git.sesse.net Git - ffmpeg/blob - libavcodec/mips/hevc_idct_msa.c

   1 /*
   2  * Copyright (c) 2015 Manojkumar Bhosale (Manojkumar.Bhosale@imgtec.com)
   3  *
   4  * This file is part of FFmpeg.
   5  *
   6  * FFmpeg is free software; you can redistribute it and/or
   7  * modify it under the terms of the GNU Lesser General Public
   8  * License as published by the Free Software Foundation; either
   9  * version 2.1 of the License, or (at your option) any later version.
  10  *
  11  * FFmpeg is distributed in the hope that it will be useful,
  12  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  13  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  14  * Lesser General Public License for more details.
  15  *
  16  * You should have received a copy of the GNU Lesser General Public
  17  * License along with FFmpeg; if not, write to the Free Software
  18  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  19  */
  20
  21 #include "libavutil/mips/generic_macros_msa.h"
  22 #include "libavcodec/mips/hevcdsp_mips.h"
  23
  24 static const int16_t gt8x8_cnst[16] = {
  25     64, 64, 83, 36, 89, 50, 18, 75, 64, -64, 36, -83, 75, -89, -50, -18
  26 };
  27
  28 static const int16_t gt16x16_cnst[64] = {
  29     64, 83, 64, 36, 89, 75, 50, 18, 90, 80, 57, 25, 70, 87, 9, 43,
  30     64, 36, -64, -83, 75, -18, -89, -50, 87, 9, -80, -70, -43, 57, -25, -90,
  31     64, -36, -64, 83, 50, -89, 18, 75, 80, -70, -25, 90, -87, 9, 43, 57,
  32     64, -83, 64, -36, 18, -50, 75, -89, 70, -87, 90, -80, 9, -43, -57, 25
  33 };
  34
  35 static const int16_t gt32x32_cnst0[256] = {
  36     90, 90, 88, 85, 82, 78, 73, 67, 61, 54, 46, 38, 31, 22, 13, 4,
  37     90, 82, 67, 46, 22, -4, -31, -54, -73, -85, -90, -88, -78, -61, -38, -13,
  38     88, 67, 31, -13, -54, -82, -90, -78, -46, -4, 38, 73, 90, 85, 61, 22,
  39     85, 46, -13, -67, -90, -73, -22, 38, 82, 88, 54, -4, -61, -90, -78, -31,
  40     82, 22, -54, -90, -61, 13, 78, 85, 31, -46, -90, -67, 4, 73, 88, 38,
  41     78, -4, -82, -73, 13, 85, 67, -22, -88, -61, 31, 90, 54, -38, -90, -46,
  42     73, -31, -90, -22, 78, 67, -38, -90, -13, 82, 61, -46, -88, -4, 85, 54,
  43     67, -54, -78, 38, 85, -22, -90, 4, 90, 13, -88, -31, 82, 46, -73, -61,
  44     61, -73, -46, 82, 31, -88, -13, 90, -4, -90, 22, 85, -38, -78, 54, 67,
  45     54, -85, -4, 88, -46, -61, 82, 13, -90, 38, 67, -78, -22, 90, -31, -73,
  46     46, -90, 38, 54, -90, 31, 61, -88, 22, 67, -85, 13, 73, -82, 4, 78,
  47     38, -88, 73, -4, -67, 90, -46, -31, 85, -78, 13, 61, -90, 54, 22, -82,
  48     31, -78, 90, -61, 4, 54, -88, 82, -38, -22, 73, -90, 67, -13, -46, 85,
  49     22, -61, 85, -90, 73, -38, -4, 46, -78, 90, -82, 54, -13, -31, 67, -88,
  50     13, -38, 61, -78, 88, -90, 85, -73, 54, -31, 4, 22, -46, 67, -82, 90,
  51     4, -13, 22, -31, 38, -46, 54, -61, 67, -73, 78, -82, 85, -88, 90, -90
  52 };
  53
  54 static const int16_t gt32x32_cnst1[64] = {
  55     90, 87, 80, 70, 57, 43, 25, 9, 87, 57, 9, -43, -80, -90, -70, -25,
  56     80, 9, -70, -87, -25, 57, 90, 43, 70, -43, -87, 9, 90, 25, -80, -57,
  57     57, -80, -25, 90, -9, -87, 43, 70, 43, -90, 57, 25, -87, 70, 9, -80,
  58     25, -70, 90, -80, 43, 9, -57, 87, 9, -25, 43, -57, 70, -80, 87, -90
  59 };
  60
  61 static const int16_t gt32x32_cnst2[16] = {
  62     89, 75, 50, 18, 75, -18, -89, -50, 50, -89, 18, 75, 18, -50, 75, -89
  63 };
  64
  65 static const int16_t gt32x32_cnst3[16] = {
  66     64, 64, 64, 64, 83, 36, -36, -83, 64, -64, -64, 64, 36, -83, 83, -36
  67 };
  68
  69 #define HEVC_IDCT4x4_COL(in_r0, in_l0, in_r1, in_l1,          \
  70                          sum0, sum1, sum2, sum3, shift)       \
  71 {                                                             \
  72     v4i32 vec0, vec1, vec2, vec3, vec4, vec5;                 \
  73     v4i32 cnst64 = __msa_ldi_w(64);                           \
  74     v4i32 cnst83 = __msa_ldi_w(83);                           \
  75     v4i32 cnst36 = __msa_ldi_w(36);                           \
  76                                                               \
  77     DOTP_SH4_SW(in_r0, in_r1, in_l0, in_l1, cnst64, cnst64,   \
  78                 cnst83, cnst36, vec0, vec2, vec1, vec3);      \
  79     DOTP_SH2_SW(in_l0, in_l1, cnst36, cnst83, vec4, vec5);    \
  80                                                               \
  81     sum0 = vec0 + vec2;                                       \
  82     sum1 = vec0 - vec2;                                       \
  83     sum3 = sum0;                                              \
  84     sum2 = sum1;                                              \
  85                                                               \
  86     vec1 += vec3;                                             \
  87     vec4 -= vec5;                                             \
  88                                                               \
  89     sum0 += vec1;                                             \
  90     sum1 += vec4;                                             \
  91     sum2 -= vec4;                                             \
  92     sum3 -= vec1;                                             \
  93                                                               \
  94     SRARI_W4_SW(sum0, sum1, sum2, sum3, shift);               \
  95     SAT_SW4_SW(sum0, sum1, sum2, sum3, 15);                   \
  96 }
  97
  98 #define HEVC_IDCT8x8_COL(in0, in1, in2, in3, in4, in5, in6, in7, shift)  \
  99 {                                                                        \
 100     v8i16 src0_r, src1_r, src2_r, src3_r;                                \
 101     v8i16 src0_l, src1_l, src2_l, src3_l;                                \
 102     v8i16 filt0, filter0, filter1, filter2, filter3;                     \
 103     v4i32 temp0_r, temp1_r, temp2_r, temp3_r, temp4_r, temp5_r;          \
 104     v4i32 temp0_l, temp1_l, temp2_l, temp3_l, temp4_l, temp5_l;          \
 105     v4i32 sum0_r, sum1_r, sum2_r, sum3_r;                                \
 106     v4i32 sum0_l, sum1_l, sum2_l, sum3_l;                                \
 107                                                                          \
 108     ILVR_H4_SH(in4, in0, in6, in2, in5, in1, in3, in7,                   \
 109                src0_r, src1_r, src2_r, src3_r);                          \
 110     ILVL_H4_SH(in4, in0, in6, in2, in5, in1, in3, in7,                   \
 111                src0_l, src1_l, src2_l, src3_l);                          \
 112                                                                          \
 113     filt0 = LD_SH(filter);                                               \
 114     SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);             \
 115     DOTP_SH4_SW(src0_r, src0_l, src1_r, src1_l, filter0, filter0,        \
 116                 filter1, filter1, temp0_r, temp0_l, temp1_r, temp1_l);   \
 117                                                                          \
 118     BUTTERFLY_4(temp0_r, temp0_l, temp1_l, temp1_r, sum0_r, sum0_l,      \
 119                 sum1_l, sum1_r);                                         \
 120     sum2_r = sum1_r;                                                     \
 121     sum2_l = sum1_l;                                                     \
 122     sum3_r = sum0_r;                                                     \
 123     sum3_l = sum0_l;                                                     \
 124                                                                          \
 125     DOTP_SH4_SW(src2_r, src2_l, src3_r, src3_l,  filter2, filter2,       \
 126                 filter3, filter3, temp2_r, temp2_l, temp3_r, temp3_l);   \
 127                                                                          \
 128     temp2_r += temp3_r;                                                  \
 129     temp2_l += temp3_l;                                                  \
 130     sum0_r += temp2_r;                                                   \
 131     sum0_l += temp2_l;                                                   \
 132     sum3_r -= temp2_r;                                                   \
 133     sum3_l -= temp2_l;                                                   \
 134                                                                          \
 135     SRARI_W4_SW(sum0_r, sum0_l, sum3_r, sum3_l, shift);                  \
 136     SAT_SW4_SW(sum0_r, sum0_l, sum3_r, sum3_l, 15);                      \
 137     PCKEV_H2_SH(sum0_l, sum0_r, sum3_l, sum3_r, in0, in7);               \
 138     DOTP_SH4_SW(src2_r, src2_l, src3_r, src3_l,  filter3, filter3,       \
 139                 filter2, filter2, temp4_r, temp4_l, temp5_r, temp5_l);   \
 140                                                                          \
 141     temp4_r -= temp5_r;                                                  \
 142     temp4_l -= temp5_l;                                                  \
 143     sum1_r += temp4_r;                                                   \
 144     sum1_l += temp4_l;                                                   \
 145     sum2_r -= temp4_r;                                                   \
 146     sum2_l -= temp4_l;                                                   \
 147                                                                          \
 148     SRARI_W4_SW(sum1_r, sum1_l, sum2_r, sum2_l, shift);                  \
 149     SAT_SW4_SW(sum1_r, sum1_l, sum2_r, sum2_l, 15);                      \
 150     PCKEV_H2_SH(sum1_l, sum1_r, sum2_l, sum2_r, in3, in4);               \
 151                                                                          \
 152     filt0 = LD_SH(filter + 8);                                           \
 153     SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);             \
 154     DOTP_SH4_SW(src0_r, src0_l, src1_r, src1_l,  filter0, filter0,       \
 155                 filter1, filter1, temp0_r, temp0_l, temp1_r, temp1_l);   \
 156                                                                          \
 157     BUTTERFLY_4(temp0_r, temp0_l, temp1_l, temp1_r, sum0_r, sum0_l,      \
 158                 sum1_l, sum1_r);                                         \
 159     sum2_r = sum1_r;                                                     \
 160     sum2_l = sum1_l;                                                     \
 161     sum3_r = sum0_r;                                                     \
 162     sum3_l = sum0_l;                                                     \
 163                                                                          \
 164     DOTP_SH4_SW(src2_r, src2_l, src3_r, src3_l, filter2, filter2,        \
 165                 filter3, filter3, temp2_r, temp2_l, temp3_r, temp3_l);   \
 166                                                                          \
 167     temp2_r += temp3_r;                                                  \
 168     temp2_l += temp3_l;                                                  \
 169     sum0_r += temp2_r;                                                   \
 170     sum0_l += temp2_l;                                                   \
 171     sum3_r -= temp2_r;                                                   \
 172     sum3_l -= temp2_l;                                                   \
 173                                                                          \
 174     SRARI_W4_SW(sum0_r, sum0_l, sum3_r, sum3_l, shift);                  \
 175     SAT_SW4_SW(sum0_r, sum0_l, sum3_r, sum3_l, 15);                      \
 176     PCKEV_H2_SH(sum0_l, sum0_r, sum3_l, sum3_r, in1, in6);               \
 177     DOTP_SH4_SW(src2_r, src2_l, src3_r, src3_l, filter3, filter3,        \
 178                 filter2, filter2, temp4_r, temp4_l, temp5_r, temp5_l);   \
 179                                                                          \
 180     temp4_r -= temp5_r;                                                  \
 181     temp4_l -= temp5_l;                                                  \
 182     sum1_r -= temp4_r;                                                   \
 183     sum1_l -= temp4_l;                                                   \
 184     sum2_r += temp4_r;                                                   \
 185     sum2_l += temp4_l;                                                   \
 186                                                                          \
 187     SRARI_W4_SW(sum1_r, sum1_l, sum2_r, sum2_l, shift);                  \
 188     SAT_SW4_SW(sum1_r, sum1_l, sum2_r, sum2_l, 15);                      \
 189     PCKEV_H2_SH(sum1_l, sum1_r, sum2_l, sum2_r, in2, in5);               \
 190 }
 191
 192 #define HEVC_IDCT16x16_COL(src0_r, src1_r, src2_r, src3_r,                \
 193                            src4_r, src5_r, src6_r, src7_r,                \
 194                            src0_l, src1_l, src2_l, src3_l,                \
 195                            src4_l, src5_l, src6_l, src7_l, shift)         \
 196 {                                                                         \
 197     int16_t *ptr0, *ptr1;                                                 \
 198     v8i16 filt0, filt1, dst0, dst1;                                       \
 199     v8i16 filter0, filter1, filter2, filter3;                             \
 200     v4i32 temp0_r, temp1_r, temp0_l, temp1_l;                             \
 201     v4i32 sum0_r, sum1_r, sum2_r, sum3_r, sum0_l, sum1_l, sum2_l;         \
 202     v4i32 sum3_l, res0_r, res1_r, res0_l, res1_l;                         \
 203                                                                           \
 204     ptr0 = (buf_ptr + 112);                                               \
 205     ptr1 = (buf_ptr + 128);                                               \
 206     k = -1;                                                               \
 207                                                                           \
 208     for (j = 0; j < 4; j++)                                               \
 209     {                                                                     \
 210         LD_SH2(filter, 8, filt0, filt1)                                   \
 211         filter += 16;                                                     \
 212         SPLATI_W2_SH(filt0, 0, filter0, filter1);                         \
 213         SPLATI_W2_SH(filt1, 0, filter2, filter3);                         \
 214         DOTP_SH4_SW(src0_r, src0_l, src4_r, src4_l,  filter0, filter0,    \
 215                     filter2, filter2, sum0_r, sum0_l, sum2_r, sum2_l);    \
 216         DOTP_SH2_SW(src7_r, src7_l, filter2, filter2, sum3_r, sum3_l);    \
 217         DPADD_SH4_SW(src1_r, src1_l, src5_r, src5_l,  filter1, filter1,   \
 218                      filter3, filter3, sum0_r, sum0_l, sum2_r, sum2_l);   \
 219         DPADD_SH2_SW(src6_r, src6_l, filter3, filter3, sum3_r, sum3_l);   \
 220                                                                           \
 221         sum1_r = sum0_r;                                                  \
 222         sum1_l = sum0_l;                                                  \
 223                                                                           \
 224         SPLATI_W2_SH(filt0, 2, filter0, filter1);                         \
 225         SPLATI_W2_SH(filt1, 2, filter2, filter3);                         \
 226         DOTP_SH2_SW(src2_r, src2_l, filter0, filter0, temp0_r, temp0_l);  \
 227         DPADD_SH2_SW(src6_r, src6_l, filter2, filter2, sum2_r, sum2_l);   \
 228         DOTP_SH2_SW(src5_r, src5_l, filter2, filter2, temp1_r, temp1_l);  \
 229                                                                           \
 230         sum0_r += temp0_r;                                                \
 231         sum0_l += temp0_l;                                                \
 232         sum1_r -= temp0_r;                                                \
 233         sum1_l -= temp0_l;                                                \
 234                                                                           \
 235         sum3_r = temp1_r - sum3_r;                                        \
 236         sum3_l = temp1_l - sum3_l;                                        \
 237                                                                           \
 238         DOTP_SH2_SW(src3_r, src3_l, filter1, filter1, temp0_r, temp0_l);  \
 239         DPADD_SH4_SW(src7_r, src7_l, src4_r, src4_l, filter3, filter3,    \
 240                      filter3, filter3, sum2_r, sum2_l, sum3_r, sum3_l);   \
 241                                                                           \
 242         sum0_r += temp0_r;                                                \
 243         sum0_l += temp0_l;                                                \
 244         sum1_r -= temp0_r;                                                \
 245         sum1_l -= temp0_l;                                                \
 246                                                                           \
 247         BUTTERFLY_4(sum0_r, sum0_l, sum2_l, sum2_r, res0_r, res0_l,       \
 248                     res1_l, res1_r);                                      \
 249         SRARI_W4_SW(res0_r, res0_l, res1_r, res1_l, shift);               \
 250         SAT_SW4_SW(res0_r, res0_l, res1_r, res1_l, 15);                   \
 251         PCKEV_H2_SH(res0_l, res0_r, res1_l, res1_r, dst0, dst1);          \
 252         ST_SH(dst0, buf_ptr);                                             \
 253         ST_SH(dst1, (buf_ptr + ((15 - (j * 2)) * 16)));                   \
 254                                                                           \
 255         BUTTERFLY_4(sum1_r, sum1_l, sum3_l, sum3_r, res0_r, res0_l,       \
 256                     res1_l, res1_r);                                      \
 257         SRARI_W4_SW(res0_r, res0_l, res1_r, res1_l, shift);               \
 258         SAT_SW4_SW(res0_r, res0_l, res1_r, res1_l, 15);                   \
 259         PCKEV_H2_SH(res0_l, res0_r, res1_l, res1_r, dst0, dst1);          \
 260         ST_SH(dst0, (ptr0 + (((j / 2 + j % 2) * 2 * k) * 16)));           \
 261         ST_SH(dst1, (ptr1 - (((j / 2 + j % 2) * 2 * k) * 16)));           \
 262                                                                           \
 263         k *= -1;                                                          \
 264         buf_ptr += 16;                                                    \
 265     }                                                                     \
 266 }
 267
 268 #define HEVC_EVEN16_CALC(input, sum0_r, sum0_l, load_idx, store_idx)  \
 269 {                                                                     \
 270     LD_SW2(input + load_idx * 8, 4, tmp0_r, tmp0_l);                  \
 271     tmp1_r = sum0_r;                                                  \
 272     tmp1_l = sum0_l;                                                  \
 273     sum0_r += tmp0_r;                                                 \
 274     sum0_l += tmp0_l;                                                 \
 275     ST_SW2(sum0_r, sum0_l, (input + load_idx * 8), 4);                \
 276     tmp1_r -= tmp0_r;                                                 \
 277     tmp1_l -= tmp0_l;                                                 \
 278     ST_SW2(tmp1_r, tmp1_l, (input + store_idx * 8), 4);               \
 279 }
 280
 281 #define HEVC_IDCT_LUMA4x4_COL(in_r0, in_l0, in_r1, in_l1,     \
 282                               res0, res1, res2, res3, shift)  \
 283 {                                                             \
 284     v4i32 vec0, vec1, vec2, vec3;                             \
 285     v4i32 cnst74 = __msa_ldi_w(74);                           \
 286     v4i32 cnst55 = __msa_ldi_w(55);                           \
 287     v4i32 cnst29 = __msa_ldi_w(29);                           \
 288                                                               \
 289     vec0 = in_r0 + in_r1;                                     \
 290     vec2 = in_r0 - in_l1;                                     \
 291     res0 = vec0 * cnst29;                                     \
 292     res1 = vec2 * cnst55;                                     \
 293     res2 = in_r0 - in_r1;                                     \
 294     vec1 = in_r1 + in_l1;                                     \
 295     res2 += in_l1;                                            \
 296     vec3 = in_l0 * cnst74;                                    \
 297     res3 = vec0 * cnst55;                                     \
 298                                                               \
 299     res0 += vec1 * cnst55;                                    \
 300     res1 -= vec1 * cnst29;                                    \
 301     res2 *= cnst74;                                           \
 302     res3 += vec2 * cnst29;                                    \
 303                                                               \
 304     res0 += vec3;                                             \
 305     res1 += vec3;                                             \
 306     res3 -= vec3;                                             \
 307                                                               \
 308     SRARI_W4_SW(res0, res1, res2, res3, shift);               \
 309     SAT_SW4_SW(res0, res1, res2, res3, 15);                   \
 310 }
 311
 312 static void hevc_idct_4x4_msa(int16_t *coeffs)
 313 {
 314     v8i16 in0, in1;
 315     v4i32 in_r0, in_l0, in_r1, in_l1;
 316     v4i32 sum0, sum1, sum2, sum3;
 317     v8i16 zeros = { 0 };
 318
 319     LD_SH2(coeffs, 8, in0, in1);
 320     ILVRL_H2_SW(zeros, in0, in_r0, in_l0);
 321     ILVRL_H2_SW(zeros, in1, in_r1, in_l1);
 322
 323     HEVC_IDCT4x4_COL(in_r0, in_l0, in_r1, in_l1, sum0, sum1, sum2, sum3, 7);
 324     TRANSPOSE4x4_SW_SW(sum0, sum1, sum2, sum3, in_r0, in_l0, in_r1, in_l1);
 325     HEVC_IDCT4x4_COL(in_r0, in_l0, in_r1, in_l1, sum0, sum1, sum2, sum3, 12);
 326     TRANSPOSE4x4_SW_SW(sum0, sum1, sum2, sum3, sum0, sum1, sum2, sum3);
 327     PCKEV_H2_SH(sum1, sum0, sum3, sum2, in0, in1);
 328     ST_SH2(in0, in1, coeffs, 8);
 329 }
 330
 331 static void hevc_idct_8x8_msa(int16_t *coeffs)
 332 {
 333     int16_t *filter = &gt8x8_cnst[0];
 334     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 335
 336     LD_SH8(coeffs, 8, in0, in1, in2, in3, in4, in5, in6, in7);
 337     HEVC_IDCT8x8_COL(in0, in1, in2, in3, in4, in5, in6, in7, 7);
 338     TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 339                        in0, in1, in2, in3, in4, in5, in6, in7);
 340     HEVC_IDCT8x8_COL(in0, in1, in2, in3, in4, in5, in6, in7, 12);
 341     TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 342                        in0, in1, in2, in3, in4, in5, in6, in7);
 343     ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, coeffs, 8);
 344 }
 345
 346 static void hevc_idct_16x16_msa(int16_t *coeffs)
 347 {
 348     int16_t i, j, k;
 349     int16_t buf[256];
 350     int16_t *buf_ptr = &buf[0];
 351     int16_t *src = coeffs;
 352     int16_t *filter = &gt16x16_cnst[0];
 353     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 354     v8i16 in8, in9, in10, in11, in12, in13, in14, in15;
 355     v8i16 vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7;
 356     v8i16 src0_r, src1_r, src2_r, src3_r, src4_r, src5_r, src6_r, src7_r;
 357     v8i16 src0_l, src1_l, src2_l, src3_l, src4_l, src5_l, src6_l, src7_l;
 358
 359     for (i = 2; i--;) {
 360         LD_SH16(src, 16, in0, in1, in2, in3, in4, in5, in6, in7,
 361                 in8, in9, in10, in11, in12, in13, in14, in15);
 362
 363         ILVR_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
 364                    src0_r, src1_r, src2_r, src3_r);
 365         ILVR_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
 366                    src4_r, src5_r, src6_r, src7_r);
 367         ILVL_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
 368                    src0_l, src1_l, src2_l, src3_l);
 369         ILVL_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
 370                    src4_l, src5_l, src6_l, src7_l);
 371         HEVC_IDCT16x16_COL(src0_r, src1_r, src2_r, src3_r, src4_r, src5_r,
 372                            src6_r, src7_r, src0_l, src1_l, src2_l, src3_l,
 373                            src4_l, src5_l, src6_l, src7_l, 7);
 374
 375         src += 8;
 376         buf_ptr = (&buf[0] + 8);
 377         filter = &gt16x16_cnst[0];
 378     }
 379
 380     src = &buf[0];
 381     buf_ptr = coeffs;
 382     filter = &gt16x16_cnst[0];
 383
 384     for (i = 2; i--;) {
 385         LD_SH16(src, 8, in0, in8, in1, in9, in2, in10, in3, in11,
 386                 in4, in12, in5, in13, in6, in14, in7, in15);
 387         TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 388                            in0, in1, in2, in3, in4, in5, in6, in7);
 389         TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
 390                            in8, in9, in10, in11, in12, in13, in14, in15);
 391         ILVR_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
 392                    src0_r, src1_r, src2_r, src3_r);
 393         ILVR_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
 394                    src4_r, src5_r, src6_r, src7_r);
 395         ILVL_H4_SH(in4, in0, in12, in8, in6, in2, in14, in10,
 396                    src0_l, src1_l, src2_l, src3_l);
 397         ILVL_H4_SH(in5, in1, in13, in9, in3, in7, in11, in15,
 398                    src4_l, src5_l, src6_l, src7_l);
 399         HEVC_IDCT16x16_COL(src0_r, src1_r, src2_r, src3_r, src4_r, src5_r,
 400                            src6_r, src7_r, src0_l, src1_l, src2_l, src3_l,
 401                            src4_l, src5_l, src6_l, src7_l, 12);
 402
 403         src += 128;
 404         buf_ptr = coeffs + 8;
 405         filter = &gt16x16_cnst[0];
 406     }
 407
 408     LD_SH8(coeffs, 16, in0, in1, in2, in3, in4, in5, in6, in7);
 409     TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 410                        vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
 411     ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, coeffs, 16);
 412
 413     LD_SH8((coeffs + 8), 16, in0, in1, in2, in3, in4, in5, in6, in7);
 414     TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 415                        vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
 416     LD_SH8((coeffs + 128), 16, in8, in9, in10, in11, in12, in13, in14, in15);
 417     ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, (coeffs + 128), 16);
 418     TRANSPOSE8x8_SH_SH(in8, in9, in10, in11, in12, in13, in14, in15,
 419                        vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
 420     ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, (coeffs + 8), 16);
 421
 422     LD_SH8((coeffs + 136), 16, in0, in1, in2, in3, in4, in5, in6, in7);
 423     TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 424                        vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7);
 425     ST_SH8(vec0, vec1, vec2, vec3, vec4, vec5, vec6, vec7, (coeffs + 136), 16);
 426 }
 427
 428 static void hevc_idct_8x32_column_msa(int16_t *coeffs, uint8_t buf_pitch,
 429                                       uint8_t round)
 430 {
 431     uint8_t i;
 432     int16_t *filter_ptr0 = &gt32x32_cnst0[0];
 433     int16_t *filter_ptr1 = &gt32x32_cnst1[0];
 434     int16_t *filter_ptr2 = &gt32x32_cnst2[0];
 435     int16_t *filter_ptr3 = &gt32x32_cnst3[0];
 436     int16_t *src0 = (coeffs + buf_pitch);
 437     int16_t *src1 = (coeffs + 2 * buf_pitch);
 438     int16_t *src2 = (coeffs + 4 * buf_pitch);
 439     int16_t *src3 = (coeffs);
 440     int32_t cnst0, cnst1;
 441     int32_t tmp_buf[8 * 32];
 442     int32_t *tmp_buf_ptr = &tmp_buf[0];
 443     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 444     v8i16 src0_r, src1_r, src2_r, src3_r, src4_r, src5_r, src6_r, src7_r;
 445     v8i16 src0_l, src1_l, src2_l, src3_l, src4_l, src5_l, src6_l, src7_l;
 446     v8i16 filt0, filter0, filter1, filter2, filter3;
 447     v4i32 sum0_r, sum0_l, sum1_r, sum1_l, tmp0_r, tmp0_l, tmp1_r, tmp1_l;
 448
 449     /* process coeff 4, 12, 20, 28 */
 450     LD_SH4(src2, 8 * buf_pitch, in0, in1, in2, in3);
 451     ILVR_H2_SH(in1, in0, in3, in2, src0_r, src1_r);
 452     ILVL_H2_SH(in1, in0, in3, in2, src0_l, src1_l);
 453
 454     /* loop for all columns of constants */
 455     for (i = 0; i < 4; i++) {
 456         /* processing single column of constants */
 457         cnst0 = LW(filter_ptr2);
 458         cnst1 = LW(filter_ptr2 + 2);
 459
 460         filter0 = (v8i16) __msa_fill_w(cnst0);
 461         filter1 = (v8i16) __msa_fill_w(cnst1);
 462
 463         DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
 464         DPADD_SH2_SW(src1_r, src1_l, filter1, filter1, sum0_r, sum0_l);
 465         ST_SW2(sum0_r, sum0_l, (tmp_buf_ptr + i * 8), 4);
 466
 467         filter_ptr2 += 4;
 468     }
 469
 470     /* process coeff 0, 8, 16, 24 */
 471     LD_SH2(src3, 16 * buf_pitch, in0, in2);
 472     LD_SH2((src3 + 8 * buf_pitch), 16 * buf_pitch, in1, in3);
 473
 474     ILVR_H2_SH(in2, in0, in3, in1, src0_r, src1_r);
 475     ILVL_H2_SH(in2, in0, in3, in1, src0_l, src1_l);
 476
 477     /* loop for all columns of constants */
 478     for (i = 0; i < 2; i++) {
 479         /* processing first column of filter constants */
 480         cnst0 = LW(filter_ptr3);
 481         cnst1 = LW(filter_ptr3 + 4);
 482
 483         filter0 = (v8i16) __msa_fill_w(cnst0);
 484         filter1 = (v8i16) __msa_fill_w(cnst1);
 485
 486         DOTP_SH4_SW(src0_r, src0_l, src1_r, src1_l, filter0, filter0, filter1,
 487                     filter1, sum0_r, sum0_l, tmp1_r, tmp1_l);
 488
 489         sum1_r = sum0_r;
 490         sum1_l = sum0_l;
 491         sum0_r += tmp1_r;
 492         sum0_l += tmp1_l;
 493
 494         sum1_r -= tmp1_r;
 495         sum1_l -= tmp1_l;
 496
 497         HEVC_EVEN16_CALC(tmp_buf_ptr, sum0_r, sum0_l, i, (7 - i));
 498         HEVC_EVEN16_CALC(tmp_buf_ptr, sum1_r, sum1_l, (3 - i), (4 + i));
 499
 500         filter_ptr3 += 8;
 501     }
 502
 503     /* process coeff 2 6 10 14 18 22 26 30 */
 504     LD_SH8(src1, 4 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
 505     ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
 506                src0_r, src1_r, src2_r, src3_r);
 507     ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
 508                src0_l, src1_l, src2_l, src3_l);
 509
 510     /* loop for all columns of constants */
 511     for (i = 0; i < 8; i++) {
 512         /* processing single column of constants */
 513         filt0 = LD_SH(filter_ptr1);
 514         SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
 515         DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
 516         DPADD_SH4_SW(src1_r, src1_l, src2_r, src2_l, filter1, filter1, filter2,
 517                      filter2, sum0_r, sum0_l, sum0_r, sum0_l);
 518         DPADD_SH2_SW(src3_r, src3_l, filter3, filter3, sum0_r, sum0_l);
 519
 520         LD_SW2(tmp_buf_ptr + i * 8, 4, tmp0_r, tmp0_l);
 521         tmp1_r = tmp0_r;
 522         tmp1_l = tmp0_l;
 523         tmp0_r += sum0_r;
 524         tmp0_l += sum0_l;
 525         ST_SW2(tmp0_r, tmp0_l, (tmp_buf_ptr + i * 8), 4);
 526         tmp1_r -= sum0_r;
 527         tmp1_l -= sum0_l;
 528         ST_SW2(tmp1_r, tmp1_l, (tmp_buf_ptr + (15 - i) * 8), 4);
 529
 530         filter_ptr1 += 8;
 531     }
 532
 533     /* process coeff 1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 */
 534     LD_SH8(src0, 2 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
 535     src0 += 16 * buf_pitch;
 536     ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
 537                src0_r, src1_r, src2_r, src3_r);
 538     ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
 539                src0_l, src1_l, src2_l, src3_l);
 540
 541     LD_SH8(src0, 2 * buf_pitch, in0, in1, in2, in3, in4, in5, in6, in7);
 542     ILVR_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
 543                src4_r, src5_r, src6_r, src7_r);
 544     ILVL_H4_SH(in1, in0, in3, in2, in5, in4, in7, in6,
 545                src4_l, src5_l, src6_l, src7_l);
 546
 547     /* loop for all columns of filter constants */
 548     for (i = 0; i < 16; i++) {
 549         /* processing single column of constants */
 550         filt0 = LD_SH(filter_ptr0);
 551         SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
 552         DOTP_SH2_SW(src0_r, src0_l, filter0, filter0, sum0_r, sum0_l);
 553         DPADD_SH4_SW(src1_r, src1_l, src2_r, src2_l, filter1, filter1, filter2,
 554                      filter2, sum0_r, sum0_l, sum0_r, sum0_l);
 555         DPADD_SH2_SW(src3_r, src3_l, filter3, filter3, sum0_r, sum0_l);
 556
 557         tmp1_r = sum0_r;
 558         tmp1_l = sum0_l;
 559
 560         filt0 = LD_SH(filter_ptr0 + 8);
 561         SPLATI_W4_SH(filt0, filter0, filter1, filter2, filter3);
 562         DOTP_SH2_SW(src4_r, src4_l, filter0, filter0, sum0_r, sum0_l);
 563         DPADD_SH4_SW(src5_r, src5_l, src6_r, src6_l, filter1, filter1, filter2,
 564                      filter2, sum0_r, sum0_l, sum0_r, sum0_l);
 565         DPADD_SH2_SW(src7_r, src7_l, filter3, filter3, sum0_r, sum0_l);
 566
 567         sum0_r += tmp1_r;
 568         sum0_l += tmp1_l;
 569
 570         LD_SW2(tmp_buf_ptr + i * 8, 4, tmp0_r, tmp0_l);
 571         tmp1_r = tmp0_r;
 572         tmp1_l = tmp0_l;
 573         tmp0_r += sum0_r;
 574         tmp0_l += sum0_l;
 575         sum1_r = __msa_fill_w(round);
 576         SRAR_W2_SW(tmp0_r, tmp0_l, sum1_r);
 577         SAT_SW2_SW(tmp0_r, tmp0_l, 15);
 578         in0 = __msa_pckev_h((v8i16) tmp0_l, (v8i16) tmp0_r);
 579         ST_SH(in0, (coeffs + i * buf_pitch));
 580         tmp1_r -= sum0_r;
 581         tmp1_l -= sum0_l;
 582         SRAR_W2_SW(tmp1_r, tmp1_l, sum1_r);
 583         SAT_SW2_SW(tmp1_r, tmp1_l, 15);
 584         in0 = __msa_pckev_h((v8i16) tmp1_l, (v8i16) tmp1_r);
 585         ST_SH(in0, (coeffs + (31 - i) * buf_pitch));
 586
 587         filter_ptr0 += 16;
 588     }
 589 }
 590
 591 static void hevc_idct_transpose_32x8_to_8x32(int16_t *coeffs, int16_t *tmp_buf)
 592 {
 593     uint8_t i;
 594     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 595
 596     for (i = 0; i < 4; i++) {
 597         LD_SH8(coeffs + i * 8, 32, in0, in1, in2, in3, in4, in5, in6, in7);
 598         TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 599                            in0, in1, in2, in3, in4, in5, in6, in7);
 600         ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, tmp_buf + i * 8 * 8, 8);
 601     }
 602 }
 603
 604 static void hevc_idct_transpose_8x32_to_32x8(int16_t *tmp_buf, int16_t *coeffs)
 605 {
 606     uint8_t i;
 607     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 608
 609     for (i = 0; i < 4; i++) {
 610         LD_SH8(tmp_buf + i * 8 * 8, 8, in0, in1, in2, in3, in4, in5, in6, in7);
 611         TRANSPOSE8x8_SH_SH(in0, in1, in2, in3, in4, in5, in6, in7,
 612                            in0, in1, in2, in3, in4, in5, in6, in7);
 613         ST_SH8(in0, in1, in2, in3, in4, in5, in6, in7, coeffs + i * 8, 32);
 614     }
 615 }
 616
 617 static void hevc_idct_32x32_msa(int16_t *coeffs)
 618 {
 619     uint8_t row_cnt, col_cnt;
 620     int16_t *src = coeffs;
 621     int16_t tmp_buf[8 * 32];
 622     int16_t *tmp_buf_ptr = &tmp_buf[0];
 623     uint8_t round;
 624     uint8_t buf_pitch;
 625
 626     /* column transform */
 627     round = 7;
 628     buf_pitch = 32;
 629     for (col_cnt = 0; col_cnt < 4; col_cnt++) {
 630         /* process 8x32 blocks */
 631         hevc_idct_8x32_column_msa((coeffs + col_cnt * 8), buf_pitch, round);
 632     }
 633
 634     /* row transform */
 635     round = 12;
 636     buf_pitch = 8;
 637     for (row_cnt = 0; row_cnt < 4; row_cnt++) {
 638         /* process 32x8 blocks */
 639         src = (coeffs + 32 * 8 * row_cnt);
 640
 641         hevc_idct_transpose_32x8_to_8x32(src, tmp_buf_ptr);
 642         hevc_idct_8x32_column_msa(tmp_buf_ptr, buf_pitch, round);
 643         hevc_idct_transpose_8x32_to_32x8(tmp_buf_ptr, src);
 644     }
 645 }
 646
 647 static void hevc_idct_dc_4x4_msa(int16_t *coeffs)
 648 {
 649     int32_t val;
 650     v8i16 dst;
 651
 652     val = (coeffs[0] + 1) >> 1;
 653     val = (val + 32) >> 6;
 654     dst = __msa_fill_h(val);
 655
 656     ST_SH2(dst, dst, coeffs, 8);
 657 }
 658
 659 static void hevc_idct_dc_8x8_msa(int16_t *coeffs)
 660 {
 661     int32_t val;
 662     v8i16 dst;
 663
 664     val = (coeffs[0] + 1) >> 1;
 665     val = (val + 32) >> 6;
 666     dst = __msa_fill_h(val);
 667
 668     ST_SH8(dst, dst, dst, dst, dst, dst, dst, dst, coeffs, 8);
 669 }
 670
 671 static void hevc_idct_dc_16x16_msa(int16_t *coeffs)
 672 {
 673     uint8_t loop;
 674     int32_t val;
 675     v8i16 dst;
 676
 677     val = (coeffs[0] + 1) >> 1;
 678     val = (val + 32) >> 6;
 679     dst = __msa_fill_h(val);
 680
 681     for (loop = 4; loop--;) {
 682         ST_SH8(dst, dst, dst, dst, dst, dst, dst, dst, coeffs, 8);
 683         coeffs += 8 * 8;
 684     }
 685 }
 686
 687 static void hevc_idct_dc_32x32_msa(int16_t *coeffs)
 688 {
 689     uint8_t loop;
 690     int32_t val;
 691     v8i16 dst;
 692
 693     val = (coeffs[0] + 1) >> 1;
 694     val = (val + 32) >> 6;
 695     dst = __msa_fill_h(val);
 696
 697     for (loop = 16; loop--;) {
 698         ST_SH8(dst, dst, dst, dst, dst, dst, dst, dst, coeffs, 8);
 699         coeffs += 8 * 8;
 700     }
 701 }
 702
 703 static void hevc_addblk_4x4_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
 704 {
 705     uint32_t dst0, dst1, dst2, dst3;
 706     v8i16 dst_r0, dst_l0, in0, in1;
 707     v4i32 dst_vec = { 0 };
 708     v16u8 zeros = { 0 };
 709
 710     LD_SH2(coeffs, 8, in0, in1);
 711     LW4(dst, stride, dst0, dst1, dst2, dst3);
 712     INSERT_W4_SW(dst0, dst1, dst2, dst3, dst_vec);
 713     ILVRL_B2_SH(zeros, dst_vec, dst_r0, dst_l0);
 714     ADD2(dst_r0, in0, dst_l0, in1, dst_r0, dst_l0);
 715     CLIP_SH2_0_255(dst_r0, dst_l0);
 716     dst_vec = (v4i32) __msa_pckev_b((v16i8) dst_l0, (v16i8) dst_r0);
 717     ST4x4_UB(dst_vec, dst_vec, 0, 1, 2, 3, dst, stride);
 718 }
 719
 720 static void hevc_addblk_8x8_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
 721 {
 722     uint8_t *temp_dst = dst;
 723     uint64_t dst0, dst1, dst2, dst3;
 724     v2i64 dst_vec0 = { 0 };
 725     v2i64 dst_vec1 = { 0 };
 726     v8i16 dst_r0, dst_l0, dst_r1, dst_l1;
 727     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 728     v16u8 zeros = { 0 };
 729
 730     LD_SH8(coeffs, 8, in0, in1, in2, in3, in4, in5, in6, in7);
 731     LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
 732     temp_dst += (4 * stride);
 733
 734     INSERT_D2_SD(dst0, dst1, dst_vec0);
 735     INSERT_D2_SD(dst2, dst3, dst_vec1);
 736     ILVRL_B2_SH(zeros, dst_vec0, dst_r0, dst_l0);
 737     ILVRL_B2_SH(zeros, dst_vec1, dst_r1, dst_l1);
 738     ADD4(dst_r0, in0, dst_l0, in1, dst_r1, in2, dst_l1, in3,
 739          dst_r0, dst_l0, dst_r1, dst_l1);
 740     CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
 741     PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
 742     ST8x4_UB(dst_r0, dst_r1, dst, stride);
 743     dst += (4 * stride);
 744
 745     LD4(temp_dst, stride, dst0, dst1, dst2, dst3);
 746     INSERT_D2_SD(dst0, dst1, dst_vec0);
 747     INSERT_D2_SD(dst2, dst3, dst_vec1);
 748     UNPCK_UB_SH(dst_vec0, dst_r0, dst_l0);
 749     UNPCK_UB_SH(dst_vec1, dst_r1, dst_l1);
 750     ADD4(dst_r0, in4, dst_l0, in5, dst_r1, in6, dst_l1, in7,
 751          dst_r0, dst_l0, dst_r1, dst_l1);
 752     CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
 753     PCKEV_B2_SH(dst_l0, dst_r0, dst_l1, dst_r1, dst_r0, dst_r1);
 754     ST8x4_UB(dst_r0, dst_r1, dst, stride);
 755 }
 756
 757 static void hevc_addblk_16x16_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
 758 {
 759     uint8_t loop_cnt;
 760     uint8_t *temp_dst = dst;
 761     v16u8 dst0, dst1, dst2, dst3;
 762     v8i16 dst_r0, dst_l0, dst_r1, dst_l1, dst_r2, dst_l2, dst_r3, dst_l3;
 763     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 764
 765     for (loop_cnt = 4; loop_cnt--;) {
 766         LD_SH4(coeffs, 16, in0, in2, in4, in6);
 767         LD_SH4((coeffs + 8), 16, in1, in3, in5, in7);
 768         coeffs += 64;
 769         LD_UB4(temp_dst, stride, dst0, dst1, dst2, dst3);
 770         temp_dst += (4 * stride);
 771
 772         UNPCK_UB_SH(dst0, dst_r0, dst_l0);
 773         UNPCK_UB_SH(dst1, dst_r1, dst_l1);
 774         UNPCK_UB_SH(dst2, dst_r2, dst_l2);
 775         UNPCK_UB_SH(dst3, dst_r3, dst_l3);
 776
 777         dst_r0 += in0;
 778         dst_l0 += in1;
 779         dst_r1 += in2;
 780         dst_l1 += in3;
 781         dst_r2 += in4;
 782         dst_l2 += in5;
 783         dst_r3 += in6;
 784         dst_l3 += in7;
 785
 786         CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
 787         CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
 788         PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
 789                     dst_r3, dst0, dst1, dst2, dst3);
 790         ST_UB4(dst0, dst1, dst2, dst3, dst, stride);
 791         dst += (4 * stride);
 792     }
 793 }
 794
 795 static void hevc_addblk_32x32_msa(int16_t *coeffs, uint8_t *dst, int32_t stride)
 796 {
 797     uint8_t loop_cnt;
 798     uint8_t *temp_dst = dst;
 799     v16u8 dst0, dst1, dst2, dst3;
 800     v8i16 dst_r0, dst_l0, dst_r1, dst_l1, dst_r2, dst_l2, dst_r3, dst_l3;
 801     v8i16 in0, in1, in2, in3, in4, in5, in6, in7;
 802
 803     for (loop_cnt = 8; loop_cnt--;) {
 804         LD_SH4(coeffs, 32, in0, in2, in4, in6);
 805         LD_SH4((coeffs + 8), 32, in1, in3, in5, in7);
 806         LD_UB4(temp_dst, stride, dst0, dst1, dst2, dst3);
 807
 808         UNPCK_UB_SH(dst0, dst_r0, dst_l0);
 809         UNPCK_UB_SH(dst1, dst_r1, dst_l1);
 810         UNPCK_UB_SH(dst2, dst_r2, dst_l2);
 811         UNPCK_UB_SH(dst3, dst_r3, dst_l3);
 812
 813         dst_r0 += in0;
 814         dst_l0 += in1;
 815         dst_r1 += in2;
 816         dst_l1 += in3;
 817         dst_r2 += in4;
 818         dst_l2 += in5;
 819         dst_r3 += in6;
 820         dst_l3 += in7;
 821
 822         CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
 823         CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
 824         PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
 825                     dst_r3, dst0, dst1, dst2, dst3);
 826         ST_UB4(dst0, dst1, dst2, dst3, dst, stride);
 827
 828         LD_SH4((coeffs + 16), 32, in0, in2, in4, in6);
 829         LD_SH4((coeffs + 24), 32, in1, in3, in5, in7);
 830         coeffs += 128;
 831         LD_UB4((temp_dst + 16), stride, dst0, dst1, dst2, dst3);
 832         temp_dst += (4 * stride);
 833
 834         UNPCK_UB_SH(dst0, dst_r0, dst_l0);
 835         UNPCK_UB_SH(dst1, dst_r1, dst_l1);
 836         UNPCK_UB_SH(dst2, dst_r2, dst_l2);
 837         UNPCK_UB_SH(dst3, dst_r3, dst_l3);
 838
 839         dst_r0 += in0;
 840         dst_l0 += in1;
 841         dst_r1 += in2;
 842         dst_l1 += in3;
 843         dst_r2 += in4;
 844         dst_l2 += in5;
 845         dst_r3 += in6;
 846         dst_l3 += in7;
 847
 848         CLIP_SH4_0_255(dst_r0, dst_l0, dst_r1, dst_l1);
 849         CLIP_SH4_0_255(dst_r2, dst_l2, dst_r3, dst_l3);
 850         PCKEV_B4_UB(dst_l0, dst_r0, dst_l1, dst_r1, dst_l2, dst_r2, dst_l3,
 851                     dst_r3, dst0, dst1, dst2, dst3);
 852
 853         ST_UB4(dst0, dst1, dst2, dst3, (dst + 16), stride);
 854         dst += (4 * stride);
 855     }
 856 }
 857
 858 static void hevc_idct_luma_4x4_msa(int16_t *coeffs)
 859 {
 860     v8i16 in0, in1, dst0, dst1;
 861     v4i32 in_r0, in_l0, in_r1, in_l1, res0, res1, res2, res3;
 862
 863     LD_SH2(coeffs, 8, in0, in1);
 864     UNPCK_SH_SW(in0, in_r0, in_l0);
 865     UNPCK_SH_SW(in1, in_r1, in_l1);
 866     HEVC_IDCT_LUMA4x4_COL(in_r0, in_l0, in_r1, in_l1, res0, res1, res2, res3,
 867                           7);
 868     TRANSPOSE4x4_SW_SW(res0, res1, res2, res3, in_r0, in_l0, in_r1, in_l1);
 869     HEVC_IDCT_LUMA4x4_COL(in_r0, in_l0, in_r1, in_l1, res0, res1, res2, res3,
 870                           12);
 871     TRANSPOSE4x4_SW_SW(res0, res1, res2, res3, res0, res1, res2, res3);
 872     PCKEV_H2_SH(res1, res0, res3, res2, dst0, dst1);
 873     ST_SH2(dst0, dst1, coeffs, 8);
 874 }
 875
 876 void ff_hevc_idct_4x4_msa(int16_t *coeffs, int col_limit)
 877 {
 878     hevc_idct_4x4_msa(coeffs);
 879 }
 880
 881 void ff_hevc_idct_8x8_msa(int16_t *coeffs, int col_limit)
 882 {
 883     hevc_idct_8x8_msa(coeffs);
 884 }
 885
 886 void ff_hevc_idct_16x16_msa(int16_t *coeffs, int col_limit)
 887 {
 888     hevc_idct_16x16_msa(coeffs);
 889 }
 890
 891 void ff_hevc_idct_32x32_msa(int16_t *coeffs, int col_limit)
 892 {
 893     hevc_idct_32x32_msa(coeffs);
 894 }
 895
 896 void ff_hevc_addblk_4x4_msa(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
 897 {
 898     hevc_addblk_4x4_msa(coeffs, dst, stride);
 899 }
 900
 901 void ff_hevc_addblk_8x8_msa(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
 902 {
 903     hevc_addblk_8x8_msa(coeffs, dst, stride);
 904 }
 905
 906 void ff_hevc_addblk_16x16_msa(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
 907 {
 908     hevc_addblk_16x16_msa(coeffs, dst, stride);
 909 }
 910
 911 void ff_hevc_addblk_32x32_msa(uint8_t *dst, int16_t *coeffs, ptrdiff_t stride)
 912 {
 913     hevc_addblk_32x32_msa(coeffs, dst, stride);
 914 }
 915
 916 void ff_hevc_idct_dc_4x4_msa(int16_t *coeffs)
 917 {
 918     hevc_idct_dc_4x4_msa(coeffs);
 919 }
 920
 921 void ff_hevc_idct_dc_8x8_msa(int16_t *coeffs)
 922 {
 923     hevc_idct_dc_8x8_msa(coeffs);
 924 }
 925
 926 void ff_hevc_idct_dc_16x16_msa(int16_t *coeffs)
 927 {
 928     hevc_idct_dc_16x16_msa(coeffs);
 929 }
 930
 931 void ff_hevc_idct_dc_32x32_msa(int16_t *coeffs)
 932 {
 933     hevc_idct_dc_32x32_msa(coeffs);
 934 }
 935
 936 void ff_hevc_idct_luma_4x4_msa(int16_t *coeffs)
 937 {
 938     hevc_idct_luma_4x4_msa(coeffs);
 939 }