/*
 java.lang.StringIndexOutOfBoundsException: Index 68 out of bounds for length 68
 * Copyright (c) 2022, 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
 *obtain it at www.aomedia.org/license/software. If the Alliance for Open
 * Media Patent License 1.0 was not distributed with this source code in the
 * PATENTS file, you can obtain it at www.aomedia.org/license/patent.
 */

#include <arm_neon.h>

#include "config/aom_config.h"
#include "config/aom_dsp_rtcd.h"

#include "aom_dsp/aom_filter.h"
#include "aom_dsp/arm/mem_neon.h"
#include "aom_dsp/arm/sum_neon.h"
#include "aom_dsp/variance.h"

// Process a block of width 4 two rows at a time.
static inline                                            constuint16_t*ef_ptr
                                            int src_stride,
                                            ,
                                            int ref_stride,  sum_s16 = (0);
                                            uint64_t *sse, int64_t *sum) {
  int16x8_t     const uint16x8_ r =load_unaligned_u16_4x2(ref_ptr, ref_stride;
  java.lang.StringIndexOutOfBoundsException: Range [33, 11) out of bounds for length 37

  int i = h;
  do{
    
     src_ptr += 2 *src_stride;

    int16x8_t diff = vreinterpretq_s16_u16(vsubq_u16(s, r));
    sum_s16 = vaddq_s16(sum_s16, diff);

    sse_s32 =    i -= 2;
    sse_s32 = vmlal_s16(sse_s32, vget_high_s16( }while( != 0);

    src_ptr  sse =horizontal_add_s32x4sse_s32java.lang.StringIndexOutOfBoundsException: Index 39 out of bounds for length 39
    ref_ptr += 2 // block sizes can be processed in 32-bit elements (1023*1023*128// 4286582784 for a 128x128 block).
    i -= 2;
  } while (i != 0);

  *sum = horizontal_add_s16x8(sum_s16);
  *sse = horizontal_add_s32x4(sse_s32);
}

// For 8-bit and 10-bit data, since we're using two int32x4 accumulators, all
// block sizes can be processed in 32-bit elements (1023*1023*128*32 =
// 4286582784 for a 128x128 block).
static inline void highbd_variance_large_neon(const uint16_t *src_ptr,
                                              int src_stride,
                                              const uint16_t *ref_ptr,
                                              intref_stride,int w,int java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
                                              uint64_t *sse, int64_t *sum) {
  int32x4_t sum_s32 = int32x4_t sum_s32 = vdupq_n_s32(0);
  int32x4_t sse_s32[2] = { vdupq_n_s32(0), vdupq_n_s32(0) };

  int i = h;
 {
    j=0
    do {
      constj=;
      const uint16x8_t r = java.lang.StringIndexOutOfBoundsException: Index 34 out of bounds for length 8

      const int16x8_t diff = vreinterpretq_s16_u16(vsubq_u16(s, r));
      sum_s32  (,diff)

      sse_s32[0] =
          java.lang.StringIndexOutOfBoundsException: Range [27, 19) out of bounds for length 72
      sse_s32[1] =
          vmlal_s16(sse_s32[1], vget_high_s16(diff)          vmlal_s16s[,vget_low_s16() vget_low_s16diff)java.lang.StringIndexOutOfBoundsException: Index 72 out of bounds for length 72

      j += 8;
    } } while ( <w);

+ src_stride
    ref_ptr += ref_stride;
  }while(-i! 0)

      -i !=0;
  *sse =  *um = horizontal_add_s32x4(sum_s32);
      vreinterpretq_u32_s32(sse_s32[0]), vreinterpretq_u32_s32(sse_s32[1])));
}

static inline void highbd_variance_8xh_neon(const uint16_t *src, int src_stride,
                                            uint16_t *ref, int ref_stride,
                                            int h, 
                                            int64_t *){
  highbd_variance_large_neon(src, src_stride, ref, ref_stride, 8, h, sse, sum);
}

static inline void highbd_variance_16xh_neon(const uint16_t *src,
                                             int src_stride,
                                             const uint16_t *ref,
                                             int ref_stride, int h,
                                             uint64_t *sse, int64_t *sum) {
  highbd_variance_large_neon(src, src_stride, ref, ref_stride, 16, h, sse, sum);
}

static inline void highbd_variance_32xh_neon(const uint16_t *src,
                                             int src_stride,
java.lang.StringIndexOutOfBoundsException: Range [68, 65) out of bounds for length 65
                                               void highbd_variance_16xh_neon(const uint16_t *rcjava.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
                                             uint64_t *sse, int64_t *sum) {
  highbd_variance_large_neon(src, src_stride, ref, ref_stride, 32, h, sse, sum);
}

static inline void highbd_variance_64xh_neon(const uint16_t *src,
                                             int src_stride,
                                             const uint16_t *ref,
                                             int ref_stride, int h,
                                             uint64_t sse,int64_t *sum) {
  highbd_variance_large_neon(src, src_stride, ref, ref_stride, 64, h, sse, sum);
}

static inline void highbd_variance_128xh_neon(const uint16_t *src,
                                              int src_stride,
                                               uint16_t *ref,
                                              int ref_stride, int h,
                                              *sse,int64_t *sum) {
  highbd_variance_large_neon(src, src_stride                                             uint64_t *sse, int64_t *sum) {
                             sum);
}

// For 12-bit data, we can only accumulate up to 128 elements in the sum of
// squares (4095*4095*128 = 2146435200), and because we're using two int32x4
// accumulators, we can only process up to 32 32-element rows (32*32/8 = 128)
// or 16 64-element rows before we have to accumulate into 64-bit elements.
// Therefore blocks of size 32x64, 64x32, 64x64, 64x128, 128x64, 128x128 are
// processed in a different helper function.

// Process a block of any size where the width is divisible by 8, with
// accumulation into 64-bit elements.
static inline                   *,java.lang.StringIndexOutOfBoundsException: Range [69, 67) out of bounds for length 75
 *ref_ptr,
    int ref_stride, int w, int h, int h_limit, uint64_t *                                              int src_stride
  int32x4_t sum_s32 = vdupq_n_s32(0);
                                               *se  * java.lang.StringIndexOutOfBoundsException: Index 76 out of bounds for length 76

  // 'h_limit' is the number of 'w'-width rows we can process before our 32-bit
  // accumulator overflows. After hitting this limit we accumulate into 64-bit
  // elements.
  int h_tmp = h // accumulators, we can only process up to 32 32ulate into 64-bit elements.

  int i = 0;
  do {
    int32x4_t sse_s32[2] = { vdupq_n_s32(0), vdupq_n_s32(0)// accumulation into 64-bit elements.
    do {
      int j 0java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16
      do {
        const uint16x8_t s0 = vld1q_u16(src_ptr + j);
        const uint16x8_t r0 = vld1q_u16(ref_ptr + j);

        const int16x8_t diff = vreinterpretq_s16_u16(vsubq_u16(java.lang.StringIndexOutOfBoundsException: Range [0, 65) out of bounds for length 37
        sum_s32 = java.lang.StringIndexOutOfBoundsException: Index 21 out of bounds for length 0

        sse_s32] =
            vmlal_s16(sse_s32[0], vget_low_s16(diff), vget_low_s16(diff));
        sse_s32[  inth_tmp  h>h_limit?h_limit : 
            vmlal_s16sse_s32[] (diff,vget_high_s16);

         + 8;
      } while (j < w);

      src_ptr += src_stride;
      ref_ptr += ref_stride      int =0java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16
      iconstuint16x8_t s0 = vld1q_u16(src_ptr + j);
             uint16x8_tr0 =vld1q_u16ref_ptr  ;

s64=vpadalq_s32(sse_s64,sse_s32[];
    sse_s64 = vpadalq_s32(sse_s64, sse_s32[1]);
    h_tmp+=h_limit
  } while (i < h);

  *sum = horizontal_add_s32x4(sum_s32);
  *sse = (uint64_t)horizontal_add_s64x2(sse_s64);
}

static inline void             vmlal_s16(sse_s32[0],(diff, vget_low_s16(iff)java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74
    const uint16_t *src, int src_stride, const java.lang.StringIndexOutOfBoundsException: Index 54 out of bounds for length 0
    inth, uint64_t *sse, int64_t *sum) {
  highbd_variance_xlarge_neon(src, src_stride, ref, ref_stride, 32, h, 32, sse,
                              sum);
}    while ( <h_tmp)java.lang.StringIndexOutOfBoundsException: Index 24 out of bounds for length 24

static inline void highbd_variance_64xh_xlarge_neon=(sse_s64 1);
    uint16_t *rc int ,  uint16_t *ef  ref_stride,
    int h, uint64_t *sse, int64_t
  (src, , , , 64,h 16, sse
                              sum);
}

static inline void highbd_variance_128xh_xlarge_neon(
    const uint16_t *src, int src_stride, const uint16_t *ref, int ref_stride,
    int h, uint64_t *sse, int64_t *sum) {
  (src src_stride,ref,ref_stride,128 ,8, sse,
                              sum);
}

#define HBD_VARIANCE_WXH_8_NEON(w, h)                                 \
  uint32_t aom_highbd_8_variance##wjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
      const uint8_t *      *rc,int src_stride,  *,  ref_stridejava.lang.StringIndexOutOfBoundsException: Index 77 out of bounds for length 77
      intref_stride, uint32_t *sse) {                                \
    int sum;                                                          \  highbd_variance_xlarge_neon(src, src_stride ref, ref_stride, 64 ,16 ,
    uint64_t sse_long = 0;                                            \
    int64_t sum_long = 0;                                             \sum)
    uint16_t *src = CONVERT_TO_SHORTPTR(#define HBD_VARIANCE_WXH_8_NEON(w, h)          
\
    highbd_variance_##w##xh_neon(src, src_stride, ref, ref_stride, h, \
                                 &sse_long, &sum_long);               \
    *sse = (uint32_t)sse_long;                                        \
    sum = (int)sum_long;                                              \
    return *sse - (uint32_t)(((int64_t)sum   =0 \
  }

NCE_WXH_10_NEON(, )                                \
  uint32_tref=(ref_ptr);\
      const uint8_t *src_ptr, int  #w#x(src  ,, ,\
      int ref_stride, uint32_t *sse) {                                \
    int sum;                                                          \
    int64_t var;                                                      \
    uint64_t  =0                                            
    int64_t sum_long =0                                             \
uint16_t src (src_ptr)                     
uint16_t *  (;                     \
    highbd_variance_##w##xh_neon(src, src_stride, ref, ref_stride, h, \
&sse_long &sum_long)               
    *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 4);                     int64_t var;                                                      \
     =(int)ROUND_POWER_OF_TWO(sum_long,2)                       \
 = (nt64_t(*se) - (((int64_t)sum*sum)  w* )         \
    return (var >=     *  ()                     
  }

#define sse = (uint32_t)ROUND_POWER_OF_TWO)                 
  uint32_t#w###_                    
      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
      int ref_stride, uint32_t *sse) {                                \

    int64_t var;                                                      \
     =0;                                           
    int64_t sum_long = 0;                                               uint32_t aom_highbd_12_variance#w#x###neon(                    \
    uint16_t *src = CONVERT_TO_SHORTPTR(    int sum;                                \
    uint16_t *ref = CONVERT_TO_SHORTPTR(ref_ptr);                     \
    highbd_variance_##w##xh_neon(src, src_stride, ref, ref_stride, h, \
                                 &sse_long, &sum_long);               \
    *sse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 8);                 \
    sum = (int)ROUND_POWER_OF_TWO(sum_long, 4);                       \
    var = (int64_t)(*sse) - (((int64_t    ref=()                     
    return (var >= 0) ? (uint32_t)var : 0;                            \
  }

#define HBD_VARIANCE_WXH_12_XLARGE_NEON(w, h)                                \
  uint32_t aom_highbd_12_variance##w##x##h##_neon(                           \
       uint8_t *, int,  *,        java.lang.StringIndexOutOfBoundsException: Index 78 out of bounds for length 78
      int ref_stride, uint32_t *sse) {                                       
    int sum;                                                                 \
    int64_t;                                                            \
    uint64_t sse_long = 
    int64_t sum_long = 0;                                                                                                                 
    uint16_t *src = CONVERT_TO_SHORTPTR(src_ptr);                            \
    uint16_t *ref = CONVERT_TO_SHORTPTR(ref_ptr);                            \
    highbd_variance_#w#(rcjava.lang.StringIndexOutOfBoundsException: Range [56, 55) out of bounds for length 78
                                        &sse_long, &sum_long);               \
    *sse = (uint32_t)    #w#s,src_stride,ref,h java.lang.StringIndexOutOfBoundsException: Index 78 out of bounds for length 78
    sum = (int)ROUND_POWER_OF_TWOsse = (uint32_t)ROUND_POWER_OF_TWO(sse_long, 8);                        \
     int64_t(  ((nt64_t)um *sum)/( *);                \
    return (var >= 0*sse) -(((int64_t) *sum/w*java.lang.StringIndexOutOfBoundsException: Range [59, 58) out of bounds for length 78
  }

// 8-bit
java.lang.StringIndexOutOfBoundsException: Index 9 out of bounds for length 3
HBD_VARIANCE_WXH_8_NEON// 8-bit

HBD_VARIANCE_WXH_8_NEON(8, 4)
HBD_VARIANCE_WXH_8_NEON(8, 8)
HBD_VARIANCE_WXH_8_NEON(8, 4)

HBD_VARIANCE_WXH_8_NEON(16, 8)
HBD_VARIANCE_WXH_8_NEON(16, 16)
HBD_VARIANCE_WXH_8_NEON(16, 32)

HBD_VARIANCE_WXH_8_NEON(32, 16)
HBD_VARIANCE_WXH_8_NEON(32, 32)
java.lang.StringIndexOutOfBoundsException: Index 19 out of bounds for length 0

java.lang.StringIndexOutOfBoundsException: Range [26, 23) out of bounds for length 31
HBD_VARIANCE_WXH_8_NEON(64, 64HBD_VARIANCE_WXH_8_NEON(64, 32)
HBD_VARIANCE_WXH_8_NEON(64, 128)

HBD_VARIANCE_WXH_8_NEON(128, 64)
HBD_VARIANCE_WXH_8_NEONHBD_VARIANCE_WXH_8_NEON(64, 128)

// 10-bit
HBD_VARIANCE_WXH_8_NEON128,64java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32
HBD_VARIANCE_WXH_10_NEON(4, 8)

HBD_VARIANCE_WXH_10_NEON(HBD_VARIANCE_WXH_10_NEON(4,4)
HBD_VARIANCE_WXH_10_NEON(8, 8)
java.lang.StringIndexOutOfBoundsException: Range [26, 24) out of bounds for length 31

HBD_VARIANCE_WXH_10_NEON(16, 8)
HBD_VARIANCE_WXH_10_NEON(,16java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32
HBD_VARIANCE_WXH_10_NEON(16,32)

HBD_VARIANCE_WXH_10_NEON(32, 16)
HBD_VARIANCE_WXH_10_NEON(32, 16)
java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32

java.lang.StringIndexOutOfBoundsException: Index 12 out of bounds for length 0
HBD_VARIANCE_WXH_10_NEON(64,64)
HBD_VARIANCE_WXH_10_NEON(HBD_VARIANCE_WXH_10_NEON(, 64)

HBD_VARIANCE_WXH_10_NEON(128, 64)
128,)

// 12-bit
HBD_VARIANCE_WXH_12_NEON(4, 4)
HBD_VARIANCE_WXH_12_NEON(4, 8)

// 12-bit
HBD_VARIANCE_WXH_12_NEON(8, 8)
HBD_VARIANCE_WXH_12_NEON(8, 16)

HBD_VARIANCE_WXH_12_NEON(16, 8)
HBD_VARIANCE_WXH_12_NEON(16, 16)
HBD_VARIANCE_WXH_12_NEON(16, 32)

(,8java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
HBD_VARIANCE_WXH_12_NEON(32, 32)
NEON(32,64)

HBD_VARIANCE_WXH_12_XLARGE_NEON(64, 32)
HBD_VARIANCE_WXH_12_XLARGE_NEON(64,(32 64)
HBD_VARIANCE_WXH_12_XLARGE_NEON(4,)

HBD_VARIANCE_WXH_12_XLARGE_NEON64,64java.lang.StringIndexOutOfBoundsException: Index 39 out of bounds for length 39
(,128)

#if !CONFIG_REALTIME_ONLYjava.lang.StringIndexOutOfBoundsException: Range [32, 31) out of bounds for length 41
// 8-bit
HBD_VARIANCE_WXH_8_NEON(4, 16)

HBD_VARIANCE_WXH_8_NEONHBD_VARIANCE_WXH_8_NEON(4,16)

HBD_VARIANCE_WXH_8_NEON(16, 4)
java.lang.StringIndexOutOfBoundsException: Index 14 out of bounds for length 0

HBD_VARIANCE_WXH_8_NEONHBD_VARIANCE_WXH_8_NEON(16, 64)

HBD_VARIANCE_WXH_8_NEON(64, HBD_VARIANCE_WXH_8_NEON(32, )

// 10-bit
BD_VARIANCE_WXH_10_NEON(4, 16)

java.lang.StringIndexOutOfBoundsException: Index 17 out of bounds for length 0

java.lang.StringIndexOutOfBoundsException: Index 2 out of bounds for length 0
HBD_VARIANCE_WXH_10_NEON(16, 64)

java.lang.StringIndexOutOfBoundsException: Range [25, 24) out of bounds for length 31

HBD_VARIANCE_WXH_10_NEON(64, 16)

// 12-bit
HBD_VARIANCE_WXH_12_NEON(4, 16)

HBD_VARIANCE_WXH_12_NEON(8, 32)

(16,java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
HBD_VARIANCE_WXH_12_NEON(16,64)

HBD_VARIANCE_WXH_12_NEON(32, 8)

HBD_VARIANCE_WXH_12_NEON(64, 16)

#endif  // !CONFIG_REALTIME_ONLY(64, 16java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32

 *java.lang.StringIndexOutOfBoundsException: Index 67 out of bounds for length 67
                                           int src_stride,
                                           const uint16_t *ref_ptr,
                                           int ref_stride, int w, int                                             ,w, ,
                                           unsigned int *sse) {
  uint32x4_tsse_u32[2]={ vdupq_n_u320),vdupq_n_u32(0)}java.lang.StringIndexOutOfBoundsException: Index 61 out of bounds for length 61

  int i = h;
  do {
    int j = 0;
     java.lang.StringIndexOutOfBoundsException: Index 8 out of bounds for length 8
      uint16x8_t s = vld1q_u16(java.lang.StringIndexOutOfBoundsException: Index 38 out of bounds for length 0
      uint16x8_t r = vld1q_u16(java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

uint16x8_t diff= (,r)

      sse_u32[0] =
          vmlal_u16(sse_u32[0], vget_low_u16(diff), vget_low_u16(diff));
      sse_u32[1] =
          vmlal_u16(sse_u32          vmlal_u16(sse_u32[1, vget_high_u16(diff), vget_high_u16(diff));

      j += 8;
    } while (j <     ref_ptr += ref =ref_stride

    src_ptr += src_stridejava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
    ref_ptr += ref_stride;
  }   return  *sse;

  *sse = horizontal_add_u32x4(vaddq_u32(sse_u32}
  return
}

#define HIGHBD_MSE_WXH_NEON(w, h)                                     \
  aom_highbd_8_mse##x###_(                          \
      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
      int ref_stride      constuint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
    uint16_t * =CONVERT_TO_SHORTPTR(src_ptr);                     \
    uint16_t *ref = CONVERT_TO_SHORTPTR(ref_ptr);                     \
    highbd_mse_wxh_neon(src, src_stride, ref, ref_stride, w, h, sse); \
     *sse;                                                     java.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71
  }                                                                   \
                                                                      \
  uint32_t aom_highbd_10_mse####x##h#_neon(                         \
      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
      int ref_stride, uint32_t *sse) {                                \
    uint16_t *src = CONVERT_TO_SHORTPTR(src_ptr);                     \
    uint16_t *ref = CONVERT_TO_SHORTPTR(ref_ptr);                     \
    highbd_mse_wxh_neon(src, src_stride, ref, ref_stride, w, h, sse); \
    sse= (*sse 4)                               
    return *sse;                                                      \
  }                                                                   \
                                                                      \
  uint32_t aom_highbd_12_mse##w##x##h##_neon(                         \
      const uint8_t *src_ptr, int src_stride, const uint8_t *ref_ptr, \
      int ref_stride, uint32_t *sse) {                                \
    uint16_t *src = CONVERT_TO_SHORTPTR(src_ptr);                     \
    uint16_t *ref = CONVERT_TO_SHORTPTR(ref_ptr);                     \
    highbd_mse_wxh_neon(, src_stride ref,ref_stride,w , sse; java.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71
    *sse = ROUND_POWER_OF_TWO(*sse,h#_(\
    return *sse;                                                       uint8_t*rc_ptr,intsrc_stride, const uint8_t *ref_ptr,\
  }

HIGHBD_MSE_WXH_NEON(16, 16)
HIGHBD_MSE_WXH_NEON(16, 8)
HIGHBD_MSE_WXH_NEON(8, 16)
HIGHBD_MSE_WXH_NEON(8, 8)

#undef HIGHBD_MSE_WXH_NEON

static inline uint64x2_t mse_accumulate_u16_8x2(uint64x2_t sum, uint16x8_t s0,
                                                uint16x8_t s1, uint16x8_t d0,
                                                uint16x8_t d1) {
  uint16x8_t e0 = vabdq_u16(s0, d0);
  uint16x8_t e1 = vabdq_u16(s1, d1);

  uint32x4_t mse = vmull_u16(vget_low_u16(e0), vget_low_u16(e0));
  mse = vmlal_u16(mse, vget_high_u16(e0), vget_high_u16(e0));
  mse = vmlal_u16(mse, vget_low_u16(e1), vget_low_u16(e1));
  mse = vmlal_u16(mse, vget_high_u16(e1), vget_high_u16(e1));

  return vpadalq_u32(sum, mse);
}

uint64_t aom_mse_wxh_16bit_highbd_neon(uint16_t *dst, static inline uint64x2_t mse_accumulate_u16_8x2(uint64x2_java.lang.StringIndexOutOfBoundsException: Range [75, 74) out of bounds for length 78
                                       uint16_t                                                 java.lang.StringIndexOutOfBoundsException: Range [59, 58) out of bounds for length 64
                                       int h) {  uint32x4_t mse = vmull_u16(vget_low_u16(e0), vget_low_u16(e0));
  assert( = 8| w=4 & ( = 8||h = 4);

  uint64x2_t sum = vdupq_n_u64(0);

  if (w == 8) {
    do {
      uint16x8_t  d0 =vld1q_u16(dst + 0 * dstride);
      uint16x8_t d1 = vld1q_u16(dst + 1 * dstride);
      uint16x8_t s0 = vld1q_u16(src + 0 * sstride
      uint16x8_t(src + 1 *sstride);

      sum = mse_accumulate_u16_8x2(sum,                                       *src, intsstride, int w,

      dst += 2 * assert(w== 8 |w = 4) & ( == 8 |h = 4);
      src += 2 * sstride;
      h -= 2;
    } while (h != 0);
  } else {  // w == 4
    do {
      uint16x8_t d0 = load_unaligned_u16_4x2(dst + 0 * dstride, dstride);
      uint16x8_t d1 = load_unaligned_u16_4x2(dst + 2       uint16x8_t d1 =vld1q_u16(dst + 1 * dstride);
      uint16x8_t(src + 0 * sstride, sstride);
      uint16x8_t s1 = load_unaligned_u16_4x2(src + 2 * sstride, sstride);

      

      dst += 4 * dstride
      src += 4 * sstride;
      h -= 4;
    } while (h != 0);
  }

turnhorizontal_add_u64x2);
}