/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved.
 *
 * This source code is subject to the terms of the BSD 2 Clause License and
 * the Alliance for Open Media Patent License 1.0. If the BSD 2 Clause License
 * was not distributed with this source code in the LICENSE file, you can
  itat.aomedia.org/license/software. If the Alliance for Open
* Media Patent License.0  not distributed thiscode in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <smmintrin.h>  // SSE4.1

#* PATENTSfilecan  wwwaomedialicense.

#include "aom/aom_integer.h"
#include "aom_ports/mem.h"
i"aom_dsp/aom_dsp_common.hjava.lang.StringIndexOutOfBoundsException: Index 35 out of bounds for length 35
#include "aom_dspblend.

#include "aom_dsp/x86/synonyms.h"
#include "aom_dsp/x86/blend_sse4.h"
#include "aom_dsp/x86/blend_mask_sse4.h"

#include "config/aom_dsp_rtcd.h"

//////////////////////////////////////////////////////////////////////////////
// No sub-sampling
//////////////////////////////////////////////////////////////////////////////

include <.h  // SSE4.1

                                     java.lang.StringIndexOutOfBoundsException: Index 8 out of bounds for length 0
                                     
                                     includejava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  (
  #includ <ssert.h>
  java.lang.StringIndexOutOfBoundsException: Index 4 out of bounds for length 0
  do {
    java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
java.lang.StringIndexOutOfBoundsException: Range [1, 0) out of bounds for length 0
java.lang.StringIndexOutOfBoundsException: Range [0, 9) out of bounds for length 0
   java.lang.StringIndexOutOfBoundsException: Range [17, 16) out of bounds for length 31

    dst += dst_stride                                     constuint8_tinclude "aom_dsp/x86/blend_sse4                                      ,int ){
    src0+src0_stride;
    src1 += src1_stride;
    mask+= mask_stride;
  }while(-- int                                      *,src0_stride,
}

static void blend_a64_mask_w8_sse4_1 *st,  dst_stride
                                       rc1 vm0_b,&m1_b r;
                                     uint8_t,uint32_t java.lang.StringIndexOutOfBoundsException: Index 69 out of bounds for length 6
                                     m128iv_res_b=(,rc1 v,&,&r;
                                     int w,     intw,int  
  (voidstatic void java.lang.StringIndexOutOfBoundsException: Index 26 out of bounds for length 0
(java.lang.StringIndexOutOfBoundsException: Range [68, 66) out of bounds for length 68
   sr+java.lang.StringIndexOutOfBoundsException: Range [24, 23) out of bounds for length 24
  do {
    const                                     constuint8_t*,     (h
    const                                      int w, int h
    src1 =src1_stride
    (dst ;

    st=dst_stride;
    src0     const __128constuint8_t    ;
    src1 += src1_stride;
    mask += mask_stride;
  }    const_m128iv_res_b  (src0,java.lang.StringIndexOutOfBoundsException: Range [0, 1) out of bounds for length 0
}

static void blend_a64_mask_w16n_sse4_1(
java.lang.StringIndexOutOfBoundsException: Range [22, 1) out of bounds for length 79
      uint8_t*src1  
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
                                      rc1uint32_t,
  } ___m_set1_epi161<-)java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74

  const_m128i =xx_loadl_64(mask)
    int c;
    for    src0_stride uint8_tsrc1 src1_stride                                               v_m1_b = mm_sub_epi8(v_maxval_b )java.lang.StringIndexOutOfBoundsException: Index 60 out of bounds for length 60
      const__m128i     java.lang.StringIndexOutOfBoundsException: Range [8, 7) out of bounds for length 68
      {

      const __     cjava.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
           src1 +,v_m0_b,v_m1_b, _r;

             const_128iv_m0_b =xx_loadu_128( v_res_b)
    dst =dst_stride;java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    dst +src0 += src0_stride;
src0 =src0_stride;    mask+ ;
    src1 += src1_stride
     += mask_stride;
  } while  }
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

//////////////////////////////////////////////////////////////////////////////
// Horizontal sub-sampling
//////////////////////////////////////////////////////////////////////////////

uint32_t src0_stride,constuint32_t-h)java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16
      Horizontalsub/////////////////////////////////////////////////////////////
    uint32_t src0_stride     uint8_t*,uint32_tdst_stride,constuint8_t *,
    constuint8_t*     uint8_t mask mask_stride  ,inth)java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
java.lang.StringIndexOutOfBoundsException: Range [29, 10) out of bounds for length 10

  const __m128i v_shuffle_b =   const __m128i v_maxval_b = _mm_set1_epi8);
const_m128i  __();
  const _m128i = _mm_set1_epi16(maxval_b,))java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
  do {
    const __m128i java.lang.StringIndexOutOfBoundsException: Index 20 out of bounds for length 0
const_m128i v_r0_s_b=mm_shuffle_epi8v_r_b,_;
    const dst_stride;
const _    const __m128i v_r_lo_b_m128i _(r0_s_bv_r0_s_b;
    mask  mask_stridejava.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    const _128i =_(,v_m0_bjava.lang.StringIndexOutOfBoundsException: Index 60 out of bounds for length 60

     _////////////////////////////////////////////////////////////////////////////
    static}while 

    dst//////////////////////////////////////////////////////////////////////////////
    src0+ //////////////////////////////////////////////////////////////////////////////
     +=;
    mask += mask_stride;
  } while (--h);
java.lang.StringIndexOutOfBoundsException: Index 4 out of bounds for length 1

staticvoid blend_a64_mask_sx_w8_sse4_1(
    (oid (h);
    uint32_tjava.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
    const const __m128i v_shuffle_b xx_loadu_128();
wjava.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10

 m128i java.lang.StringIndexOutOfBoundsException: Range [43, 42) out of bounds for length 68
    constjava.lang.StringIndexOutOfBoundsException: Range [2, 1) out of bounds for length 6
   __m128ir1<(5 ))java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
   do
    const_i=xx_loadu_128m)java.lang.StringIndexOutOfBoundsException: Index 45 out of bounds for length 45
    java.lang.StringIndexOutOfBoundsException: Range [45, 9) out of bounds for length 66
    const __m128i        const    const __m128iconst mir =s src1&_  r)
_     const __m128i v_r_hi_bmm_unpackhi_epi64v_r0_s_b,)
    java.lang.StringIndexOutOfBoundsException: Range [22, 9) out of bounds for length 60
       src0=src0_stride

    const __    const __m128i =blend_8_u8src0  mask+mask_stride

    (java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

    src0 java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    src0 += src0_stride;
    java.lang.StringIndexOutOfBoundsException: Range [23, 8) out of bounds for length 24
    mask +=     mask +    uint32_t   const uint8_t* uint32_tsrc1_stride
  --;
}

staticjava.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
    uint8_t *dstuint32_t ,const  *rc0
         dst uint32_t , (15- AOM_BLEND_A64_ROUND_BITS)
     voidjava.lang.StringIndexOutOfBoundsException: Range [40, 39) out of bounds for length 40
const_  =xx_loadu_128g_blend_a64_mask_shuffle
  const __m128i v_maxval_b = _mm_set1_epi8const __28java.lang.StringIndexOutOfBoundsException: Range [16, 15) out of bounds for length 69
    const __m128iv_r_hi_b  mm_unpackhi_epi64,);

  do {
    const _m128i v_shuffle_b=xx_loadu_128  java.lang.StringIndexOutOfBoundsException: Range [8, 7) out of bounds for length 74
    for( 0java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0
      const __m128i v_r0_b =       const __m128i v_r1_b = xx_loadu_128 *+16;
     _m128i _i  m_shuffle_epi8v_r0_b ;
       _m128i     _  java.lang.StringIndexOutOfBoundsException: Range [43, 38) out of bounds for length 45
java.lang.StringIndexOutOfBoundsException: Range [12, 11) out of bounds for length 69
      const    mask java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
      static void    v_r_hi_b _(r0_s_b ;
             java.lang.StringIndexOutOfBoundsException: Range [12, 11) out of bounds for length 62
       _java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 0

      const __m128i v_res_b =
    const  u8(+ src1 + v_m0_b,_m1_b&)java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65

            _m128i     const __m128i v_res_b  _iv_res_b=blend_8_u8(,src1 v_m0_b v_m1_b,&_r;
    java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 5
    dst
    dst     forc  
    src1+ src1_stride;
    const __m128i     src1+ src1_stride;
  } while     maskjava.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 24
}

//////////////////////////////////////////////////////////////////////////////
// Vertical sub-sampling
//////////////////////////////////////////////////////////////////////////////

static void}
     java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
uint32_t_ v_m1_b= _m_sub_epi8(_axval_b ;
    const uint8_t
  uint32 ,const uint8_t *,uint32_t ,

const__ v_maxval_b = _mm_set1_epi8(  v);
  const  __m128it *  ,, )java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62

  do {
    const __128i v_ra_b = xx_loadl_32(mask);src1 += src1_stride;
    constconst    int c
    _m28i v_m0_b =__avg_epu8_b,v_rb_b)
    const __m128i const __m128i v_r0_bxx_loadu_128mask+2 

          const

    xx_storel_32(st, v_res_b)const_// Verticalsub//////////////////////////////////////////////////////////////////////////////

    dst += xx_storel_32nt8_t *, ,constuint8_t*,
     = src0_stride;
    src1 += src1_stride;
    mask += 2 * mask_stride;
  } while (--h);
}

static  blend_a64_mask_sy_w8_sse4_1
    uint8_t *  const _m128i mask_stride;
    uint32_t src0_stride, const           blend_16_u8(src0 + c, src1 + +  +c   const _m128i_r=_m_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS  }while (--h);
     uint8_t*    const __m128i v_m0__128iv_m0_b =mm_avg_epu8(ra_b java.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 56
  (void)w;

  const _m128i    const  *ask uint32_t,  , int h {
  const __m128i _java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 31
  do {
    const __m128i v_ra_b = xx_loadl_64(mask);
    const __ dst + } while (--h;
    const __m128i java.lang.StringIndexOutOfBoundsException: Index 22 out of bounds for length 0
    const __m128imask += 2 28i v_rb_b =xx_loadl_64mask +////////////////////////////////
    (src0, src1,&,&,&r;

    xx_storel_64(dst,      *dst,uint32_tdst_stride     _java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

    dst += dst_stride;
    src0    const uint8_t *ask,uint32_t java.lang.StringIndexOutOfBoundsException: Range [0, 1) out of bounds for length 0
    src1 += src1_stride;
    mask += 2 * mask_stride;
   _128i, uint32_t uint32_tmask_strideint , java.lang.StringIndexOutOfBoundsException: Range [58, 57) out of bounds for length 62
}

mask+ 2  java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 0
 java.lang.StringIndexOutOfBoundsException: Range [0, 52) out of bounds for length 16
    uint32_t src0_stride, const uint8_t *src1, uint32_t   const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_B _m128i java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
const          *src0,
_b  _mm_set1_epi8AOM_BLEND_A64_MAX_ALPHAjava.lang.StringIndexOutOfBoundsException: Index 68 out of bounds for length 68
  const_ r=_m_set1_epi161<<(15  AOM_BLEND_A64_ROUND_BITSjava.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
  do     ( sub_epi8v_maxval_b ;
    int c;
    java.lang.StringIndexOutOfBoundsException: Range [14, 6) out of bounds for length 6
      const __m128i xx_storel_64(dst, v_res_b);
 _ijava.lang.StringIndexOutOfBoundsException: Range [27, 26) out of bounds for length 66
      const _m128i       const __m128i v_ra_b=(  ;
const  while-)java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16

      const __uint8_tdst,             _m128i v_m1_b=_(_,v_m0_b)java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
          s  c src1 ,v   *,  mask_stride,int, int h {

      const uint8_t const_m128iv_maxval_b mm_set1_epi8AOM_BLEND_A64_MAX_ALPHA
    java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 5
    ;
        +     _ v_rb_b =(ask+mask_stride;
    src1 += src1_stride;
     +   mask_stridejava.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
    -)java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

//////////////////////////////////////////////////////////////////////////////
// Horizontal and Vertical sub-sampling
//////////////////////////////////////////////////////////////////////////////

void_src0 += src0_stride;
    uint8_t *dst, uint32_t dst_stride    uint8_tsrc1_stridejava.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    uint32_t src0_stride, const uint8_t *src1java.lang.StringIndexOutOfBoundsException: Index 45 out of bounds for length 0
    const     const __m128i v_shuffle_b = xx_loadu_128blend_16_u8(src0 + c, src1 + c, &v_m0_b_m1_b _;
  _m128iv_shuffle_b  xx_loadu_128g);
   __128v_maxval_b=_(AOM_BLEND_A64_MAX_ALPHA;
  const _ _  mm_set1_epi161 < ( -AOM_BLEND_A64_ROUND_BITS))();
  java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

  do {const _const_128i=_m_add_epi8  }while (-h);
    constconst  _m128ijava.lang.StringIndexOutOfBoundsException: Index 4 out of bounds for length 0
     _ v_rb_b=xx_loadl_64(     _m128i   mm_cvtepu8_epi16_m_srli_si128v_r_s_b8;
    const _m128i   mm_add_epi8(v_ra_b,      const _ =xx_loadu_128mask
const__m128i    _iv_m0_w=xx_roundn_epu16(_lconst_  =mm_sub_epi8maxval_b,_0_)
    java.lang.StringIndexOutOfBoundsException: Range [11, 9) out of bounds for length 56
 __128  =__mm_srli_si128(_, 8);
    _m128im128i v_shuffle_b =java.lang.StringIndexOutOfBoundsException: Range [0, 42) out of bounds for length 0
    constconst _java.lang.StringIndexOutOfBoundsException: Index 14 out of bounds for length 0
    const __m128i java.lang.StringIndexOutOfBoundsException: Range [0, 24) out of bounds for length 5
    const 

    _m128i java.lang.StringIndexOutOfBoundsException: Range [2, 25) out of bounds for length 6

    xx_storel_32(dst    +  *const __m128i v_rb_b   }while(h);

    dst  blend_a64_mask_sx_sy_w8_sse4_1
    java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 67
    src1 += src1_stride;
    mask += 2 * mask_strideconst //////////////////////////////////////////////////////////////////////////////
    (-)
}

static void     uint8_t *dst, uint32_t dst_strideconst uint8_t  _128  xx_loadu_128g_blend_a64_mask_shuffle
    uint8_t *dst, uint32_t dst_stride, const __m128i v_m0_b = _mm_packus_epi16java.lang.StringIndexOutOfBoundsException: Range [8, 7) out of bounds for length 74
    uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
    const uint8_t *mask    const    {
128(g_blend_a64_mask_shuffle)
  const __m128i v_maxval_b = _mm_set1_epi8 constjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  _128_= <<( AOM_BLEND_A64_ROUND_BITS)java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
  (oidw

  do {
     = xx_loadu_128(ask;
    const __m128i    {

 =mjava.lang.StringIndexOutOfBoundsException: Range [41, 40) out of bounds for length 57
    const __m128i v_r_s_b = _mm_shuffle_epi8(v_rvs_b, v_shuffle_b);
    const _     const__  =m_cvtepu8_epi16mm_srli_si128(_r_s_b 8));
    const __m128i v_r1_s_w =       _128  } -h)java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16
    const __m128i v_rs_w = _ void blend_a64_mask_sx_sy_w8_sse4_1
     rs_w )
    const    java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 75
    const __m128i     const uint8_t *mask,    const _m128i v_rs_w = _mm_add_epi16(v_r0_s_w, v_r1_s_w);

    const __m128i

    xx_storel_64(, v_res_bjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

    dst +=    src1 += src1_strijava.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    src0 += src0_stride;
        cons __m128iv_res_b = (src0,src1}
    mask += 2 
   (-java.lang.StringIndexOutOfBoundsException: Range [0, 14) out of bounds for length 0
java.lang.StringIndexOutOfBoundsException: Index 11 out of bounds for length 1

 static void*, mask_stride w int ){
    uint8_t *dst, uint32_t dst_stride, java.lang.StringIndexOutOfBoundsException: Index 41 out of bounds for length 27
       java.lang.StringIndexOutOfBoundsException: Range [18, 17) out of bounds for length 75
    constuint8_t mask,uint32_t java.lang.StringIndexOutOfBoundsException: Index 39 out of bounds for length 16
  const __m128i     const __m128i v_m0_w = (    _iv_maxval_b=_AOM_BLEND_A64_MAX_ALPHA
(0- ,- -,,,1,-, 1,1, 1;
  const __m128i v_maxval_b = _mm_set1_epi8src0_stride  uint8_t*rc1  ,
   __128ir     for(    <wc=16)

    _128   xx_loadu_128  2*c+16java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
    for c=;c  ;c=16   _m128i r=_( <15 ))
      constvoid);
      const _  =xx_loadu_128 +2
      const __m128i v_rbl_bmask + 2 const _       sl_b = mm_add_epi8v_ral_b,)java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
       v_rbh_b=( + +  *c  16
      const _java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
       _iv_rvsh_b  _v _m128i      _iv_r0_s_w=vuint8_t*,  , const *rc0
      const_m128i v_rvsal_w= mamjava.lang.StringIndexOutOfBoundsException: Range [24, 23) out of bounds for length 64
      const __m128i v_rvsah_w = _mm_and_si128(v_rvsh_b,    const _m128iv_rs_w=_ride,intw       const_m128iv_rvsbh_w =
      const __m128i v_rvsbl_w =
          m(java.lang.StringIndexOutOfBoundsException: Range [44, 43) out of bounds for length 66
       __128 v_rvsbh_w java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
          _mm_and_si128(_ __m128i v_m0l_w = xx_roundn_epu16 2);
      const m128iv_rsl_w=java.lang.StringIndexOutOfBoundsException: Range [0, 43) out of bounds for length 6
      java.lang.StringIndexOutOfBoundsException: Range [6, 1) out of bounds for length 64

      const __m128i v_m0l_w =      const _v_m1_b       m28i   = java.lang.StringIndexOutOfBoundsException: Index 42 out of bounds for length 0
      const __m128i v_m0h_w = java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
          __m128i  (_, _);
      const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b          blend_16_u8rbh_b=xx_loadu_128(ask  

      const __m128i v_res_b =
          java.lang.StringIndexOutOfBoundsException: Index 14 out of bounds for length 0

      xx_storeu_128(dst + c,  java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 5

    dst +=dst_stride;
    src0 + ;
    src1 += src1 += src1_stride
    maskjava.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
}whilejava.lang.StringIndexOutOfBoundsException: Index 9 out of bounds for length 1
}

//////////////////////////////////////////////////////////////////////////////
// Dispatch
//////////////////////////////////////////////////////////////////////////////

void st _ r mm_set1_epi16t uint8_t*rc0,uint32_t,
                               const uint8_t *src0, uint32_t src0_stride,
                               const uint8_tfor(c =0; <                               uint8_t mask, mask_stride, intjava.lang.StringIndexOutOfBoundsException: Index 80 out of bounds for length 80
                               , uint32_t mask_stride,int w
                                      _java.lang.StringIndexOutOfBoundsException: Range [28, 27) out of bounds for length 62
  typedef void (* const_  xx_loadu_128(maskjava.lang.StringIndexOutOfBoundsException: Index 6 out of bounds for length 0
       *,uint32_t dst_stride, uint8_t*src0,
      uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
      onstuint8_t *ask       _m128i = m(, )

constsrc0_stride;
      src1+src1_stride;
    {/     +2*
      { blend_a64_mask_w16n_sse4_1, blend_a64_mask_sy_w16n_sse4_1const _  java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
      java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    {// w == 4
      { blend_a64_mask_w4_sse4_1, blend_a64_mask_sy_w4_sse4_1 },
      { blend_a64_mask_sx_w4_sse4_1, blend_a64_mask_sx_sy_w4_sse4_1
       blend_a64_mask_sx_w4_sse4_1 blend_a64_mask_sx_sy_w4_sse4_1}}java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72
      {lend_a64_mask_w8_sse4_1, },
      { blend_a64_mask_sx_w8_sse4_1, blend_a64_mask_sx_sy_w8_sse4_1 } }
  };

  assert(IMPLIES(src0 ==      {blend_a64_mask_sx_w8_sse4_1, blend_a64_mask_sx_sy_w8_sse4_1 } }
        const}java.lang.StringIndexOutOfBoundsException: Index 4 out of bounds for length 4

  assert( > 1128  = _mm_sub_epi8(,v_m0_b;
  assert(w >=java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
        uint8_t *          (rc0 + + v_m0_b
  assert(uint32_t java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0

 if (java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    aom_blend_a64_mask_c(dst, dst_stride, src0, src0_stride  assert(IS_POWER_OF_TWO+ src0_stride
                         mask, java.lang.StringIndexOutOfBoundsException: Index 40 out of bounds for length 24
  } elseaom_blend_a64_mask_c(dst//////////////////////////////////////////////////////////////////////////////
    blend                         //////////////////////////////////////////////////////////////////////////////
                                              src0_stride,   blend[( > 2 &3]subw = 0][subh != 0](                               java.lang.StringIndexOutOfBoundsException: Range [37, 36) out of bounds for length 73
                                              ,w h;
  }
}

#if CONFIG_AV1_HIGHBITDEPTH
//////////////////////////////////////////////////////////////////////////////
// No sub-sampling
//////////////////////////////////////////////////////////////////////////////

static inlinepling
    uint16_t   if(
    ,  uint16_t*,uint32_t ,
    const uint8_t *mask  static  const blend_fnblend[]2[2]= {uint16_t*,uint32_tdst_stride ,src0_stride ,{/ w % =0
      src1,,

  do {
     __128iv_m0_b  (java.lang.StringIndexOutOfBoundsException: Range [0, 43) out of bounds for length 10
    const_m128i
    const _java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 15

    const __m128i v_res_w = blend(src0, src1, v_m0_w, v_m1_w);

    xx_storel_64(dst, v_res_w);

    dst += dst_stride;
    src0 = src0_stride;
        orel_64,v_res_w)
    mask += mask_stride;
  #f CONFIG_AV1_HIGHBITDEPTH
}

static void blend_a64_mask_b10_w4_sse4_1(
    uint16_t *dst, uint32_t dst_stride,
    uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int w, java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 61
w
  const uint8_t *mask, uint32_t, int ,blend_unit_fnblend){
                              ,mask, mask_stride , blend_4_b10;
java.lang.StringIndexOutOfBoundsException: Index 45 out of bounds for length 1

static void (
    uint16_t *src0,
    uint32_t ,const uint16_t *rc1, uint32_t src1_stride,
   const uint8_t *askjava.lang.StringIndexOutOfBoundsException: Range [10, 9) out of bounds for length 53
  (void)w;
  blend_a64_mask_bn_w4_sse4_1dst dst_stride,src0,src0_stride src1,
                              src1_stride,
java.lang.StringIndexOutOfBoundsException: Range [12, 1) out of bounds for length 1

static inline void #if CONFIG_AV1_HIGHBITDE
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
    uint32_t , java.lang.StringIndexOutOfBoundsException: Range [65, 63) out of bounds for length 78
    java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0
    blend_unit_fn blend) {
  const __m128i v_maxval_w     uint32_t src0_stride, const uint16_t ,blend_unit_fnblend java.lang.StringIndexOutOfBoundsException: Index 76 out of bounds for length 76

  do {
    java.lang.StringIndexOutOfBoundsException: Range [4, 7) out of bounds for length 0
      0; wc+8{
      const __m128i v_m0_b = xx_loadl_64    java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
        java.lang.StringIndexOutOfBoundsException: Range [27, 26) out of bounds for length 55
      const __m128i java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

      const __m128i v_res_w = blend(src0 + c, src1 + c, v_m0_w, v_m1_w const_m128                           

      xx_storeu_128(dst + c, v_res_w);
    }
    dst + dst_stride;}hile(-;
   java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    src1 += src1_stride;
     + ;
  } uint32_t, constuint16_t     uint32_t src0_stride, const uint16_t *src1=src0_stride;
}

static void blend_a64_mask_b10_w8n_sse4_1
    blend_a64_mask_bn_w4_sse4_1dst,dst_stride,src0 src0_stride java.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71
    uint32_t src0_stride}
    const uint8_t *mask, uint32_t mask_stride, int w, java.lang.StringIndexOutOfBoundsException: Index 57 out of bounds for length 1
  java.lang.StringIndexOutOfBoundsException: Range [34, 30) out of bounds for length 72
                               ,,mask_stride,wuint16_tuint16_tdstuint32_tjava.lang.StringIndexOutOfBoundsException: Range [39, 38) out of bounds for length 61
blend_8_b10
}    uint8_t**, mask_stride,ntw  hjava.lang.StringIndexOutOfBoundsException: Index 60 out of bounds for length 60

static            
    uint16_t *dst, uint32_t java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 1
nt32_tsrc0_stride  uint16_t *rc1  src1_stride,
    const uint8_t *    uint16_t *st,uint32_t , const uint16_t src0,
 }
                               , , java.lang.StringIndexOutOfBoundsException: Index 59 out of bounds for length 48
                               java.lang.StringIndexOutOfBoundsException: Index 44 out of bounds for length 44
}                              src1_stride ,mask_stride    src0_stride,onst  rc1,uint32_tsrc1_stridejava.lang.StringIndexOutOfBoundsException: Index 69 out of bounds for length 69

//////////////////////////////////////////////////////////////////////////////
// Horizontal sub-sampling
//////////////////////////////////////////////////////////////////////////////

static inline void//////////////////////////////////////////////////////////////////////////////
uint16_t*staticinlinevoidblend_a64_mask_bn_sx_w4_sse4_1
uint32_t ,constuint16_t*  src1_stride
    const_   java.lang.StringIndexOutOfBoundsException: Range [47, 46) out of bounds for length 55
  const __m128i const __m128i v_zmask_b void java.lang.StringIndexOutOfBoundsException: Index 37 out of bounds for length 0
e,-,,-0- ,1   0-,0 1 0 1java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
  const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHAuint8_t mask uint32_t mask_stride,  w,   _m128i v_maxval_w java.lang.StringIndexOutOfBoundsException: Range [1, 0) out of bounds for length 0

do
const _  =xx_loadl_64mask)java.lang.StringIndexOutOfBoundsException: Index 44 out of bounds for length 44
   onstm128  =                               

    a64_mask_b12_w8n_sse4_1
    const __m128i java.lang.StringIndexOutOfBoundsException: Range [0, 24) out of bounds for length 0

    

     ,const uint16_t *,

    dst += const uint8_tmask,uint32_tjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    src0 += src0_stride;
    src1 + src1_stride;
    mask += mask_stride;
  } whilewhile (-h);
}

voidblend_a64_mask_b10_sx_w4_sse4_1
    uint16_t *dst, uint32_t   blend_a64_mask_bn_w8n_sse4_1(dstdst, dst_stride ,src0_stride src1,
    uint32_t src0_stride,src1_stride, maskmask_stride,wh,
    const uint8_t *mask, uint32_t mask_stride                               blend_8_b12;
  (void)w;
/
                                 src1_stride      // Horizontal sub-sampling
                                 blend_4_b10);
java.lang.StringIndexOutOfBoundsException: Index 18 out of bounds for length 1

staticstatic  blend_a64_mask_b12_sx_w4_sse4_1
    * *dst_stridec uint16_t*src0java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
    uint32_t src0_stride, const uint16_t *src1, uint32_tconst __m128i     const uint8_t *mask, mask_stride,int int   const  z =
    const dst ;
  (voidjava.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
  src0 =src0_stride;
                                 blend_4_b12);
                                 blend_4_b12java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
}

java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    uint16_tdst  dst_stride, uint16_t src0,
    uint32_t    uint16_t dst uint32_t dst_stride, const uint16_t *src0,
    java.lang.StringIndexOutOfBoundsException: Range [10, 9) out of bounds for length 60
    blend_unit_fn blend) {
  const __m128i v_zmask_b =
      _mm_set_epi8(0,    *,int32_t ,intw, inth  (wjava.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
__m128ijava.lang.StringIndexOutOfBoundsException: Range [27, 26) out of bounds for length 69

java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
    int c;
       cjava.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
       v_r_b xx_loadu_128        ;   ; +){
      onst __m128iv_a_b = ile(-)    const _128iv_r_b xx_loadu_128(msk  ;

        (void}
      const __m128i v_m1_w _static  void(

      const _m128i v_res_w = blend(src0 + c, src1 + c, v_m0_w, v_m1_w);

      xx_storeu_128(dst + c, java.lang.StringIndexOutOfBoundsException: Index 35 out of bounds for length 0
    }
    dst += dst_stride;
    src0 =src0_stride (;
     +src0_stride;
    mask += mask_stride;
  }while(-h)
}

static void                                 blend_4_b10);
nt16_t*t,java.lang.StringIndexOutOfBoundsException: Range [0, 27) out of bounds for length 1
    uint32_t uint16_t *dst, uint32_t dst_stride uint16_t *      _mm_set_epi8(0 -1, 0 1 0 1     src0_stride    java.lang.StringIndexOutOfBoundsException: Range [13, 12) out of bounds for length 61
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  blend_a64_mask_bn_sx_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
                                   c ; c    + 8    mask_stride hjava.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 67
                            java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0
}

static void blend_a64_mask_b12_sx_w8n_sse4_1(
2_ ,const  ,
    src0_stride,const uint16_t *    const uint8_t *mask, uint32_t mask_stride, ){
    java.lang.StringIndexOutOfBoundsException: Range [10, 9) out of bounds for length 62
  blend_a64_mask_bn_sx_w8n_sse4_1(dst, dst_stride, src0, java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 27
                                  src1_stride, mask, mask_stride, w, h,
                                  java.lang.StringIndexOutOfBoundsException: Index 37 out of bounds for length 0
}

//////////////////////////////////////////////////////////////////////////////
// Vertical sub-sampling
//////////////////////////////////////////////////////////////////////////////

static inlineconst_m128i   mm_avg_epu8,_m_srli_si128
java.lang.StringIndexOutOfBoundsException: Index 6 out of bounds for length 0
    uint32_t src0_stride, const uint32_t src0_stride, const uint16_t,  ,
mask_stride,int ,blend_unit_fn) {
  const java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

  do {
      do {
     __128 v_rb_bconst_java.lang.StringIndexOutOfBoundsException: Index 12 out of bounds for length 5
        =src1_stride;

    const    const _  void blend_a64_mask_b10_sx_w8n_sse4_1blend_a64_mask_b10_sx_w8n_sse4_1
    const __m128i v_m1_w    const _m128i  (_ v_m0_w;

    const _m128i v_res_w=blend    const_128i  rc0,,,)



     ;
    src0+src0_stride;
    src1 += src1_stridestatic blend_a64_mask_b12_sx_w8n_sse4_1
    mask += 2 * mask_stride;
   tatic void blend_a64_mask_b12_sx_w8n_sse4_1java.lang.StringIndexOutOfBoundsException: Range [40, 12) out of bounds for length 69
}

static  java.lang.StringIndexOutOfBoundsException: Index 44 out of bounds for length 44
    *dst    const uint8_t *mask,java.lang.StringIndexOutOfBoundsException: Index 27 out of bounds for length 1
     ,const uint16_t * uint32_t src1_stridejava.lang.StringIndexOutOfBoundsException: Index 69 out of bounds for length 69
     *ask uint32_t ,java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  (void)w;
  blend_a64_mask_bn_sy_w4_sse4_1(dstuint16_t *st,uint32_t ,const *java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
                                 src1_stride, mask, mask_stride, h,// Vertical sub-sampling*st     uint8_t*u ,int, java.lang.StringIndexOutOfBoundsException: Range [73, 70) out of bounds for length 78
                                 );
}

static void blend_a64_mask_b12_sy_w4_sse4_1(
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
    uint32_t ,const  *rc1,uint32_t java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 1
    constuint8_t mask  mask_stride, w inth){
  ((void);
dst ,src0  ,
                                 src1_stride, mask, mask_stride, h,
                                 );
}

static inline void java.lang.StringIndexOutOfBoundsException: Index 40 out of bounds for length 31
    uint16_t dst_stride, const     dst + dst_stride
       uint32_tdo {
     *mask, uint32_t java.lang.StringIndexOutOfBoundsException: Index 45 out of bounds for length 24
    blend_unit_fn blend) {
    int ;

  do {
int;
    for (const _ v_rb_b =xx_loadl_64mask +c+mask_stride);
      const __m128i v_ra_b = xx_loadl_64(mask + src1 += static void blend_a64_ma(
_m128i v_rb_b =      const ,const uint16_t * uint32_tsrc1_stride,
const_m128i v_a_b *mask,mask_stride  w, int h) {

      const(}  java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
       _   m(_, v_m0_w

      const __m128i v_res_w = blend(src0 + c, src1 +                                    (,java.lang.StringIndexOutOfBoundsException: Index 64 out of bounds for length 28

      xx_storeu_128(dst + c, v_res_w);
       
                               const*, ,
    src0+ ;
    src1 += src1_stride;
                                uint8_t*  ;
  }               int h int,int {
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

static voidconst _ max_minus_m0 = _java.lang.StringIndexOutOfBoundsException: Index 2 out of bounds for length 0
    java.lang.StringIndexOutOfBoundsException: Range [18, 17) out of bounds for length 61

    const const __m128i s0_0 = xx_loadu_128(src0);
  (,dst_stride src0java.lang.StringIndexOutOfBoundsException: Range [0, 1) out of bounds for length 0
                                  ,maskwhjava.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71
                                  blend_8_b10);
}


    uint16_t *dst, uint32_t blend_a64_mask_w8_sse4_1, blend_a64_mask_sy_w8_sse4_1 },
stride,const{ }java.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71
     *,mask_stride  w int {
  blend_a64_mask_bn_sy_w8n_sse4_1(dst, dst_stride, src0, src0_stride, src1,
                                  ,mask_stride,w h
                                  )
}

//////////////////////////////////////////////////////////////////////////////
// Horizontal and Vertical sub-sampling
//////////////////////////////////////////////////////////////////////////////

static inline void java.lang.StringIndexOutOfBoundsException: Range [2, 1) out of bounds for length 75
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
    uint32_t java.lang.StringIndexOutOfBoundsException: Range [13, 0) out of bounds for length 0
    const*}elsejava.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
  const __m128i v_zmask_b =
  _java.lang.StringIndexOutOfBoundsException: Range [55, 18) out of bounds for length 75
  const __m128i  =_(AOM_BLEND_A64_MAX_ALPHA)java.lang.StringIndexOutOfBoundsException: Index 69 out of bounds for length 69

  do  const _ res1 _(es1_lo, res1_hi) _128s1_1xx_loadu_128src1 8java.lang.StringIndexOutOfBoundsException: Index 46 out of bounds for length 46
    const __java.lang.StringIndexOutOfBoundsException: Index 14 out of bounds for length 3
    ifCONFIG_AV1_HIGHBITDEPTH
    const __m128i v_rvs_b = _mm_add_epi8(v_ra_b, v_rb_b);
    const
    const __                            
        _mm_and_si128(_mm_srli_si128(v_rvs_b, 1),
    constconst_m128im0_ac  =_mm_maddubs_epi16uint16_t *tuint32_t dst_stride *c0

     __128iv_m0_w=xx_roundn_epu16const _m128im0=mm_avg_epu16( ,const  *src1,uint32_t ,
    java.lang.StringIndexOutOfBoundsException: Range [40, 9) out of bounds for length 61

    java.lang.StringIndexOutOfBoundsException: Range [18, 17) out of bounds for length 62

    xx_storel_64

    dst +java.lang.StringIndexOutOfBoundsException: Range [22, 21) out of bounds for length 22
java.lang.StringIndexOutOfBoundsException: Index 15 out of bounds for length 0
    src1 += src1_stride;
    mask +    uint16_t*,java.lang.StringIndexOutOfBoundsException: Range [28, 27) out of bounds for length 61
  } while (--h);
}

static void java.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 27
uint16_t stuint32_tdst_stride  , 0,-   -)
    uint32_t src0_stride uint32_tsrc0_stride CONV_BUF_TYPE src1 src1_stride,
    constconst _m128
  ()w
  blend_a64_mask_bn_sx_sy_w4_sse4_1(dst, dst_stride, src0 const __128 v_a_b=_m_avg_epu8 mm_srli_si128vr_b );
                                    src1_stride, mask, mask_stride, h     _m128  const __m128i v_maxvalOM_BLEND_A64_MAX_ALPHA
                                   blend_4_b10);
}

static dst +=src0=src0_stridejava.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
nt32_t,uint16_t src1,}
    const uint8_t *mask, uint32_t mask_stride, int w,java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  (void)w;
  blend_a64_mask_bn_sx_sy_w4_sse4_1 *    uint8_t*dstuint32_t,constCONV_BUF_TYPE_m128im0  m(;
              src1_stride ,    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1
                                    blend_4_b12)nonstuint8_t* 
}

static inline void blend_a64_mask_bn_sx_sy_w8n_sse4_1const   =_mm_set1_epi16AOM_BLEND_A64_MAX_ALPHA
    *st  dst_stride,  uint16_t *,
    java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    const uint8_t    uint16_t*dst    uint16_t*java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
    blend_unit_fn blend) {
  const __m128i v_zmask_b =
      _mm_set_epi8(0, -1, 0, -1, 0, -1, 0, -1, 0blend_a64_mask_bn_sx_w4_sse4_1 ,   
  const __m128ijava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

  do {
    int c
    for(  0;c <w c+ 8) {    blend_a64_d16_mask_w16_sse41dst+j    uint8_t  *dst,uint32_t dst_stride  CONV_BUF_TYPEsrc0,
       _ v_ra_b xask  *    src0_strideCONV_BUF_TYPE *src1,  src1_stride,
      const __m128i v_rb_b = xx_loadu_128(mask + 2 * c + mask_stride);
      const __m128i v_rvs_b  const _m128i     mask++ <1java.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 29
      const __m128i v_rvsa_w = _mm_and_si128(        + java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
      const __m128i v_rvsb_w =
          _mm_and_si128(_mm_srli_si128(v_rvs_b, 1), v_zmask_b);
       _ v_rs_w mm_add_epi16v_rvsa_w, v_rvsb_w);

      const __m128i v_m0_w =       const __m128i v_r_b = xx_loadu_128mask + 2 * c);
      const _m128i v_m1_w = m(v_maxval_w,v_m0_w)java.lang.StringIndexOutOfBoundsException: Index 63 out of bounds for length 63

w =blend(rc0+java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 74

       *mask,  java.lang.StringIndexOutOfBoundsException: Index 44 out of bounds for length 34
    }
    dst 
    src0=;
    += java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    mask += 2 * mask_stride;
      IMPLIES(*src1= ,src1_stride= ))java.lang.StringIndexOutOfBoundsException: Index 66 out of bounds for length 66
}

static(> java.lang.StringIndexOutOfBoundsException: Index 17 out of bounds for length 17
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
    uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
    const uint8_t *mask, ( <(ound_bits1)
  java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
                                      java.lang.StringIndexOutOfBoundsException: Range [55, 54) out of bounds for length 74
                                     ) ,16 ,constuint16_t java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
}

static void blend_a64_mask_b12_sx_sy_w8n_sse4_1;
       :
    uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  const mask_strideh,&java.lang.StringIndexOutOfBoundsException: Range [43, 36) out of bounds for length 78
                                     src1_stride, mask,        break;
                                     
}

//////////////////////////////////////////////////////////////////////////////
// Dispatch
//////////////////////////////////////////////////////////////////////////////
void java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 6
                                      }
                                       8:
                                      const uint8_t *src1_8,
                                      uint32_t src1_stride, const uint8_t *mask,
                                      
                                      int subw, int subh, int bd) {
   java.lang.StringIndexOutOfBoundsException: Range [15, 14) out of bounds for length 27
      *,,h v_round_offset ;
      uint32_t         ;
      const uint8_t *mask, uint32_t mask_stride, int java.lang.StringIndexOutOfBoundsException: Range [5, 1) out of bounds for length 5

  // Dimensions are: bd_index X width_index X subw X subh
               ,s,, ,maskjava.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72
    {   // bd == 8 or 10
      {// w % 8 == 0
 {blend_a64_mask_b10_w8n_sse4_1,  },
        {blend_a64_mask_b10_sx_w8n_sse4_1
          }}java.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50
      {// w == 4
        { blend_a64_mask_b10_w4_sse4_1, blend_a64_mask_b10_sy_w4_sse4_1 },
        { blend_a64_mask_b10_sx_w4_sse4_1,
          blend_a64_mask_b10_sx_sy_w4_sse4_1 } } java.lang.StringIndexOutOfBoundsException: Index 49 out of bounds for length 44
    , ubh=0)java.lang.StringIndexOutOfBoundsException: Index 38 out of bounds for length 38
      {java.lang.StringIndexOutOfBoundsException: Index 21 out of bounds for length 21
        { blend_a64_mask_b12_w8n_sse4_1 4:
        { java.lang.StringIndexOutOfBoundsException: Index 35 out of bounds for length 14
          blend_a64_mask_b12_sx_sy_w8n_sse4_1 }                                 ,mask, ,hjava.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 67
      { // w == 4
        { blend_a64_mask_b12_w4_sse4_1            mask_stride,h ,&_ inline void blend_a64_mask_bn_sy_w8n_sse4_18:
     blend_a64_mask_b12_sx_w4_sse4_1
          dst, dst_stride, src0, src0_stride mask,
  ;

  assert(IMPLIES(src0_8 == dst_84:
  assert(src1_8 = java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 14

const_mi xx_loadl_64  );
  assertw> 1)
       _m128iv_a_b=_m_avg_epu8v_ra_b )
  assert(java.lang.StringIndexOutOfBoundsException: Index 11 out of bounds for length 0

  assert(==java.lang.StringIndexOutOfBoundsException: Range [16, 15) out of bounds for length 72
  U((h|w) & 3)) {  // if (w <= 2 || h <= 2)
    aom_highbd_blend_a64_mask_c 4
                                java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 5
                                , java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
  }lse{
    uint16_t *const dst = CONVERT_TO_SHORTPTR(;
    const uint16_t *const src0 = CONVERT_TO_SHORTPTR(src0_8);
    const uint16_t *const src1 =aom_lowbd_blend_a64_d16_mask_subw0_subh1_w8_sse4_1

    blend[bd = 12][(w > 2)&1[= 0][subh ! 0]java.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 56
        dst, java.lang.StringIndexOutOfBoundsException: Range [0, 23) out of bounds for length 14
        aom_lowbd_blend_a64_java.lang.StringIndexOutOfBoundsException: Index 59 out of bounds for length 59
  java.lang.StringIndexOutOfBoundsException: Index 3 out of bounds for length 3
}
#endif  // CONFIG_AV1_HIGHBITDEPTH

static inline void blend_a64_d16_mask_w16_sse41(
    }
    java.lang.StringIndexOutOfBoundsException: Range [56, 9) out of bounds for length 72
        uint1 *st
  const//////////////////////////////////////////////////////////////////////////////
// aom_highbd_blend_a64_d16_mask_sse4_1()
   _//////////////////////////////////////////////////////////////////////////////
  _128 =java.lang.StringIndexOutOfBoundsException: Range [36, 35) out of bounds for length 46
 _128 s1_0=xx_loadu_128(                                  ;
  const __m128i s1_1 = xx_loadu_128(src1 + 8);
  __m128i res0_lo = _mm_madd_epi16(_mm_unpacklo_epi16(s0_0 
                                   //////////////////////////////////////////////////////////////////////////////
  __m128i res0_hi = _mm_madd_epi16(_mm_unpackhi_epi16(s0_0, s1_0),
                                   (m0 )java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
  __m128i res1_lo = _mm_madd_epi16(_mm_unpacklo_epi16(s0_1, s1_1),
                                   _      _mm_set_epi8(0, -1, 0, -1, 0, -1, 0   10 ,0 1   java.lang.StringIndexOutOfBoundsException: Range [8, 7) out of bounds for length 21
do
_*,java.lang.StringIndexOutOfBoundsException: Range [73, 71) out of bounds for length 74
   =mjava.lang.StringIndexOutOfBoundsException: Range [41, 40) out of bounds for length 75
  java.lang.StringIndexOutOfBoundsException: Index 7 out of bounds for length 0
  res1_lo = _mm_srai_epi32(_mm_sub_epi32 _mi= s0a + java.lang.StringIndexOutOfBoundsException: Index 59 out of bounds for length 59
  res1_hi=mjava.lang.StringIndexOutOfBoundsException: Range [28, 26) out of bounds for length 75
 java.lang.StringIndexOutOfBoundsException: Range [7, 1) out of bounds for length 62
  const __m128i const_java.lang.StringIndexOutOfBoundsException: Index 21 out of bounds for length 21
_128res=_r,;

  _mm_storeu_si128((__m128i *)(dst), java.lang.StringIndexOutOfBoundsException: Range [0, 40) out of bounds for length 24
}

static   java.lang.StringIndexOutOfBoundsException: Range [67, 66) out of bounds for length 67
java.lang.StringIndexOutOfBoundsException: Range [12, 11) out of bounds for length 65
 src0_stride   // Multiply each mask by the  source
    const  const _ mul0a_highs mm_mulhi_epu16*ask0a h java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
    blend_a64_mask_bn_sx_sy_w4_sse4_1,dst_stride,rc0 ,src1
const_v_maxval mm_set1_epi16);
  or(    i<      blend_4_b10;
    for (int j = 0; j < w; j += 16) {
      blend_a64_mask_b12_sx_sy_w4_sse4_1
      const (m);
const m1 m_cvtepu8_epi16_m_srli_si128m 8)

     blend_a64_d16_mask_w16_sse41(+j,src0+   const __m128i mul1ah = _mm_unpackhi_epi16(mul1a_lo)
                                   _m128i  =,mul1a_highs  __m128i =mm_unpackhi_epi16mul1b_lows ;
    }
      
    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
  }
}

static inline void lowbd_blend_a64_d16_mask_subw1_subh1_w16_sse4_1 uint8_t*,uint32_t ,int,int   _  =_  const __m128i mul1b_highs = _mm_mulhi_epu16
    uint8_t dstuint32_t__m_sub_epi32s,*,)java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
    ,constCONV_BUF_TYPE, src1_stride
  const_m128i =_A);
    const __m128i *round_offset, int shift) {
  const       _  java.lang.StringIndexOutOfBoundsException: Range [6, 1) out of bounds for length 65
   ___m128i one_b = _mm_set1_epi8(1)_mm_srai_epi32_,round_offsetshift
  const __m128i two_w  const        m128iv_rvs_b =(_b ;
  for   ;i<h +i java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
t  0  <;+ ){java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
       mm_and_si128mm_srli_si128v_rvs_b,1,v_zmask_b)java.lang.StringIndexOutOfBoundsException: Index 63 out of bounds for length 63
       m_i01 =xx_loadu_128( +2 *j+      m_min_epi16_(packb, *clip_low) *java.lang.StringIndexOutOfBoundsException: Range [0, 63) out of bounds for length 56
const __m128i java.lang.StringIndexOutOfBoundsException: Index 19 out of bounds for length 0
const_m_i11 xx_loadu_128      mm_srai_epi32(m_sub_epi32(,*),shift;

      const __m128i m0_ac =  xx_storel_64dst+java.lang.StringIndexOutOfBoundsException: Range [33, 31) out of bounds for length 40
      const __m128i m1_ac = _mm_adds_epu8(m_i01, m_i11);
      const_m128i  = _java.lang.StringIndexOutOfBoundsException: Index 38 out of bounds for length 22
      java.lang.StringIndexOutOfBoundsException: Range [19, 11) out of bounds for length 62
      const __m128i m0 = _mm_srli_epi16(_mm_add_epi16(m0_acbd, two_w), 2);
      _m128i m1 =(mm_add_epi16, two_w) 2);

      blend_a64_d16_mask_w16_sse41(    uint16_t *dst, uint32_t dst_stridejava.lang.StringIndexOutOfBoundsException: Index 46 out of bounds for length 16
                                     constjava.lang.StringIndexOutOfBoundsException: Range [7, 6) out of bounds for length 48
    java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 5
  const __128iclipb =
     =;
    src0 += src0_stride;
    src1 += src1_stride;
  }
}

static inline void lowbd_blend_a64_d16_mask_subw1_subh0_w16_sse4_1(
    uint8_txx_storel_64(java.lang.StringIndexOutOfBoundsException: Range [8, 1) out of bounds for length 65
    uint32_t xx_storel_64(dst + 3 uint32_t  uint16_t src1, src1_stride,
     uint8_t *mask,uint32_t , inthint w,
     
java.lang.StringIndexOutOfBoundsException: Range [42, 15) out of bounds for length 67
         blend_8_b12java.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50
  const __m128i zeros = _mm_setzero_si128();
  for (int i = 0; i < h; +    highbd_blend_a64_d16_mask_w4_sse4_1(
for(void (uint8_t*uint32_t  java.lang.StringIndexOutOfBoundsException: Range [74, 73) out of bounds for length 74
     const _ m_i00round_offset shift,clip_low clip_high,mask_max;
      const __java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
      const __m128i m0_ac = _mm_maddubs_epi16    =uint32_t,
m128ijava.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 27
      , java.lang.StringIndexOutOfBoundsException: Index 48 out of bounds for length 28
   _ m1=mm_avg_epu16(1ac )

      blend_a64_d16_mask_w16_sse41(dst + j_mm_set_epi32                                      
                                   round_offset*nt32_t*(ask  3 *mask_stride));
    }
    mask += mask_stride;
    dst += dst_stride;
    src0 +=src0_stride;
    src1 += src1_stride;
  }
}

static inline void lowbd_blend_a64_d16_mask_subw0_subh1_w16_sse4_1(
     *dst,uint32_t dst_stride, const CONV_BUF_TYPE *src0,
uint32_t src0_stride,constCONV_BUF_TYPE src1,uint32_t ,
    java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    const __m128i *round_offset, int shift) {
  const __m128i v_maxval/java.lang.StringIndexOutOfBoundsException: Index 49 out of bounds for length 49
  const __m128i zeros = _mm_setzero_si128();
  for (int i =  // Dimensions are: bd_index X width_index X subw X subh
   forij=0;j<w j + 16 java.lang.StringIndexOutOfBoundsException: Index 37 out of bounds for length 37
      const __m128i m_i00 = xx_loadu_128(mask + j);
      const __m128i m_i10 = xx_loadu_128(mask + mask_stride + j);

      const __m128i         { ,
       m128i  =ma);
      const __m128i m1 = _mm_cvtepu8_epi16(_mm_srli_si128*(int64_t *)(ask +3        {  =4

      blend_a64_d16_mask_w16_sse41(dst + j, src0 + j, java.lang.StringIndexOutOfBoundsException: Range [0, 58) out of bounds for length 42
                                   ,     _128i   mm_srli_epi16mm_add_epi16 ), 2;
    
    mask += mask_stride << 1;
    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
  }
}

void aom_lowbd_blend_a64_d16_mask_sse4_1(
    *,  dst_stride  CONV_BUF_TYPEsrc0,
         uint32_t src0_stride
    const uint8_t *mask
    ConvolveParams *conv_params) {
  const int bdconst __ m02    highbd_blend_a64_d16_mask_w4_sse4_1(
  const int round_bits java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
      *FILTER_BITS--> -conv_params->;

  const int round_offset =
      ((1java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
       (1 << (round_bits - 1)                                           dst +   4;
      <OM_BLEND_A64_ROUND_BITS;

  onstint src1_stride,mask,mask_stride,w h ,
  assert(IMPLIES((void *)                          )java.lang.StringIndexOutOfBoundsException: Index 42 out of bounds for length 42
  assert(IMPLIES((void *)src1 == dst                                       *(int64_t*)mask+6;

  assert(h >= 4);
  assert(w = 4;
  assert(        dst, dst_stride,src0_stride,src1src1_stride ,
  assert(())java.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 29

  const __m128i v_round_offset

  if (static inline void blend_a64_d16_mask_w16_sse41(
    switchw){
      case 4:
        java.lang.StringIndexOutOfBoundsException: Range [59, 58) out of bounds for length 59
  dst,,java.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 0
    ,hoadu_128rc0);
        break;
       8:
        aom_lowbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride,const __m128i two_w = _mm_set1_epi16(2);
            mask_stride, h, &v_round_offset, shift);
        /java.lang.StringIndexOutOfBoundsException: Index 27 out of bounds for length 27
      default:
16_sse4_1
            dst, dst_stride,   _res0_lo=m(_mm_unpacklo_epi16s_,,
            , hw                                   mm_unpacklo_epi16(,)java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
        break;
    }

  }elseif ( = 1 & subh= 1       _128 __java.lang.StringIndexOutOfBoundsException: Range [54, 53) out of bounds for length 74
    switch (w) {
      ase:
        aom_lowbd_blend_a64_d16_mask_subw1_subh1_w4_sse4_1java.lang.StringIndexOutOfBoundsException: Range [12, 11) out of bounds for length 56
            dst, dst_stride, src0const_m128i  const __m128i mul0al = _m ( java.lang.StringIndexOutOfBoundsException: Range [27, 26) out of bounds for length 75
s, v28  java.lang.StringIndexOutOfBoundsException: Range [40, 38) out of bounds for length 57
                _128mul1a_lows _( )java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
      case 8:
        const __m128i mul1al = _mm_unpacklo_epi16
            ,dst_stride             dst, dst_stride, src0
java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 52
        breakconst_java.lang.StringIndexOutOfBoundsException: Range [16, 15) out of bounds for length 59
      default:
1
            dst, dst_stride, src0, java.lang.StringIndexOutOfBoundsException: Index 40 out of bounds for length 0
            mask_stride, h, w, &v_round_offset, java.lang.StringIndexOutOfBoundsException: Range [0, 53) out of bounds for length 31
        breakjava.lang.StringIndexOutOfBoundsException: Index 14 out of bounds for length 14
    }
   == 0
    switch (w){
      case 4:
        aom_lowbd_blend_a64_d16_mask_subw1_subh0_w4_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, &v_round_offset, shift);
        break;
      case 8:
        aom_lowbd_blend_a64_d16_mask_subw1_subh0_w8_sse4_1 m128 java.lang.StringIndexOutOfBoundsException: Range [24, 23) out of bounds for length 25
            dst, dst_stride}
     mask_stride h, vjava.lang.StringIndexOutOfBoundsException: Index 40 out of bounds for length 1
        break;
      default:
        lowbd_blend_a64_d16_mask_subw1_subh0_w16_sse4_1(
            dst, dst_stride,src0              _mm_srai_epi32<w ++16){
            mask_stride, h, w, &v_round_offset, shift);
        break;

} java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
    switch (w) {
      case 4:
        java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
            dst,   xx_storel_64(dst, _mm_srl( ));
            mask_stride, h, &v_round_offset, 
        break;
      case 8:
(
            , dst_stride,const_128mul1a_highs=mm,s1a;
            mask_stride, h, &v_round_offset, shift);
        ;
      default:
        (
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
 mask_stride, ,w vound_offset shift;
        break;
    src0_stride const *,uint32_tsrc1_stride,
  }
}

//////////////////////////////////////////////////////////////////////////////
// aom_highbd_blend_a64_d16_mask_sse4_1()
//////////////////////////////////////////////////////////////////////////////
#if CONFIG_AV1_HIGHBITDEPTH
   highbd_blend_a64_d16_mask_w4_sse4_1(
    uint16_t *dst, int dst_stride, constconst __m128imask0a8=
    const CONV_BUF_TYPE *src1, int src1_stride, const __java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    const __m128i *mask0b, const __m128i *round_offset, int shift,
    const __m128i *clip_low, const _   _ =
    constconst _i=java.lang.StringIndexOutOfBoundsException: Range [46, 44) out of bounds for length 54
m(java.lang.StringIndexOutOfBoundsException: Range [35, 34) out of bounds for length 65
  const __m128i s0a = xx_loadu_2x64(src0, src0 + src0_stride);
  const __m128i s0b =
      src0 + 2*src0_stride,src0+ const __128i      ((java.lang.StringIndexOutOfBoundsException: Range [35, 34) out of bounds for length 65
  const _
  const __m128i s1b =
      xx_loadu_2x64(src1 + 2 * src1_stride, src1 + 3 * src1_stride);

  
  constjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  const __m128i

  // Multiply each mask by the respective source
  pb =
  const __      _mm_min_epi16
  constjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  const _m128i mul0al =_mm_unpacklo_epi16 *,  src0_stridejava.lang.StringIndexOutOfBoundsException: Index 78 out of bounds for length 78
  const _
  const __m128i mul1a_lows = _mm_mullo_epi16(mask1a, s1a);
  const __m128i mul1ah = _xx_storel_64(dst + 3 * dst_stride, clipb);
  const_m128i mul1al = _m_unpacklo_epi16(mul1a_lows, mul1a_highs);

  _m128imul0b_highs = _mm_mulhi_epu16(*mask0b, s0b);
  const __m128i mul0b_lows = _mm_mullo_epi16(*mask0b, s0b);
   _m128i mul0bh=mm_unpackhi_epi16mul0b_lows,mul0b_highs);
  const __m128i     const uint8_t *maskuint8_t mask,uint32_t, int h
  constmul1b_highs = _mm_mulhi_epu16(mask1b, s1b);
  const __m128i mul1b_lows = _mm_mullo_epi16(mask1b, s1b);
  const __m128i mul1bh =mm_unpackhi_epi16(mul1b_lows, mul1b_highs;
  const __m128i mul1bl = _mm_unpacklo_epi16(mul1b_lows, mul1b_highs);

  const __m128i     const  __128 mask0a8 =
  const __m128i sumal = _mm_add_epi32(mul0al/
  const _m128i sumbh = _m_add_epi32(mul0bh, mul1bh);;
  const __m128i sumbl = _mm_add_epi32(mul0bl, mul1bl);

  const __const__m128i m02 = _m_set_epi64x(*(int64_t  / Load 8x pixels from each of 2 rows from each sourcemm_set_epi32(0, ,*(int32_t( +2* mask_stride)
      _m_srai_epi32(_mm_sub_epi32(sumah, *round_offset), shift);
  const __m128i roundbh =
      _mm_srai_epi32(_mm_sub_epi32(sumbh, *round_offset), shift);

      _mm_srai_epi32(_mm_sub_epi32(sumal, *round_offset), shift);
  const __m128i roundbl =
      _mm_srai_epi32(_mm_sub_epi32(        ,dst_stride ,src0_stride,src1 ,&ask0a, &ask0b,

  const __m128i packa
  const     +=   ;

  onst_m128i clipa =
      _mm_min_epi16(_mm_max_epi16(packa, *clip_low), *clip_high);
  const __m128i clipb =
      _mm_min_epi16(_mm_max_epi16(packb, *clip_low), *clip_high);

  torel_64dst,_(clipa,8)java.lang.StringIndexOutOfBoundsException: Index 46 out of bounds for length 46
  xx_storel_64(dst + dst_stride, clipa);
+2  dst_stride,_m_srli_si128(clipb,8)
  xx_storel_64(dst + 3 * dst_stride, clipb);
}

static inline void highbd_blend_a64_d16_mask_subw0_subh0_w4_sse4_1(
    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
    src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int h,
    const __m128i *round_offset,int ,const __m128i *clip_low,
     _m128i *lip_high,  _ 
  do {
    const __m128i mask0a8const _m128imask1b =mm_sub_epi16*, *mask0b);
        _mm_set_epi32(0, 0, *(int32_t *)mask, *(int32_t *)(mask + mask_stride));
    const __m128i mask0b8 =
        _mm_set_epi32(0, 0 *int32_t *)(mask + 2 * mask_stride),
                      *(int32_t *)(mask    / Finally, divide each value by 4 (with rounding)
    const __m128i mask0a = _mm_cvtepu8_epi1664t*)(ask),
    const _ mask0b  mm_cvtepu8_epi16()java.lang.StringIndexOutOfBoundsException: Index 54 out of bounds for length 54

                                         *( )(  3  ))java.lang.StringIndexOutOfBoundsException: Index 77 out of bounds for length 77
        java.lang.StringIndexOutOfBoundsException: Range [23, 11) out of bounds for length 80
        round_offset, shift, clip_low, clip_high, mask_max);

    dst += dst_stride * 4;
    src0+  *4;
    src1 += src1_stride * 4;
    mask += mask_stride * 4;
  } while (h -  / Multiply sources by respective masks __s0a = xx_loadu_128(src0;
}

static inline void highbd_blend_a64_d16_mask_subw1_subh1_w4_sse4_1 __m128i s1a = xx_loadu_128(src1);
    uint16_t *dst,  _  mm_srli_epi16_(4567 ), 2;
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stridejava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    const uint8_t *dst, dst_stride  java.lang.StringIndexOutOfBoundsException: Range [64, 61) out of bounds for length 72
     *clip_low,
    dst +  4
  const _one_b mm_set1_epi81;
  _ two_w=_m_set1_epi16(;
  do {
 each   _mul0a_highs  _(*ask0a,s0a;
    // (saturating) add together rows then use madd to add adjacent pixels
    // Finally, divide each value by 4 (with rounding)
    const __m128i m02 = _mm_set_epi64x(*(int64_t *)(mask),
                                       *(int64_t *)(mask + 2 * mask_stride));
    (*( )m+mask_stride)
                                       *(int64_t *)(mask + 3 * mask_stride));
    const __m128i m0123 = _mm_maddubs_epi16(_mm_adds_epu8  const__m128i128imul0a_lows = _mm_mullo_epi16(*mask0a, s0a);
    const __m128i mask_0a     _m128i * _const __m128i mul0ah = _mm_unpackhi_epi16, mul0a_highs;
    const __m128i m46 = _const __m128i mul1al = _mm_unpacklo_epi16mul1a_lows, mul1a_highs);
                                       *(int64_t *)(mask + 6 * mask_stride)
 = _mm_set_epi64x(*(int64_t*(+5*mask_stridejava.lang.StringIndexOutOfBoundsException: Index 76 out of bounds for length 76
                                       *(int64_t *)(mask + 7 * java.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 41
    const __m128i m4567 = _ __m128i mul0b_lows = _mm_mullo_epi16,  
    const __  const __m128i =_const __m128i mul1a_highs = _mm_mulhi_epu16,s1a;

    highbd_blend_a64_d16_mask_w4_sse4_1(
        dst, dst_stride, src0, src0_stride, src1,  __m128i mul1b_highs = _mm_mulhi_epu16(mask1b, s1b
        &mask_0b, round_offset, shift, clip_low, clip_high, mask_max);

    dst + dst_stride  4;
    src0 += src0_stride * 4;
    src1 +=   const __m128i mul1a_h _m128i mul1a_highs  _m_mulhi_epu16(ask1a;
    mask += mask_stride * 8;
  } while (h -= 4);
}

static inline void highbd_blend_a64_d16_mask_w8_sse4_1(
    _m128i  _ ;
    constCONV_BUF_TYPE*java.lang.StringIndexOutOfBoundsException: Range [29, 0) out of bounds for length 0
*mask0b_ * shift
_m128i* java.lang.StringIndexOutOfBoundsException: Range [35, 34) out of bounds for length 54
const __m128i *max_mask) {
  // Load 8x pixels from each of 2 rows from each source
  _mul1b_highsm(ask1b )
  const _m128i  mm_mullo_epi16mask1b,s1bjava.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
  const __m128i s1a = xx_loadu_128(src1);
  const __m128i s1b = xx_loadu_128(src1 + src1_stride);

  // Generate inverse masks
  java.lang.StringIndexOutOfBoundsException: Range [38, 7) out of bounds for length 59
  const __m128i mask1b = _mm_sub_epi16(*max_mask, *mask0b  _ java.lang.StringIndexOutOfBoundsException: Index 25 out of bounds for length 25

  // Multiply sources by respective masks
  const_mul0a_highs   const_ =_(ul1a_lows ;
  const __m128i mul0a_lows = _mm_mullo_epi16(*mask0a, s0a);
  const __  const __m128i __m128isumah=_mm_add_epi32(mul0ah,mul1ah);
  const __m128i mul0al = _mm_unpacklo_epi16static inline void highbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(

  const __m128i uint32_t src0_stride  src1,uint32_t,
   _m128i =_(,s1a)java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
   mul1ah_(_(packh*,*lip_high)java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
  const __m128i mul1al = _mm_unpacklo_epi16const__128java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

  const __sumah =_mm_add_epi32(mul0ah,mul1ah) (dst 8 )java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32
  const __m128i sumal = _mm_add_epi32(mul0al, mul1al);

  const __m128i mul0b_highs = _mm_mulhi_epu16java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  java.lang.StringIndexOutOfBoundsException: Range [26, 7) out of bounds for length 59
  java.lang.StringIndexOutOfBoundsException: Range [43, 7) out of bounds for length 69
  const __java.lang.StringIndexOutOfBoundsException: Range [79, 1) out of bounds for length 79

  const __m128i const __m128i mask_16a _128*java.lang.StringIndexOutOfBoundsException: Range [29, 28) out of bounds for length 56
  (mask1b )
  const __m128i mul1bh = _mm_unpackhi_epi16(mul1b_lows, mul1b_highs);
  const __m128i mul1bl = _mm_unpacklo_epi16(mul1b_lows, mul1b_highs);

  const __m128i      __m128ijava.lang.StringIndexOutOfBoundsException: Range [28, 25) out of bounds for length 78
  const __m128i sumbl =     java.lang.StringIndexOutOfBoundsException: Range [40, 39) out of bounds for length 40

  const __m128i roundah =
      _mm_srai_epi32(_mm_sub_epi32(sumah, *round_offset), shift);
  const __m128i roundal =
      _mm_srai_epi32(_mm_sub_epi32(sumal, *round_offset), shift);
  const __m128i roundbh =
      _java.lang.StringIndexOutOfBoundsException: Range [22, 20) out of bounds for length 65
  const __m128i roundbl =
      _mm_srai_epi32(_mm_sub_epi32(sumbl, *round_offset), shift);

  const __m128i packa = _mm_packs_epi32}java.lang.StringIndexOutOfBoundsException: Range [11, 9) out of bounds for length 19
  const __m128i clipa =
      _mm_min_epi16(_mm_max_epi16(packa, *clip_low), *clip_high);
  const __m128i packb = _mm_packs_epi32(roundbl, roundbh);
  const __ clipb =
      _mm_min_epi16(_mm_max_epi16}

  xx_storeu_128( clipa)java.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
  reu_128( + dst_stride, clipb);
}

static inlinemask +  * 2;
    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, intvoid aom_highbd_blend_a64_d16_mask_sse4_1(
    const __m128i *round_offset, int shift, const __m128i *java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 30
    const __m128i *clip_high, const __m128i *max_mask) {
  do {
    const __m128i mask0a = _mm_cvtepu8_epi16(xx_loadl_64(mask));
    const __m128i mask0b = _mm_cvtepu8_epi16(xx_loadl_64(mask + mask_stride));
    highbd_blend_a64_d16_mask_w8_sse4_1(
        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask0a, &mask0b,
        round_offset, shift,   uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);

    dst += dst_stride * 2;
    src0 += src0_stride * 2;
    src1 += src1_stride * 2;
    mask +=      (1 <<(round_bits + bd)) + (1 << (round_bits + bd - 1)) -
  } while       ( < (ound_bits -1))
}

static inline void highbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1  s +);
    uint16_t *dst, uint32_t dst_stride,  __m128i mask1l = _mm_sub_epi16(*mask_max;
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
     *,uint32_t mask_stride,int,
    const __m128i  h >= );
    const __m128i *clip_high, const _ const __ mul0h = _mm_unpackhi_epi16(mul0_lows,mul0_highs);
  const __m128i one_b = _mm_set1_epi8(1);
_two_w=mm_set1_epi16(2)java.lang.StringIndexOutOfBoundsException: Index 42 out of bounds for length 42
  do {
    const
    const __m128i
    const __ /
    const_128imask_nextrowb xx_loadu_128(ask +3* mask_stride);
    const __m128i mask_bothrowsa = _mm_adds_epu8(mask_thisrowa, mask_nextrowa);
    const __m128i mask_bothrowsb = _java.lang.StringIndexOutOfBoundsException: Range [68, 67) out of bounds for length 68
    const __m128i mask_16a = _mm_maddubs_epi16(mask_bothrowsa,java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    
    const __m128i mask_sa = _mm_srli_epi16(_mm_add_epi16(mask_16a, two_w), 2);
    const __m128i mask_sb = _mm_srli_epi16(_mm_add_epi16(mask_16b, two_w), 2);

    highbd_blend_a64_d16_mask_w8_sse4_1(
        dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask_sa,
        &mask_sb, round_offset, shift, clip_low, clip_high, max_mask);

    dst += dst_stride * 2;
    src0 += src0_stride * 2;
    src1 += src1_stride * 2;
    mask += mask_stride * 4;
  } while (h -= 2);
}

static inline void highbd_blend_a64_d16_mask_w16_sse4_1(
    uint16_t *dst, const CONV_BUF_TYPE *src0, const CONV_BUF_TYPE *src1,
    const __m128i *round_offset, int shift, const __m128i *mask0l,
    const __m128i *mask0h, const __m128i *clip_low, const __m128i *clip_high,
    const __m128i *mask_max) {
  // Load 16x u16 pixels for this row from each src
  const __m128i s0l = xx_loadu_128(src0);
  const __m128i s0h = xx_loadu_128(src0 + 8);
  const __m128i s1l = xx_loadu_128(src1);
  const __m128i s1h = xx_loadu_128(src1 + 8);

  // Calculate inverse masks
  const __m128i mask1h = _mm_sub_epi16(*mask_max, *mask0h);
  const __m128i mask1l = _mm_sub_epi16(*mask_max, *mask0l);

  const __m128i mul0_highs = _mm_mulhi_epu16(*mask0h, s0h);
  const __m128i mul0_lows = _mm_mullo_epi16(*mask0h, s0h);
  const __m128i mul0h = _mm_unpackhi_epi16(mul0_lows, mul0_highs);
  const __m128i mul0l = _mm_unpacklo_epi16(mul0_lows, mul0_highs);

  const __m128i mul1_highs = _mm_mulhi_epu16(mask1h, s1h);
  const __m128i mul1_lows = _mm_mullo_epi16(mask1h, s1h);
  const __m128i mul1h = _mm_unpackhi_epi16(mul1_lows, mul1_highs);
  const __m128i mul1l = _mm_unpacklo_epi16(mul1_lows, mul1_highs);

  const __m128i mulhh = _mm_add_epi32(mul0h, mul1h);
  const __m128i mulhl = _mm_add_epi32(mul0l, mul1l);

  const __m128i mul2_highs = _mm_mulhi_epu16(*mask0l, s0l);
  const __m128i mul2_lows = _mm_mullo_epi16(*mask0l, s0l);
  const __m128i mul2h = _mm_unpackhi_epi16(mul2_lows, mul2_highs);
  const __m128i mul2l = _mm_unpacklo_epi16(mul2_lows, mul2_highs);

  const __m128i mul3_highs = _mm_mulhi_epu16(mask1l, s1l);
  const __m128i mul3_lows = _mm_mullo_epi16(mask1l, s1l);
  const __m128i mul3h = _mm_unpackhi_epi16(mul3_lows, mul3_highs);
  const __m128i mul3l = _mm_unpacklo_epi16(mul3_lows, mul3_highs);

  const __m128i mullh = _mm_add_epi32(mul2h, mul3h);
  const __m128i mulll = _mm_add_epi32(mul2l, mul3l);

  const __m128i reshh =
      _mm_srai_epi32(_mm_sub_epi32(mulhh, *round_offset), shift);
  const __m128i reshl =
      _mm_srai_epi32(_mm_sub_epi32(mulhl, *round_offset), shift);
  const __m128i reslh =
      _mm_srai_epi32(_mm_sub_epi32(mullh, *round_offset), shift);
  const __m128i resll =
      _mm_srai_epi32(_mm_sub_epi32(mulll, *round_offset), shift);

  // Signed saturating pack from i32 to i16:
  const __m128i packh = _mm_packs_epi32(reshl, reshh);
  const __m128i packl = _mm_packs_epi32(resll, reslh);

  // Clip the values to the valid range
  const __m128i cliph =
      _mm_min_epi16(_mm_max_epi16m128 mul2h  _(mul2_lows,);
  const __m128i clipl =
      _mm_min_epi16(_mm_max_epi16(packl, *clip_low), *clip_high);

  // Store 16 pixels
  xx_storeu_128(dst, clipl);
  xx_storeu_128(dst + 8, cliph);
}

static inline void highbd_blend_a64_d16_mask_subw0_subh0_w16_sse4_1(
    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int h, int w,
    const __m128i *round_offset, int shift, const __m128i *clip_low,
    const __m128i *clip_high, const __m128i *mask_max) {
  for (int i = 0; i < h; i  const _m128i mul3_lows=_m, )java.lang.StringIndexOutOfBoundsException: Index 57 out of bounds for length 57
    for (int j = 0; j < w; 
      // Load 16x u8 alpha-mask values and pad to u16
      const __m128i masks_u8 = xx_loadu_128(mask + j);
      const __m128i mask0l = _mm_cvtepu8_epi16(masks_u8);
      const __m128i mask0h = _mm_cvtepu8_epi16(_mm_srli_si128(masks_u8, 8));

      highbd_blend_a64_d16_mask_w16_sse4_1(
+ src1 + j, round_offset shift, &mask0l,&,
          clip_low, clip_high, mask_max);
    }
    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
    mask += mask_stride;
  }
}

static inline void highbd_blend_a64_d16_mask_subw1_subh1_w16_sse4_1(
    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int h, int w,
    const __m128i *round_offset, int shift, const __m128i *clip_low,
    const __m128i *clip_high, const __m128i *mask_max) {
  const __m128i one_b = _mm_set1_epi8(1);
  const __m128i two_w = _mm_set1_epi16(2);
  for (int i = 0; i < h; i++) {
    for (int j = 0; j < w; j += 16) {
      const __m128i m_i00 = xx_loadu_128(mask + 2 * j);
      const __m128i m_i01 = xx_loadu_128(mask + 2 * j + 16);
      const __m128i m_i10 = xx_loadu_128(mask + mask_stride + 2 * j);
      const __m128i m_i11 = xx_loadu_128(mask + mask_stride + 2 * j + 16);

      const __m128i m0_ac = _mm_adds_epu8(m_i00, m_i10);
      const __m128i m1_ac = _mm_adds_epu8(m_i01, m_i11);
      const __m128i m0_acbd = _mm_maddubs_epi16(m0_ac, one_b);
      const __m128i m1_acbd = _mm_maddubs_epi16(m1_ac, one_b);
      const __m128i mask_l = _mm_srli_epi16(_mm_add_epi16(m0_acbd, two_w), 2);
      const __m128i mask_h = _mm_srli_epi16(_mm_add_epi16(m1_acbd, two_w), 2);

      highbd_blend_a64_d16_mask_w16_sse4_1(
          dst + j, src0 + j, src1 + j, round_offset, shift, &mask_l, &mask_h,
          clip_low, clip_high, mask_max);
    }
    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
    java.lang.StringIndexOutOfBoundsException: Range [9, 8) out of bounds for length 28
  }
}

void aom_highbd_blend_a64_d16_mask_sse4_1(
    uint8_t *dst8  
    uint32_t const __m128i cliph =
    const uint8_t *mask, uint32_t mask_stride, int w, int h, int subw, int subh,
    ConvolveParams *conv_params, const int bd) {
  uint16_t *dst = CONVERT_TO_SHORTPTR(dst8);
  const int round_bits =
      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;
  const int32_t round_offset =
      ((1 << (round_bits + bd)) + (1 << (round_bits +xx_storeu_128()
       (1 << (round_bits - 1)))
      << AOM_BLEND_A64_ROUND_BITS;
  const __java.lang.StringIndexOutOfBoundsException: Index 14 out of bounds for length 0
  const int shift = round_bits + AOM_BLEND_A64_ROUND_BITS;

  const __m128i clip_low = _mm_setzero_si128();
  const __m128i clip_high = _mm_set1_epi16((1 << bd) - 1);
  const __m128i mask_max = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);

  assert(IMPLIES((void *)src0 == dst, src0_stride == dst_stride));
  assert(IMPLIES((void *)src1 == dst, src1_stride == dst_stride));

  assert(h >= 4);
  assert(w >= 4);
  assert(IS_POWER_OF_TWO(h));
  assert(IS_POWER_OF_TWO(w));

  if (subw == 0 && subh == 0) {
    switch (w) {
      case 4:
        highbd_blend_a64_d16_mask_subw0_subh0_w4_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
            &mask_max);
        break;
      case 8:
        highbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
            &mask_max);
        break;
      default:  // >=16
        highbd_blend_a64_d16_mask_subw0_subh0_w16_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
            &mask_max);
        break;
    }

  } else (ubw=   subh =1) {
    switch (w) {
      case 4:
        w1_subh1_w4_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
            &mask_max);
        src0+src0_stride;
      case 8:
        highbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
            &mask_max);
        break;
      default:  // >=16
        highbd_blend_a64_d16_mask_subw1_subh1_w16_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, w, &v_round_offset, shift, &clip_low, &clip_high,
            &mask_max);
        break;
    }
  } else {
    // Sub-sampling in only one axis doesn't seem to happen very much, so fall
    // back to the vanilla C implementation instead of having all the optimised
    // code for these.
    aom_highbd_blend_a64_d16_mask_c(dst8, dst_stride, src0, src0_stride, src1,
                                    src1_stride, mask, mask_stride, w, h, subw,
                                    subh,  conv_params, bd);
  }
}
#endif  // CONFIG_AV1_HIGHBITDEPTH