vpx_dsp/mips/itrans32_dspr2.c

   1 /*
   2  *  Copyright (c) 2013 The WebM project authors. All Rights Reserved.
   3  *
   4  *  Use of this source code is governed by a BSD-style license
   5  *  that can be found in the LICENSE file in the root of the source
   6  *  tree. An additional intellectual property rights grant can be found
   7  *  in the file PATENTS.  All contributing project authors may
   8  *  be found in the AUTHORS file in the root of the source tree.
   9  */
  10
  11 #include <assert.h>
  12 #include <stdio.h>
  13
  14 #include "./vpx_config.h"
  15 #include "vpx_dsp/mips/inv_txfm_dspr2.h"
  16 #include "vpx_dsp/txfm_common.h"
  17
  18 #if HAVE_DSPR2
  19 static void idct32_rows_dspr2(const int16_t *input, int16_t *output,
  20                               uint32_t no_rows) {
  21   int16_t step1_0, step1_1, step1_2, step1_3, step1_4, step1_5, step1_6;
  22   int16_t step1_7, step1_8, step1_9, step1_10, step1_11, step1_12, step1_13;
  23   int16_t step1_14, step1_15, step1_16, step1_17, step1_18, step1_19, step1_20;
  24   int16_t step1_21, step1_22, step1_23, step1_24, step1_25, step1_26, step1_27;
  25   int16_t step1_28, step1_29, step1_30, step1_31;
  26   int16_t step2_0, step2_1, step2_2, step2_3, step2_4, step2_5, step2_6;
  27   int16_t step2_7, step2_8, step2_9, step2_10, step2_11, step2_12, step2_13;
  28   int16_t step2_14, step2_15, step2_16, step2_17, step2_18, step2_19, step2_20;
  29   int16_t step2_21, step2_22, step2_23, step2_24, step2_25, step2_26, step2_27;
  30   int16_t step2_28, step2_29, step2_30, step2_31;
  31   int16_t step3_8, step3_9, step3_10, step3_11, step3_12, step3_13, step3_14;
  32   int16_t step3_15, step3_16, step3_17, step3_18, step3_19, step3_20, step3_21;
  33   int16_t step3_22, step3_23, step3_24, step3_25, step3_26, step3_27, step3_28;
  34   int16_t step3_29, step3_30, step3_31;
  35   int temp0, temp1, temp2, temp3;
  36   int load1, load2, load3, load4;
  37   int result1, result2;
  38   int temp21;
  39   int i;
  40   const int const_2_power_13 = 8192;
  41   const int32_t *input_int;
  42
  43   for (i = no_rows; i--; ) {
  44     input_int = (const int32_t *)input;
  45
  46     if (!(input_int[0]  | input_int[1]  | input_int[2]  | input_int[3]  |
  47           input_int[4]  | input_int[5]  | input_int[6]  | input_int[7]  |
  48           input_int[8]  | input_int[9]  | input_int[10] | input_int[11] |
  49           input_int[12] | input_int[13] | input_int[14] | input_int[15])) {
  50       input += 32;
  51
  52       __asm__ __volatile__ (
  53           "sh     $zero,     0(%[output])     \n\t"
  54           "sh     $zero,    64(%[output])     \n\t"
  55           "sh     $zero,   128(%[output])     \n\t"
  56           "sh     $zero,   192(%[output])     \n\t"
  57           "sh     $zero,   256(%[output])     \n\t"
  58           "sh     $zero,   320(%[output])     \n\t"
  59           "sh     $zero,   384(%[output])     \n\t"
  60           "sh     $zero,   448(%[output])     \n\t"
  61           "sh     $zero,   512(%[output])     \n\t"
  62           "sh     $zero,   576(%[output])     \n\t"
  63           "sh     $zero,   640(%[output])     \n\t"
  64           "sh     $zero,   704(%[output])     \n\t"
  65           "sh     $zero,   768(%[output])     \n\t"
  66           "sh     $zero,   832(%[output])     \n\t"
  67           "sh     $zero,   896(%[output])     \n\t"
  68           "sh     $zero,   960(%[output])     \n\t"
  69           "sh     $zero,  1024(%[output])     \n\t"
  70           "sh     $zero,  1088(%[output])     \n\t"
  71           "sh     $zero,  1152(%[output])     \n\t"
  72           "sh     $zero,  1216(%[output])     \n\t"
  73           "sh     $zero,  1280(%[output])     \n\t"
  74           "sh     $zero,  1344(%[output])     \n\t"
  75           "sh     $zero,  1408(%[output])     \n\t"
  76           "sh     $zero,  1472(%[output])     \n\t"
  77           "sh     $zero,  1536(%[output])     \n\t"
  78           "sh     $zero,  1600(%[output])     \n\t"
  79           "sh     $zero,  1664(%[output])     \n\t"
  80           "sh     $zero,  1728(%[output])     \n\t"
  81           "sh     $zero,  1792(%[output])     \n\t"
  82           "sh     $zero,  1856(%[output])     \n\t"
  83           "sh     $zero,  1920(%[output])     \n\t"
  84           "sh     $zero,  1984(%[output])     \n\t"
  85
  86           :
  87           : [output] "r" (output)
  88       );
  89
  90       output += 1;
  91
  92       continue;
  93     }
  94
  95     /* prefetch row */
  96     prefetch_load((const uint8_t *)(input + 32));
  97     prefetch_load((const uint8_t *)(input + 48));
  98
  99     __asm__ __volatile__ (
 100         "lh       %[load1],             2(%[input])                     \n\t"
 101         "lh       %[load2],             62(%[input])                    \n\t"
 102         "lh       %[load3],             34(%[input])                    \n\t"
 103         "lh       %[load4],             30(%[input])                    \n\t"
 104
 105         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 106         "mthi     $zero,                $ac1                            \n\t"
 107         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 108         "mthi     $zero,                $ac3                            \n\t"
 109
 110         "madd     $ac1,                 %[load1],       %[cospi_31_64]  \n\t"
 111         "msub     $ac1,                 %[load2],       %[cospi_1_64]   \n\t"
 112         "extp     %[temp0],             $ac1,           31              \n\t"
 113
 114         "madd     $ac3,                 %[load1],       %[cospi_1_64]   \n\t"
 115         "madd     $ac3,                 %[load2],       %[cospi_31_64]  \n\t"
 116         "extp     %[temp3],             $ac3,           31              \n\t"
 117
 118         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 119         "mthi     $zero,                $ac1                            \n\t"
 120         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 121         "mthi     $zero,                $ac2                            \n\t"
 122
 123         "madd     $ac2,                 %[load3],       %[cospi_15_64]  \n\t"
 124         "msub     $ac2,                 %[load4],       %[cospi_17_64]  \n\t"
 125         "extp     %[temp1],             $ac2,           31              \n\t"
 126
 127         "madd     $ac1,                 %[load3],       %[cospi_17_64]  \n\t"
 128         "madd     $ac1,                 %[load4],       %[cospi_15_64]  \n\t"
 129         "extp     %[temp2],             $ac1,           31              \n\t"
 130
 131         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 132         "mthi     $zero,                $ac1                            \n\t"
 133         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 134         "mthi     $zero,                $ac3                            \n\t"
 135
 136         "sub      %[load1],             %[temp3],       %[temp2]        \n\t"
 137         "sub      %[load2],             %[temp0],       %[temp1]        \n\t"
 138
 139         "madd     $ac1,                 %[load1],       %[cospi_28_64]  \n\t"
 140         "msub     $ac1,                 %[load2],       %[cospi_4_64]   \n\t"
 141         "madd     $ac3,                 %[load1],       %[cospi_4_64]   \n\t"
 142         "madd     $ac3,                 %[load2],       %[cospi_28_64]  \n\t"
 143
 144         "extp     %[step1_17],          $ac1,           31              \n\t"
 145         "extp     %[step1_30],          $ac3,           31              \n\t"
 146         "add      %[step1_16],          %[temp0],       %[temp1]        \n\t"
 147         "add      %[step1_31],          %[temp2],       %[temp3]        \n\t"
 148
 149         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 150           [load3] "=&r" (load3), [load4] "=&r" (load4),
 151           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 152           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 153           [step1_16] "=r" (step1_16), [step1_17] "=r" (step1_17),
 154           [step1_30] "=r" (step1_30), [step1_31] "=r" (step1_31)
 155         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 156           [cospi_31_64] "r" (cospi_31_64), [cospi_1_64] "r" (cospi_1_64),
 157           [cospi_4_64] "r" (cospi_4_64), [cospi_17_64] "r" (cospi_17_64),
 158           [cospi_15_64] "r" (cospi_15_64), [cospi_28_64] "r" (cospi_28_64)
 159     );
 160
 161     __asm__ __volatile__ (
 162         "lh       %[load1],             18(%[input])                    \n\t"
 163         "lh       %[load2],             46(%[input])                    \n\t"
 164         "lh       %[load3],             50(%[input])                    \n\t"
 165         "lh       %[load4],             14(%[input])                    \n\t"
 166
 167         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 168         "mthi     $zero,                $ac1                            \n\t"
 169         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 170         "mthi     $zero,                $ac3                            \n\t"
 171
 172         "madd     $ac1,                 %[load1],       %[cospi_23_64]  \n\t"
 173         "msub     $ac1,                 %[load2],       %[cospi_9_64]   \n\t"
 174         "extp     %[temp0],             $ac1,           31              \n\t"
 175
 176         "madd     $ac3,                 %[load1],       %[cospi_9_64]   \n\t"
 177         "madd     $ac3,                 %[load2],       %[cospi_23_64]  \n\t"
 178         "extp     %[temp3],             $ac3,           31              \n\t"
 179
 180         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 181         "mthi     $zero,                $ac1                            \n\t"
 182         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 183         "mthi     $zero,                $ac2                            \n\t"
 184
 185         "madd     $ac2,                 %[load3],       %[cospi_7_64]   \n\t"
 186         "msub     $ac2,                 %[load4],       %[cospi_25_64]  \n\t"
 187         "extp     %[temp1],             $ac2,           31              \n\t"
 188
 189         "madd     $ac1,                 %[load3],       %[cospi_25_64]  \n\t"
 190         "madd     $ac1,                 %[load4],       %[cospi_7_64]   \n\t"
 191         "extp     %[temp2],             $ac1,           31              \n\t"
 192
 193         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 194         "mthi     $zero,                $ac1                            \n\t"
 195         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 196         "mthi     $zero,                $ac3                            \n\t"
 197
 198         "sub      %[load1],             %[temp1],       %[temp0]        \n\t"
 199         "sub      %[load2],             %[temp2],       %[temp3]        \n\t"
 200
 201         "msub     $ac1,                 %[load1],       %[cospi_28_64]  \n\t"
 202         "msub     $ac1,                 %[load2],       %[cospi_4_64]   \n\t"
 203         "msub     $ac3,                 %[load1],       %[cospi_4_64]   \n\t"
 204         "madd     $ac3,                 %[load2],       %[cospi_28_64]  \n\t"
 205
 206         "extp     %[step1_18],          $ac1,           31              \n\t"
 207         "extp     %[step1_29],          $ac3,           31              \n\t"
 208         "add      %[step1_19],          %[temp0],       %[temp1]        \n\t"
 209         "add      %[step1_28],          %[temp2],       %[temp3]        \n\t"
 210
 211         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 212           [load3] "=&r" (load3), [load4] "=&r" (load4),
 213           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 214           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 215           [step1_18] "=r" (step1_18), [step1_19] "=r" (step1_19),
 216           [step1_28] "=r" (step1_28), [step1_29] "=r" (step1_29)
 217         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 218           [cospi_23_64] "r" (cospi_23_64), [cospi_9_64] "r" (cospi_9_64),
 219           [cospi_4_64] "r" (cospi_4_64), [cospi_7_64] "r" (cospi_7_64),
 220           [cospi_25_64] "r" (cospi_25_64), [cospi_28_64] "r" (cospi_28_64)
 221     );
 222
 223     __asm__ __volatile__ (
 224         "lh       %[load1],             10(%[input])                    \n\t"
 225         "lh       %[load2],             54(%[input])                    \n\t"
 226         "lh       %[load3],             42(%[input])                    \n\t"
 227         "lh       %[load4],             22(%[input])                    \n\t"
 228
 229         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 230         "mthi     $zero,                $ac1                            \n\t"
 231         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 232         "mthi     $zero,                $ac3                            \n\t"
 233
 234         "madd     $ac1,                 %[load1],       %[cospi_27_64]  \n\t"
 235         "msub     $ac1,                 %[load2],       %[cospi_5_64]   \n\t"
 236         "extp     %[temp0],             $ac1,           31              \n\t"
 237
 238         "madd     $ac3,                 %[load1],       %[cospi_5_64]   \n\t"
 239         "madd     $ac3,                 %[load2],       %[cospi_27_64]  \n\t"
 240         "extp     %[temp3],             $ac3,           31              \n\t"
 241
 242         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 243         "mthi     $zero,                $ac1                            \n\t"
 244         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 245         "mthi     $zero,                $ac2                            \n\t"
 246
 247         "madd     $ac2,                 %[load3],       %[cospi_11_64]  \n\t"
 248         "msub     $ac2,                 %[load4],       %[cospi_21_64]  \n\t"
 249         "extp     %[temp1],             $ac2,           31              \n\t"
 250
 251         "madd     $ac1,                 %[load3],       %[cospi_21_64]  \n\t"
 252         "madd     $ac1,                 %[load4],       %[cospi_11_64]  \n\t"
 253         "extp     %[temp2],             $ac1,           31              \n\t"
 254
 255         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 256         "mthi     $zero,                $ac1                            \n\t"
 257         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 258         "mthi     $zero,                $ac3                            \n\t"
 259
 260         "sub      %[load1],             %[temp0],       %[temp1]        \n\t"
 261         "sub      %[load2],             %[temp3],       %[temp2]        \n\t"
 262
 263         "madd     $ac1,                 %[load2],       %[cospi_12_64]  \n\t"
 264         "msub     $ac1,                 %[load1],       %[cospi_20_64]  \n\t"
 265         "madd     $ac3,                 %[load1],       %[cospi_12_64]  \n\t"
 266         "madd     $ac3,                 %[load2],       %[cospi_20_64]  \n\t"
 267
 268         "extp     %[step1_21],          $ac1,           31              \n\t"
 269         "extp     %[step1_26],          $ac3,           31              \n\t"
 270         "add      %[step1_20],          %[temp0],       %[temp1]        \n\t"
 271         "add      %[step1_27],          %[temp2],       %[temp3]        \n\t"
 272
 273         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 274           [load3] "=&r" (load3), [load4] "=&r" (load4),
 275           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 276           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 277           [step1_20] "=r" (step1_20), [step1_21] "=r" (step1_21),
 278           [step1_26] "=r" (step1_26), [step1_27] "=r" (step1_27)
 279         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 280           [cospi_27_64] "r" (cospi_27_64), [cospi_5_64] "r" (cospi_5_64),
 281           [cospi_11_64] "r" (cospi_11_64), [cospi_21_64] "r" (cospi_21_64),
 282           [cospi_12_64] "r" (cospi_12_64), [cospi_20_64] "r" (cospi_20_64)
 283     );
 284
 285     __asm__ __volatile__ (
 286         "lh       %[load1],             26(%[input])                    \n\t"
 287         "lh       %[load2],             38(%[input])                    \n\t"
 288         "lh       %[load3],             58(%[input])                    \n\t"
 289         "lh       %[load4],              6(%[input])                    \n\t"
 290
 291         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 292         "mthi     $zero,                $ac1                            \n\t"
 293         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 294         "mthi     $zero,                $ac3                            \n\t"
 295
 296         "madd     $ac1,                 %[load1],       %[cospi_19_64]  \n\t"
 297         "msub     $ac1,                 %[load2],       %[cospi_13_64]  \n\t"
 298         "extp     %[temp0],             $ac1,           31              \n\t"
 299
 300         "madd     $ac3,                 %[load1],       %[cospi_13_64]  \n\t"
 301         "madd     $ac3,                 %[load2],       %[cospi_19_64]  \n\t"
 302         "extp     %[temp3],             $ac3,           31              \n\t"
 303
 304         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 305         "mthi     $zero,                $ac1                            \n\t"
 306         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 307         "mthi     $zero,                $ac2                            \n\t"
 308
 309         "madd     $ac2,                 %[load3],       %[cospi_3_64]   \n\t"
 310         "msub     $ac2,                 %[load4],       %[cospi_29_64]  \n\t"
 311         "extp     %[temp1],             $ac2,           31              \n\t"
 312
 313         "madd     $ac1,                 %[load3],       %[cospi_29_64]  \n\t"
 314         "madd     $ac1,                 %[load4],       %[cospi_3_64]   \n\t"
 315         "extp     %[temp2],             $ac1,           31              \n\t"
 316
 317         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 318         "mthi     $zero,                $ac1                            \n\t"
 319         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 320         "mthi     $zero,                $ac3                            \n\t"
 321
 322         "sub      %[load1],             %[temp1],       %[temp0]        \n\t"
 323         "sub      %[load2],             %[temp2],       %[temp3]        \n\t"
 324
 325         "msub     $ac1,                 %[load1],       %[cospi_12_64]  \n\t"
 326         "msub     $ac1,                 %[load2],       %[cospi_20_64]  \n\t"
 327         "msub     $ac3,                 %[load1],       %[cospi_20_64]  \n\t"
 328         "madd     $ac3,                 %[load2],       %[cospi_12_64]  \n\t"
 329
 330         "extp     %[step1_22],          $ac1,           31              \n\t"
 331         "extp     %[step1_25],          $ac3,           31              \n\t"
 332         "add      %[step1_23],          %[temp0],       %[temp1]        \n\t"
 333         "add      %[step1_24],          %[temp2],       %[temp3]        \n\t"
 334
 335         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 336           [load3] "=&r" (load3), [load4] "=&r" (load4),
 337           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 338           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 339           [step1_22] "=r" (step1_22), [step1_23] "=r" (step1_23),
 340           [step1_24] "=r" (step1_24), [step1_25] "=r" (step1_25)
 341         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 342           [cospi_19_64] "r" (cospi_19_64), [cospi_13_64] "r" (cospi_13_64),
 343           [cospi_3_64] "r" (cospi_3_64), [cospi_29_64] "r" (cospi_29_64),
 344           [cospi_12_64] "r" (cospi_12_64), [cospi_20_64] "r" (cospi_20_64)
 345     );
 346
 347     __asm__ __volatile__ (
 348         "lh       %[load1],              4(%[input])                    \n\t"
 349         "lh       %[load2],             60(%[input])                    \n\t"
 350         "lh       %[load3],             36(%[input])                    \n\t"
 351         "lh       %[load4],             28(%[input])                    \n\t"
 352
 353         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 354         "mthi     $zero,                $ac1                            \n\t"
 355         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 356         "mthi     $zero,                $ac3                            \n\t"
 357
 358         "madd     $ac1,                 %[load1],       %[cospi_30_64]  \n\t"
 359         "msub     $ac1,                 %[load2],       %[cospi_2_64]   \n\t"
 360         "extp     %[temp0],             $ac1,           31              \n\t"
 361
 362         "madd     $ac3,                 %[load1],       %[cospi_2_64]   \n\t"
 363         "madd     $ac3,                 %[load2],       %[cospi_30_64]  \n\t"
 364         "extp     %[temp3],             $ac3,           31              \n\t"
 365
 366         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 367         "mthi     $zero,                $ac1                            \n\t"
 368         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 369         "mthi     $zero,                $ac2                            \n\t"
 370
 371         "madd     $ac2,                 %[load3],       %[cospi_14_64]  \n\t"
 372         "msub     $ac2,                 %[load4],       %[cospi_18_64]  \n\t"
 373         "extp     %[temp1],             $ac2,           31              \n\t"
 374
 375         "madd     $ac1,                 %[load3],       %[cospi_18_64]  \n\t"
 376         "madd     $ac1,                 %[load4],       %[cospi_14_64]  \n\t"
 377         "extp     %[temp2],             $ac1,           31              \n\t"
 378
 379         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 380         "mthi     $zero,                $ac1                            \n\t"
 381         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 382         "mthi     $zero,                $ac3                            \n\t"
 383
 384         "sub      %[load1],             %[temp0],       %[temp1]        \n\t"
 385         "sub      %[load2],             %[temp3],       %[temp2]        \n\t"
 386
 387         "msub     $ac1,                 %[load1],       %[cospi_8_64]   \n\t"
 388         "madd     $ac1,                 %[load2],       %[cospi_24_64]  \n\t"
 389         "madd     $ac3,                 %[load1],       %[cospi_24_64]  \n\t"
 390         "madd     $ac3,                 %[load2],       %[cospi_8_64]   \n\t"
 391
 392         "extp     %[step2_9],           $ac1,           31              \n\t"
 393         "extp     %[step2_14],          $ac3,           31              \n\t"
 394         "add      %[step2_8],           %[temp0],       %[temp1]        \n\t"
 395         "add      %[step2_15],          %[temp2],       %[temp3]        \n\t"
 396
 397         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 398           [load3] "=&r" (load3), [load4] "=&r" (load4),
 399           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 400           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 401           [step2_8] "=r" (step2_8), [step2_9] "=r" (step2_9),
 402           [step2_14] "=r" (step2_14), [step2_15] "=r" (step2_15)
 403         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 404           [cospi_30_64] "r" (cospi_30_64), [cospi_2_64] "r" (cospi_2_64),
 405           [cospi_14_64] "r" (cospi_14_64), [cospi_18_64] "r" (cospi_18_64),
 406           [cospi_8_64] "r" (cospi_8_64), [cospi_24_64] "r" (cospi_24_64)
 407     );
 408
 409     __asm__ __volatile__ (
 410         "lh       %[load1],             20(%[input])                    \n\t"
 411         "lh       %[load2],             44(%[input])                    \n\t"
 412         "lh       %[load3],             52(%[input])                    \n\t"
 413         "lh       %[load4],             12(%[input])                    \n\t"
 414
 415         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 416         "mthi     $zero,                $ac1                            \n\t"
 417         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 418         "mthi     $zero,                $ac3                            \n\t"
 419
 420         "madd     $ac1,                 %[load1],       %[cospi_22_64]  \n\t"
 421         "msub     $ac1,                 %[load2],       %[cospi_10_64]  \n\t"
 422         "extp     %[temp0],             $ac1,           31              \n\t"
 423
 424         "madd     $ac3,                 %[load1],       %[cospi_10_64]  \n\t"
 425         "madd     $ac3,                 %[load2],       %[cospi_22_64]  \n\t"
 426         "extp     %[temp3],             $ac3,           31              \n\t"
 427
 428         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 429         "mthi     $zero,                $ac1                            \n\t"
 430         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 431         "mthi     $zero,                $ac2                            \n\t"
 432
 433         "madd     $ac2,                 %[load3],       %[cospi_6_64]   \n\t"
 434         "msub     $ac2,                 %[load4],       %[cospi_26_64]  \n\t"
 435         "extp     %[temp1],             $ac2,           31              \n\t"
 436
 437         "madd     $ac1,                 %[load3],       %[cospi_26_64]  \n\t"
 438         "madd     $ac1,                 %[load4],       %[cospi_6_64]   \n\t"
 439         "extp     %[temp2],             $ac1,           31              \n\t"
 440
 441         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 442         "mthi     $zero,                $ac1                            \n\t"
 443         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 444         "mthi     $zero,                $ac3                            \n\t"
 445
 446         "sub      %[load1],             %[temp1],       %[temp0]        \n\t"
 447         "sub      %[load2],             %[temp2],       %[temp3]        \n\t"
 448
 449         "msub     $ac1,                 %[load1],       %[cospi_24_64]  \n\t"
 450         "msub     $ac1,                 %[load2],       %[cospi_8_64]   \n\t"
 451         "madd     $ac3,                 %[load2],       %[cospi_24_64]  \n\t"
 452         "msub     $ac3,                 %[load1],       %[cospi_8_64]   \n\t"
 453
 454         "extp     %[step2_10],          $ac1,           31              \n\t"
 455         "extp     %[step2_13],          $ac3,           31              \n\t"
 456         "add      %[step2_11],          %[temp0],       %[temp1]        \n\t"
 457         "add      %[step2_12],          %[temp2],       %[temp3]        \n\t"
 458
 459         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 460           [load3] "=&r" (load3), [load4] "=&r" (load4),
 461           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 462           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 463           [step2_10] "=r" (step2_10), [step2_11] "=r" (step2_11),
 464           [step2_12] "=r" (step2_12), [step2_13] "=r" (step2_13)
 465         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 466           [cospi_22_64] "r" (cospi_22_64), [cospi_10_64] "r" (cospi_10_64),
 467           [cospi_6_64] "r" (cospi_6_64), [cospi_26_64] "r" (cospi_26_64),
 468           [cospi_8_64] "r" (cospi_8_64), [cospi_24_64] "r" (cospi_24_64)
 469     );
 470
 471     __asm__ __volatile__ (
 472         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 473         "mthi     $zero,                $ac0                            \n\t"
 474         "sub      %[temp0],             %[step2_14],    %[step2_13]     \n\t"
 475         "sub      %[temp0],             %[temp0],       %[step2_9]      \n\t"
 476         "add      %[temp0],             %[temp0],       %[step2_10]     \n\t"
 477         "madd     $ac0,                 %[temp0],       %[cospi_16_64]  \n\t"
 478
 479         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 480         "mthi     $zero,                $ac1                            \n\t"
 481         "sub      %[temp1],             %[step2_14],    %[step2_13]     \n\t"
 482         "add      %[temp1],             %[temp1],       %[step2_9]      \n\t"
 483         "sub      %[temp1],             %[temp1],       %[step2_10]     \n\t"
 484         "madd     $ac1,                 %[temp1],       %[cospi_16_64]  \n\t"
 485
 486         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 487         "mthi     $zero,                $ac2                            \n\t"
 488         "sub      %[temp0],             %[step2_15],    %[step2_12]     \n\t"
 489         "sub      %[temp0],             %[temp0],       %[step2_8]      \n\t"
 490         "add      %[temp0],             %[temp0],       %[step2_11]     \n\t"
 491         "madd     $ac2,                 %[temp0],       %[cospi_16_64]  \n\t"
 492
 493         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 494         "mthi     $zero,                $ac3                            \n\t"
 495         "sub      %[temp1],             %[step2_15],    %[step2_12]     \n\t"
 496         "add      %[temp1],             %[temp1],       %[step2_8]      \n\t"
 497         "sub      %[temp1],             %[temp1],       %[step2_11]     \n\t"
 498         "madd     $ac3,                 %[temp1],       %[cospi_16_64]  \n\t"
 499
 500         "add      %[step3_8],           %[step2_8],     %[step2_11]     \n\t"
 501         "add      %[step3_9],           %[step2_9],     %[step2_10]     \n\t"
 502         "add      %[step3_14],          %[step2_13],    %[step2_14]     \n\t"
 503         "add      %[step3_15],          %[step2_12],    %[step2_15]     \n\t"
 504
 505         "extp     %[step3_10],          $ac0,           31              \n\t"
 506         "extp     %[step3_13],          $ac1,           31              \n\t"
 507         "extp     %[step3_11],          $ac2,           31              \n\t"
 508         "extp     %[step3_12],          $ac3,           31              \n\t"
 509
 510         : [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 511           [step3_8] "=r" (step3_8), [step3_9] "=r" (step3_9),
 512           [step3_10] "=r" (step3_10), [step3_11] "=r" (step3_11),
 513           [step3_12] "=r" (step3_12), [step3_13] "=r" (step3_13),
 514           [step3_14] "=r" (step3_14), [step3_15] "=r" (step3_15)
 515         : [const_2_power_13] "r" (const_2_power_13),
 516           [step2_8] "r" (step2_8), [step2_9] "r" (step2_9),
 517           [step2_10] "r" (step2_10), [step2_11] "r" (step2_11),
 518           [step2_12] "r" (step2_12), [step2_13] "r" (step2_13),
 519           [step2_14] "r" (step2_14), [step2_15] "r" (step2_15),
 520           [cospi_16_64] "r" (cospi_16_64)
 521     );
 522
 523     step2_18 = step1_17 - step1_18;
 524     step2_29 = step1_30 - step1_29;
 525
 526     __asm__ __volatile__ (
 527         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 528         "mthi     $zero,                $ac0                            \n\t"
 529         "msub     $ac0,                 %[step2_18],    %[cospi_8_64]   \n\t"
 530         "madd     $ac0,                 %[step2_29],    %[cospi_24_64]  \n\t"
 531         "extp     %[step3_18],          $ac0,           31              \n\t"
 532
 533         : [step3_18] "=r" (step3_18)
 534         : [const_2_power_13] "r" (const_2_power_13),
 535           [step2_18] "r" (step2_18), [step2_29] "r" (step2_29),
 536           [cospi_24_64] "r" (cospi_24_64), [cospi_8_64] "r" (cospi_8_64)
 537     );
 538
 539     temp21 = step2_18 * cospi_24_64 + step2_29 * cospi_8_64;
 540     step3_29 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 541
 542     step2_19 = step1_16 - step1_19;
 543     step2_28 = step1_31 - step1_28;
 544
 545     __asm__ __volatile__ (
 546         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 547         "mthi     $zero,                $ac0                            \n\t"
 548         "msub     $ac0,                 %[step2_19],    %[cospi_8_64]   \n\t"
 549         "madd     $ac0,                 %[step2_28],    %[cospi_24_64]  \n\t"
 550         "extp     %[step3_19],          $ac0,           31              \n\t"
 551
 552         : [step3_19] "=r" (step3_19)
 553         : [const_2_power_13] "r" (const_2_power_13),
 554           [step2_19] "r" (step2_19), [step2_28] "r" (step2_28),
 555           [cospi_24_64] "r" (cospi_24_64), [cospi_8_64] "r" (cospi_8_64)
 556     );
 557
 558     temp21 = step2_19 * cospi_24_64 + step2_28 * cospi_8_64;
 559     step3_28 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 560
 561     step3_16 = step1_16 + step1_19;
 562     step3_17 = step1_17 + step1_18;
 563     step3_30 = step1_29 + step1_30;
 564     step3_31 = step1_28 + step1_31;
 565
 566     step2_20 = step1_23 - step1_20;
 567     step2_27 = step1_24 - step1_27;
 568
 569     __asm__ __volatile__ (
 570         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 571         "mthi     $zero,                $ac0                            \n\t"
 572         "msub     $ac0,                 %[step2_20],    %[cospi_24_64]  \n\t"
 573         "msub     $ac0,                 %[step2_27],    %[cospi_8_64]   \n\t"
 574         "extp     %[step3_20],          $ac0,           31              \n\t"
 575
 576         : [step3_20] "=r" (step3_20)
 577         : [const_2_power_13] "r" (const_2_power_13),
 578           [step2_20] "r" (step2_20), [step2_27] "r" (step2_27),
 579           [cospi_24_64] "r" (cospi_24_64), [cospi_8_64] "r" (cospi_8_64)
 580     );
 581
 582     temp21 = -step2_20 * cospi_8_64 + step2_27 * cospi_24_64;
 583     step3_27 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 584
 585     step2_21 = step1_22 - step1_21;
 586     step2_26 = step1_25 - step1_26;
 587
 588     __asm__ __volatile__ (
 589         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 590         "mthi     $zero,                $ac1                            \n\t"
 591         "msub     $ac1,                 %[step2_21],    %[cospi_24_64]  \n\t"
 592         "msub     $ac1,                 %[step2_26],    %[cospi_8_64]   \n\t"
 593         "extp     %[step3_21],          $ac1,           31              \n\t"
 594
 595         : [step3_21] "=r" (step3_21)
 596         : [const_2_power_13] "r" (const_2_power_13),
 597           [step2_21] "r" (step2_21), [step2_26] "r" (step2_26),
 598           [cospi_24_64] "r" (cospi_24_64), [cospi_8_64] "r" (cospi_8_64)
 599     );
 600
 601     temp21 = -step2_21 * cospi_8_64 + step2_26 * cospi_24_64;
 602     step3_26 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 603
 604     step3_22 = step1_21 + step1_22;
 605     step3_23 = step1_20 + step1_23;
 606     step3_24 = step1_24 + step1_27;
 607     step3_25 = step1_25 + step1_26;
 608
 609     step2_16 = step3_16 + step3_23;
 610     step2_17 = step3_17 + step3_22;
 611     step2_18 = step3_18 + step3_21;
 612     step2_19 = step3_19 + step3_20;
 613     step2_20 = step3_19 - step3_20;
 614     step2_21 = step3_18 - step3_21;
 615     step2_22 = step3_17 - step3_22;
 616     step2_23 = step3_16 - step3_23;
 617
 618     step2_24 = step3_31 - step3_24;
 619     step2_25 = step3_30 - step3_25;
 620     step2_26 = step3_29 - step3_26;
 621     step2_27 = step3_28 - step3_27;
 622     step2_28 = step3_28 + step3_27;
 623     step2_29 = step3_29 + step3_26;
 624     step2_30 = step3_30 + step3_25;
 625     step2_31 = step3_31 + step3_24;
 626
 627     __asm__ __volatile__ (
 628         "lh       %[load1],             0(%[input])                     \n\t"
 629         "lh       %[load2],             32(%[input])                    \n\t"
 630         "lh       %[load3],             16(%[input])                    \n\t"
 631         "lh       %[load4],             48(%[input])                    \n\t"
 632
 633         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 634         "mthi     $zero,                $ac1                            \n\t"
 635         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 636         "mthi     $zero,                $ac2                            \n\t"
 637         "add      %[result1],           %[load1],       %[load2]        \n\t"
 638         "sub      %[result2],           %[load1],       %[load2]        \n\t"
 639         "madd     $ac1,                 %[result1],     %[cospi_16_64]  \n\t"
 640         "madd     $ac2,                 %[result2],     %[cospi_16_64]  \n\t"
 641         "extp     %[temp0],             $ac1,           31              \n\t"
 642         "extp     %[temp1],             $ac2,           31              \n\t"
 643
 644         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 645         "mthi     $zero,                $ac3                            \n\t"
 646         "madd     $ac3,                 %[load3],       %[cospi_24_64]  \n\t"
 647         "msub     $ac3,                 %[load4],       %[cospi_8_64]   \n\t"
 648         "extp     %[temp2],             $ac3,           31              \n\t"
 649
 650         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 651         "mthi     $zero,                $ac1                            \n\t"
 652         "madd     $ac1,                 %[load3],       %[cospi_8_64]   \n\t"
 653         "madd     $ac1,                 %[load4],       %[cospi_24_64]  \n\t"
 654         "extp     %[temp3],             $ac1,           31              \n\t"
 655
 656         "add      %[step1_0],          %[temp0],        %[temp3]        \n\t"
 657         "add      %[step1_1],          %[temp1],        %[temp2]        \n\t"
 658         "sub      %[step1_2],          %[temp1],        %[temp2]        \n\t"
 659         "sub      %[step1_3],          %[temp0],        %[temp3]        \n\t"
 660
 661         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 662           [load3] "=&r" (load3), [load4] "=&r" (load4),
 663           [result1] "=&r" (result1), [result2] "=&r" (result2),
 664           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 665           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 666           [step1_0] "=r" (step1_0), [step1_1] "=r" (step1_1),
 667           [step1_2] "=r" (step1_2), [step1_3] "=r" (step1_3)
 668         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 669           [cospi_16_64] "r" (cospi_16_64),
 670           [cospi_24_64] "r" (cospi_24_64), [cospi_8_64] "r" (cospi_8_64)
 671
 672     );
 673
 674     __asm__ __volatile__ (
 675         "lh       %[load1],             8(%[input])                     \n\t"
 676         "lh       %[load2],             56(%[input])                    \n\t"
 677         "lh       %[load3],             40(%[input])                    \n\t"
 678         "lh       %[load4],             24(%[input])                    \n\t"
 679
 680         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 681         "mthi     $zero,                $ac1                            \n\t"
 682         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 683         "mthi     $zero,                $ac3                            \n\t"
 684
 685         "madd     $ac1,                 %[load1],       %[cospi_28_64]  \n\t"
 686         "msub     $ac1,                 %[load2],       %[cospi_4_64]   \n\t"
 687         "extp     %[temp0],             $ac1,           31              \n\t"
 688
 689         "madd     $ac3,                 %[load1],       %[cospi_4_64]   \n\t"
 690         "madd     $ac3,                 %[load2],       %[cospi_28_64]  \n\t"
 691         "extp     %[temp3],             $ac3,           31              \n\t"
 692
 693         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 694         "mthi     $zero,                $ac1                            \n\t"
 695         "mtlo     %[const_2_power_13],  $ac2                            \n\t"
 696         "mthi     $zero,                $ac2                            \n\t"
 697
 698         "madd     $ac2,                 %[load3],       %[cospi_12_64]  \n\t"
 699         "msub     $ac2,                 %[load4],       %[cospi_20_64]  \n\t"
 700         "extp     %[temp1],             $ac2,           31              \n\t"
 701
 702         "madd     $ac1,                 %[load3],       %[cospi_20_64]  \n\t"
 703         "madd     $ac1,                 %[load4],       %[cospi_12_64]  \n\t"
 704         "extp     %[temp2],             $ac1,           31              \n\t"
 705
 706         "mtlo     %[const_2_power_13],  $ac1                            \n\t"
 707         "mthi     $zero,                $ac1                            \n\t"
 708         "mtlo     %[const_2_power_13],  $ac3                            \n\t"
 709         "mthi     $zero,                $ac3                            \n\t"
 710
 711         "sub      %[load1],             %[temp3],       %[temp2]        \n\t"
 712         "sub      %[load1],             %[load1],       %[temp0]        \n\t"
 713         "add      %[load1],             %[load1],       %[temp1]        \n\t"
 714
 715         "sub      %[load2],             %[temp0],       %[temp1]        \n\t"
 716         "sub      %[load2],             %[load2],       %[temp2]        \n\t"
 717         "add      %[load2],             %[load2],       %[temp3]        \n\t"
 718
 719         "madd     $ac1,                 %[load1],       %[cospi_16_64]  \n\t"
 720         "madd     $ac3,                 %[load2],       %[cospi_16_64]  \n\t"
 721
 722         "extp     %[step1_5],           $ac1,           31              \n\t"
 723         "extp     %[step1_6],           $ac3,           31              \n\t"
 724         "add      %[step1_4],           %[temp0],       %[temp1]        \n\t"
 725         "add      %[step1_7],           %[temp3],       %[temp2]        \n\t"
 726
 727         : [load1] "=&r" (load1), [load2] "=&r" (load2),
 728           [load3] "=&r" (load3), [load4] "=&r" (load4),
 729           [temp0] "=&r" (temp0), [temp1] "=&r" (temp1),
 730           [temp2] "=&r" (temp2), [temp3] "=&r" (temp3),
 731           [step1_4] "=r" (step1_4), [step1_5] "=r" (step1_5),
 732           [step1_6] "=r" (step1_6), [step1_7] "=r" (step1_7)
 733         : [const_2_power_13] "r" (const_2_power_13), [input] "r" (input),
 734           [cospi_20_64] "r" (cospi_20_64), [cospi_12_64] "r" (cospi_12_64),
 735           [cospi_4_64] "r" (cospi_4_64), [cospi_28_64] "r" (cospi_28_64),
 736           [cospi_16_64] "r" (cospi_16_64)
 737     );
 738
 739     step2_0 = step1_0 + step1_7;
 740     step2_1 = step1_1 + step1_6;
 741     step2_2 = step1_2 + step1_5;
 742     step2_3 = step1_3 + step1_4;
 743     step2_4 = step1_3 - step1_4;
 744     step2_5 = step1_2 - step1_5;
 745     step2_6 = step1_1 - step1_6;
 746     step2_7 = step1_0 - step1_7;
 747
 748     step1_0 = step2_0 + step3_15;
 749     step1_1 = step2_1 + step3_14;
 750     step1_2 = step2_2 + step3_13;
 751     step1_3 = step2_3 + step3_12;
 752     step1_4 = step2_4 + step3_11;
 753     step1_5 = step2_5 + step3_10;
 754     step1_6 = step2_6 + step3_9;
 755     step1_7 = step2_7 + step3_8;
 756     step1_8 = step2_7 - step3_8;
 757     step1_9 = step2_6 - step3_9;
 758     step1_10 = step2_5 - step3_10;
 759     step1_11 = step2_4 - step3_11;
 760     step1_12 = step2_3 - step3_12;
 761     step1_13 = step2_2 - step3_13;
 762     step1_14 = step2_1 - step3_14;
 763     step1_15 = step2_0 - step3_15;
 764
 765     __asm__ __volatile__ (
 766         "sub      %[temp0],             %[step2_27],    %[step2_20]     \n\t"
 767         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 768         "mthi     $zero,                $ac0                            \n\t"
 769         "madd     $ac0,                 %[temp0],       %[cospi_16_64]  \n\t"
 770         "extp     %[step1_20],          $ac0,           31              \n\t"
 771
 772         : [temp0] "=&r" (temp0), [step1_20] "=r" (step1_20)
 773         : [const_2_power_13] "r" (const_2_power_13),
 774           [step2_20] "r" (step2_20), [step2_27] "r" (step2_27),
 775           [cospi_16_64] "r" (cospi_16_64)
 776     );
 777
 778     temp21 = (step2_20 + step2_27) * cospi_16_64;
 779     step1_27 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 780
 781     __asm__ __volatile__ (
 782         "sub      %[temp0],             %[step2_26],    %[step2_21]     \n\t"
 783         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 784         "mthi     $zero,                $ac0                            \n\t"
 785         "madd     $ac0,                 %[temp0],       %[cospi_16_64]  \n\t"
 786         "extp     %[step1_21],          $ac0,           31              \n\t"
 787
 788         : [temp0] "=&r" (temp0), [step1_21] "=r" (step1_21)
 789         : [const_2_power_13] "r" (const_2_power_13),
 790           [step2_26] "r" (step2_26), [step2_21] "r" (step2_21),
 791           [cospi_16_64] "r" (cospi_16_64)
 792     );
 793
 794     temp21 = (step2_21 + step2_26) * cospi_16_64;
 795     step1_26 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 796
 797     __asm__ __volatile__ (
 798         "sub      %[temp0],             %[step2_25],    %[step2_22]     \n\t"
 799         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 800         "mthi     $zero,                $ac0                            \n\t"
 801         "madd     $ac0,                 %[temp0],       %[cospi_16_64]  \n\t"
 802         "extp     %[step1_22],          $ac0,           31              \n\t"
 803
 804         : [temp0] "=&r" (temp0), [step1_22] "=r" (step1_22)
 805         : [const_2_power_13] "r" (const_2_power_13),
 806           [step2_25] "r" (step2_25), [step2_22] "r" (step2_22),
 807           [cospi_16_64] "r" (cospi_16_64)
 808     );
 809
 810     temp21 = (step2_22 + step2_25) * cospi_16_64;
 811     step1_25 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 812
 813     __asm__ __volatile__ (
 814         "sub      %[temp0],             %[step2_24],    %[step2_23]     \n\t"
 815         "mtlo     %[const_2_power_13],  $ac0                            \n\t"
 816         "mthi     $zero,                $ac0                            \n\t"
 817         "madd     $ac0,                 %[temp0],       %[cospi_16_64]  \n\t"
 818         "extp     %[step1_23],          $ac0,           31              \n\t"
 819
 820         : [temp0] "=&r" (temp0), [step1_23] "=r" (step1_23)
 821         : [const_2_power_13] "r" (const_2_power_13),
 822           [step2_24] "r" (step2_24), [step2_23] "r" (step2_23),
 823           [cospi_16_64] "r" (cospi_16_64)
 824     );
 825
 826     temp21 = (step2_23 + step2_24) * cospi_16_64;
 827     step1_24 = (temp21 + DCT_CONST_ROUNDING) >> DCT_CONST_BITS;
 828
 829     // final stage
 830     output[0 * 32] = step1_0 + step2_31;
 831     output[1 * 32] = step1_1 + step2_30;
 832     output[2 * 32] = step1_2 + step2_29;
 833     output[3 * 32] = step1_3 + step2_28;
 834     output[4 * 32] = step1_4 + step1_27;
 835     output[5 * 32] = step1_5 + step1_26;
 836     output[6 * 32] = step1_6 + step1_25;
 837     output[7 * 32] = step1_7 + step1_24;
 838     output[8 * 32] = step1_8 + step1_23;
 839     output[9 * 32] = step1_9 + step1_22;
 840     output[10 * 32] = step1_10 + step1_21;
 841     output[11 * 32] = step1_11 + step1_20;
 842     output[12 * 32] = step1_12 + step2_19;
 843     output[13 * 32] = step1_13 + step2_18;
 844     output[14 * 32] = step1_14 + step2_17;
 845     output[15 * 32] = step1_15 + step2_16;
 846     output[16 * 32] = step1_15 - step2_16;
 847     output[17 * 32] = step1_14 - step2_17;
 848     output[18 * 32] = step1_13 - step2_18;
 849     output[19 * 32] = step1_12 - step2_19;
 850     output[20 * 32] = step1_11 - step1_20;
 851     output[21 * 32] = step1_10 - step1_21;
 852     output[22 * 32] = step1_9 - step1_22;
 853     output[23 * 32] = step1_8 - step1_23;
 854     output[24 * 32] = step1_7 - step1_24;
 855     output[25 * 32] = step1_6 - step1_25;
 856     output[26 * 32] = step1_5 - step1_26;
 857     output[27 * 32] = step1_4 - step1_27;
 858     output[28 * 32] = step1_3 - step2_28;
 859     output[29 * 32] = step1_2 - step2_29;
 860     output[30 * 32] = step1_1 - step2_30;
 861     output[31 * 32] = step1_0 - step2_31;
 862
 863     input += 32;
 864     output += 1;
 865   }
 866 }
 867
 868 void vp9_idct32x32_1024_add_dspr2(const int16_t *input, uint8_t *dest,
 869                                   int dest_stride) {
 870   DECLARE_ALIGNED(32, int16_t,  out[32 * 32]);
 871   int16_t *outptr = out;
 872   uint32_t pos = 45;
 873
 874   /* bit positon for extract from acc */
 875   __asm__ __volatile__ (
 876     "wrdsp      %[pos],     1           \n\t"
 877     :
 878     : [pos] "r" (pos)
 879   );
 880
 881   // Rows
 882   idct32_rows_dspr2(input, outptr, 32);
 883
 884   // Columns
 885   vp9_idct32_cols_add_blk_dspr2(out, dest, dest_stride);
 886 }
 887
 888 void vp9_idct32x32_34_add_dspr2(const int16_t *input, uint8_t *dest,
 889                                 int stride) {
 890   DECLARE_ALIGNED(32, int16_t,  out[32 * 32]);
 891   int16_t *outptr = out;
 892   uint32_t i;
 893   uint32_t pos = 45;
 894
 895   /* bit positon for extract from acc */
 896   __asm__ __volatile__ (
 897     "wrdsp      %[pos],     1           \n\t"
 898     :
 899     : [pos] "r" (pos)
 900   );
 901
 902   // Rows
 903   idct32_rows_dspr2(input, outptr, 8);
 904
 905   outptr += 8;
 906   __asm__ __volatile__ (
 907       "sw     $zero,      0(%[outptr])     \n\t"
 908       "sw     $zero,      4(%[outptr])     \n\t"
 909       "sw     $zero,      8(%[outptr])     \n\t"
 910       "sw     $zero,     12(%[outptr])     \n\t"
 911       "sw     $zero,     16(%[outptr])     \n\t"
 912       "sw     $zero,     20(%[outptr])     \n\t"
 913       "sw     $zero,     24(%[outptr])     \n\t"
 914       "sw     $zero,     28(%[outptr])     \n\t"
 915       "sw     $zero,     32(%[outptr])     \n\t"
 916       "sw     $zero,     36(%[outptr])     \n\t"
 917       "sw     $zero,     40(%[outptr])     \n\t"
 918       "sw     $zero,     44(%[outptr])     \n\t"
 919
 920       :
 921       : [outptr] "r" (outptr)
 922   );
 923
 924   for (i = 0; i < 31; ++i) {
 925     outptr += 32;
 926
 927     __asm__ __volatile__ (
 928         "sw     $zero,      0(%[outptr])     \n\t"
 929         "sw     $zero,      4(%[outptr])     \n\t"
 930         "sw     $zero,      8(%[outptr])     \n\t"
 931         "sw     $zero,     12(%[outptr])     \n\t"
 932         "sw     $zero,     16(%[outptr])     \n\t"
 933         "sw     $zero,     20(%[outptr])     \n\t"
 934         "sw     $zero,     24(%[outptr])     \n\t"
 935         "sw     $zero,     28(%[outptr])     \n\t"
 936         "sw     $zero,     32(%[outptr])     \n\t"
 937         "sw     $zero,     36(%[outptr])     \n\t"
 938         "sw     $zero,     40(%[outptr])     \n\t"
 939         "sw     $zero,     44(%[outptr])     \n\t"
 940
 941         :
 942         : [outptr] "r" (outptr)
 943     );
 944   }
 945
 946   // Columns
 947   vp9_idct32_cols_add_blk_dspr2(out, dest, stride);
 948 }
 949
 950 void vp9_idct32x32_1_add_dspr2(const int16_t *input, uint8_t *dest,
 951                                int stride) {
 952   int       r, out;
 953   int32_t   a1, absa1;
 954   int32_t   vector_a1;
 955   int32_t   t1, t2, t3, t4;
 956   int32_t   vector_1, vector_2, vector_3, vector_4;
 957   uint32_t  pos = 45;
 958
 959   /* bit positon for extract from acc */
 960   __asm__ __volatile__ (
 961     "wrdsp      %[pos],     1           \n\t"
 962
 963     :
 964     : [pos] "r" (pos)
 965   );
 966
 967   out = DCT_CONST_ROUND_SHIFT_TWICE_COSPI_16_64(input[0]);
 968   __asm__ __volatile__ (
 969       "addi     %[out],    %[out],    32      \n\t"
 970       "sra      %[a1],     %[out],    6       \n\t"
 971
 972       : [out] "+r" (out), [a1] "=r" (a1)
 973       :
 974   );
 975
 976   if (a1 < 0) {
 977     /* use quad-byte
 978      * input and output memory are four byte aligned */
 979     __asm__ __volatile__ (
 980         "abs        %[absa1],     %[a1]         \n\t"
 981         "replv.qb   %[vector_a1], %[absa1]      \n\t"
 982
 983         : [absa1] "=r" (absa1), [vector_a1] "=r" (vector_a1)
 984         : [a1] "r" (a1)
 985     );
 986
 987     for (r = 32; r--;) {
 988       __asm__ __volatile__ (
 989           "lw             %[t1],          0(%[dest])                      \n\t"
 990           "lw             %[t2],          4(%[dest])                      \n\t"
 991           "lw             %[t3],          8(%[dest])                      \n\t"
 992           "lw             %[t4],          12(%[dest])                     \n\t"
 993           "subu_s.qb      %[vector_1],    %[t1],          %[vector_a1]    \n\t"
 994           "subu_s.qb      %[vector_2],    %[t2],          %[vector_a1]    \n\t"
 995           "subu_s.qb      %[vector_3],    %[t3],          %[vector_a1]    \n\t"
 996           "subu_s.qb      %[vector_4],    %[t4],          %[vector_a1]    \n\t"
 997           "sw             %[vector_1],    0(%[dest])                      \n\t"
 998           "sw             %[vector_2],    4(%[dest])                      \n\t"
 999           "sw             %[vector_3],    8(%[dest])                      \n\t"
1000           "sw             %[vector_4],    12(%[dest])                     \n\t"
1001
1002           "lw             %[t1],          16(%[dest])                     \n\t"
1003           "lw             %[t2],          20(%[dest])                     \n\t"
1004           "lw             %[t3],          24(%[dest])                     \n\t"
1005           "lw             %[t4],          28(%[dest])                     \n\t"
1006           "subu_s.qb      %[vector_1],    %[t1],          %[vector_a1]    \n\t"
1007           "subu_s.qb      %[vector_2],    %[t2],          %[vector_a1]    \n\t"
1008           "subu_s.qb      %[vector_3],    %[t3],          %[vector_a1]    \n\t"
1009           "subu_s.qb      %[vector_4],    %[t4],          %[vector_a1]    \n\t"
1010           "sw             %[vector_1],    16(%[dest])                     \n\t"
1011           "sw             %[vector_2],    20(%[dest])                     \n\t"
1012           "sw             %[vector_3],    24(%[dest])                     \n\t"
1013           "sw             %[vector_4],    28(%[dest])                     \n\t"
1014
1015           "add            %[dest],        %[dest],        %[stride]       \n\t"
1016
1017           : [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3), [t4] "=&r" (t4),
1018             [vector_1] "=&r" (vector_1), [vector_2] "=&r" (vector_2),
1019             [vector_3] "=&r" (vector_3), [vector_4] "=&r" (vector_4),
1020             [dest] "+&r" (dest)
1021           : [stride] "r" (stride), [vector_a1] "r" (vector_a1)
1022       );
1023     }
1024   } else {
1025     /* use quad-byte
1026      * input and output memory are four byte aligned */
1027     __asm__ __volatile__ (
1028         "replv.qb       %[vector_a1],   %[a1]     \n\t"
1029
1030         : [vector_a1] "=r" (vector_a1)
1031         : [a1] "r" (a1)
1032     );
1033
1034     for (r = 32; r--;) {
1035       __asm__ __volatile__ (
1036           "lw             %[t1],          0(%[dest])                      \n\t"
1037           "lw             %[t2],          4(%[dest])                      \n\t"
1038           "lw             %[t3],          8(%[dest])                      \n\t"
1039           "lw             %[t4],          12(%[dest])                     \n\t"
1040           "addu_s.qb      %[vector_1],    %[t1],          %[vector_a1]    \n\t"
1041           "addu_s.qb      %[vector_2],    %[t2],          %[vector_a1]    \n\t"
1042           "addu_s.qb      %[vector_3],    %[t3],          %[vector_a1]    \n\t"
1043           "addu_s.qb      %[vector_4],    %[t4],          %[vector_a1]    \n\t"
1044           "sw             %[vector_1],    0(%[dest])                      \n\t"
1045           "sw             %[vector_2],    4(%[dest])                      \n\t"
1046           "sw             %[vector_3],    8(%[dest])                      \n\t"
1047           "sw             %[vector_4],    12(%[dest])                     \n\t"
1048
1049           "lw             %[t1],          16(%[dest])                     \n\t"
1050           "lw             %[t2],          20(%[dest])                     \n\t"
1051           "lw             %[t3],          24(%[dest])                     \n\t"
1052           "lw             %[t4],          28(%[dest])                     \n\t"
1053           "addu_s.qb      %[vector_1],    %[t1],          %[vector_a1]    \n\t"
1054           "addu_s.qb      %[vector_2],    %[t2],          %[vector_a1]    \n\t"
1055           "addu_s.qb      %[vector_3],    %[t3],          %[vector_a1]    \n\t"
1056           "addu_s.qb      %[vector_4],    %[t4],          %[vector_a1]    \n\t"
1057           "sw             %[vector_1],    16(%[dest])                     \n\t"
1058           "sw             %[vector_2],    20(%[dest])                     \n\t"
1059           "sw             %[vector_3],    24(%[dest])                     \n\t"
1060           "sw             %[vector_4],    28(%[dest])                     \n\t"
1061
1062           "add            %[dest],        %[dest],        %[stride]       \n\t"
1063
1064           : [t1] "=&r" (t1), [t2] "=&r" (t2), [t3] "=&r" (t3), [t4] "=&r" (t4),
1065             [vector_1] "=&r" (vector_1), [vector_2] "=&r" (vector_2),
1066             [vector_3] "=&r" (vector_3), [vector_4] "=&r" (vector_4),
1067             [dest] "+&r" (dest)
1068           : [stride] "r" (stride), [vector_a1] "r" (vector_a1)
1069       );
1070     }
1071   }
1072 }
1073 #endif  // #if HAVE_DSPR2