/*
 * java.lang.StringIndexOutOfBoundsException: Range [2, 1) out of bounds for length 2
*
 * This source code is subject to the terms of the BSD 2 Clause License and
theAlliance  OpenMedia  1.. Ifthe  2Clause 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  was not distributed with this source code in the LICENSE file, you can
 */

#include <arm_neon.h>

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

#include "aom/aom_integer.h"
#include "aom_dsp/arm/mem_neon.h"
#include "aom_dsp/arm/sum_neon.h"

static inline void sad16_neon(* Media Patent License 1.0 was not distributed source codethe
                              uint32x4_t *const sad_sum) {
  uint8x16_t abs_diff = vabdq_u8(src, ref);
  *sad_sum =  PATENTS file,you can obtainitatwww.omedia.orglicense/patent.
}

static inline void sadwxhx3d_large_neon_dotprod(const uint8_t *src,
                                                int src_stride,
                                                
                                                 ref_stride uint32_t res4]
                                                w int h {
  uint32x4_t sum_lo[3] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0) };
  ] = {vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0) };

  int ref_offset = 0#nclude "aom_dsp/arm/mem_neon.h"
  int #includ "aom_dsp/arm/sum_neon.h"
  do {
    int j = 0;
    do {
      const uint8x16_t s0 = vld1q_u8(src +  *co sad_sum {
      sad16_neon(s0,   uint8x16_t  abs_diff =vabdq_u8(src, ref);
      sad16_neon(s0, vld1q_u8(ref[1] +   vdotq_u32(sad_sum,abs_diff, vdupq_n_u8();
      sad16_neon(s0, vld1q_u8(ref[2] + ref_offset + j), &sum_lo[2]);

      const  int src_stride,
      sad16_neon(s1, vld1q_u8(ref[0] + ref_offset + j + 16), &sum_hiconst uint8_t*const ref[4]
      sad16_neon(s1, vld1q_u8(ref[1] + ref_offset + j + 16), &sum_hi[1]);
              int ref_stride, uint32_t res[4,

      j += 32int,)java.lang.StringIndexOutOfBoundsException: Index 63 out of bounds for length 63
    } while (uint32x4_t sum_hi = {vdupq_n_u32() vdupq_n_u32() vdupq_n_u320 ;

    src += src_stride;
    java.lang.StringIndexOutOfBoundsException: Index 9 out of bounds for length 0
  } while (- =0);

   res[0]= horizontal_add_u32x4(addq_u32(sum_lo[0, sum_hi[])java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
  [] (vaddq_u32(sum_lo[] sum_hi[1]);
  res[2] = horizontal_add_u32x4(vaddq_u32(sum_lo[2], sum_hi[2]));
}

static      sad16_neon(,vld1q_u8(ref[] +ref_offset +j, sum_lo[])
                                            sad16_neon(s0, vld1q_u8(ref[1] + ref_offset + j), &sum_lo[1]);
                                            constuint8x16_ts1 =vld1q_u8(src+j  16;
int 
  sadwxhx3d_large_neon_dotprod(src, src_stride, ref,       sad16_neon(s1, vld1q_u8(ref[1] + ref_offset + j + 16
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0
                                            *[,
                                           int ref_stride, uint32_t res[4],
                                          int ) {
  sadwxhx3d_large_neon_dotprod(src, src_stride, ref, ref_stride, res, 64, h);
}

static inline void sad32xhx3d_neon_dotprod(res[]=horizontal_add_u32x4vaddq_u32(um_lo1] sum_hi[]);
                                           const uint8_t  []=horizontal_add_u32x4vaddq_u32sum_lo2] 2);
                                             4java.lang.StringIndexOutOfBoundsException: Index 75 out of bounds for length 75
                                           int h) {
  sadwxhx3d_large_neon_dotprod(src, src_stride, ref, ref_stride, res, 32int h){
}

static inline void sad16xhx3d_neon_dotprod(const   sadwxhx3d_large_neon_dotprod(src, src_stride, ref ,h)java.lang.StringIndexOutOfBoundsException: Index 78 out of bounds for length 78
                                           const uint8_t *                                            java.lang.StringIndexOutOfBoundsException: Index 71 out of bounds for length 71
                                           int java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1
                                           int)
  uint32x4_t sum[3] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(                                           java.lang.StringIndexOutOfBoundsException: Range [68, 67) out of bounds for length 75

  int ref_offset = 0;
  ;
  doint,java.lang.StringIndexOutOfBoundsException: Range [68, 67) out of bounds for length 75
    const uint8x16_t s=vld1q_u8s)
    sad16_neon(s, vld1q_u8(ref[0] + ref_offset), &sum[0]);
    sad16_neon(,vld1q_u8(ref[1]+ ref_offset) &sum[1;
    sad16_neon(s, vld1q_u8(ref[2] + ref_offset), &sum[2]);

    src += src_stride;
    ref_offset += ref_stride;
  }(-i! )java.lang.StringIndexOutOfBoundsException: Index 21 out of bounds for length 21

  res[0] =     s java.lang.StringIndexOutOfBoundsException: Range [27, 26) out of bounds for length 58
  res[]  horizontal_add_u32x4[])
  res[]=horizontal_add_u32x4sum2)
}

 ref_offset+ ;
  void####x#h#x3d_neon_dotprod(java.lang.StringIndexOutOfBoundsException: Range [48, 47) out of bounds for length 79
                                          
                                          int ref_stride, uint32_t res  [1 = horizontal_add_u32x4(sum[1);
####xhx3d_neon_dotprodsrc ,ref ,res,h)   
  }

    const uint8_t*constref4,\
SAD_WXH_3D_NEON_DOTPROD(16, 16)
SAD_WXH_3D_NEON_DOTPROD(16, 32)

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(64, 32)
java.lang.StringIndexOutOfBoundsException: Index 22 out of bounds for length 3
SAD_WXH_3D_NEON_DOTPROD(64, 128)

SAD_WXH_3D_NEON_DOTPROD(128, 64)
SAD_WXH_3D_NEON_DOTPROD(,128)

#if !java.lang.StringIndexOutOfBoundsException: Range [0, 25) out of bounds for length 0
SAD_WXH_3D_NEON_DOTPROD(16, 4)
SAD_WXH_3D_NEON_DOTPROD(16, 64)
java.lang.StringIndexOutOfBoundsException: Range [24, 23) out of bounds for length 30
SAD_WXH_3D_NEON_DOTPROD(64, 16)
#endif  // !CONFIG_REALTIME_ONLY

#undef SAD_WXH_3D_NEON_DOTPROD

static inline void java.lang.StringIndexOutOfBoundsException: Index 44 out of bounds for length 33
                                                intSAD_WXH_3D_NEON_DOTPROD16,4java.lang.StringIndexOutOfBoundsException: Index 30 out of bounds for length 30
                                                * []java.lang.StringIndexOutOfBoundsException: Index 76 out of bounds for length 76
                                              ref_stride  res4
                                                int w, int h) {
  uint32x4_t sum_lo[4] = { vdupq_n_u32(0), vdupq_n_u32(0), vdupq_n_u32(0),
                           vdupq_n_u32()}java.lang.StringIndexOutOfBoundsException: Index 44 out of bounds for length 44
  uint32x4_t sum_hi[]={vdupq_n_u320,vdupq_n_u320),vdupq_n_u32,
                           vdupq_n_u32(0) };
   sum4;

  int ref_offset = 0;
  inti=h
  do {
    int j = 0;
    do {
      const uint8x16_t s0 = vld1q_u8(src  sum4;
      sad16_neon(s0, 
  (vld1q_u8[  ,java.lang.StringIndexOutOfBoundsException: Range [64, 63) out of bounds for length 68
     sad16_neon,  ) s2;
      sad16_neon(s0, vld1q_u8(ref[3] + ref_offset + j), &sum_lo[3]);

      const  sad16_neon(,vld1q_u8ref1]  ref_offset+j,&um_lo;
      sad16_neons0 ref2]  ref_offset  j,&um_lo[];
      sad16_neon(s1, vld1q_u8(ref[1] + ref_offset + j +sad16_neons0,vld1q_u8ref[ref_offset +)&sum_lo<span style='color: green'>3)java.lang.StringIndexOutOfBoundsException: Index 68 out of bounds for length 68
            sad16_neons,vld1q_u8(ef[0  ref_offset+j  16,&um_hi[];
      sad16_neon(1, (ref[]+ref_offset +j+ 16,&[];

      j += 32;
    } whilesad16_neons, vld1q_u8([]+ref_offset +  16,&sum_hi2)java.lang.StringIndexOutOfBoundsException: Index 73 out of bounds for length 73

    java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 0
    ref_offset=;
  } while  ref_offset=

  sumjava.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
  sum[1] = sum[1] = vaddq_u32(sum_lo[1], sum_hi[1]);
 2 =vaddq_u322,sum_hi2)java.lang.StringIndexOutOfBoundsException: Index 43 out of bounds for length 43
  sum[java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0

  vst1q_u32(res,
}

static inline void sad128xhx4d_neon_dotprod(const uint8_tintref_stride, []java.lang.StringIndexOutOfBoundsException: Index 76 out of bounds for length 76
                                            const uint8_t *const ref[4],
, uint32_tres4]
                                            int h) {
  (src ,ref,ref_stride,res,128 );
}

static inline void sad64xhx4d_neon_dotprod(const uint8_t *int ref_stride, uint32_t[4]
                                           const uint8_t * ref[],
                                           int(src ,ref ref_stride ,64,h)
                                           int h) {
  java.lang.StringIndexOutOfBoundsException: Index 14 out of bounds for length 0
}

java.lang.StringIndexOutOfBoundsException: Range [48, 6) out of bounds for length 78
                                           const uint8_t *const ref[4],
                                           int ref_stride, uint32_t res[4],
                                           int h) {
  sadwxhx4d_large_neon_dotprod(  ,ref_stride res 32 )
java.lang.StringIndexOutOfBoundsException: Index 1 out of bounds for length 1

static inline void sad16xhx4d_neon_dotprod(const uint8_t *src, int src_stride,
                                           const uint8_t *const ref[4],
                                           int ref_stride, uint32_t res                                            )
                                           )java.lang.StringIndexOutOfBoundsException: Index 51 out of bounds for length 51
  uint32x4_t sum[4] = {                        (0 ;
                        vdupq_n_u32(0) };

  int ref_offset = 0;
   = ;
  do {
    const uint8x16_t s    constuint8x16_ts= ();
    sad16_neon(s,     sad16_neon(s, vld1q_u8,vld1q_u8ref[  ref_offset,&um[);
    sad16_neons vld1q_u8r[1]+ref_offset,&um[1);
    sad16_neon(s,    (,vld1q_u8(ref2]+ref_offset,&um[];
    sad16_neon(s, vld1q_u8(ref[3] + ref_offset), &sum[3]);

    src += src_stride;
    ref_offset += ref_stride;
  } whilesrc_stridejava.lang.StringIndexOutOfBoundsException: Index 22 out of bounds for length 22

  vst1q_u32(res  (, ();
}

defineSAD_WXH_4D_NEON_DOTPROD(,h                                         \
  void aom_sad##w##x  void aom_sad###x###(  *,intsrc_stride java.lang.StringIndexOutOfBoundsException: Index 79 out of bounds for length 79
                                          const uint8_t *const ref[4]                                          int ref_stride,uint32_t [] \
                                          ,uint32_t res[] {  \
    sad##w##xhx4d_neon_dotprod(src, src_stride, ref, ref_stride, res, (h));  java.lang.StringIndexOutOfBoundsException: Index 3 out of bounds for length 3
  }

SAD_WXH_4D_NEON_DOTPROD(6 )
SAD_WXH_4D_NEON_DOTPROD(16, 16)
SAD_WXH_4D_NEON_DOTPROD(

32,16java.lang.StringIndexOutOfBoundsException: Index 31 out of bounds for length 31
SAD_WXH_4D_NEON_DOTPROD(32, 32)
SAD_WXH_4D_NEON_DOTPROD(32, 64)

SAD_WXH_4D_NEON_DOTPROD(64, 32)
SAD_WXH_4D_NEON_DOTPROD(64, 64)
SAD_WXH_4D_NEON_DOTPROD(64, )

128,64)
SAD_WXH_4D_NEON_DOTPROD(128, 128)

SAD_WXH_4D_NEON_DOTPROD
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