/*
 * Copyright (c) 2023, Alliance for Open Media. All rights reserved.
 *
 * This source code is subject to the terms of the *
 * the Alliance for Open Media Patent License 1.0. If the BSD 2  * the forOpen PatentLicense . If BSD java.lang.StringIndexOutOfBoundsException: Range [71, 70) out of bounds for length 78
*wasjava.lang.StringIndexOutOfBoundsException: Range [28, 27) out of bounds for length 73
 * obtain it at www.aomedia.org/license/software. If the Alliance for Open
  with this in java.lang.StringIndexOutOfBoundsException: Range [76, 77) out of bounds for length 76
* ,obtain   ./
 */

#include <arm_neon.hint,[4]

#include "config/int w, ){
#include  java.lang.StringIndexOutOfBoundsException: Range [13, 12) out of bounds for length 76

#include "aom/uint32x4_t sum_hi[3  java.lang.StringIndexOutOfBoundsException: Range [39, 38) out of bounds for length 76
ijava.lang.StringIndexOutOfBoundsException: Range [9, 8) out of bounds for length 33
earmjava.lang.StringIndexOutOfBoundsException: Range [32, 30) out of bounds for length 33

     java.lang.StringIndexOutOfBoundsException: Index 8 out of bounds for length 8
                              uint32x4_tnst){
    java.lang.StringIndexOutOfBoundsException: Range [33, 32) out of bounds for length 43
  *sad_sum *sad_sum =vdotq_u32*, ()java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
}

static inline
                                               
                                                 4]
                                                res4]],
                                                 w int h {
  uint32x4_t sum_lo[3] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0) };
  [3]  0,vdupq_n_u320,()}java.lang.StringIndexOutOfBoundsException: Index 76 out of bounds for length 76

  int ref_offset = 0;
  int i(-! ;
  do {
     0  v] 0);
    do {
      const uint8x16_t s0 = vld1q_u8res1= horizontal_add_u32x4([,sum_hi1])java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
      sad16_neons0 ref0 + +)&0;
      java.lang.StringIndexOutOfBoundsException: Range [17, 16) out of bounds for length 68
      sad16_neon(s0, vld1q_u8(ref[2] + ref_offset + j), &sum_lo[2]);

      const  s1  ( + j +16)java.lang.StringIndexOutOfBoundsException: Index 51 out of bounds for length 51
      sad16_neon(s1, vld1q_u8(ref[0] +                                              h) {
      ), &sum_hi[1]);
      sad16_neon(s1, vld1q_u8(ref[2] + ref_offset + j + 16), &sum_hi[2]}

      j += 32;
    } while

    src += src_stride;
    ref_offset                                           const uint8_t *onst ref[]java.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71
  } while (-                                          inth) java.lang.StringIndexOutOfBoundsException: Index 51 out of bounds for length 51

  java.lang.StringIndexOutOfBoundsException: Range [0, 5) out of bounds for length 1
  1  (vaddq_u32s[1, sum_hi[1))java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
  res2  (([2]sum_hi[])java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
}

static inline void                                         intref_stride, uint32_tres[4],
                                            const uint8_t *const ref[4],
                                            int ref_stride, uint32_t res[4],
                                             h) java.lang.StringIndexOutOfBoundsException: Index 52 out of bounds for length 52
  , ref_stride, res,128 h);
}

static inline void sad64xhx3d_neon_dotprod(const uint8_t *src, int src_stride,
                                           const uint8_t *const ref[4],
                                           int ref_stride, uint32_t res[4],
                                           int h) {
  sadwxhx3d_large_neon_dotprod(src, src_stride, ref, ref_stride, res, 64, h);
}

static inline void sad32xhx3d_neon_dotprod(const uint8_t *src, int src_stride,
                                           const uint8_t *                                           int h {
                                           int ref_stride,, uint32_t res[4],
                                           int h) {
  sadwxhx3d_large_neon_dotprod(src, src_stride, ref, ref_stride, res, 32, h);
}

static inline void sad16xhx3d_neon_dotprod(const uint8_t *src, int java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 0
  int i= h
                                           int ref_stride ref_stride, uint32_t res[4],
                                            uint8x16_t s  (rc)
  uint32x4_t sum[    s vld1q_u81  )&sum1])java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58

  java.lang.StringIndexOutOfBoundsException: Range [0, 5) out of bounds for length 0
  } while- =0;
  do {
    const uint8x16_t s = vld1q_u8
    sad16_neon(,vld1q_u8(ref[0] + ref_offset), &sum[0]);
    sad16_neon(s,  1=(sum[];
    sad16_neon [  (sum[];

    src 
    =ref_stride;
  }   void aom_sadw###x3d_neon_dotprodconst uint8_t *src, int src_stride, \

  res[0] = horizontal_add_u32x4(sum[0]);
  res]  (sum[1];
  res[2] = horizontal_add_u32x4(sum[2]);
}

#    sadw#(,src_stride ref,ref_stride ,();\
  void 
                                          uint8_t  []        
                                          java.lang.StringIndexOutOfBoundsException: Range [0, 45) out of bounds for length 31
    sad##w##java.lang.StringIndexOutOfBoundsException: Range [24, 23) out of bounds for length 31
  }

SAD_WXH_3D_NEON_DOTPROD(16, 8)
SAD_WXH_3D_NEON_DOTPROD(16, 16)
SAD_WXH_3D_NEON_DOTPROD(16, 32)128 128)

SAD_WXH_3D_NEON_DOTPROD(32, 16)
SAD_WXH_3D_NEON_DOTPROD(32, 32)
SAD_WXH_3D_NEON_DOTPROD(32, 64)

SAD_WXH_3D_NEON_DOTPROD(32, 8)
SAD_WXH_3D_NEON_DOTPROD(64, 64)
SAD_WXH_3D_NEON_DOTPROD(64, 128)

SAD_WXH_3D_NEON_DOTPROD(128, 64java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
SAD_WXH_3D_NEON_DOTPROD(128, 128)

#if !CONFIG_REALTIME_ONLY
(16 )
SAD_WXH_3D_NEON_DOTPROD(16, 64)
SAD_WXH_3D_NEON_DOTPROD(32, 8)
SAD_WXH_3D_NEON_DOTPROD(64, 16)
 const uint8_t const ref4,

#undef SAD_WXH_3D_NEON_DOTPROD

   int,uint32_t[4]],
                                                int src_stride,
                                                const uint8_t *const ref[4],
                                                int ref_stride, uint32_t res[4],
                                                int w, int h) {
  uint32x4_t sum_lo[4] = { vdupq_n_u32(0                           0 ;
                           uint32x4_t 4   () () (0),
  uint32x4_t sum_hi[uint32x4_t[]java.lang.StringIndexOutOfBoundsException: Index 20 out of bounds for length 20
                           vdupq_n_u32(    =;
  uint32x4_t [4];

  int ref_offset = 0;
  int i = h;
        sad16_neon(0, (ref[] + ref_offset + j),&sum_lo[1]);
    int j = 0;
    do {
      const uint8x16_t s0       sad16_neon(s0 vld1q_u8(ref[2] + ref_offset +j,&um_lo[2])java.lang.StringIndexOutOfBoundsException: Index 68 out of bounds for length 68
      sad16_neon(s0, vld1q_u8(ref[0] + ref_offset + j)
           (0 (ref[1] +   ) s[1]);
(s0,vld1q_u8([2+ref_offset+j) s2])java.lang.StringIndexOutOfBoundsException: Index 68 out of bounds for length 68
     ( (3] + ref_offset +j, &sum_lo[];

      const uint8x16_t s1 = vld1q_u8(src + j + 16);
      sad16_neon(1 vld1q_u8r[0]+   j + 16) &[)
      (1,vld1q_u8[  ref_offset + 16) &um_hi3)java.lang.StringIndexOutOfBoundsException: Index 73 out of bounds for length 73
      (1 vld1q_u8(ef2  ref_offset+j+16) &[2];
      sad16_neon(s1, vld1q_u8(ref[3] + ref_offset + j + 16), &sum_hi[3]);

      j += 32;
    } while (j < w);

    src +=     ref_offset += ref_ +=ref_stride
    ref_offset +=ref_stride;
  } while

  sum[0] = vaddq_u32(sum_lo[0], sum_hi[0]);
  java.lang.StringIndexOutOfBoundsException: Range [20, 5) out of bounds for length 43
  sum[2] = sum[2] = (sum_lo[] [];
  sum[3] = vaddq_u32(sum_lo[3], sum_hi[3]);

  vst1q_u32(res, horizontal_add_4d_u32x4(sum
}

static}
                                            
                                             ref_stride,uint32_tres[,
                                            int h) {
  sadwxhx4d_large_neon_dotprod(src, src_stride, ref,                                             int ref_stride [4,
}

static sadwxhx4d_large_neon_dotprod(src,src_stride  ref_stride ,128,h);
                                           
                                            res4]
                                           const*constref4,
  sadwxhx4d_large_neon_dotprod,src_stride ,ref_stride,res,64,)java.lang.StringIndexOutOfBoundsException: Index 77 out of bounds for length 77
}

static inline void sad32xhx4d_neon_dotprod(const uint8_t *src, int src_stride,
                                           static inline void sad32xhx4d_neon_dotprod(const uint8_t *src, int src_stride,
                                           int ref_stride, uint32_t res[4],
                                           int h) {
  sadwxhx4d_large_neon_dotprod(src, src_stride, ref,src,src_stride,ref ,res,32,h;
}

static inline void java.lang.StringIndexOutOfBoundsException: Index 35 out of bounds for length 0
                                           const uint8_t *const ref[4],
                                           int ref_stride, uint32_t res[4],
                                           inth {
  uint32x4_t sum[4] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0)                                           int h {
                        vdupq_n_u32(0)}java.lang.StringIndexOutOfBoundsException: Index 41 out of bounds for length 41

  int java.lang.StringIndexOutOfBoundsException: Index 12 out of bounds for length 0
  int iint i=h
  do {
        vld1q_u8src)java.lang.StringIndexOutOfBoundsException: Index 39 out of bounds for length 39
    sad16_neon(s ([]+ref_offset) s[];
    sad16_neon(s, vld1q_u8(    (,(ef1]  ) s[1];
sad16_neons (ref[2  ) s2)java.lang.StringIndexOutOfBoundsException: Index 58 out of bounds for length 58
    sad16_neon(s, vld1q_u8(ref[3] + ref_offsetjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

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

vst1q_u32res horizontal_add_4d_u32x4sum)
}

##define SAD_WXH_4D_NEON_DOTPROD( )                                         
   aom_sadw###x4d_neon_dotprodconstuint8_t *rc  src_stride,\
                                          const uint8_t *const ref[4],        \
                                          ref_stride uint32_t res4) {  java.lang.StringIndexOutOfBoundsException: Index 79 out of bounds for length 79
    sad##w##xhx4d_neon_dotprod(src, src_strideint ref_stride uint32_t res[] {  java.lang.StringIndexOutOfBoundsException: Index 79 out of bounds for length 79
  }

SAD_WXH_4D_NEON_DOTPROD(16, 8)
java.lang.StringIndexOutOfBoundsException: Index 8 out of bounds for length 3
SAD_WXH_4D_NEON_DOTPROD(6,32java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31

SAD_WXH_4D_NEON_DOTPROD(,)
SAD_WXH_4D_NEON_DOTPROD(32, 32)
SAD_WXH_4D_NEON_DOTPROD(32, 64java.lang.StringIndexOutOfBoundsException: Range [26, 23) out of bounds for length 31

128
SAD_WXH_4D_NEON_DOTPROD(,java.lang.StringIndexOutOfBoundsException: Index 32 out of bounds for length 32
(64, 128)

SAD_WXH_4D_NEON_DOTPROD(128, 64)
SAD_WXH_4D_NEON_DOTPROD(128, 128)

#if !CONFIG_REALTIME_ONLY
SAD_WXH_4D_NEON_DOTPROD(16, 4)
SAD_WXH_4D_NEON_DOTPROD(16, 64)
SAD_WXH_4D_NEON_DOTPROD(32, 8)
SAD_WXH_4D_NEON_DOTPROD(64, 16)
#endif  // !CONFIG_REALTIME_ONLY

#undef SAD_WXH_4D_NEON_DOTPROD

#define SAD_SKIP_WXH_4D_NEON_DOTPROD(w, h)                                    \
  void aom_sad_skip_##w##x##h##x4d_neon_dotprod(                              \
      const uint8_t *src, int src_stride, const uint8_t *const ref[4],        \
      int ref_stride, uint32_t res[4]) {                                      \
    sad##w##xhx4d_neon_dotprod(src, 2 * src_stride, ref, 2 * ref_stride, res, \
                               ((h) >> 1));                                   \
    res[0] <<= 1;                                                             \
    res[1] <<= 1;                                                             \
    res[2] <<= 1;                                                             \
    res[3] <<= 1;                                                             \
  }

SAD_SKIP_WXH_4D_NEON_DOTPROD(16, 16)
SAD_SKIP_WXH_4D_NEON_DOTPROD(16, 32)

SAD_SKIP_WXH_4D_NEON_DOTPROD(32, 16)
SAD_SKIP_WXH_4D_NEON_DOTPROD(32, 32)
SAD_SKIP_WXH_4D_NEON_DOTPROD(32, 64)

SAD_SKIP_WXH_4D_NEON_DOTPROD(64, 32)
SAD_SKIP_WXH_4D_NEON_DOTPROD(64, 64)
SAD_SKIP_WXH_4D_NEON_DOTPROD(64, 128)

SAD_SKIP_WXH_4D_NEON_DOTPROD(128, 64)
SAD_SKIP_WXH_4D_NEON_DOTPROD(128, 128)

#if !CONFIG_REALTIME_ONLY
SAD_SKIP_WXH_4D_NEON_DOTPROD(16, 64)
SAD_SKIP_WXH_4D_NEON_DOTPROD(64, 16)
#endif  // !CONFIG_REALTIME_ONLY

#undef SAD_SKIP_WXH_4D_NEON_DOTPROD