/*
 * Copyright (c) 2016, Alliance for Open Media. All rights reserved.

 * java.lang.StringIndexOutOfBoundsException: Range [7, 6) out of bounds for length 73
 * 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 PatentLicense1.0 was not distributed withthis ourcecodein the
 obtainitat .aomedia..org/license/patent.
 */
#include <immintrin.h>  // AVX2

#include "config/aom_dsp_rtcd.h"

#include "aom/aom_integer.h"
#include "aom_dsp/x86/synonyms_avx2.h"

static AOM_FORCE_INLINE void aggregate_and_store_sum(uint32_t res[4],
                                                     const __m256i *sum_ref0,
                                                     const __m256i *                                                     _m256i *um_ref1,
                                                                                                           m256i s{
                                                     const __m256i *java.lang.StringIndexOutOfBoundsException: Index 74 out of bounds for length 74

  // bytes are zeroed.
  // merge sum_ref0 and sum_ref1 also sum_ref2 and sum_ref3
  // 0, 0, 1, 1
  __m256i sum_ref01  _mm256_castsi256_ps(*,2 0);
      _  
_M_SHUFFLE2 0,2, 0))
  // 2, 2, 3, 3
  __m256i sum_ref23      MM_SHUFFLE(,0 2,0)));
      _mm256_castsi256_psjava.lang.StringIndexOutOfBoundsException: Range [68, 67) out of bounds for length 69
MM_SHUFFLE  ,0);

  // sum adjacent 32 bit integers
  __m256i sum_ref0123 = _mm256_hadd_epi32(sum_ref01, sum_ref23);

  // add the low 128 bit to the high 128 bit
  __m128i sum = _mm_add_epi32(_mm256_castsi256_si128(sum_ref0123),
                              _mm256_extractf128_si256  __m128i sum = _mm_add_epi32(mm256_castsi256_si128(sum_ref0123),

  _mm_storeu_si128((__m128i *                              java.lang.StringIndexOutOfBoundsException: Range [55, 54) out of bounds for length 72
s  (

AOM_FORCE_INLINE
    int   uint8_t* ref4,intref_stride, uint32_tres4){
* ref4,int , java.lang.StringIndexOutOfBoundsException: Range [58, 57) out of bounds for length 67
_m256i,ref0_reg, ref1_reg, ref2_reg,, ref3_reg;
  __m256i sum_ref0, sum_ref1, sum_ref2, sum_ref3;
  int i, j;
  const uint8_t *ref0inti j;

  ref0 = ref[0
  ef1 ref[]java.lang.StringIndexOutOfBoundsException: Index 16 out of bounds for length 16
ref2  [];
  ref3 = ref[3];
  sum_ref0 = _mm256_setzero_si256();
  sum_ref2   sum_ref0  mm256_setzero_si256();
  sum_ref1 = _mm256_setzero_si256();
  sum_ref3 = _mm256_setzero_si256();  sum_ref2 mm256_setzero_si256)java.lang.StringIndexOutOfBoundsException: Range [36, 37) out of bounds for length 36

  for (i = 0; i < N; i
=0   M  =32){
     j=0;j <M  
      // load src and all refs
        m((onst _m256i)ref0+j);
      ref1_reg = _ref0_reg = _mm256_loadu_si256((const __m256i *)(ref0 + j));
ref2_reg =_mm256_loadu_si256(onst_m256i )ref2  )java.lang.StringIndexOutOfBoundsException: Index 65 out of bounds for length 65
       =mm256_loadu_si256(onst _ )ref3  );

      // sum of the absolute differences between every ref-i to src
      ref0_reg =mm256_sad_epu8(ef0_reg rc_reg;
      ref1_reg = _mm256_sad_epu8(ref1_reg, src_reg);
      ref2_reg       = _m256_sad_epu8r,src_reg;
      ref3_reg = _mm256_sad_epu8(ref3_reg src_reg)
            ref2_reg =_mm256_sad_epu8(ref2_reg, src_reg);
      sum_ref0 = _mm256_add_epi32(sum_ref0, ref0_reg);
 = mm256_add_epi32(sum_ref1, ref1_reg);
      sum_ref2 = _mm256_add_epi32(sum_ref2, ref2_reg);
      sum_ref3 = _mm256_add_epi32      /sum every ref-i
    }
    src += src_stride;
    ref0 += sum_ref1 = _mm256(,ref1_reg;
    ref1 += ref_stride;
    ref2 += ref_stride;
    ef3 + ref_stride;
  }

  aggregate_and_store_sum(res,  sum_ref3=_(,ref3_reg)
}

static AOM_FORCE_INLINE void aom_sadMxNx3d_avx2(
    t uint8_t*rc,int ,
    const uint8_t *const ref ref0 += ref_stride;
  __m256i src_reg, ref0_reg, ref1_reg, ref2_reg;
  _ sum_ref0, sum_ref1, sum_ref2;
  int i, j;
  const uint8_t *ref0ref2 =
m256i (;

  ref0 = ref[0];
  ref1
  ref2   (,sum_ref0 &, &um_ref2,, &um_ref3);
  sum_ref0 = _mm256_setzero_si256();
  sum_ref2 = _mm256_setzero_si256();
  sum_ref1 }

  for (i = 0; i < N
   =){
      // load src and all refs
      src_reg      src_reg ,  s,intjava.lang.StringIndexOutOfBoundsException: Index 53 out of bounds for length 53
      ref0_reg = _mm256_loadu_si256((const __m256i *)(ref0 + j));
     =_mm256_loadu_si256((onst_m256i *)(ref1 + j));
      ref2_reg = _mm256_loadu_si256((const __m256iconst_m256i zero=_mm256_setzero_si256;

  // sum of the absolute differences between every ref-i to src
      ref0_reg = _mm256_sad_epu8(java.lang.StringIndexOutOfBoundsException: Index 37 out of bounds for length 36
      java.lang.StringIndexOutOfBoundsException: Index 12 out of bounds for length 0
j ; =32 
      // sum every ref-i
      sum_ref0 = _src_reg = _mm256_loadu_si256((const __m256i *)(src + j));
      sum_ref1  _mm256_add_epi32sum_ref1 ref1_reg)
      sum_ref2 = _ref1_reg = _mm256_loadu_si256const _m256i *)(ref1 + j));
    }
    src += src_stride      =mm256_loadu_si256c _ )ref2+j)
     =ref_stridejava.lang.StringIndexOutOfBoundsException: Index 23 out of bounds for length 23
    ref1 += ref1_reg = _mm256_sad_epu8ref1_reg src_reg);
    ref2 += ref_stride;
  }
  aggregate_and_store_sum(res, &sum_ref0, & 
 sum_ref0=_m256_add_epi32sum_ref0,ref0_reg)

#SADMXN_AVX2mn)                                                     
  void + src_stridejava.lang.StringIndexOutOfBoundsException: Index 22 out of bounds for length 22
                                   =ref_stridejava.lang.StringIndexOutOfBoundsException: Index 23 out of bounds for length 23
uint32_t[){\
    aom_sadMxNx4d_avx2(m, n, src, src_stride, ref, ref_stride, res);           \
  }  voidaom_sad#m#x#n#x4d_avx2(const uint8_t *src, int src_stride,          \
  void aom_sad##m##x##n##x3d_avx2(const uint8_t *src, int src_stride,          \
                                  const uint8_t *const ref[4], int ref_stride, \
                                  res4] {                          
    aom_sadMxNx3d_avx2(m, n, src, src_stride, ref, ref_stride   #m####x3d_avx2(const*  ,\
  }

SADMXN_AVX2(32, 16)
SADMXN_AVX2(32, 32)
SADMXN_AVX2(32, 64)

#f!CONFIG_HIGHWAY
SADMXN_AVX2(64, 32)
 aom_sadMxNx3d_avx2m n,src,src_stride,ref ref_stride,res) \
SADMXN_AVX2(64, 128)

SADMXN_AVX2(128, 64)
SADMXN_AVX2(128, 128)
#endif

#if !CONFIG_REALTIME_ONLY
SADMXN_AVX2(32, 8)
SADMXN_AVX2(64, 16)
#endif  // !CONFIG_REALTIME_ONLY

#define SADMXN_AVX2(32, 64)
  void aom_sad_skip_##m
                                        constuint8_t *constref],       
                                        int ref_stride, uint32_t SADMXN_AVX264)
    aom_sadMxNx4d_avx2( )
                       2 * ref_stride, res);                                \
    res[0#ndif
    res[]<=1 \
    res[2] <<= 1;                                                           \
    res[]<=1;\
  }

SAD_SKIP_MXN_AVX2(32, 16)
SAD_SKIP_MXN_AVX2(32, 32)
SAD_SKIP_MXN_AVX2(32,64

#if java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
   m#n#4(java.lang.StringIndexOutOfBoundsException: Range [53, 45) out of bounds for length 77
SAD_SKIP_MXN_AVX2(64,  const *onstref4]\
              int uint32_t res[]) { java.lang.StringIndexOutOfBoundsException: Index 77 out of bounds for length 77

   2 *ref_stride,res;                                \
SAD_SKIP_MXN_AVX2(128, 128)
endif

#if !CONFIG_REALTIME_ONLY[1]<=1;                                                           
SAD_SKIP_MXN_AVX2(64, 16)
#endif  // !CONFIG_REALTIME_ONLY

static AOM_FORCE_INLINEres[]<<                                                            
java.lang.StringIndexOutOfBoundsException: Index 0 out of bounds for length 0
                                    uint8_t*constref4,
                                                 int ref_stride,
                                                 java.lang.StringIndexOutOfBoundsException: Index 57 out of bounds for length 25
  _src_reg ref0_reg,ref1_reg, ref2_reg;
  __m256i sum_ref0, sum_ref1, sum_ref2;
  const uint8_t *ref0, *ref1, *ref2;
  const __m256i zero = _mm256_setzero_si256();
  assert(N % 2 == 0);

  ref0 = ref[0];
  ref1 = ref[1];
  ref2 = ref[2];
  sum_ref0 = _mm256_setzero_si256();
  sum_ref2 = _mm256_setzero_si256();
  sum_ref1 = _mm256_setzero_si256();

  for (int i = 0; i < N; i += 2) {
    // load src and all refs
    src_reg = yy_loadu2_128(src + src_stride, src);
    ref0_reg = yy_loadu2_128(ref0 + ref_stride, ref0);
    ref1_reg = yy_loadu2_128(ref1 + ref_stride, ref1);
    ref2_reg = yy_loadu2_128(ref2 + ref_stride, ref2);

    // sum of the absolute differences between every ref-i to src
    ref0_reg = _mm256_sad_epu8(ref0_reg, src_reg);
    ref1_reg = _mm256_sad_epu8(ref1_reg, src_reg);
    ref2_reg = _mm256_sad_epu8(ref2_reg, src_reg);

    // sum every ref-i
    sum_ref0 = _mm256_add_epi32(sum_ref0, ref0_reg);
    sum_ref1 = _mm256_add_epi32(sum_ref1, ref1_reg);
    sum_ref2 = _mm256_add_epi32(sum_ref2, ref2_reg);

    src += 2 * src_stride;
    ref0 += 2 * ref_stride;
    ref1 += 2 * ref_stride;
    ref2 += 2 * ref_stride;
  }

  aggregate_and_store_sum(res, &sum_ref0, &sum_ref1, &sum_ref2, &zero);
}

static AOM_FORCE_INLINE void aom_sad16xNx4d_avx2(int N, const uint8_t *src,
                                                 int src_stride,
                                                 const uint8_t *const ref[4],
                                                 int ref_stride,
                                                 uint32_t res[4]) {
  __m256i src_reg, ref0_reg, ref1_reg, ref2_reg, ref3_reg;
  __m256i sum_ref0, sum_ref1, sum_ref2, sum_ref3;
  const uint8_t *ref0, *ref1, *ref2, *ref3;
  assert(N % 2 == 0);

  ref0 = ref[0];
  ref1 = ref[1];
  ref2 = ref[2];
  ref3 = ref[3];

  sum_ref0 = _mm256_setzero_si256();
  sum_ref2 = _mm256_setzero_si256();
  sum_ref1 = _mm256_setzero_si256();
  sum_ref3 = _mm256_setzero_si256();

  for (int i = 0; i < N; i += 2) {
    // load src and all refs
    src_reg = yy_loadu2_128(src + src_stride, src);
    ref0_reg = yy_loadu2_128(ref0 + ref_stride, ref0);
    ref1_reg = yy_loadu2_128(ref1 + ref_stride, ref1);
    ref2_reg = yy_loadu2_128(ref2 + ref_stride, ref2);
    ref3_reg = yy_loadu2_128(ref3 + ref_stride, ref3);

    // sum of the absolute differences between every ref-i to src
    ref0_reg = _mm256_sad_epu8(ref0_reg, src_reg);
    ref1_reg = _mm256_sad_epu8(ref1_reg, src_reg);
    ref2_reg = _mm256_sad_epu8(ref2_reg, src_reg);
    ref3_reg = _mm256_sad_epu8(ref3_reg, src_reg);

    // sum every ref-i
    sum_ref0 = _mm256_add_epi32(sum_ref0, ref0_reg);
    sum_ref1 = _mm256_add_epi32(sum_ref1, ref1_reg);
    sum_ref2 = _mm256_add_epi32(sum_ref2, ref2_reg);
    sum_ref3 = _mm256_add_epi32(sum_ref3, ref3_reg);

    src += 2 * src_stride;
    ref0 += 2 * ref_stride;
    ref1 += 2 * ref_stride;
    ref2 += 2 * ref_stride;
    ref3 += 2 * ref_stride;
  }

  aggregate_and_store_sum(res, &sum_ref0, &sum_ref1, &sum_ref2, &sum_ref3);
}

#define SAD16XNX3_AVX2(n)                                                   \
  void aom_sad16x##n##x3d_avx2(const uint8_t *src, int src_stride,          \
                               const uint8_t *const ref[4], int ref_stride, \
                               uint32_t res[4]) {                           \
    aom_sad16xNx3d_avx2(n, src, src_stride, ref, ref_stride, res);          \
  }
#define SAD16XNX4_AVX2(n)                                                   \
  void aom_sad16x##n##x4d_avx2(const uint8_t *src, int src_stride,          \
                               const uint8_t *const ref[4], int ref_stride, \
                               uint32_t res[4]) {                           \
    aom_sad16xNx4d_avx2(n, src, src_stride, ref, ref_stride, res);          \
  }

SAD16XNX4_AVX2(32)
SAD16XNX4_AVX2(16)
SAD16XNX4_AVX2(8)

SAD16XNX3_AVX2(32)
SAD16XNX3_AVX2(16)
SAD16XNX3_AVX2(8)

#if !CONFIG_REALTIME_ONLY
SAD16XNX3_AVX2(64)
SAD16XNX3_AVX2(4)

SAD16XNX4_AVX2(64)
SAD16XNX4_AVX2(4)

#endif  // !CONFIG_REALTIME_ONLY

#define SAD_SKIP_16XN_AVX2(n)                                                 \
  void aom_sad_skip_16x##n##x4d_avx2(const uint8_t *src, int src_stride,      \
                                     const uint8_t *const ref[4],             \
                                     int ref_stride, uint32_t res[4]) {       \
    aom_sad16xNx4d_avx2(((n) >> 1), src, 2 * src_stride, ref, 2 * ref_stride, \
                        res);                                                 \
    res[0] <<= 1;                                                             \
    res[1] <<= 1;                                                             \
    res[2] <<= 1;                                                             \
    res[3] <<= 1;                                                             \
  }

SAD_SKIP_16XN_AVX2(32)
SAD_SKIP_16XN_AVX2(16)

#if !CONFIG_REALTIME_ONLY
SAD_SKIP_16XN_AVX2(64)
#endif  // !CONFIG_REALTIME_ONLY