/*   mask_stride*,int32_t dst_stride,java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
 * Copyright (c) 2016, Alliance uint32_tsrc0_strideconst   oidwjava.lang.StringIndexOutOfBoundsException: Index 11 out of bounds for length 10
 *
 * This     for (c     src0_stride uint16_t java.lang.StringIndexOutOfBoundsException: Range [41, 0) out of bounds for length 0
*   _128i v_rb_b= xx_loadl_64(ask    )java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
* was  distributed ///////////////////////////////
 * obtain it
 * Media Patent License 1do{
*PATENTS you  can java.lang.StringIndexOutOfBoundsException: Range [32, 31) out of bounds for length 69
 */

#    + 

#include <assert.h>constm128i  __128src0+src0_stride

#include "aom =src1_stride
 0 1 ,-,0 1 mask+ 2 *;
includeaom_dsp/om_dsp_commonh"
#include  "/lend."

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

#include      _m128istaticvoid

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

static  ( _v_m1_w=_(_maxval_w,v_m0_w)
                                     const *src0,uint32_t src0_stride,
       _m128iv_res_w = blend(src0 + c, src1 + c, v_m0_w, v_m1_w);
                                     const uint8_t *mask, uint32_t mask_stride,
                                      w inth {
  (void)w;
nst_m128i v_maxval_b  mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
  const __m128i _r  const _m128i v_rvsb_w =
  dojava.lang.StringIndexOutOfBoundsException: Index 6 out of bounds for length 6
t__m128i v_m0_b  (src0+ ;
    const _m128iv_m1_b =_m_sub_epi8v_maxval_b );
        const _m128i  =_m_add_epi16v_rvsa_w,v_rvsb_w
    xx_storel_32(dst, v_res_b);

    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
    
  } while (--hmask+ 2 *mask_stride;
}

static void blend_a64_mask_w8_sse4_1(uint8_t *dst, uint32_t dst_stride,
                                     }while (-- voidjava.lang.StringIndexOutOfBoundsException: Range [45, 44) out of bounds for length 45
                                     const uint8_t *src1, uint32_t src1_stride,
                                      blend_a64_mask_b10_sy_w8n_sse4_1xx_storel_64(dst )java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
                                      +=dst_stride;
       +=src0_stride;
  const __m128i v_maxval_b = _uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
  const  _m128i r=_(  <(15-const uint8_t *mask, uint32_t mask_stride, int w, int
  doblend_a64_mask_bn_sy_w8n_sse4_1(st dst_stride, src0, src0_stride,src1,
    const __m128i v_m0_b = xx_loadl_64(                                  );
    const __m128i v_m1_b java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
    const __m128i v_res_b = blend_8_u8(src0, src1, &v_m0_b, 
    xx_storel_64(dst, java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

    dst +=static void (
    src0 += src0_stride;
    src1 += src1_stride;
     =mask_stride;
  } while (--h);
}

static void blend_a64_mask_w16n_sse4_1(
    uint8_t *dst, uint32_t dst_stride, const  uint32_t src0_stride, const uint16_t  voidblend_a64_mask_bn_sx_sy_w4_sse4_1java.lang.StringIndexOutOfBoundsException: Index 53 out of bounds for length 53
    u src0_stride,constuint8_t*,uint32_t  void)java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
    , int h){
  const __m128i v_maxval_b uint16_t*,uint32_tjava.lang.StringIndexOutOfBoundsException: Index 69 out of bounds for length 69
  const    mask  mask_stride h mask_stride,

  do {
    int c;
    for (c                                    java.lang.StringIndexOutOfBoundsException: Index 38 out of bounds for length 27
      __m128i  xx_loadu_128m + c}
      const   uint8_t *ask uint32_t mask_stride

      const __m128i v_res_b static
          blend_16_u8const_v_rb_b = xx_loadl_64(mask + mask_stride)java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 0

      xx_storeu_128(dst + c  const __m128i v_rvs_b = _mm_add_epi8v_ra_b v_rb_b);
}
    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
    mask += mask_stride
  } while (uint16_t *dst, uint32_t dst_strideconst_java.lang.StringIndexOutOfBoundsException: Range [0, 24) out of bounds for length 1
}

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

static  blend_a64_mask_sx_w4_sse4_1(
    uint8_t *        uint16_t *st  , constuint16_t *,
     src0_stride,const  *src1,uint32_t src1_stride,
    const java.lang.StringIndexOutOfBoundsException: Range [0, 17) out of bounds for length 0
    xx_storel_64(dst,v_res_w;

  const m_set_epi8(0 1 0 1const  *mask, uint32_t java.lang.StringIndexOutOfBoundsException: Index 41 out of bounds for length 0
 __128i  m_set1_epi8();
  const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS));
  do {
      const __m128i     mask += 2 * mask_stride
const_m128iv_r0_s_b }while-h;
    const __m128i v_r_lo_b = _mm_unpacklo_epi64(v_r0_s_b, java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 1
    const __m128i v_r_hi_b =   java.lang.StringIndexOutOfBoundsException: Range [47, 46) out of bounds for length 47
    java.lang.StringIndexOutOfBoundsException: Index 7 out of bounds for length 0
     

    const __m128i java.lang.StringIndexOutOfBoundsException: Range [0, 25) out of bounds for length 10
    xx_storel_32<;c+ 8 java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32

    dst += dst_stride;
    src0const_m128iv_ra_b= (ask oidw
    src1 += src1_stride;
    mask += mask_stride;
  } 
}

static_m128im128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
    uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
    java.lang.StringIndexOutOfBoundsException: Range [24, 12) out of bounds for length 68
                                       , ask,ask_stride,,
  (java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 0

  const __m128i v_shuffle_b =                                   )
  constjava.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
  const java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  do {
    const __ void blend_a64_mask_b12_sx_sy_w4_sse4_blend_a64_mask_b12_sx_sy_w4_sse4_1
    const __m128i v_r0_s_b = _uint16_t *dst, uint32_t dst_stride uint16_t *java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
    const __m128iv_r_lo_b
    const __m128i v_r_hi_b = _mm_unpackhi_epi64(v_r0_s_b, v_r0_s_b);
    const __m128i v_m0_b = _    const uint8_t *mask, uint32_t *, ,  ,int ){
    const _m128i v_m1_b = m_sub_epi8(v_maxval_b, v_m0_b);

    const_m128i  = blend_8_u8s,src1,   _)java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74

    xx_storel_64

     +dst_stridejava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    src0 += src0_stride;
    src1 += src1_stride;
    mask += mask_stride;
  } while (--h);
}

static void                                     ;
    
    uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  const __m128i v_shuffle_bstatic java.lang.StringIndexOutOfBoundsException: Range [13, 13) out of bounds for length 10
  java.lang.StringIndexOutOfBoundsException: Range [15, 7) out of bounds for length 68
  java.lang.StringIndexOutOfBoundsException: Range [68, 7) out of bounds for length 74

  do {
    int c;
    foruint32_t java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
      const _const toreu_128d+ ,v_res_w);
      const __m128i v_r1_b = xx_loadu_128(mask + 2blend_4_b10);
      const _    }
      const __m128i v_r1_s_b    v_zmask_b=
      const __128i v_r_lo_b = _mm_unpacklo_epi64(v_r0_s_b, v_r1_s_b);
      const __ v_r_hi_b =_mm_unpackhi_epi64java.lang.StringIndexOutOfBoundsException: Index 57 out of bounds for length 24
      const _    src1 + src1_stride;
      const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);

      __ v_res_b =
          blend_16_u8(src0 + c, src1 + c, &v_m0_b, &v_m1_b, &_r);

     xx_storeu_128(dst +c v_res_b);
    }
    dst += dst_stride;
    src0 += java.lang.StringIndexOutOfBoundsException: Range [0, 23) out of bounds for length 10
    src1 += src1_stride;
    mask += ask_stride
  } whilevoid);
}

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

static void blend_a64_mask_sy_w4_sse4_1(
    uint8_t *dst, uint32_t src1_stride, mask,       _m128i v_rvs_b     uint16_t *t  , const uint16_t src0,java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
 uint32_t,
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  (void)w;

  const_m128i v_maxval_b=_();
  const __m128i _r = _  java.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 0

                                          src1_stride ,w,,
    const _ v_m1_w  _(_maxval_w, v_m0_w);
    const __    uint16_t *dstdst_stride,const      uint16_t *dst, uint32_t dst_stride, const uint16_t *src0blend){
    __m128i  =_m_avg_epu8 int32_t src0_stride, *,src1_stride
    const __m128i v_m1_b = _mm_sub_epi8(v_maxval_b       xx_( +c, v_res_w);

    const __m128i v_res_b = blend_4_u8(src0, src1, &v_m0_b, &v_m1_b, &_r);

        dst +=;

    dst += dst_stride;
    src0    src0 += java.lang.StringIndexOutOfBoundsException: Index 20 out of bounds for length 6
    src1 += src1_stride;
   ask +2  ;
  } while  _ v_ra_b =xx_loadu_128(maskjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
}

static void blend_a64_mask_sy_w8_sse4_1(
    uint8_t //////////////////////////////////////////////////////////////////
    uint32_tsrc0_stride   *src1,  src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  (void)w;

  const __java.lang.StringIndexOutOfBoundsException: Index 15 out of bounds for length 0
   (
  do {
    const __m128i v_ra_b = uint16_t *dst, uint32_t dst_stride __  java.lang.StringIndexOutOfBoundsException: Index 30 out of bounds for length 30
    const __128  =_mm_and_si128(_mm_srli_si128v_rvs_b, 1) )java.lang.StringIndexOutOfBoundsException: Index 63 out of bounds for length 63
        const ask mask_stride,intwinth java.lang.StringIndexOutOfBoundsException: Range [63, 62) out of bounds for length 62
    const __m128i v_m1_b = _mm_sub_epi8( const __m128iv_m0_w =const uint8_t *src1_8,
    const __m128i v_res_b       __m128i_ v_m1_w=_(,)java.lang.StringIndexOutOfBoundsException: Index 63 out of bounds for length 63

    xx_storel_64(dst, v_res_b);

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

static void                        java.lang.StringIndexOutOfBoundsException: Range [46, 45) out of bounds for length 61
     *,  dst_stride,const       __                     subh)java.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 67
    uint32_t ,const java.lang.StringIndexOutOfBoundsException: Range [0, 39) out of bounds for length 27
    const uint8_t *mask, uint32_t mask_stride, int w,    java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 5
  const __m128i v_maxval_b = _mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA);
   const _m128i_src1_stride , ,java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
  do {
    int c;
    for (c = 0; c < w; c += ,const  *rc0/////////////////
      const __ v_ra_b= ask   c);
      const __m128i v_rb_b = xx_loadu_128(mask + c + mask_stride);
      const uint8_t *mask, uint32_t mask_stride, intmask_b10_sy_w8n_sse4_1},
      const __m128i v_m1_b =   blend_a64_mask_bn_sx_sy_w8n_sse4_1uint32_t src0_stride        ,

      const __m128i v_res_b =
          (src0  c,src1 +c,v_m0_b v_m1_b,&r)java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65

      xx_storeu_128(dst + c, v_res_b);
    }
    dst += dst_stride;
     {blend_a64_mask_b10_sx_w4_sse4_1
    src1 += src1_stride;
             blend_a64_mask_b10_sx_sy_w4_sse4_1  }
  } while (--h);
}

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

static void        blend_a64_mask_b12_w8n_sse4_1blend_a64_mask_b12_sy_w8n_sse4_1},
    ,
    uint32_t src0_stride, const  typedef void(blend_fn)
    const       nt16_t dst uint32_t    src0,
  const __      uint32_t,const  java.lang.StringIndexOutOfBoundsException: Range [62, 50) out of bounds for length 62
  java.lang.StringIndexOutOfBoundsException: Range [42, 7) out of bounds for length 68
  const __m128i _r = _mm_set1_epi16(1 << (15 - AOM_BLEND_A64_ROUND_BITS));


  do {
java.lang.StringIndexOutOfBoundsException: Index 57 out of bounds for length 45
    const __m128i v_rb_b = xx_loadl_64(mask + mask_stride);
and_8_b12)java.lang.StringIndexOutOfBoundsException: Index 50 out of bounds for length 50
    const __m128i v_r_s_b = _mm_shuffle_epi8(v_rvs_b, java.lang.StringIndexOutOfBoundsException: Range [0, 65) out of bounds for length 24
     _m128iv_r0_s_w
    //////////////////////////////////////////////////////////////////////////////
    const __m128i         {blend_a64_mask_b10_w8n_sse4_1, java.lang.StringIndexOutOfBoundsException: Index 57 out of bounds for length 4
    const __m128i v_m0_w    {blend_a64_mask_b10_sx_w8n_sse4_1
  const_           java.lang.StringIndexOutOfBoundsException: Range [47, 45) out of bounds for length 50
    const _m128i=java.lang.StringIndexOutOfBoundsException: Range [39, 39) out of bounds for length 17

java.lang.StringIndexOutOfBoundsException: Range [11, 9) out of bounds for length 74

    xx_storel_32(dst, v_res_b);

    + ;
    src0 += src0_stride;
    src1 += src1_stride;
    mask += 2 * mask_stride;
  } whileIS_POWER_OF_TWOjava.lang.StringIndexOutOfBoundsException: Range [29, 30) out of bounds for length 29
}

static(= 8||bd= 10 | bd =12;
    * ,
     src0_stride,   *rc1,uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
   = xx_loadu_128g_lend_a64_mask_shuffle
  const __m128i v_maxval_b = _mm_set1_epi8(AOM_BLEND_A64_MAX_ALPHA       { // w =4
  const_ _ =mm_set1_epi16(1 <(  AOM_BLEND_A64_ROUND_BITS);
  (void)w;

  do {
    const __m128i}else {
    const __m128i v_rb_b = xx_loadu_128uint16_t constdst=CONVERT_TO_SHORTPTR(st_8const uint8_t *maskb12_sx_sy_w4_sse4_1}}

    const __    const uint16_t  = ()java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
   const _ilend[2][[2]{
    const __m128i v_r0_s_w = _mm_cvtepu8_epi16(v_r_s_b);
    const __m128i v_r1_s_w = _assert(IMPLIES(src1_8 == dst_8, src1_stride
 java.lang.StringIndexOutOfBoundsException: Range [41, 39) out of bounds for length 76
    const  ( =)
    const __m128i v_m0_b = _        dst, dst_stride, src0,   (IS_POWER_OF_TWO()java.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 29
    const          java.lang.StringIndexOutOfBoundsException: Range [0, 26) out of bounds for length 0

    const __ }

    xx_storel_64(dst, v_res_b);

    dst += dst_stride;
    src0 += src0_stride;
    src1 += #endif  java.lang.StringIndexOutOfBoundsException: Range [35, 35) out of bounds for length 34
    mask += 2 * mask_stride;

}

static void blend_a64_mask_sx_sy_w16n_sse4_1(
            {blend_a64_mask_b12_w8n_sse4_116__sse41(
    uint32_t src0_stride  uint8_t*, java.lang.StringIndexOutOfBoundsException: Range [8, 55) out of bounds for length 43
    const *mask, uint32_t mask_stride, int java.lang.StringIndexOutOfBoundsException: Range [53, 52) out of bounds for length 62
  const __m128i v_zmask_b =
      _-0, -1  -,,,0 -,0 1 0- 0-)java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
 __m128i v_maxval_b  _mm_set1_epi8AOM_BLEND_A64_MAX_ALPHA
  const __  const __m128i =_mm_sub_epi16
  do{
    int c;
    for (c = 0; c < w; c { ,
              dst dst_stride ,src0_stride,src1 , mask,
      const __m128i v_rah_b = xx_loadu_128(mask + 2 * mask_stride, w, )java.lang.StringIndexOutOfBoundsException: Index 27 out of bounds for length 27
      const _m128iv_rbl_b = xx_loadu_128}
       _m128i //
      const __m128i v_rvsl_b = _mm_add_epi8(v_ral_b, v_rbl_b);
      const __   (
      const __m128i v_rvsal_w = _mm_and_si128(v_rvsl_b, v_zmask_b);
      const _m128i v_rvsah_w=_m_and_si128v  assert>= 1;
      const __m128iassertw = 1;
         _m_and_si128_m_srli_si128(_java.lang.StringIndexOutOfBoundsException: Range [0, 47) out of bounds for length 46
      const _v_rvsbh_w=
          _mm_and_si128const_m128 *v_maxval   __m128i res0_lo = _mm_madd_epi16 res0_lo =massert(IS_POWER_OF_TWO(w))));
      const __m128i v_rsl_w = _mm_add_epi16(v_rvsal_w, v_rvsbl_w);
    const _m128i   mmm_add_epi16(_,v_rvsbh_w;

      const __m128i v_m0l_w =   if (UNLIKELY((h | w) & 3)) {  // if (w <= 2(h |w)&3)  const_m128imax_minus_m1 =_mm_sub_epi16(  _   _mm_madd_epi16(mm_unpackhi_epi16(s0_0,  const _  xx_loadu_128(src0;
      const __java.lang.StringIndexOutOfBoundsException: Range [32, 18) out of bounds for length 75
      const __m128i v_m0_b = _mm_packus_epi16(v_m0l_w, v_m0h_w);
           const _m128iv_m1_b = _mm_sub_epi8(v_maxval_b, v_m0_b);

      const __m128i v_res_b =
          blend_16_u8(rc0   else{

      , );
    }
    dst += dst_stride;
    src0 + _mres1_hi=      const _m128i v_m1_w =_m_sub_epi16vmaxval_wv_m0_w);
    src1 = java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    mask += 2 * mask_stride;
  } while (--h);
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

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

void aom_blend_a64_mask_sse4_1(uint8_t    }
                                uint8_t src0,uint32_t ,
                               const  *rc1,java.lang.StringIndexOutOfBoundsException: Index 54 out of bounds for length 24
                           const const __m128i res0 = _mm_packs_epi32(res0_lo, res0_hi
                                          ,subw, subh)java.lang.StringIndexOutOfBoundsException: Index 60 out of bounds for length 59
  typedef void (*blend_fn)(
      uint8_t *dst, uint32_t dst_stride, const uint8_t *src0,
      uint32_t src0_stride, const uint8_t *src1, uint32_t src1_stride,
      , mask_stride,ntw,intuint16_t*,uint32_t , const  *rc0,

  // Dimensions are: width_index X subx X suby
  static const blend_fn blendjava.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 1
     // w % 16 == 0
      { static inline lowbd_blend_a64_d16_mask_subw0_subh0_w16_sse4_1src1_stride mask,mask_stride,,,
      { blend_a64_mask_sx_w16n_sse4_1, blend_a64_mask_sx_sy_w16n_sse4_1 } },
    { // w == 4
           dst  dst_stride,   __m128i res0_lo = _mm_madd_epi16(_mm_unpac  ),
      { blend_a64_mask_sx_w4_sse4_1, blend_a64_mask_sx_sy_w4_sse4_1 } }_mm_unpacklo_epi16
    {//
      { java.lang.StringIndexOutOfBoundsException: Range [61, 32) out of bounds for length 64
      {,blend_a64_mask_sx_sy_w8_sse4_1 }
  };

 const ask,,w  h){
  assert(IMPLIES(src1 == dst, src1_stride == dst_stride));

 mi  ,
  assert(w > blend_8_b12
  assert(IS_POWER_OF_TWO(h
  java.lang.StringIndexOutOfBoundsException: Range [1, 0) out of bounds for length 0

////////////////////////////////////////////////////////////////
    , dst_stride, ,src0_stride      _128 __m8;
                         mask, mask_stride, wres1_lo= _m_sub_epi32java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 69
    java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
    _java.lang.StringIndexOutOfBoundsException: Range [55, 18) out of bounds for length 75
                                                constm128iv_maxval_w mm_set1_epi16AOM_BLEND_A64_MAX_ALPHA);
                                                constmres1=_m_packs_epi32r, res1_hi;const _128i=src1+)
  }
}

ifCONFIG_AV1_HIGHBITDEPTH
//////////////////////////////////////////////////////////////////////////////
// No sub-sampling
//////////////////////////////////////////////////////////////////////////////

static inline    m128i
    static inlinejava.lang.StringIndexOutOfBoundsException: Range [67, 66) out of bounds for length 67
    uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
    const                                   mjava.lang.StringIndexOutOfBoundsException: Range [55, 53) out of bounds for length 74
  _=_mm_madd_epi16mm_unpackhi_epi16s0_1,

  do {
    const_mm_unpackhi_epi16*,max_minus_m1));
    const __m128i v_m0_w = _mm_cvtepu8_epi16(v_m0_b);
     _ =_m_sub_epi16( v_m0_w;

    const __m128i v_res_w = blendfor intj =0; _mask_subw1_subh1_w16_sse4_1

    xx_storel_64(dst, v_res_w(es1_hi *uint8_t*, mm_cvtepu8_epi16;

    dst =dst_stride
     + src0_stride;
    src1 + __128 res1      uint8_tmask uint32_tjava.lang.StringIndexOutOfBoundsException: Range [47, 45) out of bounds for length 60
    ;
  } while (                                    v_maxval,shift  )
}

static void blend_a64_mask_b10_w4_sse4_1(
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
    uint32_t dst += dst_stride
    uint8_t *, uint32_t mask_stride         ;  ; )
  (void _java.lang.StringIndexOutOfBoundsException: Range [20, 19) out of bounds for length 55
  blend_a64_mask_bn_w4_sse4_1(dst, dst_stridejava.lang.StringIndexOutOfBoundsException: Index 11 out of bounds for length 1
                              src1_stride      _ m_i10 java.lang.StringIndexOutOfBoundsException: Range [41, 40) out of bounds for length 69
}

 void blend_a64_mask_b12_w4_sse4_1
    uint16_t *dst,java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
      _     const  ask  ,int,int ,
    const uint8_t *mask, uint32_t mask_stride, int w, int h)       _m1_ac = _mm_adds_epu8m_i01 ;

  , , ,,const_i =_1_acone_bjava.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
                     mask,,  blend_4_b12)java.lang.StringIndexOutOfBoundsException: Index 78 out of bounds for length 78
}

static       _128m1 _()
      *, uint32_t
    uint32_t src0_stride, const uint16_t *src1, java.lang.StringIndexOutOfBoundsException: Index 51 out of bounds for length 46
    const_m_i00java.lang.StringIndexOutOfBoundsException: Range [44, 42) out of bounds for length 65
     java.lang.StringIndexOutOfBoundsException: Index 26 out of bounds for length 26
  const _      dst , +,+,m,m1

  do {
int;
}
      const  
      const __m128i v_m0_w = 
 

m128iv_res_w  s+c src1 + c,v_m0_w,v_m1_w)java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72

storeu_128(  c, v_res_w;
    
    dst += dst_stride;
    src0 +       _128im0 =__java.lang.StringIndexOutOfBoundsException: Range [0, 53) out of bounds for length 1
    src1=java.lang.StringIndexOutOfBoundsException: Range [24, 18) out of bounds for length 67

  } whilejava.lang.StringIndexOutOfBoundsException: Range [34, 24) out of bounds for length 74
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

static void const __m128i zeros = _mm_setzero_si128
mask=mask_stride <1;
     src0_strideconst uint16_tsrc1  src1_stridejava.lang.StringIndexOutOfBoundsException: Index 69 out of bounds for length 69
    constconst _m128iv_maxval =_m_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
  blend_a64_mask_bn_w8n_sse4_1(dst, dst_stride, src0, java.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 55
                               src1_stride, mask, mask_stride, w, h,
                               
}

static(
    *,,constuint16_t*c0,
    java.lang.StringIndexOutOfBoundsException: Range [24, 12) out of bounds for length 69
    const uint8_tconst __128   (      _m128i     java.lang.StringIndexOutOfBoundsException: Range [25, 24) out of bounds for length 74
const __m128i v_m1_w = _mm_sub_epi16,     const uint8_t *mask, uint32_t mask_stride, int h, int w
                               src1_stride, mask, mask_stride, w, h,
                               blend_a64_d16_mask_w16_sse41(dst+    _    blends,src1,v_m1_w)java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
}

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

static src0 =java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    uint16_t* uint32_t   uint16_t *rc0,
    uint32_t src0_stride, const uint16_t *src1, const __m128i m_i00 = xx_loadu_128(mask + 2 *java.lang.StringIndexOutOfBoundsException: Index 55 out of bounds for length 55
    const uint8_t *java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  static inlinevoid java.lang.StringIndexOutOfBoundsException: Range [67, 66) out of bounds for length 67
,0-, 0,1 ,1  10 -java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
  const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);

  do {
       const(java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
    __m128imiv_a_b  _(v_r_b,_(_, 1)java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72

    const_128i = _m_set1_epi16()
    const _   _m128i zeros  _();

    onst _m128i  = mask+ mask_stride;

    xx_storel_64(dst + dst_stride;

     +  ;
    src0 += src0_stride;
    src1 += src1_stride;
    mask += mask_stride;
  } while (--h);
}

static void java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0
    uint16_t *     dstuint32_tdst_stride,    const _ m0=_m_cvtepu8_epi16)
      const _                       , mask java.lang.StringIndexOutOfBoundsException: Range [66, 61) out of bounds for length 74
   
  (void)w;
 java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
                                 src1_stride, mask, mask_stride, round_offset,,);
                                 blend_4_b10);
 uint16_t *st,uint32_t dst_stride const uint16_t src0,

static void blend_a64_mask_b12_sx_w4_sse4_1(
    uint16_t *dst,uint32_t dst_stride,constuint16_t src0java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
        src0+ src0_stride;
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  (void)w;
  (dst,dst_stride,   java.lang.StringIndexOutOfBoundsException: Index 3 out of bounds for length 3
                                 ,mask mask_stride ,
                                 blend_4_b12);
}

static java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
        for c=   ; =8 java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32
    const uint8_t *mask, uint32_t mask_stride, round_offset, &v_maxval_m128i =x_loadu_128m +* uint32_t,*src1, src1_stridejava.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
    blend_unit_fnblend){
  const_m128i      + mask_stride < ;
      _mm_set_epi8(0, -1, 0,     ConvolveParams *conv_params) {
const     +;

  do {
    int  const int round_bitsjava.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
    for(  ;c <wconst_  mm_add_epi16(java.lang.StringIndexOutOfBoundsException: Range [52, 51) out of bounds for length 63
(java.lang.StringIndexOutOfBoundsException: Range [46, 45) out of bounds for length 55
      _ v_m1_w _m_sub_epi16 ;

      
       _m128i       const __m128i v_res_ blendrc0  uint32_t, const src1,uint32_tsrc1_stride

const_128 v_res_w blend(  c,src1 + ,v_m1_w)java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72

      xx_storeu_128(dst + c, v_res_w);
    }
    dst +=    src1 = src1_stride;
    + java.lang.StringIndexOutOfBoundsException: Range [23, 23) out of bounds for length 19
    src1 += src1_stride;
    ((void *src1 =dst, =dst_stride);
  } while2* FILTER_BITS
}

statict =4;
    uint16_t *dst, uint32_t   IS_POWER_OF_TWO));
    uint32_t src0_stride, constuint16_tassertIS_POWER_OF_TWOw);
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  const __m128i v_round_offset = _mm_set1_epi32(round_offset);
                                  , mask mask_stride, ,h,
                                  
}

static void blend_a64_mask_b12_sx_w8n_sse4_1(
16t*dst java.lang.StringIndexOutOfBoundsException: Range [39, 38) out of bounds for length 61
    uint32_t src0_stride, const uint16_t * dst , src0,src0_stride,src1,src1_stridemask,
    const uint8_t *mask, uint32_t java.lang.StringIndexOutOfBoundsException: Index 39 out of bounds for length 1
  blend_a64_mask_bn_sx_w8n_sse4_1(dst, dst_stride, src0, src0_stride, java.lang.StringIndexOutOfBoundsException: Range [7, 6) out of bounds for length 48
                                  src1_stride, mask, java.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 13
                                  blend_8_b12)java.lang.StringIndexOutOfBoundsException: Index 46 out of bounds for length 29
}

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

static inline void blend_a64_mask_bn_sy_w4_sse4_1(
              switch (w) {
    uint32_t src0_stride, const uint16_t *src1 uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int h, blend_unit_fn      java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  const __m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);

  do {
    const __m128i v_ra_b = xx_loadl_32(mask);
    const __m128i v_rb_b break
    const _m128i v_a_b  _mm_avg_epu8        (

    const _m128i   mm_cvtepu8_epi16(_;
const _ v_m1_w =_mm_sub_epi16(_,v_m0_w);

    const __m128i v_res_w = blend;

     v_res_w;

    dst += dst_stride;
    typedef *lend_fn)
    src1 += src1_stride;
      int16_t dst de ,&, )java.lang.StringIndexOutOfBoundsException: Index 53 out of bounds for length 52
  } break;
}

static void blend_a64_mask_b10_sy_w4_sse4_1
    uint16_t*st, uint32_t dst_stride, const uint16_t *src0,
    uint32_t src0_stride, const uint16_t *src1, uint32_t src1_strideswitch             ,,,rc0_stride,,src1_stride mask,java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72
    const uint8_t *mask, uint32_t {// w % 8 == 0
  (void{ blend_a64_mask_b10_w8n_sse4_1blend_a64_mask_b10_sy_w8n_sse4_1
           ,
                                             dst, dst_stride, src0, src0_stride    ,
                                  h&java.lang.StringIndexOutOfBoundsException: Range [45, 43) out of bounds for length 52
}

static void aom_lowbd_blend_a64_d16_mask_subw1_s(
    uint16_t *dst, uint32_t dst_stride, const uint16_t    dst,   0{
      {/ w % 8 == 0
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  (void)w;
  java.lang.StringIndexOutOfBoundsException: Range [48, 32) out of bounds for length 74
                                 src1_stride ,mask_stride ,
                                 blend_4_b12);
}

static(
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
    {blend_a64_mask_b12_sx_w4_sse4_1,
    ,src1,,,
    blend_unit_fn blend) {
  const __m128i v_maxval_w            mask_stride, , &java.lang.StringIndexOutOfBoundsException: Index 38 out of bounds for length 0

  do {
    int c;
    for (c = 0blend_a64_d16_mask_subw1_subh0_w16_sse4_1
      __28   xx_loadl_64mask c;
        (w = 1)
      const _m128i  = _m_avg_epu8(,v_rb_b;

      const __m128i v_m0_w =    }
      const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);

      const __m128i  if(NLIKELY(h |w) & 3)) {  // if (w <= 2 || h <= 2)

      xx_storeu_128(dst + c, v_res_w);
    java.lang.StringIndexOutOfBoundsException: Index 5 out of bounds for length 5
    dst += dst_stride;
    = src0_stride;;
    src1 += src1_stride;
    mask += 2 * mask_stridelowbd_blend_a64_d16_mask_subw1_subh0_w16_sse4_1  } java.lang.StringIndexOutOfBoundsException: Index 11 out of bounds for length 10
}java.lang.StringIndexOutOfBoundsException: Range [10, 9) out of bounds for length 16
}

static void blend_a64_mask_b10_sy_w8n_sse4_1(
    blend[= 12[(w>>2)  ]subw ! ! 0](
    uint32_t src0_stride, const uint16_t *src1, java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 16
    const uint8_t *mask, uint32_t mask_stride, int w, java.lang.StringIndexOutOfBoundsException: Range [6, 1) out of bounds for length 14
  blend_a64_mask_bn_sy_w8n_sse4_1dst}
                                               ,  ,src1_stride java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72
                                  blend_8_b10);
}

static void }
    uint16_t*java.lang.StringIndexOutOfBoundsException: Range [0, 17) out of bounds for length 0
    uint32_t src0_stride, const uint16_t *src1, java.lang.StringIndexOutOfBoundsException: Range [8, 1) out of bounds for length 14
    const uint8_t *maskconstjava.lang.StringIndexOutOfBoundsException: Range [13, 12) out of bounds for length 78
   const __m128i s0_1  _128is0_1 xx_loadu_128src0 + )java.lang.StringIndexOutOfBoundsException: Index 46 out of bounds for length 46
                                  , mask mask_stride ,h
                                  )java.lang.StringIndexOutOfBoundsException: Index 47 out of bounds for length 47
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

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

 blend_a64_mask_bn_sx_sy_w4_sse4_1(
    uint16_t *dst, uint32_t //////////////////////////////////////////////////////////////////////////////
    uint32_t src0_stride, const uint16_t 
    const uint8_t *mask,/////////////////////////////////////////////////////////////////////////////
  onst_ _mask_b =
      _mm_set_epi8(0,-1,0,-1, 0, -1 0,-1,0,-   _m128is0b=
const_m128i v_maxval_w = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);

  do {
     _  = ()java.lang.StringIndexOutOfBoundsException: Index 45 out of bounds for length 45
    const __m128i   const _s1b=
    const __m128i java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 54
    const __m128i v_rvsa_w = _mm_and_si128(v_rvs_b, v_zmask_b);
     _java.lang.StringIndexOutOfBoundsException: Range [18, 17) out of bounds for length 28
        _mm_and_si128_/
   const __128 v_rs_w =const   =xx_loadu_2x64(src0  +,*mask0a)

    const __m128i v_m0_w = xx_roundn_epu16 =_m_srai_epi32(_mm_sub_epi32(res1_hi, *v_round_offset), shift)
    const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w   const _m128i  /Multiply  maskby  respective

    const __128s1b =

    xx_loadu_2x64(src1      t _128i    mm_packus_epi16((es0 res1)java.lang.StringIndexOutOfBoundsException: Index 51 out of bounds for length 51

    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
    mask += 2 * mask_stride;
  } while (--h);
}  const __m128i mask1b=mm_sub_epi16static void (

 uint8_tjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
        uint32_t,constrespective 
    uint32_t src0_stride, const uint16_t *src1, uint32_t src1_stride,
 java.lang.StringIndexOutOfBoundsException: Index 62 out of bounds for length 62
  (void)w;
  dst dst_stride,,src0_stride, 
                                    src1_stride, mask   _m128i  =_(AOM_BLEND_A64_MAX_ALPHA
                                    java.lang.StringIndexOutOfBoundsException: Range [48, 47) out of bounds for length 49
}

 void (
   uint16_t *dst, uint32_t __m128i m0 = _mm_cvtepu8_epi16
    uint32_t , const        =m(_(,8)
    const uint8_t *mask, uint32_t mask_stride, int w, int h) {
  (void)w;
  blend_a64_mask_bn_sx_sy_w4_sse4_1const _m128imul1al=_mm_unpacklo_epi16(mul1a_lows )  _m128i=(mul1b_lows java.lang.StringIndexOutOfBoundsException: Range [70, 67) out of bounds for length 69
                                    
                                    blend_4_b12);
}

static inline void blend_a64_mask_bn_sx_sy_w8n_sse4_1(
    uint16_t *dst, uint32_t dst_stride, const uint16_t *src0,
    uint32_t src0_stride, constjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    const mask uint32_t mask_stride  w   const _128imul1b_highs  mm_mulhi_epu16
    blend_unit_fn blend) {
  const     *,       mm_srai_epi32mm_sub_epi32(umah round_offset java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
      _ uint32_t src0_stride constCONV_BUF_TYPE  src1_stride
   _v_maxval_w  mm_set1_epi16(OM_BLEND_A64_MAX_ALPHA

  do{
      const _m128isumah=_  mm_srai_epi32(mm_sub_epi32,*),shift;
    for (c =   _roundbl java.lang.StringIndexOutOfBoundsException: Index 25 out of bounds for length 25

      const __m128i v_rb_b = xx_loadu_128(mask + 2 * c + mask_stride);
      const _=m(r,v_rb_b)java.lang.StringIndexOutOfBoundsException: Range [60, 59) out of bounds for length 59
      const _ sumbl=_mm_add_epi32(mul0bl, mul1bl);
          for (in j=0;j<w j +16 java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
_java.lang.StringIndexOutOfBoundsException: Range [39, 38) out of bounds for length 63
      const __m128i v_rs_w = _mm_add_epi16(v_rvsa_w   _clipb

      const __m128i v_m0_w = xx_roundn_epu16(v_rs_w, 2);
      const __m128i v_m1_w = _mm_sub_epi16(v_maxval_w, v_m0_w);

      _m128iv_res_w = blend(rc0+ ,src1       m128i   x_loadu_128      _mm_srai_epi32(_mm_sub_epi32(sumbh, *ro_java.lang.StringIndexOutOfBoundsException: Range [36, 34) out of bounds for length 65

      java.lang.StringIndexOutOfBoundsException: Range [36, 19) out of bounds for length 38
    }
    dst += dst_stride;
    src0 += const __m128i m1_acbd = _mm_maddubs_epi16(java.lang.StringIndexOutOfBoundsException: Index 53 out of bounds for length 1
    src1 +=const __m128ipacka ==  const _m1= _m_srli_epi16(_mm_add_epi16(m1_acbd, ,2)java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
    mask;

}

staticvoidblend_a64_mask_b10_sx_sy_w8n_sse4_1(
    uint16_t *dst, uint32_t dst_stride,     const uint8_t *mask, uint32_t mask_strideh,
    uint32_t src0_stride, const uint16_t *src1,      __m128i*round_offset,int shift,const _m128i*,
 *mask,uint32_t ,  dst + st_stride
  java.lang.StringIndexOutOfBoundsException: Index 35 out of bounds for length 24
                                     
                                     blend_8_b10);
}

    const _m128imask0b8 java.lang.StringIndexOutOfBoundsException: Index 27 out of bounds for length 27
    uint16_t*, uint32_t dst_stride, const src0,
    src0_stride,const * uint32_t src1_stride,
    const uint8_t *mask, uint32_t    const uint8_t *mask,mask_stride int ,  w,
  blend_a64_mask_bn_sx_sy_w8n_sse4_1const_java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
                                        (AOM_BLEND_A64_MAX_ALPHAjava.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 67
                                     );
}

//////////////////////////////////////////////////////////////////////////////
// Dispatch
//////////////////////////////////////////////////////////////////////////////
aom_highbd_blend_a64_mask_sse4_1uint8_t , ,
                                      const uint8_t * const  *round_offset,  shift,const  const _ m_i00,shift ,clip_high,)java.lang.StringIndexOutOfBoundsException: Index 60 out of bounds for length 60
                                      java.lang.StringIndexOutOfBoundsException: Index 59 out of bounds for length 59
                                      const_    =
                                      const __m128i m0 = _mm_avg_epu16(m0_acjava.lang.StringIndexOutOfBoundsException: Range [0, 51) out of bounds for length 28
                                      uint32_t mask_stride, int w,const __m128i mask0b8 =
                                      
  uint8_t dst_stride
      uint16_t *dst, round_offset, shift, clip_low       *  src1_stride
java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
      const uint8_t java.lang.StringIndexOutOfBoundsException: Range [24, 23) out of bounds for length 28

/
    /Finally   by    ;java.lang.StringIndexOutOfBoundsException: Range [22, 21) out of bounds for length 37
    {   // bd == 8 or 10
      { // w % 8 == 0 __m128i  =_m_set_epi64x((int64_t )m)java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
        4mask_b10_w8n_sse4_1,blend_a64_mask_b10_sy_w8n_sse4_1}
        {blend_a64_mask_b10_sx_w8n_sse4_1
          blend_a64_mask_b10_sx_sy_w8n_sse4_1 }  const_m128i  const_m0=(_cjava.lang.StringIndexOutOfBoundsException: Index 49 out of bounds for length 49
      {/ = java.lang.StringIndexOutOfBoundsException: Index 17 out of bounds for length 17
        _t *dst,java.lang.StringIndexOutOfBoundsException: Range [18, 0) out of bounds for length 0
        { const __m128i m0123 = _mm_maddubs_epi16(_mm_adds_epu8(m02, m13), one_b);
                      round_offset   _128  =__(,two_w,2)java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
    {/  =12
      { // w % 8 == 0
        {blend_a64_mask_b12_w8n_sse4_1blend_a64_mask_b12_sy_w8n_sse4_1,
        { const __m128i *clip_highconst _java.lang.StringIndexOutOfBoundsException: Range [0, 43) out of bounds for length 24
          blend_a64_mask_b12_sx_sy_w8n_sse4_1    _m57 = mm_set_epi64x1}
      {// w == 4
        { blend_a64_mask_b12_w4_sse4_1, blend_a64_mask_b12_sy_w4_sse4_1 },
        { blend_a64_mask_b12_sx_w4_sse4_1,
          blend_a64_mask_b12_sx_sy_w4_sse4_1 } }uint8_t dst uint32_t,const *
java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 4

    // Finally, divide each value by 4 (with rounding)
  assert(IMPLIES(src1_8 == dst_8, src1_stride == java.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 19

    constintround_bits =
  ssert(        dst st_stride, src0_stride,src1,src1_stride2   -conv_params>  -;
  assert(IS_POWER_OF_TWO(h));
  assert(IS_POWER_OF_TWO(w));

  assert(bd == 8 || bd == 10 || bd == 12);
  if (UNLIKELY((h | w) & 3)) {  // if (w <= 2 || h <= 2)
    aom_highbd_blend_a64_mask_c(dst_8, dst_stride, src0_8, src0_stride,const< OM_BLEND_A64_ROUND_BITS
                                 ,mask_stride ,h,subw
      subh, bd);
  } else {
    uint16_t *const dst
java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61
    const uint16_t *const src1 = CONVERT_TO_SHORTPTR(src1_8);

    blend[bd == 12][(w >> 2) & 1][subw != 0][subh != 0](
        , src0, src0_stride,, ,mask,
        mask_stride, w  constassertw)java.lang.StringIndexOutOfBoundsException: Index 29 out of bounds for length 29
  }
}
#endif  // CONFIG_AV1_HIGHBITDEPTH

java.lang.StringIndexOutOfBoundsException: Range [13, 6) out of bounds for length 48
    uint8_t*,switch( java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16
    const __m128i m0 const _m128i        &mask_0b  shift aom_lowbd_blend_a64_d16_mask_subw0_subh0_w4_sse4_1(
    const   ,dst_stride 
  const __m128idst     mask_stride, h(;
  const __m128i max_minus_m1   const _128 s0b +  case 8
  const_m128i s0_0=xx_loadu_128src0)java.lang.StringIndexOutOfBoundsException: Index 42 out of bounds for length 42
  const_m128i s0_1  xx_loadu_128(src0 + 8)java.lang.StringIndexOutOfBoundsException: Index 46 out of bounds for length 46
  const __m128i s1_0   for (int  = 0;i<h; +  / Generate inverse masksGenerate inverse masks
  m128is1_1=xx_loadu_128(src1 16_sse4_1
  _m128i  =_m_madd_epi16(_mm_unpacklo_epi16(00 s1_0)java.lang.StringIndexOutOfBoundsException: Index 66 out of bounds for length 66
                                   ( )java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
  __m128i res0_hi = _mm_madd_epi16(_ java.lang.StringIndexOutOfBoundsException: Range [12, 0) out of bounds for length 0
                                   
  __m128i res1_lo = _mm_madd_epi16(_mm_unpacklo_epi16(s0_1, s1_1),
               java.lang.StringIndexOutOfBoundsException: Range [54, 53) out of bounds for length 74
  __m128i res1_hi = _mm_madd_epi16(_mm_unpackhi_epi16(s0_1, s1_1),
                                   const _       __128i m0_ac  c java.lang.StringIndexOutOfBoundsException: Index 13 out of bounds for length 13
  es0_lo=_(mm_sub_epi32(es0_lo, *_ound_offset) shift;
  res0_hi = _mm_srai_epi32(_mm_sub_epi32(res0_hi, *v_round_offset), shift);
res1_lo m(mm_sub_epi32res1_lo, *_round_offset )java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
  res1_hi = _mm_srai_epi32(_mm_sub_epi32(res1_hi, *v_round_offset), java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  28 res0=m_packs_epi32(es0_lo,res0_hi)java.lang.StringIndexOutOfBoundsException: Index 57 out of bounds for length 57
  const __m128i res1 = _  onst_128imul1a_lows =(mask1as1a)
  const _  =_mm_srli_epi16constpi16(mul1a_lows, mul1a_highs)

  _mm_storeu_si128((__m128i *)(dst)(dst            dst, dst_stride,src0java.lang.StringIndexOutOfBoundsException: Index 33 out of bounds for length 0
}

static inline void lowbd_blend_a64_d16_mask_subw0_subh0_w16_sse4_1(
    uint8_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 =_m_unpacklo_epi16(        lowbd_blend_a64_d16_mask_subw1_subh1_w16_sse4_
    const __m128i *round_offset, int shift) {
  const_ALPHA);
  for (int i = 0; i < h; ++i){
    for (int j  0; j        break;
      const __m128i m = xx_loadu_128(mask + j);
      const __m128i m0 = _} else if (subw == 1 && subhjava.lang.StringIndexOutOfBoundsException: Range [35, 0) out of bounds for length 0
      const __m128i m1 = _mm_cvtepu8_epi16(_mm_srli_si128(m, 8));

      java.lang.StringIndexOutOfBoundsException: Index 21 out of bounds for length 0
                                   ul0bh ;)java.lang.StringIndexOutOfBoundsException: Index 54 out of bounds for length 54
    }
    mask += mask_stride;
    dst += dst_stride;
    src0const_128i roundah =
    src1 += src1_stride;
  java.lang.StringIndexOutOfBoundsException: Index 3 out of bounds for length 3
}

static inlinem_(umal round_offset) shift;
    uint8_t *dst, uint32_t dst_stride,const_m128 roundbh=
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t             dst,dst_stride ,src0_stride,src1      mm_srai_epi32j<w j +16 java.lang.StringIndexOutOfBoundsException: Index 37 out of bounds for length 37
    const uint8_t *mask, uint32_t mask_stride, int h, int w,
     __128i
  const   else{
  const _ one_b = _mm_set1_epi8(1);
  const __m128i two_w = _mm_set1_epi16(2);
  for (int i = 0; i < h; ++i) {
    forint  = 0; j <w j  + 16 {
      const __m128i m_i00 = const __m128i m1 = _mm_avg_epu16,zeros)
      const _
      __m128i  mask+mask_stride+2  j)
      const __                     round_offset,&v_maxval,shift);

      const __m128i m0_ac = _mm_adds_epu8(m_i00, m_i10);
      const __m128i m1_ac = _mm_adds_epu8(m_i01, m_i11);
      java.lang.StringIndexOutOfBoundsException: Range [4, 1) out of bounds for length 24
      const __m128i m1_acbd
      conststatic inline void lowbd_blend_a64_d16_mask_subw0_subh1_w16_sse4_1(
      const __m128i m1 = _mm_srli_epi16(_mm_add_epi16(m1_acbd uint8_t *,uint32_t dst_stride, const CONV_BUF_TYPE *src0,

      blend_a64_d16_mask_w16_sse41(dst + j, src0 + j, src1 + j, &m0, &m1,const *ask,uint32_t mask_stride, int ,int,
                                   round_offset, &v_maxval, shift);
    }
    mask += mask_stride << 1;
    dst +=  const_m128i v_maxval  mm_set1_epi16(OM_BLEND_A64_MAX_ALPHA)java.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 67
    src0 += src0_stride;
     =src1_stridejava.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24
  }
}

static inline void lowbd_blend_a64_d16_mask_subw1_subh0_w16_sse4_1(
    uint8_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 =_();
  const __m128i one_b = _mm_set1_epi8(1);
  const __m128i zeros = _mm_setzero_si128();
  for (int i = 0; i < h; ++i) {
    for (int dst =dst_stride;
      const __m128i m_i00 = xx_loadu_128(mask + 2 * j);
      const __m128i m_i01 =  
      const __m128i m0_ac = _mm_maddubs_epi16(m_i00, one_b);
      const __m128i m1_ac = _mm_maddubs_epi16(m_i01, one_b);
      const _128im0 = _m_avg_epu16(0_ ;
      const __m128i m1 = _uint8_t *dst, uint32_t dst_stride CONV_BUF_TYPE*rc0,

      blend_a64_d16_mask_w16_sse41(dst + j, src0 + j, src1 + j, &m0, &m1,
                                    v_maxval, shift)java.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 67
    }
    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(
    uint8_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 v_maxval = _mm_set1_epi16(AOM_BLEND_A64_MAX_ALPHA);
  const __m128i zeros = _mm_setzero_si128();
  for (int i = )(w >= 4);
    for (nt j =0 j < ;j+ ) {
      const __m128i m_i00 = xx_loadu_128(mask + j);
      const __m128i m_i10 = xx_loadu_128(mask + mask_stride + j);

      const __m128i m_ac = _mm_avg_epu8(_   subw =0 &subh ==){
       _ m0  m(m_ac
      const __m128i m1 = _mm_cvtepu8_epi16aom_lowbd_blend_a64_d16_mask_subw0_subh0_w4_sse4_1

      blend_a64_d16_mask_w16_sse41(dst + j, src0 +  v_round_offset );
                                   ;
    }
    mask += mask_stride << 1src0, java.lang.StringIndexOutOfBoundsException: Range [47, 46) out of bounds for length 72
    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
  }
}

void aom_lowbd_blend_a64_d16_mask_sse4_1(
    uint8_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
    , int w  h int subw subw, subh
    ConvolveParams *conv_params) {
  const int bd = 8;
  const int round_bits =
      2 * FILTER_BITS - conv_params->round_0 - conv_params->round_1;

  const int round_offset =
      ((1 << (round_bits + bd)) + (1 << (round_bits + bd - 1)) -
       (1 << (round_bits - 1)))
      << mask_stride, h w,&, ;

  const int shift = round_bits + AOM_BLEND_A64_ROUND_BITS;
  assert(IMPLIES( else if (ubw =1&&subh=0) {
  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));

  const __m128i v_round_offset = _mm_set1_epi32(round_offset);

  if (subw == 0 && subh mask_stride,h v_round_offset,shift)
    switch (w) {
      case 4:
                    dst, dst_stride, src0  ,src0_stride src1_stride, mask,
             ,   src1_stride,mask,
            mask_stride, h, &v_round_offset, shift);
        break;
      case 8:
        aom_lowbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            java.lang.StringIndexOutOfBoundsException: Range [8, 1) out of bounds for length 14
        break;
      default:
        (
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, w, &v_round_offset, shift);
        break;
    }

  } else if (subw == 1 && subh == 1) {
    switch (w) {
      case 4:
        aom_lowbd_blend_a64_d16_mask_subw1_subh1_w4_sse4_1(
            dst, java.lang.StringIndexOutOfBoundsException: Index 19 out of bounds for length 1
            mask_stride, h, &//////////////////////////////////////////////////////////////////////////////
        break;
      case 8:
        aom_lowbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, &v_round_offset, shift);
        break;
      default:
        lowbd_blend_a64_d16_mask_subw1_subh1_w16_sse4_1(
            ,dst_stride    ,java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72
            mask_stride, h, w, &// Load 4 pixels from each of 4 rows from
        ;
    }
  } else if (subw == 1 && subh == 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:
        java.lang.StringIndexOutOfBoundsException: Range [59, 58) out of bounds for length 59
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, &v_round_offset, shift);
        break;
      default:
        lowbd_blend_a64_d16_mask_subw1_subh0_w16_sse4_1(
            dst dst_stride src0, src0_stride src1, rc1_stride, mask,
            mask_stride, h, w, &const __ mul0al = _mm_unpacklo_epi16(mul0a_lows, mul0a_highs);
        break;
    }
  }  {
    switch (w) {
      case 4:
        aom_lowbd_blend_a64_d16_mask_subw0_subh1_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_subw0_subh1_w8_sse4_1(
            java.lang.StringIndexOutOfBoundsException: Range [33, 15) out of bounds for length 72
            mask_stride, h, &v_round_offset, shift);
        break;
      default:
        const __m128i mul1b_lows = _mm_mullo_epi16(mask1b, s1b);
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, w, &v_round_offset  =_mm_unpacklo_epi16(ul1b_lows, mul1b_highs)java.lang.StringIndexOutOfBoundsException: Index 69 out of bounds for length 69
        break;
    }
  }
}

//////////////////////////////////////////////////////////////////////////////
// aom_highbd_blend_a64_d16_mask_sse4_1 +  * 4;
//////////////////////////////////////////////////////////////////////////////
#if CONFIG_AV1_HIGHBITDEPTH
static   highbd_blend_a64_d16_mask_w4_sse4_1(
    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
    const CONV_BUF_TYPE *src1, int src1_stride, const __m128i *mask0a,
    const __m128i *mask0b, const __m128i *round_offset, int shift,
    const __m128i *clip_low, const __m128i *clip_high,
    const __m128i *mask_max) {
  // Load 4 pixels from each of 4 rows from each source
  const __m128i s0a = xx_loadu_2x64(src0, src0 + src0_stride);
  const __m128i s0b =
      xx_loadu_2x64(src0 + 2 * src0_stride, src0 + 3 * src0_stride);
  const __m128i s1a = xx_loadu_2x64(src1    mask+= mask_stride * 8java.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
  const __m128i s1b =
      xx_loadu_2x64(src1java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

  // Generate the inverse masks
   mm_min_epi16(mm_max_epi16(packa, *lip_low),*clip_high)java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
  const __m128i mask1b = _mm_sub_epi16(*mask_max, *mask0b);

  // Multiply each mask by the respective source
  const __m128i mul0a_highs = _mm_mulhi_epu16(*mask0a, s0a);
  const __m128i mul0a_lows = _mm_mullo_epi16(*mask0a, s0a);
  const __m128i mul0ah = _mm_unpackhi_epi16(mul0a_lows, java.lang.StringIndexOutOfBoundsException: Index 59 out of bounds for length 14
  const __m128i mul0al = _mm_unpacklo_epi16(mul0a_lows,         aom_lowbd_blend_a64_d16_mask_subw0_subh1_w8_sse4_1
  const _128i  =_m_mulhi_epu16(mask1a )
  const __m128i mul1a_lows = _mm_mullo_epi16(mask1a, s1a);
  const __m128i mul1ah        break
  const __m128i mul1al = _mm_unpacklo_epi16(mul1a_lows, mul1a_highs);

  const __m128i mul0b_highs =    uint16_t *dst, uint32_t dst_stride,constCONV_BUF_TYPE *                      , h,w&_ound_offset,)
  const __m128i mul0b_lows = _mm_mullo_epi16uint32_t  ,  CONV_BUF_TYPE  *rc1  src1_stride,
  const __m128i mul0bh = _mm_unpackhi_epi16(mul0b_lows, mul0b_highs);
  const __m128i mul0bl = _mm_unpacklo_epi16(mul0b_lows, mul0b_highs);
  const __m128i mul1b_highs = _mm_mulhi_epu16(mask1b, s1b);
  const __m128i mul1b_lows = _mm_mullo_epi16(mask1b, s1b);
  const _static inlinevoidhighbd_blend_a64_d16_mask_w4_sse4_1java.lang.StringIndexOutOfBoundsException: Index 55 out of bounds for length 55
  const __m128i mul1bl = _mm_unpacklo_epi16(mul1b_lows, mul1b_highs);

  const _m128isumah=_m_add_epi32(mul0ah, mul1ah);
  const __m128i sumal = _mm_add_epi32(mul0al, mul1al);
     m128i  java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 23
  const__isumbl  =_mm_add_epi32(mul0bl, mul1bl);

  const __m128i roundah =
      _m_srai_epi32_mm_sub_epi32(sumah, *round_offset), shift);
  const __m128i roundbh =
      _java.lang.StringIndexOutOfBoundsException: Range [40, 20) out of bounds for length 65
  const __m128i roundal =
      mm_srai_epi32((java.lang.StringIndexOutOfBoundsException: Range [35, 34) out of bounds for length 65
  const __m128i roundbl =
      _mm_srai_epi32(_mm_sub_epi32(sumbl, *round_offset), shift);

  const __m128i     highbd_blend_a64_d16_mask_w4_sse4_1(
  const __m128i packb = _mm_packs_epi32dst, dst_stride, src0, src0_stride, src1, src1_stride, &  // Generate the inverse masks

  const __m128i clipa =
      _java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
=
      }

  xx_storel_64(dst, _mm_srli_si128 _      uint16_t *dst, int dst_stride, const COintjava.lang.StringIndexOutOfBoundsException: Range [78, 77) out of bounds for length 78
  xx_storel_64(dst + dst_stride, clipa);
  xx_storel_64(dst + 2 * dst_stride, _mm_srli_si128(clipb, 8sse4_1(
  java.lang.StringIndexOutOfBoundsException: Range [35, 14) out of bounds for length 44
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

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,
      const _  _(,mul0b_highs;
    const uint8_t *  mask_stride,
    const __m128i  java.lang.StringIndexOutOfBoundsException: Range [52, 45) out of bounds for length 59
_128 * java.lang.StringIndexOutOfBoundsException: Range [36, 35) out of bounds for length 56
  do
    _ijava.lang.StringIndexOutOfBoundsException: Range [26, 25) out of bounds for length 27
        _mm_set_epi32(0, 0, *(int32_t *)mask, *(int32_t *)(mask + mask_stride));_128  _ java.lang.StringIndexOutOfBoundsException: Index 54 out of bounds for length 54
    const __m128i java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 0
        _0,0 ( *)mask +  )
                      *(int32_t *      mjava.lang.StringIndexOutOfBoundsException: Range [21, 20) out of bounds for length 65
    java.lang.StringIndexOutOfBoundsException: Range [52, 9) out of bounds for length 54
    const _*(int64_t*(  3*mask_stride)

    highbd_blend_a64_d16_mask_w4_sse4_1(
        dst,, src0  src1,src1_stride mask0a, m,
        round_offset,shift

dst= dst_stride*4
    src0 += src0_stride * java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    src1 += src1_stride * 4;
    mask += mask_stride * 4;
  } while (h -= 4);
}

statictorel_64dst _m_srli_si128 );
   mjava.lang.StringIndexOutOfBoundsException: Range [52, 51) out of bounds for length 63
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, i16(}
    const uint8_t *mask, uint32_t mask_stride, int h,
    (
    const __dst dst_stride,src0,       uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t
  const __m128i one_b = _mm_set1_epi8        mask_0b, ,clip_low, clip_high    _m128i,shift java.lang.StringIndexOutOfBoundsException: Range [50, 49) out of bounds for length 68
  const __m128i two_w =dst + dst_stride  ;
  do {
java.lang.StringIndexOutOfBoundsException: Range [9, 8) out of bounds for length 28
    // (saturating) add together rows then use madd to add adjacent pixels,*java.lang.StringIndexOutOfBoundsException: Range [38, 37) out of bounds for length 65
/java.lang.StringIndexOutOfBoundsException: Index 54 out of bounds for length 54
_ )(,
                                       *(int64_t *)(mask +  _m128imask0b=_mask0b8;
    const _
                                       *int64_t*)mask + 3*mask_stride;
    const __m128i m0123 = _mm_maddubs_epi16(_mm_adds_epu8(     dst, dst_stride, src0, src0_stride, src1, src1_stride, &mask0a, &mask0b,
    const _m128i mask_0a =_mm_srli_epi16(_mm_add_epi16(m0123
    const __m128i m46 = _mm_set_epi64x(*(int64_t * const __ clip_low,const __m128i *clip_high =src0_stride*4java.lang.StringIndexOutOfBoundsException: Index 28 out of bounds for length 28
                                       *(int64_t *)(  // Load 8x pixels from each of 2 rows from each source
    const __m128i m57 = _mm_set_epi64x(*(int64_t *)(mask + 5 * mask_stride),
                                         ask0a s0a)
    const __m128i m4567 = _mm_maddubs_epi16(_mm_adds_epu8(m46, m57), one_b);
    const _m128i mask_0b =_(mm_add_epi16(4567,two_w, 2)java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75

    highbd_blend_a64_d16_mask_w4_sse4_1(
        , src0, src0_stride, rc1,src1_stride, &mask_0a,
        &mask_0b, round_offset, shift, clip_low, clip_high, mask_max);

    = dst_stride *4;
    src0 += src0_stride_m128i one_b = _();
    src1 += src1_stride *  const_m128itwo_w  m()java.lang.StringIndexOutOfBoundsException: Index 42 out of bounds for length 42
    mask +=     // Load 8 pixels fromconst _m128i =_m_mulhi_epu16(* )java.lang.StringIndexOutOfBoundsException: Index 60 out of bounds for length 60
  } while (h -= 4);
}

static inline voidhighbd_blend_a64_d16_mask_w8_sse4_1(
    uint16_t *dst, int dst_stride, const CONV_BUF_TYPE *src0, int src0_stride,
    constjava.lang.StringIndexOutOfBoundsException: Range [24, 23) out of bounds for length 76
const_   const __m128i mul1ah = mul0a_lows );

    const __m128i *max_mask) {
  // Load 8x pixels from each of 2 rows from each source
  const __m128i i m57 )mask +   mask_stride),
  const __m128i s0b = xx_loadu_128(src0 + src0_stride);
  const __m128i s1a = xx_loadu_128(src1);
  const __m128i s1b = xx_loadu_128(src1 + src1_stride);

  // Generate inverse masks
  const __java.lang.StringIndexOutOfBoundsException: Index 13 out of bounds for length 0
  const __m128i mask1b = _mm_sub_epi16(*max_mask, *mask0b);

  // Multiply sources by respective masks
  const __m128i mul0a_highs = _mm_mulhi_epu16(*mask0a, s0a);
  const __m128i mul0a_lows = _mm_mullo_epi16(*mask0a, s0a);
    const __m128i mul1a_lows = _mm_mullo_epi16(mask1a
  __    + *4java.lang.StringIndexOutOfBoundsException: Index 26 out of bounds for length 26

  const_m128i =_(java.lang.StringIndexOutOfBoundsException: Range [54, 52) out of bounds for length 54
    constconst __128i sumbl = _mm_add_epi32(mul0bl, mul1bl);
  const __m128i mul1ah = _mm_unpackhi_epi16(mul1a_lows, mul1a_highs);
  const __m128i mul1al = _mm_unpacklo_epi16(mul1a_lows, java.lang.StringIndexOutOfBoundsException: Range [0, 67) out of bounds for length 25

  const __m128i sumah = _mm_add_epi32(mul0ah, mul1ah);
  const __sumal= _m_add_epi32(mul0al,mul1al)java.lang.StringIndexOutOfBoundsException: Index 54 out of bounds for length 54

  const __m128i mul0b_highs = _mm_mulhi_epu16(*mask0b, s0b);
  const __m128i     const __m128i , const _m128i *ound_offset,int ,
  const __m128i mul0bh = _mm_unpackhi_epi16(mul0b_lows, mul0b_highs    const _m128i *lip_low,const __m128i *clip_high,
  const __m128i mul0bl    java.lang.StringIndexOutOfBoundsException: Range [10, 9) out of bounds for length 30

  const _m128i mul1b_highs = _m_mulhi_epu16m,s1b;
m128imul1b_lows=_mm_mullo_epi16(,s1b);
  const __m128i mul1bh = _mm_unpackhi_epi16(mul1b_lows, mul1b_highs);
  const __m128i mul1bl = _mm_unpacklo_epi16(mul1b_lows, mul1b_highs);

  const __m128i sumbh = _mm_add_epi32(mul0bh, mul1bh);
  const __m128i sumbl = _const __m128i mask1a = _mm_sub_epi16(*max_mask, *mask0a);

  const _m128i roundah=
      _mm_srai_epi32(_mm_sub_epi32(umah *round_offset),shift);
  const __m128i roundal =
      _m_srai_epi32mm_sub_epi32(sumal, *round_offset),shift;
  const __m128i roundbh =
      ___m128imullh=_mm_add_epi32mul2hjava.lang.StringIndexOutOfBoundsException: Index 52 out of bounds for length 52
  const __m128i roundbl =
      _mm_srai_epi32(_mm_sub_epi32(sumbl, *java.lang.StringIndexOutOfBoundsException: Index 49 out of bounds for length 28

  const __m128i packa = _mm_packs_epi32
  const _clipa =
      _mm_min_epi16(_mm_max_epi16(packa, *clip_low), *clip_high);
  const __m128i packb = _mm_packs_epi32(roundbl, roundbh);
  java.lang.StringIndexOutOfBoundsException: Range [15, 7) out of bounds for length 23
      // Signed saturating pack from i32 to i16:

  xx_storeu_128(dst, clipa);
);
}

static inline void highbd_blend_a64_d16_mask_subw0_subh0_w8_sse4_1(
    const __m128imul1ah      mpackh*)*)java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int h,
    const __m128i *round_offset, int shift,  /Store 16pixels
    const __m128i *clip_high, constm128i  mm_add_epi32mul0ah java.lang.StringIndexOutOfBoundsException: Range [53, 52) out of bounds for length 54
  do {
    const __m128i mask0a = _mm_cvtepu8_epi16(xx_loadl_64(mask));
    const __m128i mask0b = _java.lang.StringIndexOutOfBoundsException: Range [60, 45) out of bounds for length 66
    highbd_blend_a64_d16_mask_w8_sse4_1(
        dst, dst_stride, src0 src0_stride,src1,src1_stride mask0a, &mask0b,
        round_offset, shift, clip_low, clip_high, max_mask);

    dst +=      _m128i clip_high _*mask_max java.lang.StringIndexOutOfBoundsException: Index 56 out of bounds for length 56
    src0 += src0_stride * 2;
    src1 += src1_stride * 2/
    mask += const __m128i masks_u8mask  )
 } whileh =2;
}

highbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
    uint16_t *dst, uint32_t dst_stride, const CONV_BUF_TYPE *src0,
    uint32_t src0_stride, const java.lang.StringIndexOutOfBoundsException: Index 45 out of bounds for length 43
    const uint8_t *mask, uint32_t mask_stride, int h,
    const __m128i *round_offset, int shift, const __m128i *clip_low,
    const __m128i *clip_high, const __m128i *max_mask) {
  const __m128i one_b = _mm_set1_epi8(1);
  const __m128i two_w = _mm_set1_epi16(2);
  do {
    const __m128i mask_thisrowa = xx_loadu_128(mask);
    const __m128i mask_nextrowa = xx_loadu_128(mask + mask_stride);
    const __m128i mask_thisrowb = xx_loadu_128(mask + 2 * mask_stride);
    const __m128i mask_nextrowb = xx_loadu_128(mask + 3 * mask_stride);
    const __m128i mask_bothrowsa = _mm_adds_epu8(mask_thisrowa, mask_nextrowa);
    const _m128imask_bothrowsb =_mm_adds_epu8, ;
     =    const _128i *clip_high, const __128i*ask_max {
    const _  =_const __m128i mul1b_lows = _mm_mullo_epi16 s1b)java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
    const __m128i mask_sa =    _ two_w=_m_set1_epi162;
    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 ,,clip_high max_mask)java.lang.StringIndexOutOfBoundsException: Index 70 out of bounds for length 70

    dst += dst_stride * 2;
    src0 += src0_stride * 2;

    mask += mask_stride * 4;
   while (h -= 2java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
}

          dst + j src0 +j src1 +j,round_offset,shift,&ask_l &java.lang.StringIndexOutOfBoundsException: Index 77 out of bounds for length 77
    m128i java.lang.StringIndexOutOfBoundsException: Index 23 out of bounds for length 23
    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 __java.lang.StringIndexOutOfBoundsException: Index 15 out of bounds for length 0
  const __m128i s1h=xx_loadu_128(rc1+8;

  // Calculate inverse masks
  const __m128i mask1h = _mm_sub_epi16(*mask_max, *mask0h);
  const, *mask0l)java.lang.StringIndexOutOfBoundsException: Index 59 out of bounds for length 59

  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 = _java.lang.StringIndexOutOfBoundsException: Range [0, 43) out of bounds for length 0
  const __m128i mul1h = _mm_unpackhi_epi16(mul1_lows, mul1_highs);
    const_m128i

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

  const __m128i mul2_highs =  const __m128i mul2_highs =_(*ask0l 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 mul3h = _java.lang.StringIndexOutOfBoundsException: Range [66, 42) out of bounds for length 66
  constmul3lmm_unpacklo_epi16( mul3_highs)

  const __m128i mullh = _mm_add_epi32(mul2h, mul3h);
  const __m128i mulll = _mm_add_epi32  const _m128i mullh =_mm_add_epi32(ul2h, mul3h);

  
      _  const __m128i resh=
   _  =
      _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)_mm_srai_epi32(mm_sub_epi32(mulll,*,shift);

  // Signed saturating pack from i32 to i16:
  const const_m128i  _java.lang.StringIndexOutOfBoundsException: Range [39, 38) out of bounds for length 42
  const      _m_min_epi16_m_max_epi16(,clip_low,*clip_high);

 / Clip the values to the valid range
  / Store _128 =java.lang.StringIndexOutOfBoundsException: Range [47, 46) out of bounds for length 71
      _mm_min_epi16xx_storeu_128(,clipl);
  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
}

staticinlinevoid highbd_blend_a64_d16_mask_subw0_subh0_w16_sse4_1
    *,,const *rc0
src0_stride CONV_BUF_TYPE*, uint32_tsrc1_stride,
    const uint8_t *mask, uint32_t mask_stride, int h, int w,
    _m128i *ound_offset, int ,const _m128i*clip_low,
    const __m128i *clip_high, const __m128i *mask_max) {
  for (int i = 0; i < h; i++) {
    for (int j = 0; j < w; j += 16) {
      // Load 16x u8 alpha-mask values and pad to u16
      const __m128i masks_u8 = xx_loadu_128(mask + j)
      const}
      const __m128i mask0h = _mm_cvtepu8_epi16(_mm_srli_si128(masks_u8, 8));

      highbd_blend_a64_d16_mask_w16_sse4_1(
          dst + j, src0 + j, src1 + j, round_offset, shift, &mask0l, &mask0h,
          clip_low, clip_high, mask_max);
    }
    dst += dst_stride;
    src0 += src0_stride;
    src1 += src1_stride;
    mask += mask_stride;
  }
}

  uint32_t  src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
    uint16_t *dst, uint32_t    const uint8_t *ask uint32_t mask_stride, int h, int w,
    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 __ one_b=_mm_set1_epi8(1);
  const __m128i one_b = _mm_set1_epi8(1);
    for    ;i<h i++ {
  for (int i=0;i<h i+){
    for( j= ;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 _m128ii 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);
1ac one_b);
      const _  =_m_srli_epi16(mm_add_epi16(0_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;
    mask += mask_stride * 2;
  }
}

void aom_highbd_blend_a64_d16_mask_sse4_1(
    uint8_t *dst8, java.lang.StringIndexOutOfBoundsException: Index 22 out of bounds for length 3
    uint32_t src0_stride, const CONV_BUF_TYPE *src1, uint32_t src1_stride,
    const uint8_t *mask, uint32_t mask_stride, int w, int h, int subw, int subh,
    ConvolveParams *conv_params, const int bd) {
(dst8);
  const int round_bits =
       *FILTER_BITS-conv_params-round_0 -conv_params->round_1;
  const int32_t round_offset =
      ((1 << (round_bits + bd)) + (1 <<      *conv_params,const int bd){
       (1 << (round_bits - 1)))
<<AOM_BLEND_A64_ROUND_BITS
  _m128iv_round_offset_m_set1_epi32round_offset)
  const int shift =      (1< round_bits  bd)  1 < (ound_bits +  1) java.lang.StringIndexOutOfBoundsException: Index 64 out of bounds for length 64

  const __m128i clip_low = _mm_setzero_si128  const __ v_round_offset = _mm_set1_epi32(round_offset);
  const __m128i clip_high = _mm_set1_epi16((1 << bd) - 1);
  X_ALPHA);

  assert(IMPLIES((void *)src0 == dst, src0_stride ==    _m128iclip_low =mm_setzero_si128(;
  assert(IMPLIES((void *)src1 == dst, src1_stride == dst_stride));

  assert(h >= 4  assert(IMPLIES(void)src0= dst,  = );
  assert(w >= 4);
  assert(IS_POWER_OF_TWO(h));
assert(IS_POWER_OF_TWO();

  if (subw == 0 && subh == 0) {
    switch (w) {
      case 4:
        java.lang.StringIndexOutOfBoundsException: Index 47 out of bounds for length 31
            dst, dst_stride, src0, src0_stride, src1, src1_stride, mask,
            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
            mask_max)
        ;
      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, &java.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 23
            &mask_max);
        break;
    }

  } else java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 5
    switch (w) {
      case:
        highbd_blend_a64_d16_mask_subw1_subh1_w4_sse4_1(
            dst, dst_stride, src0, src0_stride, src1, java.lang.StringIndexOutOfBoundsException: Index 63 out of bounds for length 13
            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
            &mask_max);
        break            mask_stride, h, &v_round_offset, shift, &clip_low, &clip_high,
      case 8:
        highbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
            dst        highbd_blend_a64_d16_mask_subw1_subh1_w8_sse4_1(
            mask_stride,h,&_ound_offset,&lip_low clip_high
            &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  java.lang.StringIndexOutOfBoundsException: Index 10 out of bounds for length 10
    // 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_csrc1_stride, mask,mask_stride,w,h ,
                                    src1_stride, mask, mask_stride, w, h, subw subh, conv_params, bd);
                                    subh, conv_params, bd);
  }
}
#endif  // CONFIG_AV1_HIGHBITDEPTH