */
    245     /***************************************************************************/
    246 
    247     for(j = 0; j < TRANS_SIZE_4; j++)
    248     {
    249         WORD32 temp;
    250 
    251         /* Checking for Zero Cols */
    252         if((zero_cols & 1) == 1)
    253         {
    254             memset(pi2_dst, 0, TRANS_SIZE_4 * sizeof(WORD16));
    255         }
    256         else
    257         {
    258             /* Common operation in o[0] and o[1] */
    259             temp = (pi2_src[src_strd] + pi2_src[3 * src_strd]) * 36;
    260 
    261             o[0] = temp + 47 * pi2_src[src_strd];
    262             o[1] = temp - 119 * pi2_src[3 * src_strd];
    263             e[0] = (pi2_src[0] + pi2_src[2 * src_strd]) << 6;
    264             e[1] = (pi2_src[0] - pi2_src[2 * src_strd]) << 6;
    265 
    266             pi2_dst[0] =
    267                             CLIP_S16(((e[0] + o[0] + add) >> i4_shift));
    268             pi2_dst[1] =
    269                             CLIP_S16(((e[1] + o[1] + add) >> i4_shift));
    270             pi2_dst[2] =
    271                             CLIP_S16(((e[1] - o[1] + add) >> i4_shift));
    272             pi2_dst[3] =
    273                             CLIP_S16(((e[0] - o[0] + add) >> i4_shift));
    274         }
    275         pi2_src++;
    276         pi2_dst += dst_strd;
    277         zero_cols = zero_cols >> 1;
    278     }
    279 }
    280 #endif
    281 
    282 /**
    283  *******************************************************************************
    284  *
    285  * @brief
    286  *  This function performs Single stage  Inverse transform for 8x8 input
    287  * block
    288  *
    289  * @par Description:
    290  *  Performs single stage 8x8 inverse transform by utilizing  the symmetry of
    291  * transformation matrix and reducing number  of multiplications wherever
    292  * possible but keeping the  number of operations(addition,multiplication and
    293  * shift)  same
    294  *
    295  * @param[in] pi2_src
    296  *  Input 8x8 coefficients
    297  *
    298  * @param[out] pi2_dst
    299  *  Output 8x8 block
    300  *
    301  * @param[in] src_strd
    302  *  Input stride
    303  *
    304  * @param[in] dst_strd
    305  *  Output Stride
    306  *
    307  * @param[in] i4_shift
    308  *  Output shift
    309  *
    310  * @param[in] zero_cols
    311  *  Zero columns in pi2_src
    312  *
    313  * @returns  Void
    314  *
    315  * @remarks
    316  *  None
    317  *
    318  *******************************************************************************
    319  */
    320 
    321 #if NON_OPTIMIZED
    322 void ihevc_itrans_8x8(WORD16 *pi2_src,
    323                       WORD16 *pi2_dst,
    324                       WORD32 src_strd,
    325                       WORD32 dst_strd,
    326                       WORD32 i4_shift,
    327                       WORD32 zero_cols)
    328 {
    329     WORD32 j, k;
    330     WORD32 e[4], o[4];
    331     WORD32 ee[2], eo[2];
    332     WORD32 add;
    333 
    334     add = 1 << (i4_shift - 1);
    335 
    336     for(j = 0; j < TRANS_SIZE_8; j++)
    337     {
    338         /* Checking for Zero Cols */
    339         if((zero_cols & 1) == 1)
    340         {
    341             memset(pi2_dst, 0, TRANS_SIZE_8 * sizeof(WORD16));
    342         }
    343         else
    344         {
    345             /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
    346             for(k = 0; k < 4; k++)
    347             {
    348                 o[k] = g_ai2_ihevc_trans_8[1][k] * pi2_src[src_strd]
    349                                 + g_ai2_ihevc_trans_8[3][k]
    350                                                 * pi2_src[3 * src_strd]
    351                                 + g_ai2_ihevc_trans_8[5][k]
    352                                                 * pi2_src[5 * src_strd]
    353                                 + g_ai2_ihevc_trans_8[7][k]
    354                                                 * pi2_src[7 * src_strd];
    355             }
    356 
    357             eo[0] = g_ai2_ihevc_trans_8[2][0] * pi2_src[2 * src_strd]
    358                             + g_ai2_ihevc_trans_8[6][0] * pi2_src[6 * src_strd];
    359             eo[1] = g_ai2_ihevc_trans_8[2][1] * pi2_src[2 * src_strd]
    360                             + g_ai2_ihevc_trans_8[6][1] * pi2_src[6 * src_strd];
    361             ee[0] = g_ai2_ihevc_trans_8[0][0] * pi2_src[0]
    362                             + g_ai2_ihevc_trans_8[4][0] * pi2_src[4 * src_strd];
    363             ee[1] = g_ai2_ihevc_trans_8[0][1] * pi2_src[0]
    364                             + g_ai2_ihevc_trans_8[4][1] * pi2_src[4 * src_strd];
    365 
    366             /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
    367             e[0] = ee[0] + eo[0];
    368             e[3] = ee[0] - eo[0];
    369             e[1] = ee[1] + eo[1];
    370             e[2] = ee[1] - eo[1];
    371             for(k = 0; k < 4; k++)
    372             {
    373                 pi2_dst[k] =
    374                                 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
    375                 pi2_dst[k + 4] =
    376                                 CLIP_S16(((e[3 - k] - o[3 - k] + add) >> i4_shift));
    377             }
    378         }
    379         pi2_src++;
    380         pi2_dst += dst_strd;
    381         zero_cols = zero_cols >> 1;
    382     }
    383 }
    384 
    385 #else
    386 void ihevc_itrans_8x8(WORD16 *pi2_src,
    387                       WORD16 *pi2_dst,
    388                       WORD32 src_strd,
    389                       WORD32 dst_strd,
    390                       WORD32 i4_shift,
    391                       WORD32 zero_cols)
    392 {
    393     /* Transform Matrix 8x8                          */
    394     /*              0    1    2   3   4   5   6   7  */
    395     /*     0 -      64   64   64  64  64  64  64  64 */
    396     /*     1 -      89   75   50  18 -18 -50 -75 -89 */
    397     /*     2 -      83   36  -36 -83 -83 -36  36  83 */
    398     /*     3 -      75  -18  -89 -50  50  89  18 -75 */
    399     /*     4 -      64  -64  -64  64  64 -64 -64  64 */
    400     /*     5 -      50  -89   18  75 -75 -18  89 -50 */
    401     /*     6 -      36  -83   83 -36 -36  83 -83  36 */
    402     /*     7 -      18  -50   75 -89  89 -75  50 -18 */
    403 
    404     /* 0th and 4th row will have no multiplications */
    405     /* 2nd and 6th row has only two coefff multiplies */
    406     /* 1st, 3rd, 5th and 7th rows have o mirror symmetry */
    407     WORD32 j, k;
    408     WORD32 temp1, temp2;
    409     WORD32 e[4], o[4];
    410     WORD32 ee[2], eo[2];
    411     WORD32 add;
    412 
    413     add = 1 << (i4_shift - 1);
    414 
    415     for(j = 0; j < TRANS_SIZE_8; j++)
    416     {
    417         /* Checking for Zero Cols */
    418         if((zero_cols & 1) == 1)
    419         {
    420             memset(pi2_dst, 0, TRANS_SIZE_8 * sizeof(WORD16));
    421         }
    422         else
    423         {
    424 
    425             /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
    426             /*
    427              o[0] = 89 *pi2_src[8] +  75 *pi2_src[3*8] +  50 *pi2_src[5*8] +  18 *pi2_src[7*8];
    428              o[1] = 75 *pi2_src[8] + -18 *pi2_src[3*8] + -89 *pi2_src[5*8] + -50 *pi2_src[7*8];
    429              o[2] = 50 *pi2_src[8] + -89 *pi2_src[3*8] +  18 *pi2_src[5*8] +  75 *pi2_src[7*8];
    430              o[3] = 18 *pi2_src[8] + -50 *pi2_src[3*8] +  75 *pi2_src[5*8] + -89 *pi2_src[7*8];
    431              */
    432 
    433             /* Optimization: 4 mul + 2 add  ---> 3 mul + 3 add */
    434             /*
    435              temp1 = (pi2_src[8  ] + pi2_src[3*8]) * 75;
    436              temp2 = (pi2_src[5*8] + pi2_src[7*8]) * 50;
    437 
    438              o[0] = temp1 + 14 * pi2_src[8  ] + temp2 - 32 * pi2_src[7*8];
    439              o[1] = temp1 - 93 * pi2_src[3*8] - temp2 - 39 * pi2_src[5*8];
    440              */
    441 
    442             temp1 = (pi2_src[src_strd] + pi2_src[3 * src_strd]) * 75;
    443             temp2 = (pi2_src[5 * src_strd] + pi2_src[7 * src_strd]) * 50;
    444 
    445             o[0] = temp1 + 14 * pi2_src[src_strd] + temp2
    446                             - (pi2_src[7 * src_strd] << 5);
    447             o[1] = temp1 - 93 * pi2_src[3 * src_strd] - temp2
    448                             - 39 * pi2_src[5 * src_strd];
    449 
    450             /* Optimization: 4 mul + 2 add  ---> 3 mul + 3 add */
    451             /*
    452              temp1 = (pi2_src[8  ] - pi2_src[3*8]) * 50;
    453              temp2 = (pi2_src[5*8] + pi2_src[7*8]) * 75;
    454 
    455              o[2] = temp1 - 39 * pi2_src[3*8] + temp2 -  57 * pi2_src[5*8];
    456              o[3] = temp1 - 32 * pi2_src[8  ] + temp2 - 164 * pi2_src[7*8];
    457              */
    458 
    459             temp1 = (pi2_src[src_strd] - pi2_src[3 * src_strd]) * 50;
    460             temp2 = (pi2_src[5 * src_strd] + pi2_src[7 * src_strd]) * 75;
    461 
    462             o[2] = temp1 - 39 * pi2_src[3 * src_strd] + temp2
    463                             - 57 * pi2_src[5 * src_strd];
    464             o[3] = temp1 - (pi2_src[src_strd] << 5) + temp2
    465                             - 164 * pi2_src[7 * src_strd];
    466 
    467             /*
    468              eo[0] = 83 *pi2_src[ 2*8 ] +  36 *pi2_src[ 6*8 ];
    469              eo[1] = 36 *pi2_src[ 2*8 ] + -83 *pi2_src[ 6*8 ];
    470              ee[0] = 64 *pi2_src[ 0   ] +  64 *pi2_src[ 4*8 ];
    471              ee[1] = 64 *pi2_src[ 0   ] + -64 *pi2_src[ 4*8 ];
    472              */
    473 
    474             /* Optimization: 4 mul + 2 add  ---> 3 mul + 3 add */
    475             temp1 = (pi2_src[2 * src_strd] + pi2_src[6 * src_strd]) * 36;
    476             eo[0] = temp1 + 47 * pi2_src[2 * src_strd];
    477             eo[1] = temp1 - 119 * pi2_src[6 * src_strd];
    478 
    479             /* Optimization: 4 mul + 2 add  ---> 2 i4_shift + 2 add */
    480             ee[0] = (pi2_src[0] + pi2_src[4 * src_strd]) << 6;
    481             ee[1] = (pi2_src[0] - pi2_src[4 * src_strd]) << 6;
    482 
    483             e[0] = ee[0] + eo[0];
    484             e[3] = ee[0] - eo[0];
    485             e[1] = ee[1] + eo[1];
    486             e[2] = ee[1] - eo[1];
    487 
    488             for(k = 0; k < 4; k++)
    489             {
    490                 pi2_dst[k] =
    491                                 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
    492                 pi2_dst[k + 4] =
    493                                 CLIP_S16(((e[3 - k] - o[3 - k] + add) >> i4_shift));
    494             }
    495         }
    496         pi2_src++;
    497         pi2_dst += dst_strd;
    498         zero_cols = zero_cols >> 1;
    499     }
    500 
    501 }
    502 #endif
    503 
    504 
    505 /**
    506  *******************************************************************************
    507  *
    508  * @brief
    509  *  This function performs Single stage  Inverse transform for 16x16 input
    510  * block
    511  *
    512  * @par Description:
    513  *  Performs single stage 16x16 inverse transform by  utilizing the symmetry
    514  * of transformation matrix  and reducing number of multiplications wherever
    515  * possible  but keeping the number of operations  (addition,multiplication
    516  * and shift) same
    517  *
    518  * @param[in] pi2_src
    519  *  Input 16x16 coefficients
    520  *
    521  * @param[out] pi2_dst
    522  *  Output 16x16 block
    523  *
    524  * @param[in] src_strd
    525  *  Input stride
    526  *
    527  * @param[in] dst_strd
    528  *  Output Stride
    529  *
    530  * @param[in] i4_shift
    531  *  Output shift
    532  *
    533  * @param[in] zero_cols
    534  *  Zero columns in pi2_src
    535  *
    536  * @returns  Void
    537  *
    538  * @remarks
    539  *  None
    540  *
    541  *******************************************************************************
    542  */
    543 
    544 #if NON_OPTIMIZED
    545 void ihevc_itrans_16x16(WORD16 *pi2_src,
    546                         WORD16 *pi2_dst,
    547                         WORD32 src_strd,
    548                         WORD32 dst_strd,
    549                         WORD32 i4_shift,
    550                         WORD32 zero_cols)
    551 {
    552     WORD32 j, k;
    553     WORD32 e[8], o[8];
    554     WORD32 ee[4], eo[4];
    555     WORD32 eee[2], eeo[2];
    556     WORD32 add;
    557 
    558     add = 1 << (i4_shift - 1);
    559 
    560     for(j = 0; j < TRANS_SIZE_16; j++)
    561     {
    562         /* Checking for Zero Cols */
    563         if((zero_cols & 1) == 1)
    564         {
    565             memset(pi2_dst, 0, TRANS_SIZE_16 * sizeof(WORD16));
    566         }
    567         else
    568         {
    569             /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
    570             for(k = 0; k < 8; k++)
    571             {
    572                 o[k] = g_ai2_ihevc_trans_16[1][k] * pi2_src[src_strd]
    573                                 + g_ai2_ihevc_trans_16[3][k]
    574                                                 * pi2_src[3 * src_strd]
    575                                 + g_ai2_ihevc_trans_16[5][k]
    576                                                 * pi2_src[5 * src_strd]
    577                                 + g_ai2_ihevc_trans_16[7][k]
    578                                                 * pi2_src[7 * src_strd]
    579                                 + g_ai2_ihevc_trans_16[9][k]
    580                                                 * pi2_src[9 * src_strd]
    581                                 + g_ai2_ihevc_trans_16[11][k]
    582                                                 * pi2_src[11 * src_strd]
    583                                 + g_ai2_ihevc_trans_16[13][k]
    584                                                 * pi2_src[13 * src_strd]
    585                                 + g_ai2_ihevc_trans_16[15][k]
    586                                                 * pi2_src[15 * src_strd];
    587             }
    588             for(k = 0; k < 4; k++)
    589             {
    590                 eo[k] = g_ai2_ihevc_trans_16[2][k] * pi2_src[2 * src_strd]
    591                                 + g_ai2_ihevc_trans_16[6][k]
    592                                                 * pi2_src[6 * src_strd]
    593                                 + g_ai2_ihevc_trans_16[10][k]
    594                                                 * pi2_src[10 * src_strd]
    595                                 + g_ai2_ihevc_trans_16[14][k]
    596                                                 * pi2_src[14 * src_strd];
    597             }
    598             eeo[0] = g_ai2_ihevc_trans_16[4][0] * pi2_src[4 * src_strd]
    599                             + g_ai2_ihevc_trans_16[12][0]
    600                                             * pi2_src[12 * src_strd];
    601             eee[0] =
    602                             g_ai2_ihevc_trans_16[0][0] * pi2_src[0]
    603                                             + g_ai2_ihevc_trans_16[8][0]
    604                                                             * pi2_src[8
    605                                                                             * src_strd];
    606             eeo[1] = g_ai2_ihevc_trans_16[4][1] * pi2_src[4 * src_strd]
    607                             + g_ai2_ihevc_trans_16[12][1]
    608                                             * pi2_src[12 * src_strd];
    609             eee[1] =
    610                             g_ai2_ihevc_trans_16[0][1] * pi2_src[0]
    611                                             + g_ai2_ihevc_trans_16[8][1]
    612                                                             * pi2_src[8
    613                                                                             * src_strd];
    614 
    615             /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
    616             for(k = 0; k < 2; k++)
    617             {
    618                 ee[k] = eee[k] + eeo[k];
    619                 ee[k + 2] = eee[1 - k] - eeo[1 - k];
    620             }
    621             for(k = 0; k < 4; k++)
    622             {
    623                 e[k] = ee[k] + eo[k];
    624                 e[k + 4] = ee[3 - k] - eo[3 - k];
    625             }
    626             for(k = 0; k < 8; k++)
    627             {
    628                 pi2_dst[k] =
    629                                 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
    630                 pi2_dst[k + 8] =
    631                                 CLIP_S16(((e[7 - k] - o[7 - k] + add) >> i4_shift));
    632             }
    633         }
    634         pi2_src++;
    635         pi2_dst += dst_strd;
    636         zero_cols = zero_cols >> 1;
    637     }
    638 }
    639 #else
    640 void ihevc_itrans_16x16(WORD16 *pi2_src,
    641                         WORD16 *pi2_dst,
    642                         WORD32 src_strd,
    643                         WORD32 dst_strd,
    644                         WORD32 i4_shift,
    645                         WORD32 zero_cols)
    646 {
    647     WORD32 j, k;
    648     WORD32 e[8], o[8];
    649     WORD32 ee[4], eo[4];
    650     WORD32 eee[2], eeo[2];
    651     WORD32 add;
    652     WORD32 temp1, temp2;
    653 
    654     add = 1 << (i4_shift - 1);
    655     /***************************************************************************/
    656     /* Transform Matrix 16x16                                                  */
    657     /*       0   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15     */
    658     /* 0  { 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64, 64},   */
    659     /* 1  { 90, 87, 80, 70, 57, 43, 25,  9, -9,-25,-43,-57,-70,-80,-87,-90},   */
    660     /* 2  { 89, 75, 50, 18,-18,-50,-75,-89,-89,-75,-50,-18, 18, 50, 75, 89},   */
    661     /* 3  { 87, 57,  9,-43,-80,-90,-70,-25, 25, 70, 90, 80, 43, -9,-57,-87},   */
    662     /* 4  { 83, 36,-36,-83,-83,-36, 36, 83, 83, 36,-36,-83,-83,-36, 36, 83},   */
    663     /* 5  { 80,  9,-70,-87,-25, 57, 90, 43,-43,-90,-57, 25, 87, 70, -9,-80},   */
    664     /* 6  { 75,-18,-89,-50, 50, 89, 18,-75,-75, 18, 89, 50,-50,-89,-18, 75},   */
    665     /* 7  { 70,-43,-87,  9, 90, 25,-80,-57, 57, 80,-25,-90, -9, 87, 43,-70},   */
    666     /* 8  { 64,-64,-64, 64, 64,-64,-64, 64, 64,-64,-64, 64, 64,-64,-64, 64},   */
    667     /* 9  { 57,-80,-25, 90, -9,-87, 43, 70,-70,-43, 87,  9,-90, 25, 80,-57},   */
    668     /* 10 { 50,-89, 18, 75,-75,-18, 89,-50,-50, 89,-18,-75, 75, 18,-89, 50},   */
    669     /* 11 { 43,-90, 57, 25,-87, 70,  9,-80, 80, -9,-70, 87,-25,-57, 90,-43},   */
    670     /* 12 { 36,-83, 83,-36,-36, 83,-83, 36, 36,-83, 83,-36,-36, 83,-83, 36},   */
    671     /* 13 { 25,-70, 90,-80, 43,  9,-57, 87,-87, 57, -9,-43, 80,-90, 70,-25},   */
    672     /* 14 { 18,-50, 75,-89, 89,-75, 50,-18,-18, 50,-75, 89,-89, 75,-50, 18},   */
    673     /* 15 {  9,-25, 43,-57, 70,-80, 87,-90, 90,-87, 80,-70, 57,-43, 25, -9}    */
    674     /***************************************************************************/
    675 
    676     for(j = 0; j < TRANS_SIZE_16; j++)
    677     {
    678         /* Checking for Zero Cols */
    679         if((zero_cols & 1) == 1)
    680         {
    681             memset(pi2_dst, 0, TRANS_SIZE_16 * sizeof(WORD16));
    682         }
    683         else
    684         {
    685             /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
    686             {
    687                 /*
    688                  o[k] = g_ai2_ihevc_trans_16[ 1][k]*pi2_src[ src_strd   ] + g_ai2_ihevc_trans_16[ 3][k]*pi2_src[ 3*src_strd   ] + g_ai2_ihevc_trans_16[ 5][k]*pi2_src[ 5*src_strd   ] + g_ai2_ihevc_trans_16[ 7][k]*pi2_src[ 7*src_strd   ] +
    689                  g_ai2_ihevc_trans_16[ 9][k]*pi2_src[ 9*src_strd   ] + g_ai2_ihevc_trans_16[11][k]*pi2_src[11*src_strd   ] + g_ai2_ihevc_trans_16[13][k]*pi2_src[13*src_strd   ] + g_ai2_ihevc_trans_16[15][k]*pi2_src[15*src_strd   ];
    690                  */
    691 
    692                 o[0] = 90 * pi2_src[src_strd] + 87 * pi2_src[3 * src_strd]
    693                                 + 80 * pi2_src[5 * src_strd]
    694                                 + 70 * pi2_src[7 * src_strd]
    695                                 + 57 * pi2_src[9 * src_strd]
    696                                 + 43 * pi2_src[11 * src_strd]
    697                                 + 25 * pi2_src[13 * src_strd]
    698                                 + 9 * pi2_src[15 * src_strd];
    699 
    700                 o[1] = 87 * pi2_src[src_strd] + 57 * pi2_src[3 * src_strd]
    701                                 + 9 * pi2_src[5 * src_strd]
    702                                 + -43 * pi2_src[7 * src_strd]
    703                                 + -80 * pi2_src[9 * src_strd]
    704                                 + -90 * pi2_src[11 * src_strd]
    705                                 + -70 * pi2_src[13 * src_strd]
    706                                 + -25 * pi2_src[15 * src_strd];
    707 
    708                 o[2] = 80 * pi2_src[src_strd] + 9 * pi2_src[3 * src_strd]
    709                                 + -70 * pi2_src[5 * src_strd]
    710                                 + -87 * pi2_src[7 * src_strd]
    711                                 + -25 * pi2_src[9 * src_strd]
    712                                 + 57 * pi2_src[11 * src_strd]
    713                                 + 90 * pi2_src[13 * src_strd]
    714                                 + 43 * pi2_src[15 * src_strd];
    715 
    716                 o[3] = 70 * pi2_src[src_strd] + -43 * pi2_src[3 * src_strd]
    717                                 + -87 * pi2_src[5 * src_strd]
    718                                 + 9 * pi2_src[7 * src_strd]
    719                                 + 90 * pi2_src[9 * src_strd]
    720                                 + 25 * pi2_src[11 * src_strd]
    721                                 + -80 * pi2_src[13 * src_strd]
    722                                 + -57 * pi2_src[15 * src_strd];
    723 
    724                 o[4] = 57 * pi2_src[src_strd] + -80 * pi2_src[3 * src_strd]
    725                                 + -25 * pi2_src[5 * src_strd]
    726                                 + 90 * pi2_src[7 * src_strd]
    727                                 + -9 * pi2_src[9 * src_strd]
    728                                 + -87 * pi2_src[11 * src_strd]
    729                                 + 43 * pi2_src[13 * src_strd]
    730                                 + 70 * pi2_src[15 * src_strd];
    731 
    732                 o[5] = 43 * pi2_src[src_strd] + -90 * pi2_src[3 * src_strd]
    733                                 + 57 * pi2_src[5 * src_strd]
    734                                 + 25 * pi2_src[7 * src_strd]
    735                                 + -87 * pi2_src[9 * src_strd]
    736                                 + 70 * pi2_src[11 * src_strd]
    737                                 + 9 * pi2_src[13 * src_strd]
    738                                 + -80 * pi2_src[15 * src_strd];
    739 
    740                 o[6] = 25 * pi2_src[src_strd] + -70 * pi2_src[3 * src_strd]
    741                                 + 90 * pi2_src[5 * src_strd]
    742                                 + -80 * pi2_src[7 * src_strd]
    743                                 + 43 * pi2_src[9 * src_strd]
    744                                 + 9 * pi2_src[11 * src_strd]
    745                                 + -57 * pi2_src[13 * src_strd]
    746                                 + 87 * pi2_src[15 * src_strd];
    747 
    748                 o[7] = 9 * pi2_src[src_strd] + -25 * pi2_src[3 * src_strd]
    749                                 + 43 * pi2_src[5 * src_strd]
    750                                 + -57 * pi2_src[7 * src_strd]
    751                                 + 70 * pi2_src[9 * src_strd]
    752                                 + -80 * pi2_src[11 * src_strd]
    753                                 + 87 * pi2_src[13 * src_strd]
    754                                 + -90 * pi2_src[15 * src_strd];
    755             }
    756             {
    757                 temp1 = (pi2_src[2 * src_strd] + pi2_src[6 * src_strd]) * 75;
    758                 temp2 = (pi2_src[10 * src_strd] + pi2_src[14 * src_strd]) * 50;
    759                 eo[0] = temp1 + 14 * pi2_src[2 * src_strd] + temp2
    760                                 - (pi2_src[14 * src_strd] << 5);
    761                 eo[1] = temp1 - 93 * pi2_src[6 * src_strd] - temp2
    762                                 - 39 * pi2_src[10 * src_strd];
    763 
    764                 temp1 = (pi2_src[2 * src_strd] - pi2_src[6 * src_strd]) * 50;
    765                 temp2 = (pi2_src[10 * src_strd] + pi2_src[14 * src_strd]) * 75;
    766                 eo[2] = temp1 - 39 * pi2_src[6 * src_strd] + temp2
    767                                 - 57 * pi2_src[10 * src_strd];
    768                 eo[3] = temp1 - (pi2_src[2 * src_strd] << 5) + temp2
    769                                 - 164 * pi2_src[14 * src_strd];
    770             }
    771 
    772             temp1 = (pi2_src[4 * src_strd] + pi2_src[12 * src_strd]) * 36;
    773             eeo[0] = temp1 + 47 * pi2_src[4 * src_strd];
    774             eeo[1] = temp1 - 119 * pi2_src[12 * src_strd];
    775 
    776             eee[0] = (pi2_src[0] + pi2_src[8 * src_strd]) << 6;
    777             eee[1] = (pi2_src[0] - pi2_src[8 * src_strd]) << 6;
    778 
    779             /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
    780             for(k = 0; k < 2; k++)
    781             {
    782                 ee[k] = eee[k] + eeo[k];
    783                 ee[k + 2] = eee[1 - k] - eeo[1 - k];
    784             }
    785             for(k = 0; k < 4; k++)
    786             {
    787                 e[k] = ee[k] + eo[k];
    788                 e[k + 4] = ee[3 - k] - eo[3 - k];
    789             }
    790             for(k = 0; k < 8; k++)
    791             {
    792                 pi2_dst[k] =
    793                                 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
    794                 pi2_dst[k + 8] =
    795                                 CLIP_S16(((e[7 - k] - o[7 - k] + add) >> i4_shift));
    796             }
    797         }
    798         pi2_src++;
    799         pi2_dst += dst_strd;
    800         zero_cols = zero_cols >> 1;
    801     }
    802 }
    803 #endif
    804 
    805 /**
    806  *******************************************************************************
    807  *
    808  * @brief
    809  *  This function performs Single stage  Inverse transform for 32x32 input
    810  * block
    811  *
    812  * @par Description:
    813  *  Performs single stage 32x32 inverse transform by  utilizing the symmetry
    814  * of transformation matrix and  reducing number of multiplications wherever
    815  * possible  but keeping the number of operations  (addition,multiplication
    816  * and shift) same
    817  *
    818  * @param[in] pi2_src
    819  *  Input 32x32 coefficients
    820  *
    821  * @param[out] pi2_dst
    822  *  Output 32x32 block
    823  *
    824  * @param[in] src_strd
    825  *  Input stride
    826  *
    827  * @param[in] dst_strd
    828  *  Output Stride
    829  *
    830  * @param[in] i4_shift
    831  *  Output shift
    832  *
    833  * @param[in] zero_cols
    834  *  Zero columns in pi2_src
    835  *
    836  * @returns  Void
    837  *
    838  * @remarks
    839  *  None
    840  *
    841  *******************************************************************************
    842  */
    843 
    844 
    845 void ihevc_itrans_32x32(WORD16 *pi2_src,
    846                         WORD16 *pi2_dst,
    847                         WORD32 src_strd,
    848                         WORD32 dst_strd,
    849                         WORD32 i4_shift,
    850                         WORD32 zero_cols)
    851 {
    852     WORD32 j, k;
    853     WORD32 e[16], o[16];
    854     WORD32 ee[8], eo[8];
    855     WORD32 eee[4], eeo[4];
    856     WORD32 eeee[2], eeeo[2];
    857     WORD32 add;
    858 
    859     add = 1 << (i4_shift - 1);
    860 
    861     for(j = 0; j < TRANS_SIZE_32; j++)
    862     {
    863         /* Checking for Zero Cols */
    864         if((zero_cols & 1) == 1)
    865         {
    866             memset(pi2_dst, 0, TRANS_SIZE_32 * sizeof(WORD16));
    867         }
    868         else
    869         {
    870             /* Utilizing symmetry properties to the maximum to minimize the number of multiplications */
    871             for(k = 0; k < 16; k++)
    872             {
    873                 o[k] = g_ai2_ihevc_trans_32[1][k] * pi2_src[src_strd]
    874                                 + g_ai2_ihevc_trans_32[3][k]
    875                                                 * pi2_src[3 * src_strd]
    876                                 + g_ai2_ihevc_trans_32[5][k]
    877                                                 * pi2_src[5 * src_strd]
    878                                 + g_ai2_ihevc_trans_32[7][k]
    879                                                 * pi2_src[7 * src_strd]
    880                                 + g_ai2_ihevc_trans_32[9][k]
    881                                                 * pi2_src[9 * src_strd]
    882                                 + g_ai2_ihevc_trans_32[11][k]
    883                                                 * pi2_src[11 * src_strd]
    884                                 + g_ai2_ihevc_trans_32[13][k]
    885                                                 * pi2_src[13 * src_strd]
    886                                 + g_ai2_ihevc_trans_32[15][k]
    887                                                 * pi2_src[15 * src_strd]
    888                                 + g_ai2_ihevc_trans_32[17][k]
    889                                                 * pi2_src[17 * src_strd]
    890                                 + g_ai2_ihevc_trans_32[19][k]
    891                                                 * pi2_src[19 * src_strd]
    892                                 + g_ai2_ihevc_trans_32[21][k]
    893                                                 * pi2_src[21 * src_strd]
    894                                 + g_ai2_ihevc_trans_32[23][k]
    895                                                 * pi2_src[23 * src_strd]
    896                                 + g_ai2_ihevc_trans_32[25][k]
    897                                                 * pi2_src[25 * src_strd]
    898                                 + g_ai2_ihevc_trans_32[27][k]
    899                                                 * pi2_src[27 * src_strd]
    900                                 + g_ai2_ihevc_trans_32[29][k]
    901                                                 * pi2_src[29 * src_strd]
    902                                 + g_ai2_ihevc_trans_32[31][k]
    903                                                 * pi2_src[31 * src_strd];
    904             }
    905             for(k = 0; k < 8; k++)
    906             {
    907                 eo[k] = g_ai2_ihevc_trans_32[2][k] * pi2_src[2 * src_strd]
    908                                 + g_ai2_ihevc_trans_32[6][k]
    909                                                 * pi2_src[6 * src_strd]
    910                                 + g_ai2_ihevc_trans_32[10][k]
    911                                                 * pi2_src[10 * src_strd]
    912                                 + g_ai2_ihevc_trans_32[14][k]
    913                                                 * pi2_src[14 * src_strd]
    914                                 + g_ai2_ihevc_trans_32[18][k]
    915                                                 * pi2_src[18 * src_strd]
    916                                 + g_ai2_ihevc_trans_32[22][k]
    917                                                 * pi2_src[22 * src_strd]
    918                                 + g_ai2_ihevc_trans_32[26][k]
    919                                                 * pi2_src[26 * src_strd]
    920                                 + g_ai2_ihevc_trans_32[30][k]
    921                                                 * pi2_src[30 * src_strd];
    922             }
    923             for(k = 0; k < 4; k++)
    924             {
    925                 eeo[k] = g_ai2_ihevc_trans_32[4][k] * pi2_src[4 * src_strd]
    926                                 + g_ai2_ihevc_trans_32[12][k]
    927                                                 * pi2_src[12 * src_strd]
    928                                 + g_ai2_ihevc_trans_32[20][k]
    929                                                 * pi2_src[20 * src_strd]
    930                                 + g_ai2_ihevc_trans_32[28][k]
    931                                                 * pi2_src[28 * src_strd];
    932             }
    933             eeeo[0] = g_ai2_ihevc_trans_32[8][0] * pi2_src[8 * src_strd]
    934                             + g_ai2_ihevc_trans_32[24][0]
    935                                             * pi2_src[24 * src_strd];
    936             eeeo[1] = g_ai2_ihevc_trans_32[8][1] * pi2_src[8 * src_strd]
    937                             + g_ai2_ihevc_trans_32[24][1]
    938                                             * pi2_src[24 * src_strd];
    939             eeee[0] = g_ai2_ihevc_trans_32[0][0] * pi2_src[0]
    940                             + g_ai2_ihevc_trans_32[16][0]
    941                                             * pi2_src[16 * src_strd];
    942             eeee[1] = g_ai2_ihevc_trans_32[0][1] * pi2_src[0]
    943                             + g_ai2_ihevc_trans_32[16][1]
    944                                             * pi2_src[16 * src_strd];
    945 
    946             /* Combining e and o terms at each hierarchy levels to calculate the final spatial domain vector */
    947             eee[0] = eeee[0] + eeeo[0];
    948             eee[3] = eeee[0] - eeeo[0];
    949             eee[1] = eeee[1] + eeeo[1];
    950             eee[2] = eeee[1] - eeeo[1];
    951             for(k = 0; k < 4; k++)
    952             {
    953                 ee[k] = eee[k] + eeo[k];
    954                 ee[k + 4] = eee[3 - k] - eeo[3 - k];
    955             }
    956             for(k = 0; k < 8; k++)
    957             {
    958                 e[k] = ee[k] + eo[k];
    959                 e[k + 8] = ee[7 - k] - eo[7 - k];
    960             }
    961             for(k = 0; k < 16; k++)
    962             {
    963                 pi2_dst[k] =
    964                                 CLIP_S16(((e[k] + o[k] + add) >> i4_shift));
    965                 pi2_dst[k + 16] =
    966                                 CLIP_S16(((e[15 - k] - o[15 - k] + add) >> i4_shift));
    967             }
    968         }
    969         pi2_src++;
    970         pi2_dst += dst_strd;
    971         zero_cols = zero_cols >> 1;
    972     }
    973 }
    974 
    975