git.sesse.net Git - x264/blob - common/ppc/mc.c

   1 /*****************************************************************************
   2  * mc.c: ppc motion compensation
   3  *****************************************************************************
   4  * Copyright (C) 2003-2015 x264 project
   5  *
   6  * Authors: Eric Petit <eric.petit@lapsus.org>
   7  *          Guillaume Poirier <gpoirier@mplayerhq.hu>
   8  *
   9  * This program is free software; you can redistribute it and/or modify
  10  * it under the terms of the GNU General Public License as published by
  11  * the Free Software Foundation; either version 2 of the License, or
  12  * (at your option) any later version.
  13  *
  14  * This program is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  17  * GNU General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU General Public License
  20  * along with this program; if not, write to the Free Software
  21  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02111, USA.
  22  *
  23  * This program is also available under a commercial proprietary license.
  24  * For more information, contact us at licensing@x264.com.
  25  *****************************************************************************/
  26
  27 #include <stdlib.h>
  28 #include <stdio.h>
  29 #include <string.h>
  30 #include <stdint.h>
  31 #include <stdarg.h>
  32
  33 #include "x264.h"
  34 #include "common/common.h"
  35 #include "common/mc.h"
  36 #include "mc.h"
  37 #include "ppccommon.h"
  38
  39 #if !HIGH_BIT_DEPTH
  40 typedef void (*pf_mc_t)( uint8_t *src, intptr_t i_src,
  41                          uint8_t *dst, intptr_t i_dst, int i_height );
  42
  43 static inline int x264_tapfilter( uint8_t *pix, int i_pix_next )
  44 {
  45     return pix[-2*i_pix_next] - 5*pix[-1*i_pix_next] + 20*(pix[0] +
  46            pix[1*i_pix_next]) - 5*pix[ 2*i_pix_next] +
  47            pix[ 3*i_pix_next];
  48 }
  49
  50 static inline int x264_tapfilter1( uint8_t *pix )
  51 {
  52     return pix[-2] - 5*pix[-1] + 20*(pix[0] + pix[1]) - 5*pix[ 2] +
  53            pix[ 3];
  54 }
  55
  56 static inline void x264_pixel_avg2_w4_altivec( uint8_t *dst,  intptr_t i_dst,
  57                                                uint8_t *src1, intptr_t i_src1,
  58                                                uint8_t *src2, int i_height )
  59 {
  60     for( int y = 0; y < i_height; y++ )
  61     {
  62         for( int x = 0; x < 4; x++ )
  63             dst[x] = ( src1[x] + src2[x] + 1 ) >> 1;
  64         dst  += i_dst;
  65         src1 += i_src1;
  66         src2 += i_src1;
  67     }
  68 }
  69
  70 static inline void x264_pixel_avg2_w8_altivec( uint8_t *dst,  intptr_t i_dst,
  71                                                uint8_t *src1, intptr_t i_src1,
  72                                                uint8_t *src2, int i_height )
  73 {
  74     vec_u8_t src1v, src2v;
  75     PREP_LOAD;
  76     PREP_STORE8;
  77     PREP_LOAD_SRC( src1 );
  78     PREP_LOAD_SRC( src2 );
  79
  80     for( int y = 0; y < i_height; y++ )
  81     {
  82         VEC_LOAD( src1, src1v, 8, vec_u8_t, src1 );
  83         VEC_LOAD( src2, src2v, 8, vec_u8_t, src2 );
  84         src1v = vec_avg( src1v, src2v );
  85         VEC_STORE8( src1v, dst );
  86
  87         dst  += i_dst;
  88         src1 += i_src1;
  89         src2 += i_src1;
  90     }
  91 }
  92
  93 static inline void x264_pixel_avg2_w16_altivec( uint8_t *dst,  intptr_t i_dst,
  94                                                 uint8_t *src1, intptr_t i_src1,
  95                                                 uint8_t *src2, int i_height )
  96 {
  97     vec_u8_t src1v, src2v;
  98     PREP_LOAD;
  99     PREP_LOAD_SRC( src1 );
 100     PREP_LOAD_SRC( src2 );
 101
 102     for( int y = 0; y < i_height; y++ )
 103     {
 104         VEC_LOAD( src1, src1v, 16, vec_u8_t, src1 );
 105         VEC_LOAD( src2, src2v, 16, vec_u8_t, src2 );
 106         src1v = vec_avg( src1v, src2v );
 107         vec_st(src1v, 0, dst);
 108
 109         dst  += i_dst;
 110         src1 += i_src1;
 111         src2 += i_src1;
 112     }
 113 }
 114
 115 static inline void x264_pixel_avg2_w20_altivec( uint8_t *dst,  intptr_t i_dst,
 116                                                 uint8_t *src1, intptr_t i_src1,
 117                                                 uint8_t *src2, int i_height )
 118 {
 119     x264_pixel_avg2_w16_altivec(dst, i_dst, src1, i_src1, src2, i_height);
 120     x264_pixel_avg2_w4_altivec(dst+16, i_dst, src1+16, i_src1, src2+16, i_height);
 121 }
 122
 123 /* mc_copy: plain c */
 124
 125 #define MC_COPY( name, a )                                \
 126 static void name( uint8_t *dst, intptr_t i_dst,           \
 127                   uint8_t *src, intptr_t i_src, int i_height ) \
 128 {                                                         \
 129     int y;                                                \
 130     for( y = 0; y < i_height; y++ )                       \
 131     {                                                     \
 132         memcpy( dst, src, a );                            \
 133         src += i_src;                                     \
 134         dst += i_dst;                                     \
 135     }                                                     \
 136 }
 137 MC_COPY( x264_mc_copy_w4_altivec,  4  )
 138 MC_COPY( x264_mc_copy_w8_altivec,  8  )
 139
 140 static void x264_mc_copy_w16_altivec( uint8_t *dst, intptr_t i_dst,
 141                                       uint8_t *src, intptr_t i_src, int i_height )
 142 {
 143     vec_u8_t cpyV;
 144     PREP_LOAD;
 145     PREP_LOAD_SRC( src );
 146
 147     for( int y = 0; y < i_height; y++ )
 148     {
 149         VEC_LOAD( src, cpyV, 16, vec_u8_t, src );
 150         vec_st(cpyV, 0, dst);
 151
 152         src += i_src;
 153         dst += i_dst;
 154     }
 155 }
 156
 157
 158 static void x264_mc_copy_w16_aligned_altivec( uint8_t *dst, intptr_t i_dst,
 159                                               uint8_t *src, intptr_t i_src, int i_height )
 160 {
 161     for( int y = 0; y < i_height; ++y )
 162     {
 163         vec_u8_t cpyV = vec_ld( 0, src );
 164         vec_st(cpyV, 0, dst);
 165
 166         src += i_src;
 167         dst += i_dst;
 168     }
 169 }
 170
 171
 172 static void mc_luma_altivec( uint8_t *dst,    intptr_t i_dst_stride,
 173                              uint8_t *src[4], intptr_t i_src_stride,
 174                              int mvx, int mvy,
 175                              int i_width, int i_height, const x264_weight_t *weight )
 176 {
 177     int qpel_idx = ((mvy&3)<<2) + (mvx&3);
 178     intptr_t offset = (mvy>>2)*i_src_stride + (mvx>>2);
 179     uint8_t *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
 180     if( qpel_idx & 5 ) /* qpel interpolation needed */
 181     {
 182         uint8_t *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
 183
 184         switch( i_width )
 185         {
 186             case 4:
 187                 x264_pixel_avg2_w4_altivec( dst, i_dst_stride, src1, i_src_stride, src2, i_height );
 188                 break;
 189             case 8:
 190                 x264_pixel_avg2_w8_altivec( dst, i_dst_stride, src1, i_src_stride, src2, i_height );
 191                 break;
 192             case 16:
 193             default:
 194                 x264_pixel_avg2_w16_altivec( dst, i_dst_stride, src1, i_src_stride, src2, i_height );
 195         }
 196         if( weight->weightfn )
 197             weight->weightfn[i_width>>2]( dst, i_dst_stride, dst, i_dst_stride, weight, i_height );
 198     }
 199     else if( weight->weightfn )
 200         weight->weightfn[i_width>>2]( dst, i_dst_stride, src1, i_src_stride, weight, i_height );
 201     else
 202     {
 203         switch( i_width )
 204         {
 205             case 4:
 206                 x264_mc_copy_w4_altivec( dst, i_dst_stride, src1, i_src_stride, i_height );
 207                 break;
 208             case 8:
 209                 x264_mc_copy_w8_altivec( dst, i_dst_stride, src1, i_src_stride, i_height );
 210                 break;
 211             case 16:
 212                 x264_mc_copy_w16_altivec( dst, i_dst_stride, src1, i_src_stride, i_height );
 213                 break;
 214         }
 215     }
 216 }
 217
 218
 219
 220 static uint8_t *get_ref_altivec( uint8_t *dst,   intptr_t *i_dst_stride,
 221                                  uint8_t *src[4], intptr_t i_src_stride,
 222                                  int mvx, int mvy,
 223                                  int i_width, int i_height, const x264_weight_t *weight )
 224 {
 225     int qpel_idx = ((mvy&3)<<2) + (mvx&3);
 226     intptr_t offset = (mvy>>2)*i_src_stride + (mvx>>2);
 227     uint8_t *src1 = src[x264_hpel_ref0[qpel_idx]] + offset + ((mvy&3) == 3) * i_src_stride;
 228     if( qpel_idx & 5 ) /* qpel interpolation needed */
 229     {
 230         uint8_t *src2 = src[x264_hpel_ref1[qpel_idx]] + offset + ((mvx&3) == 3);
 231         switch( i_width )
 232         {
 233             case 4:
 234                 x264_pixel_avg2_w4_altivec( dst, *i_dst_stride, src1, i_src_stride, src2, i_height );
 235                 break;
 236             case 8:
 237                 x264_pixel_avg2_w8_altivec( dst, *i_dst_stride, src1, i_src_stride, src2, i_height );
 238                 break;
 239             case 12:
 240             case 16:
 241             default:
 242                 x264_pixel_avg2_w16_altivec( dst, *i_dst_stride, src1, i_src_stride, src2, i_height );
 243                 break;
 244             case 20:
 245                 x264_pixel_avg2_w20_altivec( dst, *i_dst_stride, src1, i_src_stride, src2, i_height );
 246                 break;
 247         }
 248         if( weight->weightfn )
 249             weight->weightfn[i_width>>2]( dst, *i_dst_stride, dst, *i_dst_stride, weight, i_height );
 250         return dst;
 251     }
 252     else if( weight->weightfn )
 253     {
 254         weight->weightfn[i_width>>2]( dst, *i_dst_stride, src1, i_src_stride, weight, i_height );
 255         return dst;
 256     }
 257     else
 258     {
 259         *i_dst_stride = i_src_stride;
 260         return src1;
 261     }
 262 }
 263
 264 static void mc_chroma_2xh( uint8_t *dstu, uint8_t *dstv, intptr_t i_dst_stride,
 265                            uint8_t *src, intptr_t i_src_stride,
 266                            int mvx, int mvy, int i_height )
 267 {
 268     uint8_t *srcp;
 269     int d8x = mvx&0x07;
 270     int d8y = mvy&0x07;
 271
 272     int cA = (8-d8x)*(8-d8y);
 273     int cB = d8x    *(8-d8y);
 274     int cC = (8-d8x)*d8y;
 275     int cD = d8x    *d8y;
 276
 277     src += (mvy >> 3) * i_src_stride + (mvx >> 3)*2;
 278     srcp = &src[i_src_stride];
 279
 280     for( int y = 0; y < i_height; y++ )
 281     {
 282         dstu[0] = ( cA*src[0] + cB*src[2] + cC*srcp[0] + cD*srcp[2] + 32 ) >> 6;
 283         dstv[0] = ( cA*src[1] + cB*src[3] + cC*srcp[1] + cD*srcp[3] + 32 ) >> 6;
 284         dstu[1] = ( cA*src[2] + cB*src[4] + cC*srcp[2] + cD*srcp[4] + 32 ) >> 6;
 285         dstv[1] = ( cA*src[3] + cB*src[5] + cC*srcp[3] + cD*srcp[5] + 32 ) >> 6;
 286
 287         src  += i_src_stride;
 288         srcp += i_src_stride;
 289         dstu += i_dst_stride;
 290         dstv += i_dst_stride;
 291     }
 292  }
 293
 294 static void mc_chroma_altivec_4xh( uint8_t *dstu, uint8_t *dstv, intptr_t i_dst_stride,
 295                                    uint8_t *src, intptr_t i_src_stride,
 296                                    int mvx, int mvy, int i_height )
 297 {
 298     uint8_t *srcp;
 299     int d8x = mvx & 0x07;
 300     int d8y = mvy & 0x07;
 301
 302     ALIGNED_16( uint16_t coeff[4] );
 303     coeff[0] = (8-d8x)*(8-d8y);
 304     coeff[1] = d8x    *(8-d8y);
 305     coeff[2] = (8-d8x)*d8y;
 306     coeff[3] = d8x    *d8y;
 307
 308     src += (mvy >> 3) * i_src_stride + (mvx >> 3)*2;
 309     srcp = &src[i_src_stride];
 310
 311     LOAD_ZERO;
 312     PREP_LOAD;
 313     PREP_LOAD_SRC( src );
 314     vec_u16_t   coeff0v, coeff1v, coeff2v, coeff3v;
 315     vec_u8_t    src2v_8, dstuv, dstvv;
 316     vec_u16_t   src0v_16, src1v_16, src2v_16, src3v_16, dstv16;
 317     vec_u16_t   shiftv, k32v;
 318
 319     static const vec_u8_t perm0v = CV(1,5,9,13,1,5,9,13,1,5,9,13,1,5,9,13);
 320     static const vec_u8_t perm1v = CV(3,7,11,15,3,7,11,15,3,7,11,15,3,7,11,15);
 321
 322     coeff0v = vec_ld( 0, coeff );
 323     coeff3v = vec_splat( coeff0v, 3 );
 324     coeff2v = vec_splat( coeff0v, 2 );
 325     coeff1v = vec_splat( coeff0v, 1 );
 326     coeff0v = vec_splat( coeff0v, 0 );
 327     k32v    = vec_sl( vec_splat_u16( 1 ), vec_splat_u16( 5 ) );
 328     shiftv  = vec_splat_u16( 6 );
 329
 330     VEC_LOAD( src, src2v_8, 9, vec_u8_t, src );
 331     src2v_16 = vec_u8_to_u16( src2v_8 );
 332     src3v_16 = vec_u8_to_u16( vec_sld( src2v_8, src2v_8, 2 ) );
 333
 334     for( int y = 0; y < i_height; y += 2 )
 335     {
 336         src0v_16 = src2v_16;
 337         src1v_16 = src3v_16;
 338         VEC_LOAD( srcp, src2v_8, 9, vec_u8_t, src );
 339         src2v_16 = vec_u8_to_u16( src2v_8 );
 340         src3v_16 = vec_u8_to_u16( vec_sld( src2v_8, src2v_8, 2 ) );
 341
 342         dstv16 = vec_mladd( coeff0v, src0v_16, k32v );
 343         dstv16 = vec_mladd( coeff1v, src1v_16, dstv16 );
 344         dstv16 = vec_mladd( coeff2v, src2v_16, dstv16 );
 345         dstv16 = vec_mladd( coeff3v, src3v_16, dstv16 );
 346
 347         dstv16 = vec_sr( dstv16, shiftv );
 348
 349         dstuv = (vec_u8_t)vec_perm( dstv16, dstv16, perm0v );
 350         dstvv = (vec_u8_t)vec_perm( dstv16, dstv16, perm1v );
 351         vec_ste( (vec_u32_t)dstuv, 0, (uint32_t*) dstu );
 352         vec_ste( (vec_u32_t)dstvv, 0, (uint32_t*) dstv );
 353
 354         srcp += i_src_stride;
 355         dstu += i_dst_stride;
 356         dstv += i_dst_stride;
 357
 358         src0v_16 = src2v_16;
 359         src1v_16 = src3v_16;
 360         VEC_LOAD( srcp, src2v_8, 9, vec_u8_t, src );
 361         src2v_16 = vec_u8_to_u16( src2v_8 );
 362         src3v_16 = vec_u8_to_u16( vec_sld( src2v_8, src2v_8, 2 ) );
 363
 364         dstv16 = vec_mladd( coeff0v, src0v_16, k32v );
 365         dstv16 = vec_mladd( coeff1v, src1v_16, dstv16 );
 366         dstv16 = vec_mladd( coeff2v, src2v_16, dstv16 );
 367         dstv16 = vec_mladd( coeff3v, src3v_16, dstv16 );
 368
 369         dstv16 = vec_sr( dstv16, shiftv );
 370
 371         dstuv = (vec_u8_t)vec_perm( dstv16, dstv16, perm0v );
 372         dstvv = (vec_u8_t)vec_perm( dstv16, dstv16, perm1v );
 373         vec_ste( (vec_u32_t)dstuv, 0, (uint32_t*) dstu );
 374         vec_ste( (vec_u32_t)dstvv, 0, (uint32_t*) dstv );
 375
 376         srcp += i_src_stride;
 377         dstu += i_dst_stride;
 378         dstv += i_dst_stride;
 379     }
 380 }
 381
 382 static void mc_chroma_altivec_8xh( uint8_t *dstu, uint8_t *dstv, intptr_t i_dst_stride,
 383                                    uint8_t *src, intptr_t i_src_stride,
 384                                    int mvx, int mvy, int i_height )
 385 {
 386     uint8_t *srcp;
 387     int d8x = mvx & 0x07;
 388     int d8y = mvy & 0x07;
 389
 390     ALIGNED_16( uint16_t coeff[4] );
 391     coeff[0] = (8-d8x)*(8-d8y);
 392     coeff[1] = d8x    *(8-d8y);
 393     coeff[2] = (8-d8x)*d8y;
 394     coeff[3] = d8x    *d8y;
 395
 396     src += (mvy >> 3) * i_src_stride + (mvx >> 3)*2;
 397     srcp = &src[i_src_stride];
 398
 399     LOAD_ZERO;
 400     PREP_LOAD;
 401     PREP_LOAD_SRC( src );
 402     PREP_STORE8;
 403     vec_u16_t   coeff0v, coeff1v, coeff2v, coeff3v;
 404     vec_u8_t    src0v_8, src1v_8, src2v_8, src3v_8;
 405     vec_u8_t    dstuv, dstvv;
 406     vec_u16_t   src0v_16h, src1v_16h, src2v_16h, src3v_16h, dstv_16h;
 407     vec_u16_t   src0v_16l, src1v_16l, src2v_16l, src3v_16l, dstv_16l;
 408     vec_u16_t   shiftv, k32v;
 409
 410     coeff0v = vec_ld( 0, coeff );
 411     coeff3v = vec_splat( coeff0v, 3 );
 412     coeff2v = vec_splat( coeff0v, 2 );
 413     coeff1v = vec_splat( coeff0v, 1 );
 414     coeff0v = vec_splat( coeff0v, 0 );
 415     k32v    = vec_sl( vec_splat_u16( 1 ), vec_splat_u16( 5 ) );
 416     shiftv  = vec_splat_u16( 6 );
 417
 418     static const vec_u8_t perm0v = CV(1,5,9,13,17,21,25,29,0,0,0,0,0,0,0,0);
 419     static const vec_u8_t perm1v = CV(3,7,11,15,19,23,27,31,0,0,0,0,0,0,0,0);
 420
 421     VEC_LOAD( src, src2v_8, 16, vec_u8_t, src );
 422     VEC_LOAD( src+16, src3v_8, 2, vec_u8_t, src );
 423     src3v_8 = vec_sld( src2v_8, src3v_8, 2 );
 424
 425     for( int y = 0; y < i_height; y += 2 )
 426     {
 427         src0v_8 = src2v_8;
 428         src1v_8 = src3v_8;
 429         VEC_LOAD( srcp, src2v_8, 16, vec_u8_t, src );
 430         VEC_LOAD( srcp+16, src3v_8, 2, vec_u8_t, src );
 431
 432         src3v_8 = vec_sld( src2v_8, src3v_8, 2 );
 433
 434         src0v_16h = vec_u8_to_u16_h( src0v_8 );
 435         src0v_16l = vec_u8_to_u16_l( src0v_8 );
 436         src1v_16h = vec_u8_to_u16_h( src1v_8 );
 437         src1v_16l = vec_u8_to_u16_l( src1v_8 );
 438         src2v_16h = vec_u8_to_u16_h( src2v_8 );
 439         src2v_16l = vec_u8_to_u16_l( src2v_8 );
 440         src3v_16h = vec_u8_to_u16_h( src3v_8 );
 441         src3v_16l = vec_u8_to_u16_l( src3v_8 );
 442
 443         dstv_16h = vec_mladd( coeff0v, src0v_16h, k32v );
 444         dstv_16l = vec_mladd( coeff0v, src0v_16l, k32v );
 445         dstv_16h = vec_mladd( coeff1v, src1v_16h, dstv_16h );
 446         dstv_16l = vec_mladd( coeff1v, src1v_16l, dstv_16l );
 447         dstv_16h = vec_mladd( coeff2v, src2v_16h, dstv_16h );
 448         dstv_16l = vec_mladd( coeff2v, src2v_16l, dstv_16l );
 449         dstv_16h = vec_mladd( coeff3v, src3v_16h, dstv_16h );
 450         dstv_16l = vec_mladd( coeff3v, src3v_16l, dstv_16l );
 451
 452         dstv_16h = vec_sr( dstv_16h, shiftv );
 453         dstv_16l = vec_sr( dstv_16l, shiftv );
 454
 455         dstuv = (vec_u8_t)vec_perm( dstv_16h, dstv_16l, perm0v );
 456         dstvv = (vec_u8_t)vec_perm( dstv_16h, dstv_16l, perm1v );
 457
 458         VEC_STORE8( dstuv, dstu );
 459         VEC_STORE8( dstvv, dstv );
 460
 461         srcp += i_src_stride;
 462         dstu += i_dst_stride;
 463         dstv += i_dst_stride;
 464
 465         src0v_8 = src2v_8;
 466         src1v_8 = src3v_8;
 467         VEC_LOAD( srcp, src2v_8, 16, vec_u8_t, src );
 468         VEC_LOAD( srcp+16, src3v_8, 2, vec_u8_t, src );
 469
 470         src3v_8 = vec_sld( src2v_8, src3v_8, 2 );
 471
 472         src0v_16h = vec_u8_to_u16_h( src0v_8 );
 473         src0v_16l = vec_u8_to_u16_l( src0v_8 );
 474         src1v_16h = vec_u8_to_u16_h( src1v_8 );
 475         src1v_16l = vec_u8_to_u16_l( src1v_8 );
 476         src2v_16h = vec_u8_to_u16_h( src2v_8 );
 477         src2v_16l = vec_u8_to_u16_l( src2v_8 );
 478         src3v_16h = vec_u8_to_u16_h( src3v_8 );
 479         src3v_16l = vec_u8_to_u16_l( src3v_8 );
 480
 481         dstv_16h = vec_mladd( coeff0v, src0v_16h, k32v );
 482         dstv_16l = vec_mladd( coeff0v, src0v_16l, k32v );
 483         dstv_16h = vec_mladd( coeff1v, src1v_16h, dstv_16h );
 484         dstv_16l = vec_mladd( coeff1v, src1v_16l, dstv_16l );
 485         dstv_16h = vec_mladd( coeff2v, src2v_16h, dstv_16h );
 486         dstv_16l = vec_mladd( coeff2v, src2v_16l, dstv_16l );
 487         dstv_16h = vec_mladd( coeff3v, src3v_16h, dstv_16h );
 488         dstv_16l = vec_mladd( coeff3v, src3v_16l, dstv_16l );
 489
 490         dstv_16h = vec_sr( dstv_16h, shiftv );
 491         dstv_16l = vec_sr( dstv_16l, shiftv );
 492
 493         dstuv = (vec_u8_t)vec_perm( dstv_16h, dstv_16l, perm0v );
 494         dstvv = (vec_u8_t)vec_perm( dstv_16h, dstv_16l, perm1v );
 495
 496         VEC_STORE8( dstuv, dstu );
 497         VEC_STORE8( dstvv, dstv );
 498
 499         srcp += i_src_stride;
 500         dstu += i_dst_stride;
 501         dstv += i_dst_stride;
 502     }
 503 }
 504
 505 static void mc_chroma_altivec( uint8_t *dstu, uint8_t *dstv, intptr_t i_dst_stride,
 506                                uint8_t *src, intptr_t i_src_stride,
 507                                int mvx, int mvy, int i_width, int i_height )
 508 {
 509     if( i_width == 8 )
 510         mc_chroma_altivec_8xh( dstu, dstv, i_dst_stride, src, i_src_stride,
 511                                mvx, mvy, i_height );
 512     else if( i_width == 4 )
 513         mc_chroma_altivec_4xh( dstu, dstv, i_dst_stride, src, i_src_stride,
 514                                mvx, mvy, i_height );
 515     else
 516         mc_chroma_2xh( dstu, dstv, i_dst_stride, src, i_src_stride,
 517                        mvx, mvy, i_height );
 518 }
 519
 520 #define HPEL_FILTER_1( t1v, t2v, t3v, t4v, t5v, t6v ) \
 521 {                                                     \
 522     t1v = vec_add( t1v, t6v );                        \
 523     t2v = vec_add( t2v, t5v );                        \
 524     t3v = vec_add( t3v, t4v );                        \
 525                                                       \
 526     t1v = vec_sub( t1v, t2v );   /* (a-b) */          \
 527     t2v = vec_sub( t2v, t3v );   /* (b-c) */          \
 528     t2v = vec_sl(  t2v, twov );  /* (b-c)*4 */        \
 529     t1v = vec_sub( t1v, t2v );   /* a-5*b+4*c */      \
 530     t3v = vec_sl(  t3v, fourv ); /* 16*c */           \
 531     t1v = vec_add( t1v, t3v );   /* a-5*b+20*c */     \
 532 }
 533
 534 #define HPEL_FILTER_2( t1v, t2v, t3v, t4v, t5v, t6v ) \
 535 {                                                     \
 536     t1v = vec_add( t1v, t6v );                        \
 537     t2v = vec_add( t2v, t5v );                        \
 538     t3v = vec_add( t3v, t4v );                        \
 539                                                       \
 540     t1v = vec_sub( t1v, t2v );  /* (a-b) */           \
 541     t1v = vec_sra( t1v, twov ); /* (a-b)/4 */         \
 542     t1v = vec_sub( t1v, t2v );  /* (a-b)/4-b */       \
 543     t1v = vec_add( t1v, t3v );  /* (a-b)/4-b+c */     \
 544     t1v = vec_sra( t1v, twov ); /* ((a-b)/4-b+c)/4 */ \
 545     t1v = vec_add( t1v, t3v );  /* ((a-b)/4-b+c)/4+c = (a-5*b+20*c)/16 */ \
 546 }
 547
 548 #define HPEL_FILTER_HORIZONTAL()                             \
 549 {                                                            \
 550     VEC_LOAD_G( &src[x- 2+i_stride*y], src1v, 16, vec_u8_t); \
 551     VEC_LOAD_G( &src[x+14+i_stride*y], src6v, 16, vec_u8_t); \
 552                                                              \
 553     src2v = vec_sld( src1v, src6v,  1 );                     \
 554     src3v = vec_sld( src1v, src6v,  2 );                     \
 555     src4v = vec_sld( src1v, src6v,  3 );                     \
 556     src5v = vec_sld( src1v, src6v,  4 );                     \
 557     src6v = vec_sld( src1v, src6v,  5 );                     \
 558                                                              \
 559     temp1v = vec_u8_to_s16_h( src1v );                       \
 560     temp2v = vec_u8_to_s16_h( src2v );                       \
 561     temp3v = vec_u8_to_s16_h( src3v );                       \
 562     temp4v = vec_u8_to_s16_h( src4v );                       \
 563     temp5v = vec_u8_to_s16_h( src5v );                       \
 564     temp6v = vec_u8_to_s16_h( src6v );                       \
 565                                                              \
 566     HPEL_FILTER_1( temp1v, temp2v, temp3v,                   \
 567                    temp4v, temp5v, temp6v );                 \
 568                                                              \
 569     dest1v = vec_add( temp1v, sixteenv );                    \
 570     dest1v = vec_sra( dest1v, fivev );                       \
 571                                                              \
 572     temp1v = vec_u8_to_s16_l( src1v );                       \
 573     temp2v = vec_u8_to_s16_l( src2v );                       \
 574     temp3v = vec_u8_to_s16_l( src3v );                       \
 575     temp4v = vec_u8_to_s16_l( src4v );                       \
 576     temp5v = vec_u8_to_s16_l( src5v );                       \
 577     temp6v = vec_u8_to_s16_l( src6v );                       \
 578                                                              \
 579     HPEL_FILTER_1( temp1v, temp2v, temp3v,                   \
 580                    temp4v, temp5v, temp6v );                 \
 581                                                              \
 582     dest2v = vec_add( temp1v, sixteenv );                    \
 583     dest2v = vec_sra( dest2v, fivev );                       \
 584                                                              \
 585     destv = vec_packsu( dest1v, dest2v );                    \
 586                                                              \
 587     VEC_STORE16( destv, &dsth[x+i_stride*y], dsth );         \
 588 }
 589
 590 #define HPEL_FILTER_VERTICAL()                                    \
 591 {                                                                 \
 592     VEC_LOAD( &src[x+i_stride*(y-2)], src1v, 16, vec_u8_t, src ); \
 593     VEC_LOAD( &src[x+i_stride*(y-1)], src2v, 16, vec_u8_t, src ); \
 594     VEC_LOAD( &src[x+i_stride*(y-0)], src3v, 16, vec_u8_t, src ); \
 595     VEC_LOAD( &src[x+i_stride*(y+1)], src4v, 16, vec_u8_t, src ); \
 596     VEC_LOAD( &src[x+i_stride*(y+2)], src5v, 16, vec_u8_t, src ); \
 597     VEC_LOAD( &src[x+i_stride*(y+3)], src6v, 16, vec_u8_t, src ); \
 598                                                                   \
 599     temp1v = vec_u8_to_s16_h( src1v );                            \
 600     temp2v = vec_u8_to_s16_h( src2v );                            \
 601     temp3v = vec_u8_to_s16_h( src3v );                            \
 602     temp4v = vec_u8_to_s16_h( src4v );                            \
 603     temp5v = vec_u8_to_s16_h( src5v );                            \
 604     temp6v = vec_u8_to_s16_h( src6v );                            \
 605                                                                   \
 606     HPEL_FILTER_1( temp1v, temp2v, temp3v,                        \
 607                    temp4v, temp5v, temp6v );                      \
 608                                                                   \
 609     dest1v = vec_add( temp1v, sixteenv );                         \
 610     dest1v = vec_sra( dest1v, fivev );                            \
 611                                                                   \
 612     temp4v = vec_u8_to_s16_l( src1v );                            \
 613     temp5v = vec_u8_to_s16_l( src2v );                            \
 614     temp6v = vec_u8_to_s16_l( src3v );                            \
 615     temp7v = vec_u8_to_s16_l( src4v );                            \
 616     temp8v = vec_u8_to_s16_l( src5v );                            \
 617     temp9v = vec_u8_to_s16_l( src6v );                            \
 618                                                                   \
 619     HPEL_FILTER_1( temp4v, temp5v, temp6v,                        \
 620                    temp7v, temp8v, temp9v );                      \
 621                                                                   \
 622     dest2v = vec_add( temp4v, sixteenv );                         \
 623     dest2v = vec_sra( dest2v, fivev );                            \
 624                                                                   \
 625     destv = vec_packsu( dest1v, dest2v );                         \
 626                                                                   \
 627     VEC_STORE16( destv, &dstv[x+i_stride*y], dsth );              \
 628 }
 629
 630 #define HPEL_FILTER_CENTRAL()                           \
 631 {                                                       \
 632     temp1v = vec_sld( tempav, tempbv, 12 );             \
 633     temp2v = vec_sld( tempav, tempbv, 14 );             \
 634     temp3v = tempbv;                                    \
 635     temp4v = vec_sld( tempbv, tempcv,  2 );             \
 636     temp5v = vec_sld( tempbv, tempcv,  4 );             \
 637     temp6v = vec_sld( tempbv, tempcv,  6 );             \
 638                                                         \
 639     HPEL_FILTER_2( temp1v, temp2v, temp3v,              \
 640                    temp4v, temp5v, temp6v );            \
 641                                                         \
 642     dest1v = vec_add( temp1v, thirtytwov );             \
 643     dest1v = vec_sra( dest1v, sixv );                   \
 644                                                         \
 645     temp1v = vec_sld( tempbv, tempcv, 12 );             \
 646     temp2v = vec_sld( tempbv, tempcv, 14 );             \
 647     temp3v = tempcv;                                    \
 648     temp4v = vec_sld( tempcv, tempdv,  2 );             \
 649     temp5v = vec_sld( tempcv, tempdv,  4 );             \
 650     temp6v = vec_sld( tempcv, tempdv,  6 );             \
 651                                                         \
 652     HPEL_FILTER_2( temp1v, temp2v, temp3v,              \
 653                    temp4v, temp5v, temp6v );            \
 654                                                         \
 655     dest2v = vec_add( temp1v, thirtytwov );             \
 656     dest2v = vec_sra( dest2v, sixv );                   \
 657                                                         \
 658     destv = vec_packsu( dest1v, dest2v );               \
 659                                                         \
 660     VEC_STORE16( destv, &dstc[x-16+i_stride*y], dsth ); \
 661 }
 662
 663 void x264_hpel_filter_altivec( uint8_t *dsth, uint8_t *dstv, uint8_t *dstc, uint8_t *src,
 664                                intptr_t i_stride, int i_width, int i_height, int16_t *buf )
 665 {
 666     vec_u8_t destv;
 667     vec_u8_t src1v, src2v, src3v, src4v, src5v, src6v;
 668     vec_s16_t dest1v, dest2v;
 669     vec_s16_t temp1v, temp2v, temp3v, temp4v, temp5v, temp6v, temp7v, temp8v, temp9v;
 670     vec_s16_t tempav, tempbv, tempcv, tempdv, tempev;
 671
 672     PREP_LOAD;
 673     PREP_LOAD_SRC( src);
 674     PREP_STORE16;
 675     PREP_STORE16_DST( dsth );
 676     LOAD_ZERO;
 677
 678     vec_u16_t twov, fourv, fivev, sixv;
 679     vec_s16_t sixteenv, thirtytwov;
 680     vec_u16_u temp_u;
 681
 682     temp_u.s[0]=2;
 683     twov = vec_splat( temp_u.v, 0 );
 684     temp_u.s[0]=4;
 685     fourv = vec_splat( temp_u.v, 0 );
 686     temp_u.s[0]=5;
 687     fivev = vec_splat( temp_u.v, 0 );
 688     temp_u.s[0]=6;
 689     sixv = vec_splat( temp_u.v, 0 );
 690     temp_u.s[0]=16;
 691     sixteenv = (vec_s16_t)vec_splat( temp_u.v, 0 );
 692     temp_u.s[0]=32;
 693     thirtytwov = (vec_s16_t)vec_splat( temp_u.v, 0 );
 694
 695     for( int y = 0; y < i_height; y++ )
 696     {
 697         int x = 0;
 698
 699         /* horizontal_filter */
 700         HPEL_FILTER_HORIZONTAL();
 701
 702         /* vertical_filter */
 703         HPEL_FILTER_VERTICAL();
 704
 705         /* central_filter */
 706         tempav = tempcv;
 707         tempbv = tempdv;
 708         tempcv = vec_splat( temp1v, 0 ); /* first only */
 709         tempdv = temp1v;
 710         tempev = temp4v;
 711
 712         for( x = 16; x < i_width; x+=16 )
 713         {
 714             /* horizontal_filter */
 715             HPEL_FILTER_HORIZONTAL();
 716
 717             /* vertical_filter */
 718             HPEL_FILTER_VERTICAL();
 719
 720             /* central_filter */
 721             tempav = tempcv;
 722             tempbv = tempdv;
 723             tempcv = tempev;
 724             tempdv = temp1v;
 725             tempev = temp4v;
 726
 727             HPEL_FILTER_CENTRAL();
 728         }
 729
 730         /* Partial vertical filter */
 731         VEC_LOAD_PARTIAL( &src[x+i_stride*(y-2)], src1v, 16, vec_u8_t, src );
 732         VEC_LOAD_PARTIAL( &src[x+i_stride*(y-1)], src2v, 16, vec_u8_t, src );
 733         VEC_LOAD_PARTIAL( &src[x+i_stride*(y-0)], src3v, 16, vec_u8_t, src );
 734         VEC_LOAD_PARTIAL( &src[x+i_stride*(y+1)], src4v, 16, vec_u8_t, src );
 735         VEC_LOAD_PARTIAL( &src[x+i_stride*(y+2)], src5v, 16, vec_u8_t, src );
 736         VEC_LOAD_PARTIAL( &src[x+i_stride*(y+3)], src6v, 16, vec_u8_t, src );
 737
 738         temp1v = vec_u8_to_s16_h( src1v );
 739         temp2v = vec_u8_to_s16_h( src2v );
 740         temp3v = vec_u8_to_s16_h( src3v );
 741         temp4v = vec_u8_to_s16_h( src4v );
 742         temp5v = vec_u8_to_s16_h( src5v );
 743         temp6v = vec_u8_to_s16_h( src6v );
 744
 745         HPEL_FILTER_1( temp1v, temp2v, temp3v, temp4v, temp5v, temp6v );
 746
 747         /* central_filter */
 748         tempav = tempcv;
 749         tempbv = tempdv;
 750         tempcv = tempev;
 751         tempdv = temp1v;
 752         /* tempev is not used */
 753
 754         HPEL_FILTER_CENTRAL();
 755     }
 756 }
 757
 758 static void frame_init_lowres_core_altivec( uint8_t *src0, uint8_t *dst0, uint8_t *dsth, uint8_t *dstv, uint8_t *dstc,
 759                                             intptr_t src_stride, intptr_t dst_stride, int width, int height )
 760 {
 761     int w = width >> 4;
 762     int end = (width & 15);
 763     vec_u8_t src0v, src1v, src2v;
 764     vec_u8_t lv, hv, src1p1v;
 765     vec_u8_t avg0v, avg1v, avghv, avghp1v, avgleftv, avgrightv;
 766     static const vec_u8_t inverse_bridge_shuffle = CV(0x00, 0x02, 0x04, 0x06, 0x08, 0x0A, 0x0C, 0x0E, 0x10, 0x12, 0x14, 0x16, 0x18, 0x1A, 0x1C, 0x1E );
 767
 768     for( int y = 0; y < height; y++ )
 769     {
 770         int x;
 771         uint8_t *src1 = src0+src_stride;
 772         uint8_t *src2 = src1+src_stride;
 773
 774         src0v = vec_ld(0, src0);
 775         src1v = vec_ld(0, src1);
 776         src2v = vec_ld(0, src2);
 777
 778         avg0v = vec_avg(src0v, src1v);
 779         avg1v = vec_avg(src1v, src2v);
 780
 781         for( x = 0; x < w; x++ )
 782         {
 783             lv = vec_ld(16*(x*2+1), src0);
 784             src1v = vec_ld(16*(x*2+1), src1);
 785             avghv = vec_avg(lv, src1v);
 786
 787             lv = vec_ld(16*(x*2+2), src0);
 788             src1p1v = vec_ld(16*(x*2+2), src1);
 789             avghp1v = vec_avg(lv, src1p1v);
 790
 791             avgleftv = vec_avg(vec_sld(avg0v, avghv, 1), avg0v);
 792             avgrightv = vec_avg(vec_sld(avghv, avghp1v, 1), avghv);
 793
 794             vec_st(vec_perm(avgleftv, avgrightv, inverse_bridge_shuffle), 16*x, dst0);
 795             vec_st((vec_u8_t)vec_pack((vec_u16_t)avgleftv,(vec_u16_t)avgrightv), 16*x, dsth);
 796
 797             avg0v = avghp1v;
 798
 799             hv = vec_ld(16*(x*2+1), src2);
 800             avghv = vec_avg(src1v, hv);
 801
 802             hv = vec_ld(16*(x*2+2), src2);
 803             avghp1v = vec_avg(src1p1v, hv);
 804
 805             avgleftv = vec_avg(vec_sld(avg1v, avghv, 1), avg1v);
 806             avgrightv = vec_avg(vec_sld(avghv, avghp1v, 1), avghv);
 807
 808             vec_st(vec_perm(avgleftv, avgrightv, inverse_bridge_shuffle), 16*x, dstv);
 809             vec_st((vec_u8_t)vec_pack((vec_u16_t)avgleftv,(vec_u16_t)avgrightv), 16*x, dstc);
 810
 811             avg1v = avghp1v;
 812
 813         }
 814         if( end )
 815         {
 816             lv = vec_ld(16*(x*2+1), src0);
 817             src1v = vec_ld(16*(x*2+1), src1);
 818             avghv = vec_avg(lv, src1v);
 819
 820             lv = vec_ld(16*(x*2+1), src2);
 821             avghp1v = vec_avg(src1v, lv);
 822
 823             avgleftv = vec_avg(vec_sld(avg0v, avghv, 1), avg0v);
 824             avgrightv = vec_avg(vec_sld(avg1v, avghp1v, 1), avg1v);
 825
 826             lv = vec_perm(avgleftv, avgrightv, inverse_bridge_shuffle);
 827             hv = (vec_u8_t)vec_pack((vec_u16_t)avgleftv,(vec_u16_t)avgrightv);
 828
 829             vec_ste((vec_u32_t)lv,16*x,(uint32_t*)dst0);
 830             vec_ste((vec_u32_t)lv,16*x+4,(uint32_t*)dst0);
 831             vec_ste((vec_u32_t)hv,16*x,(uint32_t*)dsth);
 832             vec_ste((vec_u32_t)hv,16*x+4,(uint32_t*)dsth);
 833
 834             lv = vec_sld(lv, lv, 8);
 835             hv = vec_sld(hv, hv, 8);
 836
 837             vec_ste((vec_u32_t)lv,16*x,(uint32_t*)dstv);
 838             vec_ste((vec_u32_t)lv,16*x+4,(uint32_t*)dstv);
 839             vec_ste((vec_u32_t)hv,16*x,(uint32_t*)dstc);
 840             vec_ste((vec_u32_t)hv,16*x+4,(uint32_t*)dstc);
 841         }
 842
 843         src0 += src_stride*2;
 844         dst0 += dst_stride;
 845         dsth += dst_stride;
 846         dstv += dst_stride;
 847         dstc += dst_stride;
 848     }
 849 }
 850
 851 static void mc_weight_w2_altivec( uint8_t *dst, intptr_t i_dst, uint8_t *src, intptr_t i_src,
 852                                   const x264_weight_t *weight, int i_height )
 853 {
 854     LOAD_ZERO;
 855     PREP_LOAD;
 856     PREP_LOAD_SRC( src );
 857     vec_u8_t srcv;
 858     vec_s16_t weightv;
 859     vec_s16_t scalev, offsetv, denomv, roundv;
 860     vec_s16_u loadv;
 861
 862     int denom = weight->i_denom;
 863
 864     loadv.s[0] = weight->i_scale;
 865     scalev = vec_splat( loadv.v, 0 );
 866
 867     loadv.s[0] = weight->i_offset;
 868     offsetv = vec_splat( loadv.v, 0 );
 869
 870     if( denom >= 1 )
 871     {
 872         loadv.s[0] = denom;
 873         denomv = vec_splat( loadv.v, 0 );
 874
 875         loadv.s[0] = 1<<(denom - 1);
 876         roundv = vec_splat( loadv.v, 0 );
 877
 878         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
 879         {
 880             VEC_LOAD( src, srcv, 2, vec_u8_t, src );
 881             weightv = vec_u8_to_s16( srcv );
 882
 883             weightv = vec_mladd( weightv, scalev, roundv );
 884             weightv = vec_sra( weightv, (vec_u16_t)denomv );
 885             weightv = vec_add( weightv, offsetv );
 886
 887             srcv = vec_packsu( weightv, zero_s16v );
 888             vec_ste( vec_splat( (vec_u16_t)srcv, 0 ), 0, (uint16_t*)dst );
 889         }
 890     }
 891     else
 892     {
 893         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
 894         {
 895             VEC_LOAD( src, srcv, 2, vec_u8_t, src );
 896             weightv = vec_u8_to_s16( srcv );
 897
 898             weightv = vec_mladd( weightv, scalev, offsetv );
 899
 900             srcv = vec_packsu( weightv, zero_s16v );
 901             vec_ste( vec_splat( (vec_u16_t)srcv, 0 ), 0, (uint16_t*)dst );
 902         }
 903     }
 904 }
 905 static void mc_weight_w4_altivec( uint8_t *dst, intptr_t i_dst, uint8_t *src, intptr_t i_src,
 906                                   const x264_weight_t *weight, int i_height )
 907 {
 908     LOAD_ZERO;
 909     PREP_LOAD;
 910     PREP_LOAD_SRC( src );
 911     vec_u8_t srcv;
 912     vec_s16_t weightv;
 913     vec_s16_t scalev, offsetv, denomv, roundv;
 914     vec_s16_u loadv;
 915
 916     int denom = weight->i_denom;
 917
 918     loadv.s[0] = weight->i_scale;
 919     scalev = vec_splat( loadv.v, 0 );
 920
 921     loadv.s[0] = weight->i_offset;
 922     offsetv = vec_splat( loadv.v, 0 );
 923
 924     if( denom >= 1 )
 925     {
 926         loadv.s[0] = denom;
 927         denomv = vec_splat( loadv.v, 0 );
 928
 929         loadv.s[0] = 1<<(denom - 1);
 930         roundv = vec_splat( loadv.v, 0 );
 931
 932         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
 933         {
 934             VEC_LOAD( src, srcv, 4, vec_u8_t, src );
 935             weightv = vec_u8_to_s16( srcv );
 936
 937             weightv = vec_mladd( weightv, scalev, roundv );
 938             weightv = vec_sra( weightv, (vec_u16_t)denomv );
 939             weightv = vec_add( weightv, offsetv );
 940
 941             srcv = vec_packsu( weightv, zero_s16v );
 942             vec_ste( vec_splat( (vec_u32_t)srcv, 0 ), 0, (uint32_t*)dst );
 943         }
 944     }
 945     else
 946     {
 947         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
 948         {
 949             VEC_LOAD( src, srcv, 4, vec_u8_t, src );
 950             weightv = vec_u8_to_s16( srcv );
 951
 952             weightv = vec_mladd( weightv, scalev, offsetv );
 953
 954             srcv = vec_packsu( weightv, zero_s16v );
 955             vec_ste( vec_splat( (vec_u32_t)srcv, 0 ), 0, (uint32_t*)dst );
 956         }
 957     }
 958 }
 959 static void mc_weight_w8_altivec( uint8_t *dst, intptr_t i_dst, uint8_t *src, intptr_t i_src,
 960                                   const x264_weight_t *weight, int i_height )
 961 {
 962     LOAD_ZERO;
 963     PREP_LOAD;
 964     PREP_LOAD_SRC( src );
 965     PREP_STORE8;
 966     vec_u8_t srcv;
 967     vec_s16_t weightv;
 968     vec_s16_t scalev, offsetv, denomv, roundv;
 969     vec_s16_u loadv;
 970
 971     int denom = weight->i_denom;
 972
 973     loadv.s[0] = weight->i_scale;
 974     scalev = vec_splat( loadv.v, 0 );
 975
 976     loadv.s[0] = weight->i_offset;
 977     offsetv = vec_splat( loadv.v, 0 );
 978
 979     if( denom >= 1 )
 980     {
 981         loadv.s[0] = denom;
 982         denomv = vec_splat( loadv.v, 0 );
 983
 984         loadv.s[0] = 1<<(denom - 1);
 985         roundv = vec_splat( loadv.v, 0 );
 986
 987         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
 988         {
 989             VEC_LOAD( src, srcv, 8, vec_u8_t, src );
 990             weightv = vec_u8_to_s16( srcv );
 991
 992             weightv = vec_mladd( weightv, scalev, roundv );
 993             weightv = vec_sra( weightv, (vec_u16_t)denomv );
 994             weightv = vec_add( weightv, offsetv );
 995
 996             srcv = vec_packsu( weightv, zero_s16v );
 997             VEC_STORE8( srcv, dst );
 998         }
 999     }
1000     else
1001     {
1002         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
1003         {
1004             VEC_LOAD( src, srcv, 8, vec_u8_t, src );
1005             weightv = vec_u8_to_s16( srcv );
1006
1007             weightv = vec_mladd( weightv, scalev, offsetv );
1008
1009             srcv = vec_packsu( weightv, zero_s16v );
1010             VEC_STORE8( srcv, dst );
1011         }
1012     }
1013 }
1014 static void mc_weight_w16_altivec( uint8_t *dst, intptr_t i_dst, uint8_t *src, intptr_t i_src,
1015                                    const x264_weight_t *weight, int i_height )
1016 {
1017     LOAD_ZERO;
1018     PREP_LOAD;
1019     PREP_LOAD_SRC( src );
1020     vec_u8_t srcv;
1021     vec_s16_t weight_lv, weight_hv;
1022     vec_s16_t scalev, offsetv, denomv, roundv;
1023     vec_s16_u loadv;
1024
1025     int denom = weight->i_denom;
1026
1027     loadv.s[0] = weight->i_scale;
1028     scalev = vec_splat( loadv.v, 0 );
1029
1030     loadv.s[0] = weight->i_offset;
1031     offsetv = vec_splat( loadv.v, 0 );
1032
1033     if( denom >= 1 )
1034     {
1035         loadv.s[0] = denom;
1036         denomv = vec_splat( loadv.v, 0 );
1037
1038         loadv.s[0] = 1<<(denom - 1);
1039         roundv = vec_splat( loadv.v, 0 );
1040
1041         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
1042         {
1043             VEC_LOAD( src, srcv, 16, vec_u8_t, src );
1044             weight_hv = vec_u8_to_s16_h( srcv );
1045             weight_lv = vec_u8_to_s16_l( srcv );
1046
1047             weight_hv = vec_mladd( weight_hv, scalev, roundv );
1048             weight_lv = vec_mladd( weight_lv, scalev, roundv );
1049             weight_hv = vec_sra( weight_hv, (vec_u16_t)denomv );
1050             weight_lv = vec_sra( weight_lv, (vec_u16_t)denomv );
1051             weight_hv = vec_add( weight_hv, offsetv );
1052             weight_lv = vec_add( weight_lv, offsetv );
1053
1054             srcv = vec_packsu( weight_hv, weight_lv );
1055             vec_st( srcv, 0, dst );
1056         }
1057     }
1058     else
1059     {
1060         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
1061         {
1062             VEC_LOAD( src, srcv, 16, vec_u8_t, src );
1063             weight_hv = vec_u8_to_s16_h( srcv );
1064             weight_lv = vec_u8_to_s16_l( srcv );
1065
1066             weight_hv = vec_mladd( weight_hv, scalev, offsetv );
1067             weight_lv = vec_mladd( weight_lv, scalev, offsetv );
1068
1069             srcv = vec_packsu( weight_hv, weight_lv );
1070             vec_st( srcv, 0, dst );
1071         }
1072     }
1073 }
1074 static void mc_weight_w20_altivec( uint8_t *dst, intptr_t i_dst, uint8_t *src, intptr_t i_src,
1075                                    const x264_weight_t *weight, int i_height )
1076 {
1077     LOAD_ZERO;
1078     PREP_LOAD_SRC( src );
1079     vec_u8_t src_1v, src_2v, src_3v;
1080     vec_s16_t weight_lv, weight_hv, weight_3v;
1081     vec_s16_t scalev, offsetv, denomv, roundv;
1082     vec_s16_u loadv;
1083
1084     int denom = weight->i_denom;
1085
1086     loadv.s[0] = weight->i_scale;
1087     scalev = vec_splat( loadv.v, 0 );
1088
1089     loadv.s[0] = weight->i_offset;
1090     offsetv = vec_splat( loadv.v, 0 );
1091
1092     if( denom >= 1 )
1093     {
1094         loadv.s[0] = denom;
1095         denomv = vec_splat( loadv.v, 0 );
1096
1097         loadv.s[0] = 1<<(denom - 1);
1098         roundv = vec_splat( loadv.v, 0 );
1099
1100         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
1101         {
1102             src_1v = vec_ld( 0,  src );
1103             src_2v = vec_ld( 16, src );
1104             src_3v = vec_ld( 19, src );
1105             src_1v = vec_perm( src_1v, src_2v, _src_ );
1106             src_3v = vec_perm( src_2v, src_3v, _src_ );
1107             weight_hv = vec_u8_to_s16_h( src_1v );
1108             weight_lv = vec_u8_to_s16_l( src_1v );
1109             weight_3v = vec_u8_to_s16_h( src_3v );
1110
1111             weight_hv = vec_mladd( weight_hv, scalev, roundv );
1112             weight_lv = vec_mladd( weight_lv, scalev, roundv );
1113             weight_3v = vec_mladd( weight_3v, scalev, roundv );
1114             weight_hv = vec_sra( weight_hv, (vec_u16_t)denomv );
1115             weight_lv = vec_sra( weight_lv, (vec_u16_t)denomv );
1116             weight_3v = vec_sra( weight_3v, (vec_u16_t)denomv );
1117             weight_hv = vec_add( weight_hv, offsetv );
1118             weight_lv = vec_add( weight_lv, offsetv );
1119             weight_3v = vec_add( weight_3v, offsetv );
1120
1121             src_1v = vec_packsu( weight_hv, weight_lv );
1122             src_3v = vec_packsu( weight_3v, zero_s16v );
1123             vec_st( src_1v, 0, dst );
1124             vec_ste( (vec_u32_t)src_3v, 16, (uint32_t*)dst );
1125         }
1126     }
1127     else
1128     {
1129         for( int y = 0; y < i_height; y++, dst += i_dst, src += i_src )
1130         {
1131             src_1v = vec_ld( 0,  src );
1132             src_2v = vec_ld( 16, src );
1133             src_3v = vec_ld( 19, src );
1134             src_1v = vec_perm( src_1v, src_2v, _src_ );
1135             src_3v = vec_perm( src_2v, src_3v, _src_ );
1136             weight_hv = vec_u8_to_s16_h( src_1v );
1137             weight_lv = vec_u8_to_s16_l( src_1v );
1138             weight_3v = vec_u8_to_s16_h( src_3v );
1139
1140             weight_hv = vec_mladd( weight_hv, scalev, offsetv );
1141             weight_lv = vec_mladd( weight_lv, scalev, offsetv );
1142             weight_3v = vec_mladd( weight_3v, scalev, offsetv );
1143
1144             src_1v = vec_packsu( weight_hv, weight_lv );
1145             src_3v = vec_packsu( weight_3v, zero_s16v );
1146             vec_st( src_1v, 0, dst );
1147             vec_ste( (vec_u32_t)src_3v, 16, (uint32_t*)dst );
1148         }
1149     }
1150 }
1151
1152 static weight_fn_t x264_mc_weight_wtab_altivec[6] =
1153 {
1154     mc_weight_w2_altivec,
1155     mc_weight_w4_altivec,
1156     mc_weight_w8_altivec,
1157     mc_weight_w16_altivec,
1158     mc_weight_w16_altivec,
1159     mc_weight_w20_altivec,
1160 };
1161
1162 #endif // !HIGH_BIT_DEPTH
1163
1164 void x264_mc_altivec_init( x264_mc_functions_t *pf )
1165 {
1166 #if !HIGH_BIT_DEPTH
1167     pf->mc_luma   = mc_luma_altivec;
1168     pf->get_ref   = get_ref_altivec;
1169     pf->mc_chroma = mc_chroma_altivec;
1170
1171     pf->copy_16x16_unaligned = x264_mc_copy_w16_altivec;
1172     pf->copy[PIXEL_16x16] = x264_mc_copy_w16_aligned_altivec;
1173
1174     pf->hpel_filter = x264_hpel_filter_altivec;
1175     pf->frame_init_lowres_core = frame_init_lowres_core_altivec;
1176
1177     pf->weight = x264_mc_weight_wtab_altivec;
1178 #endif // !HIGH_BIT_DEPTH
1179 }