xref: /aosp_15_r20/external/ComputeLibrary/cl_kernels/common/gemmlowp.clembed (revision c217d954acce2dbc11938adb493fc0abd69584f3)

R"(




#ifndef ARM_COMPUTE_HELPER_H
#define ARM_COMPUTE_HELPER_H




#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE(N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                     \
    (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));




#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_PARTIAL_16(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                            \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                           \
    }                                                                                                                                                     \
    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else                                                                                                                                                  \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                               \
    }

#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }

#define STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
    if(!(PARTIAL_COND_Y))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }


#if defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)


#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)

#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)

#else

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)

#endif

#endif


#if defined(PARTIAL_STORE_M0)

#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(max(0, (int)(y * M0) - (int)((M0 - PARTIAL_STORE_M0) % M0))))
#else
#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(y * M0))
#endif



#define STORE_VECTOR_SELECT(basename, data_type, ptr, vec_size, leftover, cond) \
    STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, cond)


#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : enable
#endif

#if defined(ARM_COMPUTE_DEBUG_ENABLED) && defined(cl_arm_printf)
#pragma OPENCL EXTENSION cl_arm_printf : enable
#endif

#define GPU_ARCH_MIDGARD 0x100
#define GPU_ARCH_BIFROST 0x200
#define GPU_ARCH_VALHALL 0x300


#define CONCAT(a, b) a##b


#define EXPAND(x) x


#define CLAMP(x, min_val, max_val) min(max(x, min_val), max_val)


#define REV1(x) ((x))
#define REV2(x) ((x).s10)
#define REV3(x) ((x).s210)
#define REV4(x) ((x).s3210)
#define REV8(x) ((x).s76543210)
#define REV16(x) ((x).sFEDCBA9876543210)



#define REVERSE_STR(x, s) REV##s((x))
#define REVERSE(x, s) REVERSE_STR(x, s)



#define ROT1_0(x) ((x))
#define ROT1_1(x) ((x))

#define ROT2_0(x) ((x))
#define ROT2_1(x) ((x).s10)
#define ROT2_2(x) ((x))

#define ROT3_0(x) ((x))
#define ROT3_1(x) ((x).s201)
#define ROT3_2(x) ((x).s120)
#define ROT3_3(x) ((x))

#define ROT4_0(x) ((x))
#define ROT4_1(x) ((x).s3012)
#define ROT4_2(x) ((x).s2301)
#define ROT4_3(x) ((x).s1230)
#define ROT4_4(x) ((x))

#define ROT8_0(x) ((x))
#define ROT8_1(x) ((x).s70123456)
#define ROT8_2(x) ((x).s67012345)
#define ROT8_3(x) ((x).s56701234)
#define ROT8_4(x) ((x).s45670123)
#define ROT8_5(x) ((x).s34567012)
#define ROT8_6(x) ((x).s23456701)
#define ROT8_7(x) ((x).s12345670)
#define ROT8_8(x) ((x))

#define ROT16_0(x) ((x))
#define ROT16_1(x) ((x).sF0123456789ABCDE)
#define ROT16_2(x) ((x).sEF0123456789ABCD)
#define ROT16_3(x) ((x).sDEF0123456789ABC)
#define ROT16_4(x) ((x).sCDEF0123456789AB)
#define ROT16_5(x) ((x).sBCDEF0123456789A)
#define ROT16_6(x) ((x).sABCDEF0123456789)
#define ROT16_7(x) ((x).s9ABCDEF012345678)
#define ROT16_8(x) ((x).s89ABCDEF01234567)
#define ROT16_9(x) ((x).s789ABCDEF0123456)
#define ROT16_10(x) ((x).s6789ABCDEF012345)
#define ROT16_11(x) ((x).s56789ABCDEF01234)
#define ROT16_12(x) ((x).s456789ABCDEF0123)
#define ROT16_13(x) ((x).s3456789ABCDEF012)
#define ROT16_14(x) ((x).s23456789ABCDEF01)
#define ROT16_15(x) ((x).s123456789ABCDEF0)
#define ROT16_16(x) ((x))



#define ROTATE_STR(x, s, n) ROT##s##_##n(x)
#define ROTATE(x, s, n) ROTATE_STR(x, s, n)



#define V_OFFS1(dt) (dt##1)(0)
#define V_OFFS2(dt) (dt##2)(0, 1)
#define V_OFFS3(dt) (dt##3)(0, 1, 2)
#define V_OFFS4(dt) (dt##4)(0, 1, 2, 3)
#define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7)
#define V_OFFS16(dt) (dt##16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)



#define VEC_OFFS_STR(dt, s) V_OFFS##s(dt)
#define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s)


#define VLOAD_STR(size) vload##size
#define VLOAD(size) VLOAD_STR(size)


#define VLOAD_PARTIAL_STR(size, load_size) vload_partial_##size##_##load_size
#define VLOAD_PARTIAL(size, load_size) VLOAD_PARTIAL_STR(size, load_size)

#define NO_LOAD(data, offs, ptr) \
    {                            \
    }


#define vload_partial_1_0 NO_LOAD
#define vload_partial_1_1 vload1
#define vload_partial_1_2 NO_LOAD
#define vload_partial_1_3 NO_LOAD
#define vload_partial_1_4 NO_LOAD
#define vload_partial_1_5 NO_LOAD
#define vload_partial_1_6 NO_LOAD
#define vload_partial_1_7 NO_LOAD
#define vload_partial_1_8 NO_LOAD
#define vload_partial_1_9 NO_LOAD
#define vload_partial_1_10 NO_LOAD
#define vload_partial_1_11 NO_LOAD
#define vload_partial_1_12 NO_LOAD
#define vload_partial_1_13 NO_LOAD
#define vload_partial_1_14 NO_LOAD
#define vload_partial_1_15 NO_LOAD
#define vload_partial_1_16 NO_LOAD

#define vload_partial_2_0 NO_LOAD
#define vload_partial_2_1 vload_partial_1
#define vload_partial_2_2 vload_partial_2
#define vload_partial_2_3 NO_LOAD
#define vload_partial_2_4 NO_LOAD
#define vload_partial_2_5 NO_LOAD
#define vload_partial_2_6 NO_LOAD
#define vload_partial_2_7 NO_LOAD
#define vload_partial_2_8 NO_LOAD
#define vload_partial_2_9 NO_LOAD
#define vload_partial_2_10 NO_LOAD
#define vload_partial_2_11 NO_LOAD
#define vload_partial_2_12 NO_LOAD
#define vload_partial_2_13 NO_LOAD
#define vload_partial_2_14 NO_LOAD
#define vload_partial_2_15 NO_LOAD
#define vload_partial_2_16 NO_LOAD

#define vload_partial_3_0 NO_LOAD
#define vload_partial_3_1 vload_partial_1
#define vload_partial_3_2 vload_partial_2
#define vload_partial_3_3 vload_partial_3
#define vload_partial_3_4 NO_LOAD
#define vload_partial_3_5 NO_LOAD
#define vload_partial_3_6 NO_LOAD
#define vload_partial_3_7 NO_LOAD
#define vload_partial_3_8 NO_LOAD
#define vload_partial_3_9 NO_LOAD
#define vload_partial_3_10 NO_LOAD
#define vload_partial_3_11 NO_LOAD
#define vload_partial_3_12 NO_LOAD
#define vload_partial_3_13 NO_LOAD
#define vload_partial_3_14 NO_LOAD
#define vload_partial_3_15 NO_LOAD
#define vload_partial_3_16 NO_LOAD

#define vload_partial_4_0 NO_LOAD
#define vload_partial_4_1 vload_partial_1
#define vload_partial_4_2 vload_partial_2
#define vload_partial_4_3 vload_partial_3
#define vload_partial_4_4 vload_partial_4
#define vload_partial_4_5 NO_LOAD
#define vload_partial_4_6 NO_LOAD
#define vload_partial_4_7 NO_LOAD
#define vload_partial_4_8 NO_LOAD
#define vload_partial_4_9 NO_LOAD
#define vload_partial_4_10 NO_LOAD
#define vload_partial_4_11 NO_LOAD
#define vload_partial_4_12 NO_LOAD
#define vload_partial_4_13 NO_LOAD
#define vload_partial_4_14 NO_LOAD
#define vload_partial_4_15 NO_LOAD
#define vload_partial_4_16 NO_LOAD

#define vload_partial_8_0 NO_LOAD
#define vload_partial_8_1 vload_partial_1
#define vload_partial_8_2 vload_partial_2
#define vload_partial_8_3 vload_partial_3
#define vload_partial_8_4 vload_partial_4
#define vload_partial_8_5 vload_partial_5
#define vload_partial_8_6 vload_partial_6
#define vload_partial_8_7 vload_partial_7
#define vload_partial_8_8 vload_partial_8
#define vload_partial_8_9 NO_LOAD
#define vload_partial_8_10 NO_LOAD
#define vload_partial_8_11 NO_LOAD
#define vload_partial_8_12 NO_LOAD
#define vload_partial_8_13 NO_LOAD
#define vload_partial_8_14 NO_LOAD
#define vload_partial_8_15 NO_LOAD
#define vload_partial_8_16 NO_LOAD

#define vload_partial_16_0 NO_LOAD
#define vload_partial_16_1 vload_partial_1
#define vload_partial_16_2 vload_partial_2
#define vload_partial_16_3 vload_partial_3
#define vload_partial_16_4 vload_partial_4
#define vload_partial_16_5 vload_partial_5
#define vload_partial_16_6 vload_partial_6
#define vload_partial_16_7 vload_partial_7
#define vload_partial_16_8 vload_partial_8
#define vload_partial_16_9 vload_partial_9
#define vload_partial_16_10 vload_partial_10
#define vload_partial_16_11 vload_partial_11
#define vload_partial_16_12 vload_partial_12
#define vload_partial_16_13 vload_partial_13
#define vload_partial_16_14 vload_partial_14
#define vload_partial_16_15 vload_partial_15
#define vload_partial_16_16 vload_partial_16


#define vload_partial_1(DATA, OFFSET, PTR) \
    DATA.s0 = vload1(OFFSET, PTR);

#define vload_partial_2(DATA, OFFSET, PTR) \
    DATA.s01 = vload2(OFFSET, PTR);

#define vload_partial_3(DATA, OFFSET, PTR) \
    DATA.s012 = vload3(OFFSET, PTR);

#define vload_partial_4(DATA, OFFSET, PTR) \
    DATA.s0123 = vload4(OFFSET, PTR);

#define vload_partial_5(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    DATA.s4 = vload1(OFFSET, PTR + 4);

#define vload_partial_6(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vload_partial_7(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vload_partial_8(DATA, OFFSET, PTR) \
    DATA.s01234567 = vload8(OFFSET, PTR);

#define vload_partial_9(DATA, OFFSET, PTR)        \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    DATA.s8 = vload1(OFFSET, PTR + 8);

#define vload_partial_10(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vload_partial_11(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_3(DATA.s89A, OFFSET, PTR + 8);

#define vload_partial_12(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_4(DATA.s89AB, OFFSET, PTR + 8);

#define vload_partial_13(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_5(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_14(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_6(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_15(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_7(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_16(DATA, OFFSET, PTR) \
    DATA = vload16(OFFSET, PTR);



#define PIXEL_UNIT4 1
#define PIXEL_UNIT8 2
#define PIXEL_UNIT16 4


#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) PIXEL_UNIT##vec_size
#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size)


#define read_image2d_floatx1(img, x_coord, y_coord) (float4)(read_imagef(img, (int2)(x_coord, y_coord)));
#define read_image2d_floatx2(img, x_coord, y_coord) (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_floatx4(img, x_coord, y_coord) (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord)));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define read_image2d_halfx1(img, x_coord, y_coord) (half4)(read_imageh(img, (int2)(x_coord, y_coord)));
#define read_image2d_halfx2(img, x_coord, y_coord) (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_halfx4(img, x_coord, y_coord) (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord)));
#endif

#define write_image2d_floatx1(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values));
#define write_image2d_floatx2(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_floatx4(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imagef(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imagef(img, (int2)(x_coord + 3, y_coord), values.sCDEF));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define write_image2d_halfx1(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values));
#define write_image2d_halfx2(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_halfx4(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imageh(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imageh(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
#endif


#define READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) read_image2d_##data_type##x##n0(img, x_coord, y_coord)
#define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord)


#define WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values) write_image2d_##data_type##x##n0(img, x_coord, y_coord, values)
#define WRITE_IMAGE2D(data_type, n0, img, x_coord, y_coord, values) WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values)

#define VSTORE_STR(size) vstore##size
#define VSTORE(size) VSTORE_STR(size)

#define float1 float
#define half1 half
#define char1 char
#define uchar1 uchar
#define short1 short
#define ushort1 ushort
#define int1 int
#define uint1 uint
#define long1 long
#define ulong1 ulong
#define double1 double

#define vload1(OFFSET, PTR) *(OFFSET + PTR)
#define vstore1(DATA, OFFSET, PTR) *(OFFSET + PTR) = DATA


#define VSTORE_PARTIAL_STR(size, store_size) vstore_partial_##size##_##store_size
#define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size)

#define NO_STORE(data, offs, ptr) \
    {                             \
    }


#define vstore_partial_1_0 NO_STORE
#define vstore_partial_1_1 vstore1
#define vstore_partial_1_2 NO_STORE
#define vstore_partial_1_3 NO_STORE
#define vstore_partial_1_4 NO_STORE
#define vstore_partial_1_5 NO_STORE
#define vstore_partial_1_6 NO_STORE
#define vstore_partial_1_7 NO_STORE
#define vstore_partial_1_8 NO_STORE
#define vstore_partial_1_9 NO_STORE
#define vstore_partial_1_10 NO_STORE
#define vstore_partial_1_11 NO_STORE
#define vstore_partial_1_12 NO_STORE
#define vstore_partial_1_13 NO_STORE
#define vstore_partial_1_14 NO_STORE
#define vstore_partial_1_15 NO_STORE
#define vstore_partial_1_16 NO_STORE

#define vstore_partial_2_0 NO_STORE
#define vstore_partial_2_1 vstore_partial_1
#define vstore_partial_2_2 vstore_partial_2
#define vstore_partial_2_3 NO_STORE
#define vstore_partial_2_4 NO_STORE
#define vstore_partial_2_5 NO_STORE
#define vstore_partial_2_6 NO_STORE
#define vstore_partial_2_7 NO_STORE
#define vstore_partial_2_8 NO_STORE
#define vstore_partial_2_9 NO_STORE
#define vstore_partial_2_10 NO_STORE
#define vstore_partial_2_11 NO_STORE
#define vstore_partial_2_12 NO_STORE
#define vstore_partial_2_13 NO_STORE
#define vstore_partial_2_14 NO_STORE
#define vstore_partial_2_15 NO_STORE
#define vstore_partial_2_16 NO_STORE

#define vstore_partial_3_0 NO_STORE
#define vstore_partial_3_1 vstore_partial_1
#define vstore_partial_3_2 vstore_partial_2
#define vstore_partial_3_3 vstore_partial_3
#define vstore_partial_3_4 NO_STORE
#define vstore_partial_3_5 NO_STORE
#define vstore_partial_3_6 NO_STORE
#define vstore_partial_3_7 NO_STORE
#define vstore_partial_3_8 NO_STORE
#define vstore_partial_3_9 NO_STORE
#define vstore_partial_3_10 NO_STORE
#define vstore_partial_3_11 NO_STORE
#define vstore_partial_3_12 NO_STORE
#define vstore_partial_3_13 NO_STORE
#define vstore_partial_3_14 NO_STORE
#define vstore_partial_3_15 NO_STORE
#define vstore_partial_3_16 NO_STORE

#define vstore_partial_4_0 NO_STORE
#define vstore_partial_4_1 vstore_partial_1
#define vstore_partial_4_2 vstore_partial_2
#define vstore_partial_4_3 vstore_partial_3
#define vstore_partial_4_4 vstore_partial_4
#define vstore_partial_4_5 NO_STORE
#define vstore_partial_4_6 NO_STORE
#define vstore_partial_4_7 NO_STORE
#define vstore_partial_4_8 NO_STORE
#define vstore_partial_4_9 NO_STORE
#define vstore_partial_4_10 NO_STORE
#define vstore_partial_4_11 NO_STORE
#define vstore_partial_4_12 NO_STORE
#define vstore_partial_4_13 NO_STORE
#define vstore_partial_4_14 NO_STORE
#define vstore_partial_4_15 NO_STORE
#define vstore_partial_4_16 NO_STORE

#define vstore_partial_8_0 NO_STORE
#define vstore_partial_8_1 vstore_partial_1
#define vstore_partial_8_2 vstore_partial_2
#define vstore_partial_8_3 vstore_partial_3
#define vstore_partial_8_4 vstore_partial_4
#define vstore_partial_8_5 vstore_partial_5
#define vstore_partial_8_6 vstore_partial_6
#define vstore_partial_8_7 vstore_partial_7
#define vstore_partial_8_8 vstore_partial_8
#define vstore_partial_8_9 NO_STORE
#define vstore_partial_8_10 NO_STORE
#define vstore_partial_8_11 NO_STORE
#define vstore_partial_8_12 NO_STORE
#define vstore_partial_8_13 NO_STORE
#define vstore_partial_8_14 NO_STORE
#define vstore_partial_8_15 NO_STORE
#define vstore_partial_8_16 NO_STORE

#define vstore_partial_16_0 NO_STORE
#define vstore_partial_16_1 vstore_partial_1
#define vstore_partial_16_2 vstore_partial_2
#define vstore_partial_16_3 vstore_partial_3
#define vstore_partial_16_4 vstore_partial_4
#define vstore_partial_16_5 vstore_partial_5
#define vstore_partial_16_6 vstore_partial_6
#define vstore_partial_16_7 vstore_partial_7
#define vstore_partial_16_8 vstore_partial_8
#define vstore_partial_16_9 vstore_partial_9
#define vstore_partial_16_10 vstore_partial_10
#define vstore_partial_16_11 vstore_partial_11
#define vstore_partial_16_12 vstore_partial_12
#define vstore_partial_16_13 vstore_partial_13
#define vstore_partial_16_14 vstore_partial_14
#define vstore_partial_16_15 vstore_partial_15
#define vstore_partial_16_16 vstore_partial_16


#define vstore_partial_1(DATA, OFFSET, PTR) \
    vstore1(DATA.s0, OFFSET, PTR);

#define vstore_partial_2(DATA, OFFSET, PTR) \
    vstore2(DATA.s01, OFFSET, PTR);

#define vstore_partial_3(DATA, OFFSET, PTR) \
    vstore3(DATA.s012, OFFSET, PTR);

#define vstore_partial_4(DATA, OFFSET, PTR) \
    vstore4(DATA.s0123, OFFSET, PTR);

#define vstore_partial_5(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore1(DATA.s4, OFFSET, PTR + 4);

#define vstore_partial_6(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vstore_partial_7(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vstore_partial_8(DATA, OFFSET, PTR) \
    vstore8(DATA.s01234567, OFFSET, PTR);

#define vstore_partial_9(DATA, OFFSET, PTR)        \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore1(DATA.s8, OFFSET, PTR + 8);

#define vstore_partial_10(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vstore_partial_11(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_3(DATA.s89a, OFFSET, PTR + 8);

#define vstore_partial_12(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_4(DATA.s89ab, OFFSET, PTR + 8);

#define vstore_partial_13(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_5(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_14(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_6(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_15(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_7(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_16(DATA, OFFSET, PTR) \
    vstore16(DATA, OFFSET, PTR);





#define convert_float_sat convert_float
#define convert_float1_sat convert_float
#define convert_float2_sat convert_float2
#define convert_float3_sat convert_float3
#define convert_float4_sat convert_float4
#define convert_float8_sat convert_float8
#define convert_float16_sat convert_float16
#define convert_half_sat convert_float
#define convert_half1_sat convert_half
#define convert_half2_sat convert_half2
#define convert_half3_sat convert_half3
#define convert_half4_sat convert_half4
#define convert_half8_sat convert_half8
#define convert_half16_sat convert_half16

#define convert_float1 convert_float
#define convert_half1 convert_half
#define convert_char1 convert_char
#define convert_uchar1 convert_uchar
#define convert_short1 convert_short
#define convert_ushort1 convert_ushort
#define convert_int1 convert_int
#define convert_uint1 convert_uint
#define convert_long1 convert_long
#define convert_ulong1 convert_ulong
#define convert_double1 convert_double

#define convert_char1_sat convert_char_sat
#define convert_uchar1_sat convert_uchar_sat
#define convert_uchar2_sat convert_uchar2_sat
#define convert_uchar3_sat convert_uchar3_sat
#define convert_uchar4_sat convert_uchar4_sat
#define convert_uchar8_sat convert_uchar8_sat
#define convert_uchar16_sat convert_uchar16_sat
#define convert_short1_sat convert_short_sat
#define convert_ushort1_sat convert_ushort_sat
#define convert_int1_sat convert_int_sat
#define convert_uint1_sat convert_uint_sat
#define convert_long1_sat convert_long_sat
#define convert_ulong1_sat convert_ulong_sat
#define convert_double1_sat convert_double_sat

#define VEC_DATA_TYPE_STR(type, size) type##size
#define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size)

#define CONVERT_STR(x, type) (convert_##type((x)))
#define CONVERT(x, type) CONVERT_STR(x, type)

#define CONVERT_SAT_STR(x, type) (convert_##type##_sat((x)))
#define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type)

#define CONVERT_SAT_ROUND_STR(x, type, round) (convert_##type##_sat_##round((x)))
#define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round)

#define select_vec_dt_uchar(size) uchar##size
#define select_vec_dt_char(size) char##size
#define select_vec_dt_ushort(size) ushort##size
#define select_vec_dt_short(size) short##size
#define select_vec_dt_half(size) short##size
#define select_vec_dt_uint(size) uint##size
#define select_vec_dt_int(size) int##size
#define select_vec_dt_float(size) int##size
#define select_vec_dt_ulong(size) ulong##size
#define select_vec_dt_long(size) long##size

#define SELECT_VEC_DATA_TYPE_STR(type, size) select_vec_dt_##type(size)
#define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size)
#define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1)

#define signed_int_vec_dt_uchar(size) char##size
#define signed_int_vec_dt_char(size) char##size
#define signed_int_vec_dt_ushort(size) short##size
#define signed_int_vec_dt_short(size) short##size
#define signed_int_vec_dt_half(size) short##size
#define signed_int_vec_dt_uint(size) int##size
#define signed_int_vec_dt_int(size) int##size
#define signed_int_vec_dt_float(size) int##size
#define signed_int_vec_dt_ulong(size) long##size
#define signed_int_vec_dt_long(size) long##size

#define SIGNED_INT_VEC_DATA_TYPE_STR(type, size) signed_int_vec_dt_##type(size)
#define SIGNED_INT_VEC_DATA_TYPE(type, size) SIGNED_INT_VEC_DATA_TYPE_STR(type, size)
#define SIGNED_INT_DATA_TYPE(type) SIGNED_INT_VEC_DATA_TYPE_STR(type, 1)

#define sum_reduce_1(x) (x)
#define sum_reduce_2(x) ((x).s0) + ((x).s1)
#define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2)
#define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23)
#define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567)
#define sum_reduce_16(x) sum_reduce_8((x).s01234567) + sum_reduce_8((x).s89ABCDEF)

#define SUM_REDUCE_STR(x, size) sum_reduce_##size(x)
#define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size)

#define prod_reduce_1(x) (x)
#define prod_reduce_2(x) ((x).s0) * ((x).s1)
#define prod_reduce_3(x) prod_reduce_2((x).s01) * ((x).s2)
#define prod_reduce_4(x) prod_reduce_2((x).s01) * prod_reduce_2((x).s23)
#define prod_reduce_8(x) prod_reduce_4((x).s0123) * prod_reduce_4((x).s4567)
#define prod_reduce_16(x) prod_reduce_8((x).s01234567) * prod_reduce_8((x).s89ABCDEF)

#define PROD_REDUCE_STR(x, size) prod_reduce_##size(x)
#define PROD_REDUCE(x, size) PROD_REDUCE_STR(x, size)

#define max_reduce_1(x) (x)
#define max_reduce_2(x) max(((x).s0), ((x).s1))
#define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2))
#define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23))
#define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567))
#define max_reduce_16(x) max(max_reduce_8((x).s01234567), max_reduce_8((x).s89ABCDEF))

#define MAX_REDUCE_STR(x, size) max_reduce_##size(x)
#define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size)

#define VECTOR_DECLARATION(name)     \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_offset_first_element_in_bytes

#define IMAGE_DECLARATION(name)      \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR3D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR4D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR5D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_stride_v, \
    uint        name##_step_v,   \
    uint        name##_offset_first_element_in_bytes

#define CONVERT_TO_VECTOR_STRUCT(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x)

#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0)

#define CONVERT_TO_IMAGE_STRUCT(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y)

#define CONVERT_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT(name)                                                                                                           \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0)

#define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size)                                                                                                 \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z, name##_stride_w, name##_step_w, mod_size)

#define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name)                                                                                       \
    tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                           name##_stride_z, name##_step_z)


typedef struct Vector
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
} Vector;


typedef struct Image
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
} Image;


typedef struct Tensor3D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
} Tensor3D;


typedef struct Tensor4D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
    int             stride_w;
} Tensor4D;


inline Vector update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x)
{
    Vector vector =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
    };
    vector.ptr += vector.offset_first_element_in_bytes + get_global_id(0) * step_x;
    return vector;
}


inline Image update_image_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y;
    return img;
}


inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return img;
}


inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return tensor;
}


inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    return tensor;
}

inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z, uint stride_w,
                                             uint step_w,
                                             uint mod_size)
{
    Tensor4D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z,
        .stride_w                      = stride_w
    };

    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w;
    return tensor;
}


inline __global const uchar *vector_offset(const Vector *vec, int x)
{
    return vec->ptr + x * vec->stride_x;
}


inline __global uchar *offset(const Image *img, int x, int y)
{
    return img->ptr + x * img->stride_x + y * img->stride_y;
}


inline __global const uchar *tensor3D_offset(const Tensor3D *tensor, int x, int y, int z)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z;
}


inline __global const uchar *tensor4D_offset(const Tensor4D *tensor, int x, int y, int z, int w)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + w * tensor->stride_w;
}


inline __global const uchar *tensor3D_index2ptr(const Tensor3D *tensor, uint width, uint height, uint depth, uint index)
{
    uint num_elements = width * height;

    const uint z = index / num_elements;

    index %= num_elements;

    const uint y = index / width;

    index %= width;

    const uint x = index;

    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + tensor->offset_first_element_in_bytes;
}

#endif

#if GPU_ARCH == GPU_ARCH_BIFROST
#define MLA(a, b, c) (fma(c, b, a))
#else
#define MLA(a, b, c) ((b) * (c) + (a))
#endif


#define hard_swish_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (x * ((min(max((x + (DATA_TYPE)3.0), (DATA_TYPE)0.0), (DATA_TYPE)6.0)) * (DATA_TYPE)0.166666667))


#define logistic_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) ((DATA_TYPE)1.0 / ((DATA_TYPE)1.0 + exp(-x)))


#define tanh_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) ((DATA_TYPE)A_VAL * tanh((DATA_TYPE)B_VAL * x))


#define relu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (max((DATA_TYPE)0.0, x))


#define brelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (min((DATA_TYPE)A_VAL, max((DATA_TYPE)0.0, x)))


#define lu_brelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (min(max(x, (DATA_TYPE)B_VAL), (DATA_TYPE)A_VAL))


#define lrelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) ((min(x, (DATA_TYPE)0.0) * (DATA_TYPE)A_VAL) + max(x, (DATA_TYPE)0.0))


#define srelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (log((DATA_TYPE)1.0 + exp(x)))


#define elu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (select(((DATA_TYPE)A_VAL * (exp(x) - (DATA_TYPE)1.0)), x, (SELECT_VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE))isgreaterequal(x, (DATA_TYPE)0.0)))


#define abs_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (fabs(x))


#define square_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (x * x)


#define sqrt_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (sqrt(x))


#define linear_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (MLA((DATA_TYPE)B_VAL, (DATA_TYPE)A_VAL, x))


#define gelu_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (x * (DATA_TYPE)0.5 * ((DATA_TYPE)1.0 + erf(x / (DATA_TYPE)1.41421356237)))


#define identity_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) (x)

#define ACT_OP(op, DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) op##_op(DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL)

#define ACTIVATION(op, DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL) ACT_OP(op, DATA_TYPE, VEC_SIZE, x, A_VAL, B_VAL)

#ifndef ARM_COMPUTE_HELPER_H
#define ARM_COMPUTE_HELPER_H




#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE(N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                     \
    (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));




#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_PARTIAL_16(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                            \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                           \
    }                                                                                                                                                     \
    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else                                                                                                                                                  \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                               \
    }

#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }

#define STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
    if(!(PARTIAL_COND_Y))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }


#if defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)


#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)

#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)

#else

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)

#endif

#endif


#if defined(PARTIAL_STORE_M0)

#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(max(0, (int)(y * M0) - (int)((M0 - PARTIAL_STORE_M0) % M0))))
#else
#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(y * M0))
#endif



#define STORE_VECTOR_SELECT(basename, data_type, ptr, vec_size, leftover, cond) \
    STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, cond)


#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : enable
#endif

#if defined(ARM_COMPUTE_DEBUG_ENABLED) && defined(cl_arm_printf)
#pragma OPENCL EXTENSION cl_arm_printf : enable
#endif

#define GPU_ARCH_MIDGARD 0x100
#define GPU_ARCH_BIFROST 0x200
#define GPU_ARCH_VALHALL 0x300


#define CONCAT(a, b) a##b


#define EXPAND(x) x


#define CLAMP(x, min_val, max_val) min(max(x, min_val), max_val)


#define REV1(x) ((x))
#define REV2(x) ((x).s10)
#define REV3(x) ((x).s210)
#define REV4(x) ((x).s3210)
#define REV8(x) ((x).s76543210)
#define REV16(x) ((x).sFEDCBA9876543210)



#define REVERSE_STR(x, s) REV##s((x))
#define REVERSE(x, s) REVERSE_STR(x, s)



#define ROT1_0(x) ((x))
#define ROT1_1(x) ((x))

#define ROT2_0(x) ((x))
#define ROT2_1(x) ((x).s10)
#define ROT2_2(x) ((x))

#define ROT3_0(x) ((x))
#define ROT3_1(x) ((x).s201)
#define ROT3_2(x) ((x).s120)
#define ROT3_3(x) ((x))

#define ROT4_0(x) ((x))
#define ROT4_1(x) ((x).s3012)
#define ROT4_2(x) ((x).s2301)
#define ROT4_3(x) ((x).s1230)
#define ROT4_4(x) ((x))

#define ROT8_0(x) ((x))
#define ROT8_1(x) ((x).s70123456)
#define ROT8_2(x) ((x).s67012345)
#define ROT8_3(x) ((x).s56701234)
#define ROT8_4(x) ((x).s45670123)
#define ROT8_5(x) ((x).s34567012)
#define ROT8_6(x) ((x).s23456701)
#define ROT8_7(x) ((x).s12345670)
#define ROT8_8(x) ((x))

#define ROT16_0(x) ((x))
#define ROT16_1(x) ((x).sF0123456789ABCDE)
#define ROT16_2(x) ((x).sEF0123456789ABCD)
#define ROT16_3(x) ((x).sDEF0123456789ABC)
#define ROT16_4(x) ((x).sCDEF0123456789AB)
#define ROT16_5(x) ((x).sBCDEF0123456789A)
#define ROT16_6(x) ((x).sABCDEF0123456789)
#define ROT16_7(x) ((x).s9ABCDEF012345678)
#define ROT16_8(x) ((x).s89ABCDEF01234567)
#define ROT16_9(x) ((x).s789ABCDEF0123456)
#define ROT16_10(x) ((x).s6789ABCDEF012345)
#define ROT16_11(x) ((x).s56789ABCDEF01234)
#define ROT16_12(x) ((x).s456789ABCDEF0123)
#define ROT16_13(x) ((x).s3456789ABCDEF012)
#define ROT16_14(x) ((x).s23456789ABCDEF01)
#define ROT16_15(x) ((x).s123456789ABCDEF0)
#define ROT16_16(x) ((x))



#define ROTATE_STR(x, s, n) ROT##s##_##n(x)
#define ROTATE(x, s, n) ROTATE_STR(x, s, n)



#define V_OFFS1(dt) (dt##1)(0)
#define V_OFFS2(dt) (dt##2)(0, 1)
#define V_OFFS3(dt) (dt##3)(0, 1, 2)
#define V_OFFS4(dt) (dt##4)(0, 1, 2, 3)
#define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7)
#define V_OFFS16(dt) (dt##16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)



#define VEC_OFFS_STR(dt, s) V_OFFS##s(dt)
#define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s)


#define VLOAD_STR(size) vload##size
#define VLOAD(size) VLOAD_STR(size)


#define VLOAD_PARTIAL_STR(size, load_size) vload_partial_##size##_##load_size
#define VLOAD_PARTIAL(size, load_size) VLOAD_PARTIAL_STR(size, load_size)

#define NO_LOAD(data, offs, ptr) \
    {                            \
    }


#define vload_partial_1_0 NO_LOAD
#define vload_partial_1_1 vload1
#define vload_partial_1_2 NO_LOAD
#define vload_partial_1_3 NO_LOAD
#define vload_partial_1_4 NO_LOAD
#define vload_partial_1_5 NO_LOAD
#define vload_partial_1_6 NO_LOAD
#define vload_partial_1_7 NO_LOAD
#define vload_partial_1_8 NO_LOAD
#define vload_partial_1_9 NO_LOAD
#define vload_partial_1_10 NO_LOAD
#define vload_partial_1_11 NO_LOAD
#define vload_partial_1_12 NO_LOAD
#define vload_partial_1_13 NO_LOAD
#define vload_partial_1_14 NO_LOAD
#define vload_partial_1_15 NO_LOAD
#define vload_partial_1_16 NO_LOAD

#define vload_partial_2_0 NO_LOAD
#define vload_partial_2_1 vload_partial_1
#define vload_partial_2_2 vload_partial_2
#define vload_partial_2_3 NO_LOAD
#define vload_partial_2_4 NO_LOAD
#define vload_partial_2_5 NO_LOAD
#define vload_partial_2_6 NO_LOAD
#define vload_partial_2_7 NO_LOAD
#define vload_partial_2_8 NO_LOAD
#define vload_partial_2_9 NO_LOAD
#define vload_partial_2_10 NO_LOAD
#define vload_partial_2_11 NO_LOAD
#define vload_partial_2_12 NO_LOAD
#define vload_partial_2_13 NO_LOAD
#define vload_partial_2_14 NO_LOAD
#define vload_partial_2_15 NO_LOAD
#define vload_partial_2_16 NO_LOAD

#define vload_partial_3_0 NO_LOAD
#define vload_partial_3_1 vload_partial_1
#define vload_partial_3_2 vload_partial_2
#define vload_partial_3_3 vload_partial_3
#define vload_partial_3_4 NO_LOAD
#define vload_partial_3_5 NO_LOAD
#define vload_partial_3_6 NO_LOAD
#define vload_partial_3_7 NO_LOAD
#define vload_partial_3_8 NO_LOAD
#define vload_partial_3_9 NO_LOAD
#define vload_partial_3_10 NO_LOAD
#define vload_partial_3_11 NO_LOAD
#define vload_partial_3_12 NO_LOAD
#define vload_partial_3_13 NO_LOAD
#define vload_partial_3_14 NO_LOAD
#define vload_partial_3_15 NO_LOAD
#define vload_partial_3_16 NO_LOAD

#define vload_partial_4_0 NO_LOAD
#define vload_partial_4_1 vload_partial_1
#define vload_partial_4_2 vload_partial_2
#define vload_partial_4_3 vload_partial_3
#define vload_partial_4_4 vload_partial_4
#define vload_partial_4_5 NO_LOAD
#define vload_partial_4_6 NO_LOAD
#define vload_partial_4_7 NO_LOAD
#define vload_partial_4_8 NO_LOAD
#define vload_partial_4_9 NO_LOAD
#define vload_partial_4_10 NO_LOAD
#define vload_partial_4_11 NO_LOAD
#define vload_partial_4_12 NO_LOAD
#define vload_partial_4_13 NO_LOAD
#define vload_partial_4_14 NO_LOAD
#define vload_partial_4_15 NO_LOAD
#define vload_partial_4_16 NO_LOAD

#define vload_partial_8_0 NO_LOAD
#define vload_partial_8_1 vload_partial_1
#define vload_partial_8_2 vload_partial_2
#define vload_partial_8_3 vload_partial_3
#define vload_partial_8_4 vload_partial_4
#define vload_partial_8_5 vload_partial_5
#define vload_partial_8_6 vload_partial_6
#define vload_partial_8_7 vload_partial_7
#define vload_partial_8_8 vload_partial_8
#define vload_partial_8_9 NO_LOAD
#define vload_partial_8_10 NO_LOAD
#define vload_partial_8_11 NO_LOAD
#define vload_partial_8_12 NO_LOAD
#define vload_partial_8_13 NO_LOAD
#define vload_partial_8_14 NO_LOAD
#define vload_partial_8_15 NO_LOAD
#define vload_partial_8_16 NO_LOAD

#define vload_partial_16_0 NO_LOAD
#define vload_partial_16_1 vload_partial_1
#define vload_partial_16_2 vload_partial_2
#define vload_partial_16_3 vload_partial_3
#define vload_partial_16_4 vload_partial_4
#define vload_partial_16_5 vload_partial_5
#define vload_partial_16_6 vload_partial_6
#define vload_partial_16_7 vload_partial_7
#define vload_partial_16_8 vload_partial_8
#define vload_partial_16_9 vload_partial_9
#define vload_partial_16_10 vload_partial_10
#define vload_partial_16_11 vload_partial_11
#define vload_partial_16_12 vload_partial_12
#define vload_partial_16_13 vload_partial_13
#define vload_partial_16_14 vload_partial_14
#define vload_partial_16_15 vload_partial_15
#define vload_partial_16_16 vload_partial_16


#define vload_partial_1(DATA, OFFSET, PTR) \
    DATA.s0 = vload1(OFFSET, PTR);

#define vload_partial_2(DATA, OFFSET, PTR) \
    DATA.s01 = vload2(OFFSET, PTR);

#define vload_partial_3(DATA, OFFSET, PTR) \
    DATA.s012 = vload3(OFFSET, PTR);

#define vload_partial_4(DATA, OFFSET, PTR) \
    DATA.s0123 = vload4(OFFSET, PTR);

#define vload_partial_5(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    DATA.s4 = vload1(OFFSET, PTR + 4);

#define vload_partial_6(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vload_partial_7(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vload_partial_8(DATA, OFFSET, PTR) \
    DATA.s01234567 = vload8(OFFSET, PTR);

#define vload_partial_9(DATA, OFFSET, PTR)        \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    DATA.s8 = vload1(OFFSET, PTR + 8);

#define vload_partial_10(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vload_partial_11(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_3(DATA.s89A, OFFSET, PTR + 8);

#define vload_partial_12(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_4(DATA.s89AB, OFFSET, PTR + 8);

#define vload_partial_13(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_5(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_14(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_6(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_15(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_7(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_16(DATA, OFFSET, PTR) \
    DATA = vload16(OFFSET, PTR);



#define PIXEL_UNIT4 1
#define PIXEL_UNIT8 2
#define PIXEL_UNIT16 4


#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) PIXEL_UNIT##vec_size
#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size)


#define read_image2d_floatx1(img, x_coord, y_coord) (float4)(read_imagef(img, (int2)(x_coord, y_coord)));
#define read_image2d_floatx2(img, x_coord, y_coord) (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_floatx4(img, x_coord, y_coord) (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord)));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define read_image2d_halfx1(img, x_coord, y_coord) (half4)(read_imageh(img, (int2)(x_coord, y_coord)));
#define read_image2d_halfx2(img, x_coord, y_coord) (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_halfx4(img, x_coord, y_coord) (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord)));
#endif

#define write_image2d_floatx1(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values));
#define write_image2d_floatx2(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_floatx4(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imagef(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imagef(img, (int2)(x_coord + 3, y_coord), values.sCDEF));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define write_image2d_halfx1(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values));
#define write_image2d_halfx2(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_halfx4(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imageh(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imageh(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
#endif


#define READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) read_image2d_##data_type##x##n0(img, x_coord, y_coord)
#define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord)


#define WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values) write_image2d_##data_type##x##n0(img, x_coord, y_coord, values)
#define WRITE_IMAGE2D(data_type, n0, img, x_coord, y_coord, values) WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values)

#define VSTORE_STR(size) vstore##size
#define VSTORE(size) VSTORE_STR(size)

#define float1 float
#define half1 half
#define char1 char
#define uchar1 uchar
#define short1 short
#define ushort1 ushort
#define int1 int
#define uint1 uint
#define long1 long
#define ulong1 ulong
#define double1 double

#define vload1(OFFSET, PTR) *(OFFSET + PTR)
#define vstore1(DATA, OFFSET, PTR) *(OFFSET + PTR) = DATA


#define VSTORE_PARTIAL_STR(size, store_size) vstore_partial_##size##_##store_size
#define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size)

#define NO_STORE(data, offs, ptr) \
    {                             \
    }


#define vstore_partial_1_0 NO_STORE
#define vstore_partial_1_1 vstore1
#define vstore_partial_1_2 NO_STORE
#define vstore_partial_1_3 NO_STORE
#define vstore_partial_1_4 NO_STORE
#define vstore_partial_1_5 NO_STORE
#define vstore_partial_1_6 NO_STORE
#define vstore_partial_1_7 NO_STORE
#define vstore_partial_1_8 NO_STORE
#define vstore_partial_1_9 NO_STORE
#define vstore_partial_1_10 NO_STORE
#define vstore_partial_1_11 NO_STORE
#define vstore_partial_1_12 NO_STORE
#define vstore_partial_1_13 NO_STORE
#define vstore_partial_1_14 NO_STORE
#define vstore_partial_1_15 NO_STORE
#define vstore_partial_1_16 NO_STORE

#define vstore_partial_2_0 NO_STORE
#define vstore_partial_2_1 vstore_partial_1
#define vstore_partial_2_2 vstore_partial_2
#define vstore_partial_2_3 NO_STORE
#define vstore_partial_2_4 NO_STORE
#define vstore_partial_2_5 NO_STORE
#define vstore_partial_2_6 NO_STORE
#define vstore_partial_2_7 NO_STORE
#define vstore_partial_2_8 NO_STORE
#define vstore_partial_2_9 NO_STORE
#define vstore_partial_2_10 NO_STORE
#define vstore_partial_2_11 NO_STORE
#define vstore_partial_2_12 NO_STORE
#define vstore_partial_2_13 NO_STORE
#define vstore_partial_2_14 NO_STORE
#define vstore_partial_2_15 NO_STORE
#define vstore_partial_2_16 NO_STORE

#define vstore_partial_3_0 NO_STORE
#define vstore_partial_3_1 vstore_partial_1
#define vstore_partial_3_2 vstore_partial_2
#define vstore_partial_3_3 vstore_partial_3
#define vstore_partial_3_4 NO_STORE
#define vstore_partial_3_5 NO_STORE
#define vstore_partial_3_6 NO_STORE
#define vstore_partial_3_7 NO_STORE
#define vstore_partial_3_8 NO_STORE
#define vstore_partial_3_9 NO_STORE
#define vstore_partial_3_10 NO_STORE
#define vstore_partial_3_11 NO_STORE
#define vstore_partial_3_12 NO_STORE
#define vstore_partial_3_13 NO_STORE
#define vstore_partial_3_14 NO_STORE
#define vstore_partial_3_15 NO_STORE
#define vstore_partial_3_16 NO_STORE

#define vstore_partial_4_0 NO_STORE
#define vstore_partial_4_1 vstore_partial_1
#define vstore_partial_4_2 vstore_partial_2
#define vstore_partial_4_3 vstore_partial_3
#define vstore_partial_4_4 vstore_partial_4
#define vstore_partial_4_5 NO_STORE
#define vstore_partial_4_6 NO_STORE
#define vstore_partial_4_7 NO_STORE
#define vstore_partial_4_8 NO_STORE
#define vstore_partial_4_9 NO_STORE
#define vstore_partial_4_10 NO_STORE
#define vstore_partial_4_11 NO_STORE
#define vstore_partial_4_12 NO_STORE
#define vstore_partial_4_13 NO_STORE
#define vstore_partial_4_14 NO_STORE
#define vstore_partial_4_15 NO_STORE
#define vstore_partial_4_16 NO_STORE

#define vstore_partial_8_0 NO_STORE
#define vstore_partial_8_1 vstore_partial_1
#define vstore_partial_8_2 vstore_partial_2
#define vstore_partial_8_3 vstore_partial_3
#define vstore_partial_8_4 vstore_partial_4
#define vstore_partial_8_5 vstore_partial_5
#define vstore_partial_8_6 vstore_partial_6
#define vstore_partial_8_7 vstore_partial_7
#define vstore_partial_8_8 vstore_partial_8
#define vstore_partial_8_9 NO_STORE
#define vstore_partial_8_10 NO_STORE
#define vstore_partial_8_11 NO_STORE
#define vstore_partial_8_12 NO_STORE
#define vstore_partial_8_13 NO_STORE
#define vstore_partial_8_14 NO_STORE
#define vstore_partial_8_15 NO_STORE
#define vstore_partial_8_16 NO_STORE

#define vstore_partial_16_0 NO_STORE
#define vstore_partial_16_1 vstore_partial_1
#define vstore_partial_16_2 vstore_partial_2
#define vstore_partial_16_3 vstore_partial_3
#define vstore_partial_16_4 vstore_partial_4
#define vstore_partial_16_5 vstore_partial_5
#define vstore_partial_16_6 vstore_partial_6
#define vstore_partial_16_7 vstore_partial_7
#define vstore_partial_16_8 vstore_partial_8
#define vstore_partial_16_9 vstore_partial_9
#define vstore_partial_16_10 vstore_partial_10
#define vstore_partial_16_11 vstore_partial_11
#define vstore_partial_16_12 vstore_partial_12
#define vstore_partial_16_13 vstore_partial_13
#define vstore_partial_16_14 vstore_partial_14
#define vstore_partial_16_15 vstore_partial_15
#define vstore_partial_16_16 vstore_partial_16


#define vstore_partial_1(DATA, OFFSET, PTR) \
    vstore1(DATA.s0, OFFSET, PTR);

#define vstore_partial_2(DATA, OFFSET, PTR) \
    vstore2(DATA.s01, OFFSET, PTR);

#define vstore_partial_3(DATA, OFFSET, PTR) \
    vstore3(DATA.s012, OFFSET, PTR);

#define vstore_partial_4(DATA, OFFSET, PTR) \
    vstore4(DATA.s0123, OFFSET, PTR);

#define vstore_partial_5(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore1(DATA.s4, OFFSET, PTR + 4);

#define vstore_partial_6(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vstore_partial_7(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vstore_partial_8(DATA, OFFSET, PTR) \
    vstore8(DATA.s01234567, OFFSET, PTR);

#define vstore_partial_9(DATA, OFFSET, PTR)        \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore1(DATA.s8, OFFSET, PTR + 8);

#define vstore_partial_10(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vstore_partial_11(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_3(DATA.s89a, OFFSET, PTR + 8);

#define vstore_partial_12(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_4(DATA.s89ab, OFFSET, PTR + 8);

#define vstore_partial_13(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_5(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_14(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_6(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_15(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_7(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_16(DATA, OFFSET, PTR) \
    vstore16(DATA, OFFSET, PTR);





#define convert_float_sat convert_float
#define convert_float1_sat convert_float
#define convert_float2_sat convert_float2
#define convert_float3_sat convert_float3
#define convert_float4_sat convert_float4
#define convert_float8_sat convert_float8
#define convert_float16_sat convert_float16
#define convert_half_sat convert_float
#define convert_half1_sat convert_half
#define convert_half2_sat convert_half2
#define convert_half3_sat convert_half3
#define convert_half4_sat convert_half4
#define convert_half8_sat convert_half8
#define convert_half16_sat convert_half16

#define convert_float1 convert_float
#define convert_half1 convert_half
#define convert_char1 convert_char
#define convert_uchar1 convert_uchar
#define convert_short1 convert_short
#define convert_ushort1 convert_ushort
#define convert_int1 convert_int
#define convert_uint1 convert_uint
#define convert_long1 convert_long
#define convert_ulong1 convert_ulong
#define convert_double1 convert_double

#define convert_char1_sat convert_char_sat
#define convert_uchar1_sat convert_uchar_sat
#define convert_uchar2_sat convert_uchar2_sat
#define convert_uchar3_sat convert_uchar3_sat
#define convert_uchar4_sat convert_uchar4_sat
#define convert_uchar8_sat convert_uchar8_sat
#define convert_uchar16_sat convert_uchar16_sat
#define convert_short1_sat convert_short_sat
#define convert_ushort1_sat convert_ushort_sat
#define convert_int1_sat convert_int_sat
#define convert_uint1_sat convert_uint_sat
#define convert_long1_sat convert_long_sat
#define convert_ulong1_sat convert_ulong_sat
#define convert_double1_sat convert_double_sat

#define VEC_DATA_TYPE_STR(type, size) type##size
#define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size)

#define CONVERT_STR(x, type) (convert_##type((x)))
#define CONVERT(x, type) CONVERT_STR(x, type)

#define CONVERT_SAT_STR(x, type) (convert_##type##_sat((x)))
#define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type)

#define CONVERT_SAT_ROUND_STR(x, type, round) (convert_##type##_sat_##round((x)))
#define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round)

#define select_vec_dt_uchar(size) uchar##size
#define select_vec_dt_char(size) char##size
#define select_vec_dt_ushort(size) ushort##size
#define select_vec_dt_short(size) short##size
#define select_vec_dt_half(size) short##size
#define select_vec_dt_uint(size) uint##size
#define select_vec_dt_int(size) int##size
#define select_vec_dt_float(size) int##size
#define select_vec_dt_ulong(size) ulong##size
#define select_vec_dt_long(size) long##size

#define SELECT_VEC_DATA_TYPE_STR(type, size) select_vec_dt_##type(size)
#define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size)
#define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1)

#define signed_int_vec_dt_uchar(size) char##size
#define signed_int_vec_dt_char(size) char##size
#define signed_int_vec_dt_ushort(size) short##size
#define signed_int_vec_dt_short(size) short##size
#define signed_int_vec_dt_half(size) short##size
#define signed_int_vec_dt_uint(size) int##size
#define signed_int_vec_dt_int(size) int##size
#define signed_int_vec_dt_float(size) int##size
#define signed_int_vec_dt_ulong(size) long##size
#define signed_int_vec_dt_long(size) long##size

#define SIGNED_INT_VEC_DATA_TYPE_STR(type, size) signed_int_vec_dt_##type(size)
#define SIGNED_INT_VEC_DATA_TYPE(type, size) SIGNED_INT_VEC_DATA_TYPE_STR(type, size)
#define SIGNED_INT_DATA_TYPE(type) SIGNED_INT_VEC_DATA_TYPE_STR(type, 1)

#define sum_reduce_1(x) (x)
#define sum_reduce_2(x) ((x).s0) + ((x).s1)
#define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2)
#define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23)
#define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567)
#define sum_reduce_16(x) sum_reduce_8((x).s01234567) + sum_reduce_8((x).s89ABCDEF)

#define SUM_REDUCE_STR(x, size) sum_reduce_##size(x)
#define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size)

#define prod_reduce_1(x) (x)
#define prod_reduce_2(x) ((x).s0) * ((x).s1)
#define prod_reduce_3(x) prod_reduce_2((x).s01) * ((x).s2)
#define prod_reduce_4(x) prod_reduce_2((x).s01) * prod_reduce_2((x).s23)
#define prod_reduce_8(x) prod_reduce_4((x).s0123) * prod_reduce_4((x).s4567)
#define prod_reduce_16(x) prod_reduce_8((x).s01234567) * prod_reduce_8((x).s89ABCDEF)

#define PROD_REDUCE_STR(x, size) prod_reduce_##size(x)
#define PROD_REDUCE(x, size) PROD_REDUCE_STR(x, size)

#define max_reduce_1(x) (x)
#define max_reduce_2(x) max(((x).s0), ((x).s1))
#define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2))
#define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23))
#define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567))
#define max_reduce_16(x) max(max_reduce_8((x).s01234567), max_reduce_8((x).s89ABCDEF))

#define MAX_REDUCE_STR(x, size) max_reduce_##size(x)
#define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size)

#define VECTOR_DECLARATION(name)     \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_offset_first_element_in_bytes

#define IMAGE_DECLARATION(name)      \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR3D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR4D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR5D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_stride_v, \
    uint        name##_step_v,   \
    uint        name##_offset_first_element_in_bytes

#define CONVERT_TO_VECTOR_STRUCT(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x)

#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0)

#define CONVERT_TO_IMAGE_STRUCT(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y)

#define CONVERT_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT(name)                                                                                                           \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0)

#define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size)                                                                                                 \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z, name##_stride_w, name##_step_w, mod_size)

#define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name)                                                                                       \
    tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                           name##_stride_z, name##_step_z)


typedef struct Vector
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
} Vector;


typedef struct Image
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
} Image;


typedef struct Tensor3D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
} Tensor3D;


typedef struct Tensor4D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
    int             stride_w;
} Tensor4D;


inline Vector update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x)
{
    Vector vector =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
    };
    vector.ptr += vector.offset_first_element_in_bytes + get_global_id(0) * step_x;
    return vector;
}


inline Image update_image_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y;
    return img;
}


inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return img;
}


inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return tensor;
}


inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    return tensor;
}

inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z, uint stride_w,
                                             uint step_w,
                                             uint mod_size)
{
    Tensor4D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z,
        .stride_w                      = stride_w
    };

    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w;
    return tensor;
}


inline __global const uchar *vector_offset(const Vector *vec, int x)
{
    return vec->ptr + x * vec->stride_x;
}


inline __global uchar *offset(const Image *img, int x, int y)
{
    return img->ptr + x * img->stride_x + y * img->stride_y;
}


inline __global const uchar *tensor3D_offset(const Tensor3D *tensor, int x, int y, int z)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z;
}


inline __global const uchar *tensor4D_offset(const Tensor4D *tensor, int x, int y, int z, int w)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + w * tensor->stride_w;
}


inline __global const uchar *tensor3D_index2ptr(const Tensor3D *tensor, uint width, uint height, uint depth, uint index)
{
    uint num_elements = width * height;

    const uint z = index / num_elements;

    index %= num_elements;

    const uint y = index / width;

    index %= width;

    const uint x = index;

    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + tensor->offset_first_element_in_bytes;
}

#endif


#define SCALAR_ACCESS_STR(offset, n0, x) scalar_access_##offset##_##n0(x)
#define SCALAR_ACCESS(offset, n0, x) SCALAR_ACCESS_STR(offset, n0, x)


#define scalar_access_0_1(x) ((x).s0)
#define scalar_access_0_2(x) ((x).s01)
#define scalar_access_0_3(x) ((x).s012)
#define scalar_access_0_4(x) ((x).s0123)
#define scalar_access_0_8(x) ((x).s01234567)
#define scalar_access_0_16(x) ((x).s0123456789ABCDEF)


#define scalar_access_1_1(x) ((x).s1)
#define scalar_access_1_2(x) ((x).s12)
#define scalar_access_1_3(x) ((x).s123)
#define scalar_access_1_4(x) ((x).s1234)
#define scalar_access_1_8(x) ((x).s12345678)


#define scalar_access_2_1(x) ((x).s2)
#define scalar_access_2_2(x) ((x).s23)
#define scalar_access_2_3(x) ((x).s234)
#define scalar_access_2_4(x) ((x).s2345)
#define scalar_access_2_8(x) ((x).s23456789)


#define scalar_access_3_1(x) ((x).s3)
#define scalar_access_3_2(x) ((x).s34)
#define scalar_access_3_3(x) ((x).s345)
#define scalar_access_3_4(x) ((x).s3456)
#define scalar_access_3_8(x) ((x).s3456789A)


#define scalar_access_4_1(x) ((x).s4)
#define scalar_access_4_2(x) ((x).s45)
#define scalar_access_4_3(x) ((x).s456)
#define scalar_access_4_4(x) ((x).s4567)
#define scalar_access_4_8(x) ((x).s456789AB)


#define scalar_access_8_1(x) ((x).s8)
#define scalar_access_8_2(x) ((x).s89)
#define scalar_access_8_3(x) ((x).s89A)
#define scalar_access_8_4(x) ((x).s89AB)
#define scalar_access_8_8(x) ((x).s89ABCDEF)


#define scalar_access_12_1(x) ((x).sC)
#define scalar_access_12_2(x) ((x).sCD)
#define scalar_access_12_3(x) ((x).sCDE)
#define scalar_access_12_4(x) ((x).sCDEF)


#define scalar_access_16_1(x) ((x).sF)


#define LOAD_TENSOR_ROW_0(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    ({})

#define LOAD_TENSOR_ROW_1(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##0) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define LOAD_TENSOR_ROW_2(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_1(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##1) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define LOAD_TENSOR_ROW_3(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_2(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##2) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define LOAD_TENSOR_ROW_4(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_3(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##3) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define LOAD_TENSOR_ROW_5(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_4(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##4) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define LOAD_TENSOR_ROW_6(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_5(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##5) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define LOAD_TENSOR_ROW_7(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_6(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##6) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define LOAD_TENSOR_ROW_8(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_7(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##7) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define LOAD_TENSOR_ROW_9(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_8(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##8) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define LOAD_TENSOR_ROW_10(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_9(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)      \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##9) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define LOAD_TENSOR_ROW_11(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_10(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##A) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define LOAD_TENSOR_ROW_12(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_11(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##B) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define LOAD_TENSOR_ROW_13(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_12(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##C) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define LOAD_TENSOR_ROW_14(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_13(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##D) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define LOAD_TENSOR_ROW_15(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_14(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##E) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define LOAD_TENSOR_ROW_16(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) \
    LOAD_TENSOR_ROW_15(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)     \
    SCALAR_ACCESS(COL_OFFSET, N0, BASENAME##F) = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define LOAD_TENSOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) LOAD_TENSOR_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)
#define LOAD_TENSOR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z) LOAD_TENSOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, COL_OFFSET, STRIDE_Y, Z)



#define LOAD_TENSOR_M0X0(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    ({})

#define LOAD_TENSOR_M0X1(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, N0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);

#define LOAD_TENSOR_M0X2(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, N0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);

#define LOAD_TENSOR_M0X3(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, N0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);

#define LOAD_TENSOR_M0X4(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, N0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);

#define LOAD_TENSOR_M0X5(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, 4, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);       \
    LOAD_TENSOR(M0, 1, DATA_TYPE, a, input_ptr + 4 * sizeof(DATA_TYPE), 4, src_stride_y, zin);

#define LOAD_TENSOR_M0X6(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, 4, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);       \
    LOAD_TENSOR(M0, 2, DATA_TYPE, a, input_ptr + 4 * sizeof(DATA_TYPE), 4, src_stride_y, zin);

#define LOAD_TENSOR_M0X7(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, 4, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);       \
    LOAD_TENSOR(M0, 3, DATA_TYPE, a, input_ptr + 4 * sizeof(DATA_TYPE), 4, src_stride_y, zin);

#define LOAD_TENSOR_M0X8(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, N0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);

#define LOAD_TENSOR_M0X9(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, 8, DATA_TYPE, a, input_ptr 0, src_stride_y, zin);        \
    LOAD_TENSOR(M0, 1, DATA_TYPE, a, input_ptr + 8 * sizeof(DATA_TYPE), 8, src_stride_y, zin);

#define LOAD_TENSOR_M0X10(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, 8, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);        \
    LOAD_TENSOR(M0, 2, DATA_TYPE, a, input_ptr + 8 * sizeof(DATA_TYPE), 8, src_stride_y, zin);

#define LOAD_TENSOR_M0X11(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, 8, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);        \
    LOAD_TENSOR(M0, 3, DATA_TYPE, a, input_ptr + 8 * sizeof(DATA_TYPE), 8, src_stride_y, zin);

#define LOAD_TENSOR_M0X12(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, 8, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);        \
    LOAD_TENSOR(M0, 4, DATA_TYPE, a, input_ptr + 8 * sizeof(DATA_TYPE), 8, src_stride_y, zin);

#define LOAD_TENSOR_M0X13(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin)                  \
    LOAD_TENSOR(M0, 8, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);                         \
    LOAD_TENSOR(M0, 4, DATA_TYPE, a, input_ptr + 8 * sizeof(DATA_TYPE), 8, src_stride_y, zin); \
    LOAD_TENSOR(M0, 1, DATA_TYPE, a, input_ptr + 12 * sizeof(DATA_TYPE), 12, src_stride_y, zin);

#define LOAD_TENSOR_M0X14(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin)                  \
    LOAD_TENSOR(M0, 8, DATA_TYPE, a, input_ptr 0, src_stride_y, zin);                          \
    LOAD_TENSOR(M0, 4, DATA_TYPE, a, input_ptr + 8 * sizeof(DATA_TYPE), 8, src_stride_y, zin); \
    LOAD_TENSOR(M0, 2, DATA_TYPE, a, input_ptr + 12 * sizeof(DATA_TYPE), 12, src_stride_y, zin);

#define LOAD_TENSOR_M0X15(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin)                  \
    LOAD_TENSOR(M0, 8, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);                         \
    LOAD_TENSOR(M0, 4, DATA_TYPE, a, input_ptr + 8 * sizeof(DATA_TYPE), 8, src_stride_y, zin); \
    LOAD_TENSOR(M0, 3, DATA_TYPE, a, input_ptr + 12 * sizeof(DATA_TYPE), 12, src_stride_y, zin);

#define LOAD_TENSOR_M0X16(M0, N0, DATA_TYPE, a, input_ptr, src_stride_y, zin) \
    LOAD_TENSOR(M0, N0, DATA_TYPE, a, input_ptr, 0, src_stride_y, zin);



#define LOAD_TENSOR_M0XN0_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) LOAD_TENSOR_M0X##N0(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define LOAD_TENSOR_M0XN0(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) LOAD_TENSOR_M0XN0_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)


#define LOAD_ROW_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##0 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 0 * STRIDE_Y + Z##0));

#define LOAD_ROW_2(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##1 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 1 * STRIDE_Y + Z##1));

#define LOAD_ROW_3(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_2(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##2 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 2 * STRIDE_Y + Z##2));

#define LOAD_ROW_4(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_3(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##3 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 3 * STRIDE_Y + Z##3));

#define LOAD_ROW_5(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_4(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##4 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 4 * STRIDE_Y + Z##4));

#define LOAD_ROW_6(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_5(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##5 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 5 * STRIDE_Y + Z##5));

#define LOAD_ROW_7(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_6(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##6 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 6 * STRIDE_Y + Z##6));

#define LOAD_ROW_8(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_7(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##7 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 7 * STRIDE_Y + Z##7));

#define LOAD_ROW_9(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_8(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                      \
    BASENAME##8 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 8 * STRIDE_Y + Z##8));

#define LOAD_ROW_10(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_9(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)      \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##9 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 9 * STRIDE_Y + Z##9));

#define LOAD_ROW_11(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_10(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##A = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 10 * STRIDE_Y + Z##A));

#define LOAD_ROW_12(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_11(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##B = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 11 * STRIDE_Y + Z##B));

#define LOAD_ROW_13(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_12(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##C = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 12 * STRIDE_Y + Z##C));

#define LOAD_ROW_14(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_13(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##D = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 13 * STRIDE_Y + Z##D));

#define LOAD_ROW_15(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_14(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##E = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 14 * STRIDE_Y + Z##E));

#define LOAD_ROW_16(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_15(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##F = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + 15 * STRIDE_Y + Z##F));




#define LOAD_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
#define LOAD_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)



#define LOAD_ROW_PARTIAL_1(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + OFFSET + 0 * STRIDE_Y + Z##0));

#define LOAD_ROW_PARTIAL_2(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_1(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + OFFSET + 1 * STRIDE_Y + Z##1));

#define LOAD_ROW_PARTIAL_3(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_2(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + OFFSET + 2 * STRIDE_Y + Z##2));

#define LOAD_ROW_PARTIAL_4(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_3(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + OFFSET + 3 * STRIDE_Y + Z##3));

#define LOAD_ROW_PARTIAL_5(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_4(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + OFFSET + 4 * STRIDE_Y + Z##4));

#define LOAD_ROW_PARTIAL_6(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_5(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + OFFSET + 5 * STRIDE_Y + Z##5));

#define LOAD_ROW_PARTIAL_7(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_6(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + OFFSET + 6 * STRIDE_Y + Z##6));

#define LOAD_ROW_PARTIAL_8(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_7(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + OFFSET + 7 * STRIDE_Y + Z##7));

#define LOAD_ROW_PARTIAL_9(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_8(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                         \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + OFFSET + 8 * STRIDE_Y + Z##8));

#define LOAD_ROW_PARTIAL_10(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_9(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)      \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                          \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + OFFSET + 9 * STRIDE_Y + Z##9));

#define LOAD_ROW_PARTIAL_11(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_10(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                          \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + OFFSET + 10 * STRIDE_Y + Z##A));

#define LOAD_ROW_PARTIAL_12(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_11(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                          \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + OFFSET + 11 * STRIDE_Y + Z##B));

#define LOAD_ROW_PARTIAL_13(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_12(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                          \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + OFFSET + 12 * STRIDE_Y + Z##C));

#define LOAD_ROW_PARTIAL_14(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_13(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                          \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + OFFSET + 13 * STRIDE_Y + Z##D));

#define LOAD_ROW_PARTIAL_15(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_14(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                          \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + OFFSET + 14 * STRIDE_Y + Z##E));

#define LOAD_ROW_PARTIAL_16(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) \
    LOAD_ROW_PARTIAL_15(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)     \
    VLOAD_PARTIAL(N0, LOAD_N0)                                                          \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + OFFSET + 15 * STRIDE_Y + Z##F));



#define LOAD_BLOCK_PARTIAL_STR(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_ROW_PARTIAL_##LOAD_M0(N0, LOAD_N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)
#define LOAD_BLOCK_PARTIAL(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z) LOAD_BLOCK_PARTIAL_STR(LOAD_M0, LOAD_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)

#define LOAD_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                                   \
    {                                                                                                                                                            \
        LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                                                                           \
    }                                                                                                                                                            \
    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                               \
    {                                                                                                                                                            \
        LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                            \
    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                               \
    {                                                                                                                                                            \
        LOAD_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                            \
    else                                                                                                                                                         \
    {                                                                                                                                                            \
        LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                                               \
    }

#define LOAD_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X))                                                                                                \
    {                                                                                                                    \
        LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                                   \
    }                                                                                                                    \
    else                                                                                                                 \
    {                                                                                                                    \
        LOAD_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                     \
    }

#define LOAD_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
    if(!(PARTIAL_COND_Y))                                                                                                \
    {                                                                                                                    \
        LOAD_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                                   \
    }                                                                                                                    \
    else                                                                                                                 \
    {                                                                                                                    \
        LOAD_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z);                     \
    }


#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0

#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    LOAD_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z)

#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0

#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0);                                                                                 \
    LOAD_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)

#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0

#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0);                                                                                 \
    LOAD_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)

#else

#define LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE, N0), BASENAME, 0);                                                                                 \
    LOAD_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)

#endif


#define LOAD_TEXTURE2D_ROW_1(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    BASENAME##0 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 0 * X_STEP_ROW), (Y_COORD + 0 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_2(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_1(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##1 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 1 * X_STEP_ROW), (Y_COORD + 1 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_3(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_2(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##2 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 2 * X_STEP_ROW), (Y_COORD + 2 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_4(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_3(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##3 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 3 * X_STEP_ROW), (Y_COORD + 3 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_5(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_4(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##4 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 4 * X_STEP_ROW), (Y_COORD + 4 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_6(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_5(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##5 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 5 * X_STEP_ROW), (Y_COORD + 5 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_7(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_6(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##6 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 6 * X_STEP_ROW), (Y_COORD + 6 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_8(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_7(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##7 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 7 * X_STEP_ROW), (Y_COORD + 7 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_9(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_8(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##8 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 8 * X_STEP_ROW), (Y_COORD + 8 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_10(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_9(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)      \
    BASENAME##9 = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 9 * X_STEP_ROW), (Y_COORD + 9 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_11(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_10(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##A = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 10 * X_STEP_ROW), (Y_COORD + 10 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_12(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_11(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##B = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 11 * X_STEP_ROW), (Y_COORD + 11 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_13(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_12(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##C = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 12 * X_STEP_ROW), (Y_COORD + 12 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_14(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_13(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##D = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 13 * X_STEP_ROW), (Y_COORD + 13 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_15(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_14(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##E = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 14 * X_STEP_ROW), (Y_COORD + 14 * Y_STEP_ROW))

#define LOAD_TEXTURE2D_ROW_16(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) \
    LOAD_TEXTURE2D_ROW_15(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)     \
    BASENAME##F = READ_IMAGE2D(DATA_TYPE, N0, IMG, (X_COORD + 15 * X_STEP_ROW), (Y_COORD + 15 * Y_STEP_ROW))



#define LOAD_TEXTURE2D_STR(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) LOAD_TEXTURE2D_ROW_##M0(N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)
#define LOAD_TEXTURE2D(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW) LOAD_TEXTURE2D_STR(M0, N0, DATA_TYPE, BASENAME, IMG, X_COORD, Y_COORD, X_STEP_ROW, Y_STEP_ROW)



#define LOAD_ROW_INDIRECT_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##0;                                                                            \
    if(Y_MASK##0 != 0)                                                                      \
        BASENAME##0 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##0 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##0 = 0;

#define LOAD_ROW_INDIRECT_2(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##1;                                                                            \
    if(Y_MASK##1 != 0)                                                                      \
        BASENAME##1 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##1 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##1 = 0;

#define LOAD_ROW_INDIRECT_3(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_2(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##2;                                                                            \
    if(Y_MASK##2 != 0)                                                                      \
        BASENAME##2 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##2 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##2 = 0;

#define LOAD_ROW_INDIRECT_4(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_3(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##3;                                                                            \
    if(Y_MASK##3 != 0)                                                                      \
        BASENAME##3 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##3 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##3 = 0;

#define LOAD_ROW_INDIRECT_5(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_4(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##4;                                                                            \
    if(Y_MASK##4 != 0)                                                                      \
        BASENAME##4 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##4 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##4 = 0;

#define LOAD_ROW_INDIRECT_6(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_5(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##5;                                                                            \
    if(Y_MASK##5 != 0)                                                                      \
        BASENAME##5 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##5 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##5 = 0;

#define LOAD_ROW_INDIRECT_7(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_6(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##6;                                                                            \
    if(Y_MASK##6 != 0)                                                                      \
        BASENAME##6 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##6 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##6 = 0;

#define LOAD_ROW_INDIRECT_8(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_7(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##7;                                                                            \
    if(Y_MASK##7 != 0)                                                                      \
        BASENAME##7 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##7 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##7 = 0;

#define LOAD_ROW_INDIRECT_9(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)      \
    LOAD_ROW_INDIRECT_8(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##8;                                                                            \
    if(Y_MASK##8 != 0)                                                                      \
        BASENAME##8 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##8 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##8 = 0;

#define LOAD_ROW_INDIRECT_10(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)     \
    LOAD_ROW_INDIRECT_9(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)          \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##9;                                                                            \
    if(Y_MASK##9 != 0)                                                                      \
        BASENAME##9 = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##9 * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##9 = 0;

#define LOAD_ROW_INDIRECT_11(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)     \
    LOAD_ROW_INDIRECT_10(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)         \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##A;                                                                            \
    if(Y_MASK##A != 0)                                                                      \
        BASENAME##A = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##A * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##A = 0;

#define LOAD_ROW_INDIRECT_12(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)     \
    LOAD_ROW_INDIRECT_11(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)         \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##B;                                                                            \
    if(Y_MASK##B != 0)                                                                      \
        BASENAME##B = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##B * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##B = 0;

#define LOAD_ROW_INDIRECT_13(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)     \
    LOAD_ROW_INDIRECT_12(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)         \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##C;                                                                            \
    if(Y_MASK##C != 0)                                                                      \
        BASENAME##C = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##C * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##C = 0;

#define LOAD_ROW_INDIRECT_14(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)     \
    LOAD_ROW_INDIRECT_13(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)         \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##D;                                                                            \
    if(Y_MASK##D != 0)                                                                      \
        BASENAME##D = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##D * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##D = 0;

#define LOAD_ROW_INDIRECT_15(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)     \
    LOAD_ROW_INDIRECT_14(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)         \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##E;                                                                            \
    if(Y_MASK##E != 0)                                                                      \
        BASENAME##E = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##E * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##E = 0;

#define LOAD_ROW_INDIRECT_16(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)     \
    LOAD_ROW_INDIRECT_15(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)         \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                                            \
    BASENAME##F;                                                                            \
    if(Y_MASK##F != 0)                                                                      \
        BASENAME##F = VLOAD(N0)(0, (__global DATA_TYPE *)(PTR + OFFSET + Y##F * STRIDE_Y)); \
    else                                                                                    \
        BASENAME##F = 0;


#define LOAD_BLOCK_INDIRECT_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) LOAD_ROW_INDIRECT_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)
#define LOAD_BLOCK_INDIRECT(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK) LOAD_BLOCK_INDIRECT_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y, Y, Y_MASK)


#define LOAD_ELEMENT_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##0 = *((__global DATA_TYPE *)(PTR + OFFSET + 0 * STRIDE_Y));

#define LOAD_ELEMENT_2(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_1(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##1 = *((__global DATA_TYPE *)(PTR + OFFSET + 1 * STRIDE_Y));

#define LOAD_ELEMENT_3(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_2(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##2 = *((__global DATA_TYPE *)(PTR + OFFSET + 2 * STRIDE_Y));

#define LOAD_ELEMENT_4(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_3(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##3 = *((__global DATA_TYPE *)(PTR + OFFSET + 3 * STRIDE_Y));

#define LOAD_ELEMENT_5(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_4(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##4 = *((__global DATA_TYPE *)(PTR + OFFSET + 4 * STRIDE_Y));

#define LOAD_ELEMENT_6(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_5(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##5 = *((__global DATA_TYPE *)(PTR + OFFSET + 5 * STRIDE_Y));

#define LOAD_ELEMENT_7(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_6(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##6 = *((__global DATA_TYPE *)(PTR + OFFSET + 6 * STRIDE_Y));

#define LOAD_ELEMENT_8(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_7(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##7 = *((__global DATA_TYPE *)(PTR + OFFSET + 7 * STRIDE_Y));

#define LOAD_ELEMENT_9(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_8(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                       \
    BASENAME##8 = *((__global DATA_TYPE *)(PTR + OFFSET + 8 * STRIDE_Y));

#define LOAD_ELEMENT_10(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_9(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)      \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                        \
    BASENAME##9 = *((__global DATA_TYPE *)(PTR + OFFSET + 9 * STRIDE_Y));

#define LOAD_ELEMENT_11(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_10(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                        \
    BASENAME##A = *((__global DATA_TYPE *)(PTR + OFFSET + 10 * STRIDE_Y));

#define LOAD_ELEMENT_12(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_11(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                        \
    BASENAME##B = *((__global DATA_TYPE *)(PTR + OFFSET + 11 * STRIDE_Y));

#define LOAD_ELEMENT_13(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_12(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                        \
    BASENAME##C = *((__global DATA_TYPE *)(PTR + OFFSET + 12 * STRIDE_Y));

#define LOAD_ELEMENT_14(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_13(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                        \
    BASENAME##D = *((__global DATA_TYPE *)(PTR + OFFSET + 13 * STRIDE_Y));

#define LOAD_ELEMENT_15(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_14(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                        \
    BASENAME##E = *((__global DATA_TYPE *)(PTR + OFFSET + 14 * STRIDE_Y));

#define LOAD_ELEMENT_16(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) \
    LOAD_ELEMENT_15(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)     \
    VEC_DATA_TYPE(DATA_TYPE, N0)                                        \
    BASENAME##F = *((__global DATA_TYPE *)(PTR + OFFSET + 15 * STRIDE_Y));




#define LOAD_SCALAR_AS_VECTOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) LOAD_ELEMENT_##M0(N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)
#define LOAD_SCALAR_AS_VECTOR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y) LOAD_SCALAR_AS_VECTOR_STR(M0, N0, DATA_TYPE, BASENAME, PTR, OFFSET, STRIDE_Y)



#define CALCULATE_Z_OFFSET_1(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    Z##0 = (0 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##0 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##0);                                                      \
    Z##0 *= (CROSS_PLANE_PAD * STRIDE_Y);

#define CALCULATE_Z_OFFSET_2(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    CALCULATE_Z_OFFSET_1(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)     \
    Z##1 = (1 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##1 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##1);                                                      \
    Z##1 *= (CROSS_PLANE_PAD * STRIDE_Y);

#define CALCULATE_Z_OFFSET_3(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    CALCULATE_Z_OFFSET_2(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)     \
    Z##2 = (2 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##2 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##2);                                                      \
    Z##2 *= (CROSS_PLANE_PAD * STRIDE_Y);

#define CALCULATE_Z_OFFSET_4(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    CALCULATE_Z_OFFSET_3(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)     \
    Z##3 = (3 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##3 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##3);                                                      \
    Z##3 *= (CROSS_PLANE_PAD * STRIDE_Y);

#define CALCULATE_Z_OFFSET_5(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    CALCULATE_Z_OFFSET_4(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)     \
    Z##4 = (4 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##4 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##4);                                                      \
    Z##4 *= (CROSS_PLANE_PAD * STRIDE_Y);

#define CALCULATE_Z_OFFSET_6(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    CALCULATE_Z_OFFSET_5(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)     \
    Z##5 = (5 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##5 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##5);                                                      \
    Z##5 *= (CROSS_PLANE_PAD * STRIDE_Y);

#define CALCULATE_Z_OFFSET_7(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    CALCULATE_Z_OFFSET_6(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)     \
    Z##6 = (6 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##6 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##6);                                                      \
    Z##6 *= (CROSS_PLANE_PAD * STRIDE_Y);

#define CALCULATE_Z_OFFSET_8(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) \
    CALCULATE_Z_OFFSET_7(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)     \
    Z##7 = (7 + (DATA_TYPE)(Y)) / (DATA_TYPE)HEIGHT_GEMM3D;                                               \
    Z##7 = min((DATA_TYPE)(DEPTH_GEMM3D - 1), Z##7);                                                      \
    Z##7 *= (CROSS_PLANE_PAD * STRIDE_Y);




#define CALCULATE_Z_OFFSET_STR(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) CALCULATE_Z_OFFSET_##M0(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)
#define CALCULATE_Z_OFFSET(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y) CALCULATE_Z_OFFSET_STR(M0, DATA_TYPE, Z, Y, HEIGHT_GEMM3D, DEPTH_GEMM3D, CROSS_PLANE_PAD, STRIDE_Y)



#define SCALE_ROW_1(DATA_TYPE, BASENAME, SCALE) \
    BASENAME##0 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_2(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_1(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##1 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_3(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_2(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##2 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_4(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_3(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##3 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_5(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_4(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##4 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_6(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_5(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##5 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_7(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_6(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##6 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_8(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_7(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##7 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_9(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_8(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##8 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_10(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_9(DATA_TYPE, BASENAME, SCALE)      \
    BASENAME##9 *= (DATA_TYPE)SCALE;

#define SCALE_ROW_11(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_10(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##A *= (DATA_TYPE)SCALE;

#define SCALE_ROW_12(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_11(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##B *= (DATA_TYPE)SCALE;

#define SCALE_ROW_13(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_12(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##C *= (DATA_TYPE)SCALE;

#define SCALE_ROW_14(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_13(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##D *= (DATA_TYPE)SCALE;

#define SCALE_ROW_15(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_14(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##E *= (DATA_TYPE)SCALE;

#define SCALE_ROW_16(DATA_TYPE, BASENAME, SCALE) \
    SCALE_ROW_15(DATA_TYPE, BASENAME, SCALE)     \
    BASENAME##F *= (DATA_TYPE)SCALE;



#define SCALE_BLOCK_STR(N, DATA_TYPE, BASENAME, SCALE) SCALE_ROW_##N(DATA_TYPE, BASENAME, SCALE)
#define SCALE_BLOCK(N, DATA_TYPE, BASENAME, SCALE) SCALE_BLOCK_STR(N, DATA_TYPE, BASENAME, SCALE)



#define COLUMN_VECTOR1(IDX_COL, BASENAME, X, TYPE) \
    TYPE BASENAME##IDX_COL = (TYPE)((X##0).s##IDX_COL);
#define COLUMN_VECTOR2(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 2)                         \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 2))((X##0).s##IDX_COL, (X##1).s##IDX_COL);
#define COLUMN_VECTOR3(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 3)                         \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 3))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL);
#define COLUMN_VECTOR4(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 4)                         \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 4))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL);
#define COLUMN_VECTOR8(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 8)                         \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 8))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL, (X##4).s##IDX_COL, (X##5).s##IDX_COL, (X##6).s##IDX_COL, (X##7).s##IDX_COL);
#define COLUMN_VECTOR16(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 16)                         \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 16))((X##0).s##IDX_COL, (X##1).s##IDX_COL, (X##2).s##IDX_COL, (X##3).s##IDX_COL, (X##4).s##IDX_COL, (X##5).s##IDX_COL, (X##6).s##IDX_COL, (X##7).s##IDX_COL, (X##8).s##IDX_COL, (X##9).s##IDX_COL, (X##A).s##IDX_COL, (X##B).s##IDX_COL, (X##C).s##IDX_COL, (X##D).s##IDX_COL, (X##E).s##IDX_COL, (X##F).s##IDX_COL);



#define COLUMN_VECTOR_SCALAR1(IDX_COL, BASENAME, X, TYPE) \
    TYPE BASENAME##IDX_COL = (TYPE)((X##0));
#define COLUMN_VECTOR_SCALAR2(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 2)                                \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 2))((X##0), (X##1));
#define COLUMN_VECTOR_SCALAR3(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 3)                                \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 3))((X##0), (X##1), (X##2));
#define COLUMN_VECTOR_SCALAR4(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 4)                                \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 4))((X##0), (X##1), (X##2), (X##3));
#define COLUMN_VECTOR_SCALAR8(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 8)                                \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 8))((X##0), (X##1), (X##2), (X##3), (X##4), (X##5), (X##6), (X##7));
#define COLUMN_VECTOR_SCALAR16(IDX_COL, BASENAME, X, TYPE) \
    VEC_DATA_TYPE(TYPE, 16)                                \
    BASENAME##IDX_COL = (VEC_DATA_TYPE(TYPE, 16))((X##0), (X##1), (X##2), (X##3), (X##4), (X##5), (X##6), (X##7), (X##8), (X##9), (X##A), (X##B), (X##C), (X##D), (X##E), (X##F));



#define TRANSPOSE_K0X1(K0, BASENAME, BS, TYPE) \
    COLUMN_VECTOR_SCALAR(K0, 0, BASENAME, BS, TYPE);
#define TRANSPOSE_K0X2(K0, BASENAME, BS, TYPE) \
    COLUMN_VECTOR(K0, 0, BASENAME, BS, TYPE);  \
    COLUMN_VECTOR(K0, 1, BASENAME, BS, TYPE);
#define TRANSPOSE_K0X3(K0, BASENAME, BS, TYPE) \
    TRANSPOSE_K0X2(K0, BASENAME, BS, TYPE);    \
    COLUMN_VECTOR(K0, 2, BASENAME, BS, TYPE);
#define TRANSPOSE_K0X4(K0, BASENAME, BS, TYPE) \
    TRANSPOSE_K0X3(K0, BASENAME, BS, TYPE);    \
    COLUMN_VECTOR(K0, 3, BASENAME, BS, TYPE);
#define TRANSPOSE_K0X8(K0, BASENAME, BS, TYPE) \
    TRANSPOSE_K0X4(K0, BASENAME, BS, TYPE);    \
    COLUMN_VECTOR(K0, 4, BASENAME, BS, TYPE);  \
    COLUMN_VECTOR(K0, 5, BASENAME, BS, TYPE);  \
    COLUMN_VECTOR(K0, 6, BASENAME, BS, TYPE);  \
    COLUMN_VECTOR(K0, 7, BASENAME, BS, TYPE);
#define TRANSPOSE_K0X16(K0, BASENAME, BS, TYPE) \
    TRANSPOSE_K0X8(K0, BASENAME, BS, TYPE);     \
    COLUMN_VECTOR(K0, 8, BASENAME, BS, TYPE);   \
    COLUMN_VECTOR(K0, 9, BASENAME, BS, TYPE);   \
    COLUMN_VECTOR(K0, A, BASENAME, BS, TYPE);   \
    COLUMN_VECTOR(K0, B, BASENAME, BS, TYPE);   \
    COLUMN_VECTOR(K0, C, BASENAME, BS, TYPE);   \
    COLUMN_VECTOR(K0, D, BASENAME, BS, TYPE);   \
    COLUMN_VECTOR(K0, E, BASENAME, BS, TYPE);   \
    COLUMN_VECTOR(K0, F, BASENAME, BS, TYPE);




#define COLUMN_VECTOR(K0, IDX_COL, BASENAME, BS, TYPE) \
    CONCAT(COLUMN_VECTOR, K0)                          \
    (IDX_COL, BASENAME, BS, TYPE);


#define COLUMN_VECTOR_SCALAR(K0, IDX_COL, BASENAME, BS, TYPE) \
    CONCAT(COLUMN_VECTOR_SCALAR, K0)                          \
    (IDX_COL, BASENAME, BS, TYPE);


#define TRANSPOSE_K0XN0(K0, N0, BASENAME, BS, TYPE) \
    CONCAT(TRANSPOSE_K0X, N0)                       \
    (K0, BASENAME, BS, TYPE);


#define ADD_ROW_1(BASENAME, BIAS) \
    BASENAME##0 += BIAS##0;

#define ADD_ROW_2(BASENAME, BIAS) \
    ADD_ROW_1(BASENAME, BIAS)     \
    BASENAME##1 += BIAS##1;

#define ADD_ROW_3(BASENAME, BIAS) \
    ADD_ROW_2(BASENAME, BIAS)     \
    BASENAME##2 += BIAS##2;

#define ADD_ROW_4(BASENAME, BIAS) \
    ADD_ROW_3(BASENAME, BIAS)     \
    BASENAME##3 += BIAS##3;

#define ADD_ROW_5(BASENAME, BIAS) \
    ADD_ROW_4(BASENAME, BIAS)     \
    BASENAME##4 += BIAS##4;

#define ADD_ROW_6(BASENAME, BIAS) \
    ADD_ROW_5(BASENAME, BIAS)     \
    BASENAME##5 += BIAS##5;

#define ADD_ROW_7(BASENAME, BIAS) \
    ADD_ROW_6(BASENAME, BIAS)     \
    BASENAME##6 += BIAS##6;

#define ADD_ROW_8(BASENAME, BIAS) \
    ADD_ROW_7(BASENAME, BIAS)     \
    BASENAME##7 += BIAS##7;

#define ADD_ROW_9(BASENAME, BIAS) \
    ADD_ROW_8(BASENAME, BIAS)     \
    BASENAME##8 += BIAS##8;

#define ADD_ROW_10(BASENAME, BIAS) \
    ADD_ROW_9(BASENAME, BIAS)      \
    BASENAME##9 += BIAS##9;

#define ADD_ROW_11(BASENAME, BIAS) \
    ADD_ROW_10(BASENAME, BIAS)     \
    BASENAME##A += BIAS##A;

#define ADD_ROW_12(BASENAME, BIAS) \
    ADD_ROW_11(BASENAME, BIAS)     \
    BASENAME##B += BIAS##B;

#define ADD_ROW_13(BASENAME, BIAS) \
    ADD_ROW_12(BASENAME, BIAS)     \
    BASENAME##C += BIAS##C;

#define ADD_ROW_14(BASENAME, BIAS) \
    ADD_ROW_13(BASENAME, BIAS)     \
    BASENAME##D += BIAS##D;

#define ADD_ROW_15(BASENAME, BIAS) \
    ADD_ROW_14(BASENAME, BIAS)     \
    BASENAME##E += BIAS##E;

#define ADD_ROW_16(BASENAME, BIAS) \
    ADD_ROW_15(BASENAME, BIAS)     \
    BASENAME##F += BIAS##F;




#define ADD_BLOCK_STR(N, BASENAME, BIAS) ADD_ROW_##N(BASENAME, BIAS)
#define ADD_BLOCK(N, BASENAME, BIAS) ADD_BLOCK_STR(N, BASENAME, BIAS)



#define ADD_ROW_BROADCAST_1(BASENAME, BIAS) \
    BASENAME##0 += BIAS;

#define ADD_ROW_BROADCAST_2(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_1(BASENAME, BIAS)     \
    BASENAME##1 += BIAS;

#define ADD_ROW_BROADCAST_3(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_2(BASENAME, BIAS)     \
    BASENAME##2 += BIAS;

#define ADD_ROW_BROADCAST_4(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_3(BASENAME, BIAS)     \
    BASENAME##3 += BIAS;

#define ADD_ROW_BROADCAST_5(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_4(BASENAME, BIAS)     \
    BASENAME##4 += BIAS;

#define ADD_ROW_BROADCAST_6(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_5(BASENAME, BIAS)     \
    BASENAME##5 += BIAS;

#define ADD_ROW_BROADCAST_7(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_6(BASENAME, BIAS)     \
    BASENAME##6 += BIAS;

#define ADD_ROW_BROADCAST_8(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_7(BASENAME, BIAS)     \
    BASENAME##7 += BIAS;

#define ADD_ROW_BROADCAST_9(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_8(BASENAME, BIAS)     \
    BASENAME##8 += BIAS;

#define ADD_ROW_BROADCAST_10(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_9(BASENAME, BIAS)      \
    BASENAME##9 += BIAS;

#define ADD_ROW_BROADCAST_11(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_10(BASENAME, BIAS)     \
    BASENAME##A += BIAS;

#define ADD_ROW_BROADCAST_12(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_11(BASENAME, BIAS)     \
    BASENAME##B += BIAS;

#define ADD_ROW_BROADCAST_13(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_12(BASENAME, BIAS)     \
    BASENAME##C += BIAS;

#define ADD_ROW_BROADCAST_14(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_13(BASENAME, BIAS)     \
    BASENAME##D += BIAS;

#define ADD_ROW_BROADCAST_15(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_14(BASENAME, BIAS)     \
    BASENAME##E += BIAS;

#define ADD_ROW_BROADCAST_16(BASENAME, BIAS) \
    ADD_ROW_BROADCAST_15(BASENAME, BIAS)     \
    BASENAME##F += BIAS;


#define ADD_BLOCK_BROADCAST_STR(N, BASENAME, BIAS) ADD_ROW_BROADCAST_##N(BASENAME, BIAS)
#define ADD_BLOCK_BROADCAST(N, BASENAME, BIAS) ADD_BLOCK_BROADCAST_STR(N, BASENAME, BIAS)



#define ACTIVATION_ROW_1(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    BASENAME##0 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##0, A_VAL, B_VAL);

#define ACTIVATION_ROW_2(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_1(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##1 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##1, A_VAL, B_VAL);

#define ACTIVATION_ROW_3(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_2(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##2 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##2, A_VAL, B_VAL);

#define ACTIVATION_ROW_4(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_3(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##3 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##3, A_VAL, B_VAL);

#define ACTIVATION_ROW_5(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_4(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##4 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##4, A_VAL, B_VAL);

#define ACTIVATION_ROW_6(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_5(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##5 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##5, A_VAL, B_VAL);

#define ACTIVATION_ROW_7(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_6(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##6 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##6, A_VAL, B_VAL);

#define ACTIVATION_ROW_8(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_7(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##7 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##7, A_VAL, B_VAL);

#define ACTIVATION_ROW_9(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_8(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##8 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##8, A_VAL, B_VAL);

#define ACTIVATION_ROW_10(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_9(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)      \
    BASENAME##9 = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##9, A_VAL, B_VAL);

#define ACTIVATION_ROW_11(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_10(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##A = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##A, A_VAL, B_VAL);

#define ACTIVATION_ROW_12(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_11(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##B = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##B, A_VAL, B_VAL);

#define ACTIVATION_ROW_13(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_12(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##C = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##C, A_VAL, B_VAL);

#define ACTIVATION_ROW_14(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_13(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##D = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##D, A_VAL, B_VAL);

#define ACTIVATION_ROW_15(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_14(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##E = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##E, A_VAL, B_VAL);

#define ACTIVATION_ROW_16(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) \
    ACTIVATION_ROW_15(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)     \
    BASENAME##F = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME##F, A_VAL, B_VAL);



#define ACTIVATION_BLOCK_STR(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) ACTIVATION_ROW_##N(ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)
#define ACTIVATION_BLOCK(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL) ACTIVATION_BLOCK_STR(N, ACTIVATION_TYPE, DATA_TYPE, VEC_SIZE, BASENAME, A_VAL, B_VAL)



#define CONVERT_ROW_1(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##0 = CONVERT(BASENAME_SRC##0, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_2(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_1(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##1 = CONVERT(BASENAME_SRC##1, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_3(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_2(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##2 = CONVERT(BASENAME_SRC##2, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_4(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_3(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##3 = CONVERT(BASENAME_SRC##3, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_5(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_4(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##4 = CONVERT(BASENAME_SRC##4, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_6(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_5(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##5 = CONVERT(BASENAME_SRC##5, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_7(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_6(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##6 = CONVERT(BASENAME_SRC##6, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_8(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_7(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##7 = CONVERT(BASENAME_SRC##7, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_9(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_8(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                 \
    BASENAME_DST##8 = CONVERT(BASENAME_SRC##8, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_10(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_9(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)      \
    VEC_DATA_TYPE(DATA_TYPE, N)                                  \
    BASENAME_DST##9 = CONVERT(BASENAME_SRC##9, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_11(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_10(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                  \
    BASENAME_DST##A = CONVERT(BASENAME_SRC##A, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_12(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_11(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                  \
    BASENAME_DST##B = CONVERT(BASENAME_SRC##B, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_13(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_12(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                  \
    BASENAME_DST##C = CONVERT(BASENAME_SRC##C, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_14(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_13(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                  \
    BASENAME_DST##D = CONVERT(BASENAME_SRC##D, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_15(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_14(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                  \
    BASENAME_DST##E = CONVERT(BASENAME_SRC##E, VEC_DATA_TYPE(DATA_TYPE, N));

#define CONVERT_ROW_16(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) \
    CONVERT_ROW_15(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)     \
    VEC_DATA_TYPE(DATA_TYPE, N)                                  \
    BASENAME_DST##F = CONVERT(BASENAME_SRC##F, VEC_DATA_TYPE(DATA_TYPE, N));



#define CONVERT_BLOCK_STR(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) CONVERT_ROW_##M(N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)
#define CONVERT_BLOCK(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST) CONVERT_BLOCK_STR(M, N, DATA_TYPE, BASENAME_SRC, BASENAME_DST)


#ifndef ARM_COMPUTE_HELPERS_ASYMM_H
#define ARM_COMPUTE_HELPERS_ASYMM_H


#ifndef ARM_COMPUTE_HELPER_H
#define ARM_COMPUTE_HELPER_H




#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE(N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                     \
    (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));




#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_PARTIAL_16(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                            \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                           \
    }                                                                                                                                                     \
    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else                                                                                                                                                  \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                               \
    }

#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }

#define STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
    if(!(PARTIAL_COND_Y))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }


#if defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)


#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)

#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)

#else

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)

#endif

#endif


#if defined(PARTIAL_STORE_M0)

#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(max(0, (int)(y * M0) - (int)((M0 - PARTIAL_STORE_M0) % M0))))
#else
#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(y * M0))
#endif



#define STORE_VECTOR_SELECT(basename, data_type, ptr, vec_size, leftover, cond) \
    STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, cond)


#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : enable
#endif

#if defined(ARM_COMPUTE_DEBUG_ENABLED) && defined(cl_arm_printf)
#pragma OPENCL EXTENSION cl_arm_printf : enable
#endif

#define GPU_ARCH_MIDGARD 0x100
#define GPU_ARCH_BIFROST 0x200
#define GPU_ARCH_VALHALL 0x300


#define CONCAT(a, b) a##b


#define EXPAND(x) x


#define CLAMP(x, min_val, max_val) min(max(x, min_val), max_val)


#define REV1(x) ((x))
#define REV2(x) ((x).s10)
#define REV3(x) ((x).s210)
#define REV4(x) ((x).s3210)
#define REV8(x) ((x).s76543210)
#define REV16(x) ((x).sFEDCBA9876543210)



#define REVERSE_STR(x, s) REV##s((x))
#define REVERSE(x, s) REVERSE_STR(x, s)



#define ROT1_0(x) ((x))
#define ROT1_1(x) ((x))

#define ROT2_0(x) ((x))
#define ROT2_1(x) ((x).s10)
#define ROT2_2(x) ((x))

#define ROT3_0(x) ((x))
#define ROT3_1(x) ((x).s201)
#define ROT3_2(x) ((x).s120)
#define ROT3_3(x) ((x))

#define ROT4_0(x) ((x))
#define ROT4_1(x) ((x).s3012)
#define ROT4_2(x) ((x).s2301)
#define ROT4_3(x) ((x).s1230)
#define ROT4_4(x) ((x))

#define ROT8_0(x) ((x))
#define ROT8_1(x) ((x).s70123456)
#define ROT8_2(x) ((x).s67012345)
#define ROT8_3(x) ((x).s56701234)
#define ROT8_4(x) ((x).s45670123)
#define ROT8_5(x) ((x).s34567012)
#define ROT8_6(x) ((x).s23456701)
#define ROT8_7(x) ((x).s12345670)
#define ROT8_8(x) ((x))

#define ROT16_0(x) ((x))
#define ROT16_1(x) ((x).sF0123456789ABCDE)
#define ROT16_2(x) ((x).sEF0123456789ABCD)
#define ROT16_3(x) ((x).sDEF0123456789ABC)
#define ROT16_4(x) ((x).sCDEF0123456789AB)
#define ROT16_5(x) ((x).sBCDEF0123456789A)
#define ROT16_6(x) ((x).sABCDEF0123456789)
#define ROT16_7(x) ((x).s9ABCDEF012345678)
#define ROT16_8(x) ((x).s89ABCDEF01234567)
#define ROT16_9(x) ((x).s789ABCDEF0123456)
#define ROT16_10(x) ((x).s6789ABCDEF012345)
#define ROT16_11(x) ((x).s56789ABCDEF01234)
#define ROT16_12(x) ((x).s456789ABCDEF0123)
#define ROT16_13(x) ((x).s3456789ABCDEF012)
#define ROT16_14(x) ((x).s23456789ABCDEF01)
#define ROT16_15(x) ((x).s123456789ABCDEF0)
#define ROT16_16(x) ((x))



#define ROTATE_STR(x, s, n) ROT##s##_##n(x)
#define ROTATE(x, s, n) ROTATE_STR(x, s, n)



#define V_OFFS1(dt) (dt##1)(0)
#define V_OFFS2(dt) (dt##2)(0, 1)
#define V_OFFS3(dt) (dt##3)(0, 1, 2)
#define V_OFFS4(dt) (dt##4)(0, 1, 2, 3)
#define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7)
#define V_OFFS16(dt) (dt##16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)



#define VEC_OFFS_STR(dt, s) V_OFFS##s(dt)
#define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s)


#define VLOAD_STR(size) vload##size
#define VLOAD(size) VLOAD_STR(size)


#define VLOAD_PARTIAL_STR(size, load_size) vload_partial_##size##_##load_size
#define VLOAD_PARTIAL(size, load_size) VLOAD_PARTIAL_STR(size, load_size)

#define NO_LOAD(data, offs, ptr) \
    {                            \
    }


#define vload_partial_1_0 NO_LOAD
#define vload_partial_1_1 vload1
#define vload_partial_1_2 NO_LOAD
#define vload_partial_1_3 NO_LOAD
#define vload_partial_1_4 NO_LOAD
#define vload_partial_1_5 NO_LOAD
#define vload_partial_1_6 NO_LOAD
#define vload_partial_1_7 NO_LOAD
#define vload_partial_1_8 NO_LOAD
#define vload_partial_1_9 NO_LOAD
#define vload_partial_1_10 NO_LOAD
#define vload_partial_1_11 NO_LOAD
#define vload_partial_1_12 NO_LOAD
#define vload_partial_1_13 NO_LOAD
#define vload_partial_1_14 NO_LOAD
#define vload_partial_1_15 NO_LOAD
#define vload_partial_1_16 NO_LOAD

#define vload_partial_2_0 NO_LOAD
#define vload_partial_2_1 vload_partial_1
#define vload_partial_2_2 vload_partial_2
#define vload_partial_2_3 NO_LOAD
#define vload_partial_2_4 NO_LOAD
#define vload_partial_2_5 NO_LOAD
#define vload_partial_2_6 NO_LOAD
#define vload_partial_2_7 NO_LOAD
#define vload_partial_2_8 NO_LOAD
#define vload_partial_2_9 NO_LOAD
#define vload_partial_2_10 NO_LOAD
#define vload_partial_2_11 NO_LOAD
#define vload_partial_2_12 NO_LOAD
#define vload_partial_2_13 NO_LOAD
#define vload_partial_2_14 NO_LOAD
#define vload_partial_2_15 NO_LOAD
#define vload_partial_2_16 NO_LOAD

#define vload_partial_3_0 NO_LOAD
#define vload_partial_3_1 vload_partial_1
#define vload_partial_3_2 vload_partial_2
#define vload_partial_3_3 vload_partial_3
#define vload_partial_3_4 NO_LOAD
#define vload_partial_3_5 NO_LOAD
#define vload_partial_3_6 NO_LOAD
#define vload_partial_3_7 NO_LOAD
#define vload_partial_3_8 NO_LOAD
#define vload_partial_3_9 NO_LOAD
#define vload_partial_3_10 NO_LOAD
#define vload_partial_3_11 NO_LOAD
#define vload_partial_3_12 NO_LOAD
#define vload_partial_3_13 NO_LOAD
#define vload_partial_3_14 NO_LOAD
#define vload_partial_3_15 NO_LOAD
#define vload_partial_3_16 NO_LOAD

#define vload_partial_4_0 NO_LOAD
#define vload_partial_4_1 vload_partial_1
#define vload_partial_4_2 vload_partial_2
#define vload_partial_4_3 vload_partial_3
#define vload_partial_4_4 vload_partial_4
#define vload_partial_4_5 NO_LOAD
#define vload_partial_4_6 NO_LOAD
#define vload_partial_4_7 NO_LOAD
#define vload_partial_4_8 NO_LOAD
#define vload_partial_4_9 NO_LOAD
#define vload_partial_4_10 NO_LOAD
#define vload_partial_4_11 NO_LOAD
#define vload_partial_4_12 NO_LOAD
#define vload_partial_4_13 NO_LOAD
#define vload_partial_4_14 NO_LOAD
#define vload_partial_4_15 NO_LOAD
#define vload_partial_4_16 NO_LOAD

#define vload_partial_8_0 NO_LOAD
#define vload_partial_8_1 vload_partial_1
#define vload_partial_8_2 vload_partial_2
#define vload_partial_8_3 vload_partial_3
#define vload_partial_8_4 vload_partial_4
#define vload_partial_8_5 vload_partial_5
#define vload_partial_8_6 vload_partial_6
#define vload_partial_8_7 vload_partial_7
#define vload_partial_8_8 vload_partial_8
#define vload_partial_8_9 NO_LOAD
#define vload_partial_8_10 NO_LOAD
#define vload_partial_8_11 NO_LOAD
#define vload_partial_8_12 NO_LOAD
#define vload_partial_8_13 NO_LOAD
#define vload_partial_8_14 NO_LOAD
#define vload_partial_8_15 NO_LOAD
#define vload_partial_8_16 NO_LOAD

#define vload_partial_16_0 NO_LOAD
#define vload_partial_16_1 vload_partial_1
#define vload_partial_16_2 vload_partial_2
#define vload_partial_16_3 vload_partial_3
#define vload_partial_16_4 vload_partial_4
#define vload_partial_16_5 vload_partial_5
#define vload_partial_16_6 vload_partial_6
#define vload_partial_16_7 vload_partial_7
#define vload_partial_16_8 vload_partial_8
#define vload_partial_16_9 vload_partial_9
#define vload_partial_16_10 vload_partial_10
#define vload_partial_16_11 vload_partial_11
#define vload_partial_16_12 vload_partial_12
#define vload_partial_16_13 vload_partial_13
#define vload_partial_16_14 vload_partial_14
#define vload_partial_16_15 vload_partial_15
#define vload_partial_16_16 vload_partial_16


#define vload_partial_1(DATA, OFFSET, PTR) \
    DATA.s0 = vload1(OFFSET, PTR);

#define vload_partial_2(DATA, OFFSET, PTR) \
    DATA.s01 = vload2(OFFSET, PTR);

#define vload_partial_3(DATA, OFFSET, PTR) \
    DATA.s012 = vload3(OFFSET, PTR);

#define vload_partial_4(DATA, OFFSET, PTR) \
    DATA.s0123 = vload4(OFFSET, PTR);

#define vload_partial_5(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    DATA.s4 = vload1(OFFSET, PTR + 4);

#define vload_partial_6(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vload_partial_7(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vload_partial_8(DATA, OFFSET, PTR) \
    DATA.s01234567 = vload8(OFFSET, PTR);

#define vload_partial_9(DATA, OFFSET, PTR)        \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    DATA.s8 = vload1(OFFSET, PTR + 8);

#define vload_partial_10(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vload_partial_11(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_3(DATA.s89A, OFFSET, PTR + 8);

#define vload_partial_12(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_4(DATA.s89AB, OFFSET, PTR + 8);

#define vload_partial_13(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_5(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_14(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_6(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_15(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_7(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_16(DATA, OFFSET, PTR) \
    DATA = vload16(OFFSET, PTR);



#define PIXEL_UNIT4 1
#define PIXEL_UNIT8 2
#define PIXEL_UNIT16 4


#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) PIXEL_UNIT##vec_size
#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size)


#define read_image2d_floatx1(img, x_coord, y_coord) (float4)(read_imagef(img, (int2)(x_coord, y_coord)));
#define read_image2d_floatx2(img, x_coord, y_coord) (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_floatx4(img, x_coord, y_coord) (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord)));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define read_image2d_halfx1(img, x_coord, y_coord) (half4)(read_imageh(img, (int2)(x_coord, y_coord)));
#define read_image2d_halfx2(img, x_coord, y_coord) (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_halfx4(img, x_coord, y_coord) (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord)));
#endif

#define write_image2d_floatx1(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values));
#define write_image2d_floatx2(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_floatx4(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imagef(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imagef(img, (int2)(x_coord + 3, y_coord), values.sCDEF));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define write_image2d_halfx1(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values));
#define write_image2d_halfx2(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_halfx4(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imageh(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imageh(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
#endif


#define READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) read_image2d_##data_type##x##n0(img, x_coord, y_coord)
#define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord)


#define WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values) write_image2d_##data_type##x##n0(img, x_coord, y_coord, values)
#define WRITE_IMAGE2D(data_type, n0, img, x_coord, y_coord, values) WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values)

#define VSTORE_STR(size) vstore##size
#define VSTORE(size) VSTORE_STR(size)

#define float1 float
#define half1 half
#define char1 char
#define uchar1 uchar
#define short1 short
#define ushort1 ushort
#define int1 int
#define uint1 uint
#define long1 long
#define ulong1 ulong
#define double1 double

#define vload1(OFFSET, PTR) *(OFFSET + PTR)
#define vstore1(DATA, OFFSET, PTR) *(OFFSET + PTR) = DATA


#define VSTORE_PARTIAL_STR(size, store_size) vstore_partial_##size##_##store_size
#define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size)

#define NO_STORE(data, offs, ptr) \
    {                             \
    }


#define vstore_partial_1_0 NO_STORE
#define vstore_partial_1_1 vstore1
#define vstore_partial_1_2 NO_STORE
#define vstore_partial_1_3 NO_STORE
#define vstore_partial_1_4 NO_STORE
#define vstore_partial_1_5 NO_STORE
#define vstore_partial_1_6 NO_STORE
#define vstore_partial_1_7 NO_STORE
#define vstore_partial_1_8 NO_STORE
#define vstore_partial_1_9 NO_STORE
#define vstore_partial_1_10 NO_STORE
#define vstore_partial_1_11 NO_STORE
#define vstore_partial_1_12 NO_STORE
#define vstore_partial_1_13 NO_STORE
#define vstore_partial_1_14 NO_STORE
#define vstore_partial_1_15 NO_STORE
#define vstore_partial_1_16 NO_STORE

#define vstore_partial_2_0 NO_STORE
#define vstore_partial_2_1 vstore_partial_1
#define vstore_partial_2_2 vstore_partial_2
#define vstore_partial_2_3 NO_STORE
#define vstore_partial_2_4 NO_STORE
#define vstore_partial_2_5 NO_STORE
#define vstore_partial_2_6 NO_STORE
#define vstore_partial_2_7 NO_STORE
#define vstore_partial_2_8 NO_STORE
#define vstore_partial_2_9 NO_STORE
#define vstore_partial_2_10 NO_STORE
#define vstore_partial_2_11 NO_STORE
#define vstore_partial_2_12 NO_STORE
#define vstore_partial_2_13 NO_STORE
#define vstore_partial_2_14 NO_STORE
#define vstore_partial_2_15 NO_STORE
#define vstore_partial_2_16 NO_STORE

#define vstore_partial_3_0 NO_STORE
#define vstore_partial_3_1 vstore_partial_1
#define vstore_partial_3_2 vstore_partial_2
#define vstore_partial_3_3 vstore_partial_3
#define vstore_partial_3_4 NO_STORE
#define vstore_partial_3_5 NO_STORE
#define vstore_partial_3_6 NO_STORE
#define vstore_partial_3_7 NO_STORE
#define vstore_partial_3_8 NO_STORE
#define vstore_partial_3_9 NO_STORE
#define vstore_partial_3_10 NO_STORE
#define vstore_partial_3_11 NO_STORE
#define vstore_partial_3_12 NO_STORE
#define vstore_partial_3_13 NO_STORE
#define vstore_partial_3_14 NO_STORE
#define vstore_partial_3_15 NO_STORE
#define vstore_partial_3_16 NO_STORE

#define vstore_partial_4_0 NO_STORE
#define vstore_partial_4_1 vstore_partial_1
#define vstore_partial_4_2 vstore_partial_2
#define vstore_partial_4_3 vstore_partial_3
#define vstore_partial_4_4 vstore_partial_4
#define vstore_partial_4_5 NO_STORE
#define vstore_partial_4_6 NO_STORE
#define vstore_partial_4_7 NO_STORE
#define vstore_partial_4_8 NO_STORE
#define vstore_partial_4_9 NO_STORE
#define vstore_partial_4_10 NO_STORE
#define vstore_partial_4_11 NO_STORE
#define vstore_partial_4_12 NO_STORE
#define vstore_partial_4_13 NO_STORE
#define vstore_partial_4_14 NO_STORE
#define vstore_partial_4_15 NO_STORE
#define vstore_partial_4_16 NO_STORE

#define vstore_partial_8_0 NO_STORE
#define vstore_partial_8_1 vstore_partial_1
#define vstore_partial_8_2 vstore_partial_2
#define vstore_partial_8_3 vstore_partial_3
#define vstore_partial_8_4 vstore_partial_4
#define vstore_partial_8_5 vstore_partial_5
#define vstore_partial_8_6 vstore_partial_6
#define vstore_partial_8_7 vstore_partial_7
#define vstore_partial_8_8 vstore_partial_8
#define vstore_partial_8_9 NO_STORE
#define vstore_partial_8_10 NO_STORE
#define vstore_partial_8_11 NO_STORE
#define vstore_partial_8_12 NO_STORE
#define vstore_partial_8_13 NO_STORE
#define vstore_partial_8_14 NO_STORE
#define vstore_partial_8_15 NO_STORE
#define vstore_partial_8_16 NO_STORE

#define vstore_partial_16_0 NO_STORE
#define vstore_partial_16_1 vstore_partial_1
#define vstore_partial_16_2 vstore_partial_2
#define vstore_partial_16_3 vstore_partial_3
#define vstore_partial_16_4 vstore_partial_4
#define vstore_partial_16_5 vstore_partial_5
#define vstore_partial_16_6 vstore_partial_6
#define vstore_partial_16_7 vstore_partial_7
#define vstore_partial_16_8 vstore_partial_8
#define vstore_partial_16_9 vstore_partial_9
#define vstore_partial_16_10 vstore_partial_10
#define vstore_partial_16_11 vstore_partial_11
#define vstore_partial_16_12 vstore_partial_12
#define vstore_partial_16_13 vstore_partial_13
#define vstore_partial_16_14 vstore_partial_14
#define vstore_partial_16_15 vstore_partial_15
#define vstore_partial_16_16 vstore_partial_16


#define vstore_partial_1(DATA, OFFSET, PTR) \
    vstore1(DATA.s0, OFFSET, PTR);

#define vstore_partial_2(DATA, OFFSET, PTR) \
    vstore2(DATA.s01, OFFSET, PTR);

#define vstore_partial_3(DATA, OFFSET, PTR) \
    vstore3(DATA.s012, OFFSET, PTR);

#define vstore_partial_4(DATA, OFFSET, PTR) \
    vstore4(DATA.s0123, OFFSET, PTR);

#define vstore_partial_5(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore1(DATA.s4, OFFSET, PTR + 4);

#define vstore_partial_6(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vstore_partial_7(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vstore_partial_8(DATA, OFFSET, PTR) \
    vstore8(DATA.s01234567, OFFSET, PTR);

#define vstore_partial_9(DATA, OFFSET, PTR)        \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore1(DATA.s8, OFFSET, PTR + 8);

#define vstore_partial_10(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vstore_partial_11(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_3(DATA.s89a, OFFSET, PTR + 8);

#define vstore_partial_12(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_4(DATA.s89ab, OFFSET, PTR + 8);

#define vstore_partial_13(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_5(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_14(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_6(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_15(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_7(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_16(DATA, OFFSET, PTR) \
    vstore16(DATA, OFFSET, PTR);





#define convert_float_sat convert_float
#define convert_float1_sat convert_float
#define convert_float2_sat convert_float2
#define convert_float3_sat convert_float3
#define convert_float4_sat convert_float4
#define convert_float8_sat convert_float8
#define convert_float16_sat convert_float16
#define convert_half_sat convert_float
#define convert_half1_sat convert_half
#define convert_half2_sat convert_half2
#define convert_half3_sat convert_half3
#define convert_half4_sat convert_half4
#define convert_half8_sat convert_half8
#define convert_half16_sat convert_half16

#define convert_float1 convert_float
#define convert_half1 convert_half
#define convert_char1 convert_char
#define convert_uchar1 convert_uchar
#define convert_short1 convert_short
#define convert_ushort1 convert_ushort
#define convert_int1 convert_int
#define convert_uint1 convert_uint
#define convert_long1 convert_long
#define convert_ulong1 convert_ulong
#define convert_double1 convert_double

#define convert_char1_sat convert_char_sat
#define convert_uchar1_sat convert_uchar_sat
#define convert_uchar2_sat convert_uchar2_sat
#define convert_uchar3_sat convert_uchar3_sat
#define convert_uchar4_sat convert_uchar4_sat
#define convert_uchar8_sat convert_uchar8_sat
#define convert_uchar16_sat convert_uchar16_sat
#define convert_short1_sat convert_short_sat
#define convert_ushort1_sat convert_ushort_sat
#define convert_int1_sat convert_int_sat
#define convert_uint1_sat convert_uint_sat
#define convert_long1_sat convert_long_sat
#define convert_ulong1_sat convert_ulong_sat
#define convert_double1_sat convert_double_sat

#define VEC_DATA_TYPE_STR(type, size) type##size
#define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size)

#define CONVERT_STR(x, type) (convert_##type((x)))
#define CONVERT(x, type) CONVERT_STR(x, type)

#define CONVERT_SAT_STR(x, type) (convert_##type##_sat((x)))
#define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type)

#define CONVERT_SAT_ROUND_STR(x, type, round) (convert_##type##_sat_##round((x)))
#define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round)

#define select_vec_dt_uchar(size) uchar##size
#define select_vec_dt_char(size) char##size
#define select_vec_dt_ushort(size) ushort##size
#define select_vec_dt_short(size) short##size
#define select_vec_dt_half(size) short##size
#define select_vec_dt_uint(size) uint##size
#define select_vec_dt_int(size) int##size
#define select_vec_dt_float(size) int##size
#define select_vec_dt_ulong(size) ulong##size
#define select_vec_dt_long(size) long##size

#define SELECT_VEC_DATA_TYPE_STR(type, size) select_vec_dt_##type(size)
#define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size)
#define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1)

#define signed_int_vec_dt_uchar(size) char##size
#define signed_int_vec_dt_char(size) char##size
#define signed_int_vec_dt_ushort(size) short##size
#define signed_int_vec_dt_short(size) short##size
#define signed_int_vec_dt_half(size) short##size
#define signed_int_vec_dt_uint(size) int##size
#define signed_int_vec_dt_int(size) int##size
#define signed_int_vec_dt_float(size) int##size
#define signed_int_vec_dt_ulong(size) long##size
#define signed_int_vec_dt_long(size) long##size

#define SIGNED_INT_VEC_DATA_TYPE_STR(type, size) signed_int_vec_dt_##type(size)
#define SIGNED_INT_VEC_DATA_TYPE(type, size) SIGNED_INT_VEC_DATA_TYPE_STR(type, size)
#define SIGNED_INT_DATA_TYPE(type) SIGNED_INT_VEC_DATA_TYPE_STR(type, 1)

#define sum_reduce_1(x) (x)
#define sum_reduce_2(x) ((x).s0) + ((x).s1)
#define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2)
#define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23)
#define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567)
#define sum_reduce_16(x) sum_reduce_8((x).s01234567) + sum_reduce_8((x).s89ABCDEF)

#define SUM_REDUCE_STR(x, size) sum_reduce_##size(x)
#define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size)

#define prod_reduce_1(x) (x)
#define prod_reduce_2(x) ((x).s0) * ((x).s1)
#define prod_reduce_3(x) prod_reduce_2((x).s01) * ((x).s2)
#define prod_reduce_4(x) prod_reduce_2((x).s01) * prod_reduce_2((x).s23)
#define prod_reduce_8(x) prod_reduce_4((x).s0123) * prod_reduce_4((x).s4567)
#define prod_reduce_16(x) prod_reduce_8((x).s01234567) * prod_reduce_8((x).s89ABCDEF)

#define PROD_REDUCE_STR(x, size) prod_reduce_##size(x)
#define PROD_REDUCE(x, size) PROD_REDUCE_STR(x, size)

#define max_reduce_1(x) (x)
#define max_reduce_2(x) max(((x).s0), ((x).s1))
#define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2))
#define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23))
#define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567))
#define max_reduce_16(x) max(max_reduce_8((x).s01234567), max_reduce_8((x).s89ABCDEF))

#define MAX_REDUCE_STR(x, size) max_reduce_##size(x)
#define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size)

#define VECTOR_DECLARATION(name)     \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_offset_first_element_in_bytes

#define IMAGE_DECLARATION(name)      \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR3D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR4D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR5D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_stride_v, \
    uint        name##_step_v,   \
    uint        name##_offset_first_element_in_bytes

#define CONVERT_TO_VECTOR_STRUCT(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x)

#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0)

#define CONVERT_TO_IMAGE_STRUCT(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y)

#define CONVERT_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT(name)                                                                                                           \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0)

#define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size)                                                                                                 \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z, name##_stride_w, name##_step_w, mod_size)

#define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name)                                                                                       \
    tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                           name##_stride_z, name##_step_z)


typedef struct Vector
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
} Vector;


typedef struct Image
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
} Image;


typedef struct Tensor3D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
} Tensor3D;


typedef struct Tensor4D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
    int             stride_w;
} Tensor4D;


inline Vector update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x)
{
    Vector vector =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
    };
    vector.ptr += vector.offset_first_element_in_bytes + get_global_id(0) * step_x;
    return vector;
}


inline Image update_image_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y;
    return img;
}


inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return img;
}


inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return tensor;
}


inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    return tensor;
}

inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z, uint stride_w,
                                             uint step_w,
                                             uint mod_size)
{
    Tensor4D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z,
        .stride_w                      = stride_w
    };

    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w;
    return tensor;
}


inline __global const uchar *vector_offset(const Vector *vec, int x)
{
    return vec->ptr + x * vec->stride_x;
}


inline __global uchar *offset(const Image *img, int x, int y)
{
    return img->ptr + x * img->stride_x + y * img->stride_y;
}


inline __global const uchar *tensor3D_offset(const Tensor3D *tensor, int x, int y, int z)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z;
}


inline __global const uchar *tensor4D_offset(const Tensor4D *tensor, int x, int y, int z, int w)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + w * tensor->stride_w;
}


inline __global const uchar *tensor3D_index2ptr(const Tensor3D *tensor, uint width, uint height, uint depth, uint index)
{
    uint num_elements = width * height;

    const uint z = index / num_elements;

    index %= num_elements;

    const uint y = index / width;

    index %= width;

    const uint x = index;

    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + tensor->offset_first_element_in_bytes;
}

#endif


#define CONVERT_DOWN_RTE_STR(x, type) (convert_##type##_rte((x)))
#define CONVERT_DOWN_RTE(x, type) CONVERT_DOWN_RTE_STR(x, type)


inline uchar quantize_qasymm8(float input, float offset, float scale)
{
    float out_f32 = input / scale + offset;
    uchar res_u8  = CONVERT_SAT(CONVERT_DOWN_RTE(out_f32, int), uchar);
    return res_u8;
}


inline float dequantize_qasymm8(uchar input, float offset, float scale)
{
    return ((float)input - offset) * scale;
}


inline float dequantize_qasymm8_signed(char input, float offset, float scale)
{
    return ((float)input - offset) * scale;
}


#define QUANTIZE_IMPL(type, size)                                                                                       \
    inline VEC_DATA_TYPE(type, size) quantize_##type##size(VEC_DATA_TYPE(float, size) input, float offset, float scale) \
    {                                                                                                                   \
        VEC_DATA_TYPE(float, size)                                                                                      \
        out_f32 = input / (VEC_DATA_TYPE(float, size))(scale) + (VEC_DATA_TYPE(float, size))(offset);                   \
        VEC_DATA_TYPE(type, size)                                                                                       \
        res = CONVERT_SAT(CONVERT_DOWN_RTE(out_f32, VEC_DATA_TYPE(int, size)), VEC_DATA_TYPE(type, size));              \
        return res;                                                                                                     \
    }


#define DEQUANTIZE_IMPL(type, size)                                                                                       \
    inline VEC_DATA_TYPE(float, size) dequantize_##type##size(VEC_DATA_TYPE(type, size) input, float offset, float scale) \
    {                                                                                                                     \
        return (CONVERT(input, VEC_DATA_TYPE(float, size)) - offset) * scale;                                             \
    }


#define ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(size)                                                                                        \
    inline VEC_DATA_TYPE(int, size) asymm_rounding_divide_by_POW2_##size(VEC_DATA_TYPE(int, size) x, VEC_DATA_TYPE(int, size) exponent) \
    {                                                                                                                                   \
        const VEC_DATA_TYPE(int, size)                                                                                                  \
        zero = (VEC_DATA_TYPE(int, size))0;                                                                                         \
        const VEC_DATA_TYPE(int, size)                                                                                                  \
        one = (VEC_DATA_TYPE(int, size))1;                                                                                          \
        VEC_DATA_TYPE(int, size)                                                                                                        \
        mask = (one << exponent) - one;                                                                                                 \
        VEC_DATA_TYPE(int, size)                                                                                                        \
        threshold = (mask >> 1) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))(x < 0));                                          \
        return (x >> exponent) + select(zero, one, (SELECT_VEC_DATA_TYPE(int, size))((x & mask) > threshold));                          \
    }


#define ASYMM_MULT_IMPL(size)                                                                                \
    inline VEC_DATA_TYPE(int, size) asymm_mult##size(VEC_DATA_TYPE(int, size) a, VEC_DATA_TYPE(int, size) b) \
    {                                                                                                        \
        VEC_DATA_TYPE(int, size)                                                                             \
        overflow = a == b && a == INT_MIN;                                                                   \
        VEC_DATA_TYPE(long, size)                                                                            \
        a_64 = convert_long##size(a);                                                                        \
        VEC_DATA_TYPE(long, size)                                                                            \
        b_64 = convert_long##size(b);                                                                        \
        VEC_DATA_TYPE(long, size)                                                                            \
        ab_64 = a_64 * b_64;                                                                                 \
                                                                                      \
        VEC_DATA_TYPE(long, size)                                                                            \
        mask1 = 1 << 30;                                                                                     \
        VEC_DATA_TYPE(long, size)                                                                            \
        mask2 = 1 - (1 << 30);                                                                               \
        VEC_DATA_TYPE(long, size)                                                                            \
        is_positive_or_zero = ab_64 >= 0;                                                                    \
        VEC_DATA_TYPE(long, size)                                                                            \
        nudge = select(mask2, mask1, (SELECT_VEC_DATA_TYPE(long, size))(is_positive_or_zero));               \
        VEC_DATA_TYPE(long, size)                                                                            \
        mask = 1ll << 31;                                                                                    \
        VEC_DATA_TYPE(int, size)                                                                             \
        ab_x2_high32 = convert_int##size((ab_64 + nudge) / mask);                                            \
        return select(ab_x2_high32, INT_MAX, (SELECT_VEC_DATA_TYPE(int, size))(overflow));                   \
    }


#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(size)                                                    \
    inline VEC_DATA_TYPE(int, size) asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(VEC_DATA_TYPE(int, size) a) \
    {                                                                                                                               \
        const VEC_DATA_TYPE(int, size) constant_term     = 1895147668;                                                              \
        const VEC_DATA_TYPE(int, size) constant_1_over_3 = 715827883;                                                               \
        const int k_fractional_bits = 31;                                                                                           \
        VEC_DATA_TYPE(int, size)                                                                                                    \
        x = a + (1 << (k_fractional_bits - 3));                                                                                     \
        VEC_DATA_TYPE(int, size)                                                                                                    \
        x2 = ASYMM_MULT(x, x, size);                                                                                                \
        VEC_DATA_TYPE(int, size)                                                                                                    \
        x3 = ASYMM_MULT(x2, x, size);                                                                                               \
        VEC_DATA_TYPE(int, size)                                                                                                    \
        x4 = ASYMM_MULT(x2, x2, size);                                                                                              \
        VEC_DATA_TYPE(int, size)                                                                                                    \
        x4_over_4 = ASYMM_ROUNDING_DIVIDE_BY_POW2(x4, 2, size);                                                                     \
        VEC_DATA_TYPE(int, size)                                                                                                    \
        x4_over_24_plus_x3_over_6_plus_x2 = ASYMM_MULT((x4_over_4 + x3), constant_1_over_3, size) + x2;                             \
        VEC_DATA_TYPE(int, size)                                                                                                    \
        x4_over_24_plus_x3_over_6_plus_x2_over_2 = ASYMM_ROUNDING_DIVIDE_BY_POW2(x4_over_24_plus_x3_over_6_plus_x2, 1, size);       \
        return constant_term + ASYMM_MULT(constant_term, x + x4_over_24_plus_x3_over_6_plus_x2_over_2, size);                       \
    }


#define ASYMM_SELECT_USING_MASK_IMPL(size)                                                                                                                                \
    inline VEC_DATA_TYPE(int, size) asymm_select_using_mask##size(VEC_DATA_TYPE(int, size) if_mask, VEC_DATA_TYPE(int, size) then_val, VEC_DATA_TYPE(int, size) else_val) \
    {                                                                                                                                                                     \
        return (if_mask & then_val) ^ (~if_mask & else_val);                                                                                                              \
    }


#define ASYMM_MASK_IF_ZERO_IMPL(size)                                                    \
    inline VEC_DATA_TYPE(int, size) asymm_mask_if_zero##size(VEC_DATA_TYPE(int, size) a) \
    {                                                                                    \
        const VEC_DATA_TYPE(int, size) all_zeros = 0;                                    \
        const VEC_DATA_TYPE(int, size) all_ones  = ~0;                                   \
        return select(all_zeros, all_ones, (SELECT_VEC_DATA_TYPE(int, size))(a == 0));   \
    }


#define ASYMM_MASK_IF_NON_ZERO_IMPL(size)                                                    \
    inline VEC_DATA_TYPE(int, size) asymm_mask_if_non_zero##size(VEC_DATA_TYPE(int, size) a) \
    {                                                                                        \
        const VEC_DATA_TYPE(int, size) all_zeros = 0;                                        \
        const VEC_DATA_TYPE(int, size) all_ones  = ~0;                                       \
        return select(all_zeros, all_ones, (SELECT_VEC_DATA_TYPE(int, size))(a != 0));       \
    }

#define EXP_BARREL_SHIFTER_IMPL(size)                                                                                                                                                                         \
    inline VEC_DATA_TYPE(int, size) exp_barrel_shifter##size(VEC_DATA_TYPE(int, size) result, int exponent, int fp_multiplier, int k_integer_bits, int k_fractional_bits, VEC_DATA_TYPE(int, size) remainder) \
    {                                                                                                                                                                                                         \
        if(k_integer_bits > exponent)                                                                                                                                                                         \
        {                                                                                                                                                                                                     \
            const int k_shift_amount = k_integer_bits > exponent ? k_fractional_bits + exponent : 0;                                                                                                          \
            return ASYMM_SELECT_USING_MASK(                                                                                                                                                                   \
                    ASYMM_MASK_IF_NON_ZERO(remainder & (1 << k_shift_amount), size),                                                                                                                              \
                    ASYMM_MULT(result, fp_multiplier, size), result, size);                                                                                                                                       \
        }                                                                                                                                                                                                     \
        \
        return result;                                                                                                                                                                                        \
    }


#define ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(size)                                                                               \
    inline VEC_DATA_TYPE(int, size) asymm_exp_on_negative_values##size(VEC_DATA_TYPE(int, size) a, int k_integer_bits)        \
    {                                                                                                                         \
        const int k_fractional_bits = 31 - k_integer_bits;                                                                    \
        VEC_DATA_TYPE(int, size)                                                                                              \
        k_one_quarter = 1 << (k_fractional_bits - 2);                                                                         \
        VEC_DATA_TYPE(int, size)                                                                                              \
        mask = k_one_quarter - 1;                                                                                             \
        VEC_DATA_TYPE(int, size)                                                                                              \
        a_mod_quarter_minus_one_quarter = (a & mask) - k_one_quarter;                                                         \
        VEC_DATA_TYPE(int, size)                                                                                              \
        a_mod_quarter_minus_one_quarter_scaled = a_mod_quarter_minus_one_quarter << k_integer_bits;                           \
        VEC_DATA_TYPE(int, size)                                                                                              \
        result = ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a_mod_quarter_minus_one_quarter_scaled, size); \
        VEC_DATA_TYPE(int, size)                                                                                              \
        remainder = a_mod_quarter_minus_one_quarter - a;                                                                      \
        \
        result = EXP_BARREL_SHIFTER(result, -2, 1672461947, k_integer_bits, k_fractional_bits, remainder, size);              \
        result = EXP_BARREL_SHIFTER(result, -1, 1302514674, k_integer_bits, k_fractional_bits, remainder, size);              \
        result = EXP_BARREL_SHIFTER(result, +0, 790015084, k_integer_bits, k_fractional_bits, remainder, size);               \
        result = EXP_BARREL_SHIFTER(result, +1, 290630308, k_integer_bits, k_fractional_bits, remainder, size);               \
        result = EXP_BARREL_SHIFTER(result, +2, 39332535, k_integer_bits, k_fractional_bits, remainder, size);                \
        result = EXP_BARREL_SHIFTER(result, +3, 720401, k_integer_bits, k_fractional_bits, remainder, size);                  \
        result = EXP_BARREL_SHIFTER(result, +4, 242, k_integer_bits, k_fractional_bits, remainder, size);                     \
        \
        if(k_integer_bits > 5)                                                                                                \
        {                                                                                                                     \
            const VEC_DATA_TYPE(int, size) clamp = -(1 << (k_fractional_bits + 5));                                           \
            result = ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_NON_ZERO(a < clamp, size), 0, result, size);                       \
        }                                                                                                                     \
        \
        const VEC_DATA_TYPE(int, size) Q0_one = INT_MAX;                                                                      \
        return ASYMM_SELECT_USING_MASK(ASYMM_MASK_IF_ZERO(a, size), Q0_one, result, size);                                    \
    }


#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(size)                                                                  \
    inline VEC_DATA_TYPE(int, size) asymm_saturating_rounding_mult_by_pow2##size(VEC_DATA_TYPE(int, size) x, int exponent) \
    {                                                                                                                      \
        if(exponent < 0)                                                                                                   \
        {                                                                                                                  \
            return ASYMM_ROUNDING_DIVIDE_BY_POW2(x, -exponent, size);                                                      \
        }                                                                                                                  \
        \
        const VEC_DATA_TYPE(int, size) min = INT_MIN;                                                                      \
        const VEC_DATA_TYPE(int, size) max = INT_MAX;                                                                      \
        int threshold = ((1 << (31 - exponent)) - 1);                                                                      \
        VEC_DATA_TYPE(int, size)                                                                                           \
        positive_mask = ASYMM_MASK_IF_NON_ZERO(x > threshold, size);                                                       \
        VEC_DATA_TYPE(int, size)                                                                                           \
        negative_mask = ASYMM_MASK_IF_NON_ZERO(x < -threshold, size);                                                      \
        VEC_DATA_TYPE(int, size)                                                                                           \
        result = x << exponent;                                                                                            \
        result = ASYMM_SELECT_USING_MASK(positive_mask, max, result, size);                                                \
        result = ASYMM_SELECT_USING_MASK(negative_mask, min, result, size);                                                \
        return result;                                                                                                     \
    }


#define ASYMM_ROUNDING_HALF_SUM_IMPL(size)                                                                                \
    inline VEC_DATA_TYPE(int, size) asymm_rounding_half_sum##size(VEC_DATA_TYPE(int, size) a, VEC_DATA_TYPE(int, size) b) \
    {                                                                                                                     \
        VEC_DATA_TYPE(long, size)                                                                                         \
        a64 = convert_long##size(a);                                                                                      \
        VEC_DATA_TYPE(long, size)                                                                                         \
        b64 = convert_long##size(b);                                                                                      \
        VEC_DATA_TYPE(long, size)                                                                                         \
        sum = a64 + b64;                                                                                                  \
        const VEC_DATA_TYPE(long, size) one       = 1;                                                                    \
        const VEC_DATA_TYPE(long, size) minus_one = -1;                                                                   \
        VEC_DATA_TYPE(long, size)                                                                                         \
        sign = select(minus_one, one, (SELECT_VEC_DATA_TYPE(long, size))(sum >= 0));                                      \
        return convert_int##size((sum + sign) / 2);                                                                       \
    }


#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(size)                                                    \
    inline VEC_DATA_TYPE(int, size) asymm_one_over_one_plus_x_for_x_in_0_1##size(VEC_DATA_TYPE(int, size) a) \
    {                                                                                                        \
        const VEC_DATA_TYPE(int, size) Q0_one = INT_MAX;                                                     \
        const VEC_DATA_TYPE(int, size) Q2_one = 1 << (31 - 2);                                               \
        VEC_DATA_TYPE(int, size)                                                                             \
        half_denominator = ASYMM_ROUNDING_HALF_SUM(a, Q0_one, size);                                         \
        const VEC_DATA_TYPE(int, size) Q2_48_over_17     = 1515870810;                                       \
        const VEC_DATA_TYPE(int, size) Q2_neg_32_over_17 = -1010580540;                                      \
        VEC_DATA_TYPE(int, size)                                                                             \
        x = Q2_48_over_17 + ASYMM_MULT(half_denominator, Q2_neg_32_over_17, size);                           \
        for(int i = 0; i < 3; i++)                                                                           \
        {                                                                                                    \
            VEC_DATA_TYPE(int, size)                                                                         \
            half_denominator_times_x = ASYMM_MULT(half_denominator, x, size);                                \
            VEC_DATA_TYPE(int, size)                                                                         \
            one_minus_half_denominator_times_x = Q2_one - half_denominator_times_x;                          \
            VEC_DATA_TYPE(int, size)                                                                         \
            tmp = ASYMM_MULT(x, one_minus_half_denominator_times_x, size);                                   \
            x   = x + ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(tmp, 2, size);                                  \
        }                                                                                                    \
        return ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(x, 1, size);                                           \
    }


#define ASYMM_RESCALE_IMPL(size)                                                                                                    \
    inline VEC_DATA_TYPE(int, size) asymm_rescale##size(VEC_DATA_TYPE(int, size) value, int src_integer_bits, int dst_integer_bits) \
    {                                                                                                                               \
        int exponent = src_integer_bits - dst_integer_bits;                                                                         \
        return ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(value, exponent, size);                                                       \
    }

#define QUANTIZE_STR(input, offset, scale, type, size) quantize_##type##size(input, offset, scale)
#define QUANTIZE(input, offset, scale, type, size) QUANTIZE_STR(input, offset, scale, type, size)
#define DEQUANTIZE_STR(input, offset, scale, type, size) dequantize_##type##size(input, offset, scale)
#define DEQUANTIZE(input, offset, scale, type, size) DEQUANTIZE_STR(input, offset, scale, type, size)

#define ASYMM_ROUNDING_DIVIDE_BY_POW2_STR(x, exponent, size) asymm_rounding_divide_by_POW2_##size(x, exponent)
#define ASYMM_ROUNDING_DIVIDE_BY_POW2(x, exponent, size) ASYMM_ROUNDING_DIVIDE_BY_POW2_STR(x, exponent, size)
#define ASYMM_MULT_STR(a, b, size) asymm_mult##size(a, b)
#define ASYMM_MULT(a, b, size) ASYMM_MULT_STR(a, b, size)
#define ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(x, quantized_multiplier, left_shift, size) \
    ASYMM_MULT(x *((VEC_DATA_TYPE(int, size))(1) << (-left_shift)), quantized_multiplier, size)
#define ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(x, quantized_multiplier, right_shift, size) \
    ASYMM_ROUNDING_DIVIDE_BY_POW2(ASYMM_MULT(x, quantized_multiplier, size), right_shift, size)
#define ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL(a, size) asymm_exp_on_interval_between_negative_one_quarter_and_0_excl##size(a)
#define ASYMM_SELECT_USING_MASK(if_mask, then_val, else_val, size) asymm_select_using_mask##size(if_mask, then_val, else_val)
#define ASYMM_MASK_IF_ZERO(a, size) asymm_mask_if_zero##size(a)
#define ASYMM_MASK_IF_NON_ZERO(a, size) asymm_mask_if_non_zero##size(a)
#define EXP_BARREL_SHIFTER(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder, size) exp_barrel_shifter##size(result, exponent, fp_multiplier, k_integer_bits, k_fractional_bits, remainder)
#define ASYMM_EXP_ON_NEGATIVE_VALUES_STR(a, k_integer_bits, size) asymm_exp_on_negative_values##size(a, k_integer_bits)
#define ASYMM_EXP_ON_NEGATIVE_VALUES(a, k_integer_bits, size) ASYMM_EXP_ON_NEGATIVE_VALUES_STR(a, k_integer_bits, size)
#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size) asymm_one_over_one_plus_x_for_x_in_0_1##size(a)
#define ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1(a, size) ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_STR(a, size)
#define ASYMM_SATURATING_ROUNDING_MULT_BY_POW2(x, exponent, size) asymm_saturating_rounding_mult_by_pow2##size(x, exponent)
#define ASYMM_ROUNDING_HALF_SUM(a, b, size) asymm_rounding_half_sum##size(a, b)
#define ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size) asymm_rescale##size(value, src_integer_bits, dst_integer_bits)
#define ASYMM_RESCALE(value, src_integer_bits, dst_integer_bits, size) ASYMM_RESCALE_STR(value, src_integer_bits, dst_integer_bits, size)

#define MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(size)                                                                             \
    inline VEC_DATA_TYPE(int, size) multiply_by_quantized_multiplier##size(VEC_DATA_TYPE(int, size) input, int qmul, int shift) \
    {                                                                                                                           \
        const int left_shift  = shift > 0 ? shift : 0;                                                                          \
        const int right_shift = shift > 0 ? 0 : -shift;                                                                         \
        return ASYMM_ROUNDING_DIVIDE_BY_POW2(ASYMM_MULT(input * (1 << left_shift), qmul, size), right_shift, size);             \
    }
#define MULTIPLY_BY_QUANTIZED_MULTIPLIER(input, qmul, shift, size) multiply_by_quantized_multiplier##size(input, qmul, shift)

QUANTIZE_IMPL(uchar, 1)
QUANTIZE_IMPL(char, 1)
QUANTIZE_IMPL(uint, 1)
QUANTIZE_IMPL(int, 1)
QUANTIZE_IMPL(uchar, 2)
QUANTIZE_IMPL(char, 2)
QUANTIZE_IMPL(uint, 2)
QUANTIZE_IMPL(int, 2)
QUANTIZE_IMPL(uchar, 3)
QUANTIZE_IMPL(char, 3)
QUANTIZE_IMPL(uint, 3)
QUANTIZE_IMPL(int, 3)
QUANTIZE_IMPL(uchar, 4)
QUANTIZE_IMPL(ushort, 4)
QUANTIZE_IMPL(short, 4)
QUANTIZE_IMPL(int, 4)
QUANTIZE_IMPL(uchar, 8)
QUANTIZE_IMPL(char, 8)
QUANTIZE_IMPL(uint, 8)
QUANTIZE_IMPL(int, 8)
QUANTIZE_IMPL(uchar, 16)
QUANTIZE_IMPL(char, 16)
QUANTIZE_IMPL(ushort, 16)
QUANTIZE_IMPL(short, 16)
QUANTIZE_IMPL(uint, 16)
QUANTIZE_IMPL(int, 16)

DEQUANTIZE_IMPL(uchar, 1)
DEQUANTIZE_IMPL(char, 1)
DEQUANTIZE_IMPL(uint, 1)
DEQUANTIZE_IMPL(int, 1)
DEQUANTIZE_IMPL(uchar, 2)
DEQUANTIZE_IMPL(char, 2)
DEQUANTIZE_IMPL(uint, 2)
DEQUANTIZE_IMPL(int, 2)
DEQUANTIZE_IMPL(uchar, 3)
DEQUANTIZE_IMPL(char, 3)
DEQUANTIZE_IMPL(uint, 3)
DEQUANTIZE_IMPL(int, 3)
DEQUANTIZE_IMPL(uchar, 4)
DEQUANTIZE_IMPL(ushort, 4)
DEQUANTIZE_IMPL(short, 4)
DEQUANTIZE_IMPL(int, 4)
DEQUANTIZE_IMPL(uchar, 8)
DEQUANTIZE_IMPL(char, 8)
DEQUANTIZE_IMPL(uint, 8)
DEQUANTIZE_IMPL(int, 8)
DEQUANTIZE_IMPL(uchar, 16)
DEQUANTIZE_IMPL(char, 16)
DEQUANTIZE_IMPL(ushort, 16)
DEQUANTIZE_IMPL(short, 16)
DEQUANTIZE_IMPL(uint, 16)
DEQUANTIZE_IMPL(int, 16)

ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(1)
ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(2)
ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(3)
ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(4)
ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(8)
ASYMM_ROUNDING_DIVIDE_BY_POW2_IMPL(16)

ASYMM_MULT_IMPL(1)
ASYMM_MULT_IMPL(2)
ASYMM_MULT_IMPL(3)
ASYMM_MULT_IMPL(4)
ASYMM_MULT_IMPL(8)
ASYMM_MULT_IMPL(16)

ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(1)
ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(2)
ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(3)
ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(4)
ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(8)
ASYMM_EXP_ON_INTERVAL_BETWEEN_NEGATIVE_ONE_QUARTER_AND_0_EXCL_IMPL(16)

ASYMM_SELECT_USING_MASK_IMPL(1)
ASYMM_SELECT_USING_MASK_IMPL(2)
ASYMM_SELECT_USING_MASK_IMPL(3)
ASYMM_SELECT_USING_MASK_IMPL(4)
ASYMM_SELECT_USING_MASK_IMPL(8)
ASYMM_SELECT_USING_MASK_IMPL(16)

ASYMM_MASK_IF_ZERO_IMPL(1)
ASYMM_MASK_IF_ZERO_IMPL(2)
ASYMM_MASK_IF_ZERO_IMPL(3)
ASYMM_MASK_IF_ZERO_IMPL(4)
ASYMM_MASK_IF_ZERO_IMPL(8)
ASYMM_MASK_IF_ZERO_IMPL(16)

ASYMM_MASK_IF_NON_ZERO_IMPL(1)
ASYMM_MASK_IF_NON_ZERO_IMPL(2)
ASYMM_MASK_IF_NON_ZERO_IMPL(3)
ASYMM_MASK_IF_NON_ZERO_IMPL(4)
ASYMM_MASK_IF_NON_ZERO_IMPL(8)
ASYMM_MASK_IF_NON_ZERO_IMPL(16)

EXP_BARREL_SHIFTER_IMPL(1)
EXP_BARREL_SHIFTER_IMPL(2)
EXP_BARREL_SHIFTER_IMPL(3)
EXP_BARREL_SHIFTER_IMPL(4)
EXP_BARREL_SHIFTER_IMPL(8)
EXP_BARREL_SHIFTER_IMPL(16)

ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(1)
ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(2)
ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(3)
ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(4)
ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(8)
ASYMM_EXP_ON_NEGATIVE_VALUES_IMPL(16)

ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(1)
ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(2)
ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(3)
ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(4)
ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(8)
ASYMM_SATURATING_ROUNDING_MULT_BY_POW2_IMPL(16)

ASYMM_ROUNDING_HALF_SUM_IMPL(1)
ASYMM_ROUNDING_HALF_SUM_IMPL(2)
ASYMM_ROUNDING_HALF_SUM_IMPL(3)
ASYMM_ROUNDING_HALF_SUM_IMPL(4)
ASYMM_ROUNDING_HALF_SUM_IMPL(8)
ASYMM_ROUNDING_HALF_SUM_IMPL(16)

ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(1)
ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(2)
ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(3)
ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(4)
ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(8)
ASYMM_ONE_OVER_ONE_PLUS_X_FOR_X_IN_0_1_IMPL(16)

ASYMM_RESCALE_IMPL(1)
ASYMM_RESCALE_IMPL(2)
ASYMM_RESCALE_IMPL(3)
ASYMM_RESCALE_IMPL(4)
ASYMM_RESCALE_IMPL(8)
ASYMM_RESCALE_IMPL(16)

MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(1)
MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(2)
MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(3)
MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(4)
MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(8)
MULTIPLY_BY_QUANTIZED_MULTIPLIER_IMPL(16)

#endif

#ifndef ARM_COMPUTE_REPEAT_H
#define ARM_COMPUTE_REPEAT_H


#ifndef ARM_COMPUTE_HELPER_H
#define ARM_COMPUTE_HELPER_H




#define STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE(N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##0), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_1(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##1), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_2(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##2), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_3(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##3), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_4(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##4), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_5(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##5), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_6(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##6), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_7(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##7), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_8(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                         \
    (CONVERT_SAT((BASENAME##8), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define CONVERT_STORE_ROW_10(N0, DATA, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_9(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE(N0)                                                     \
    (CONVERT_SAT((BASENAME##9), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_10(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##A), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_11(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##B), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_12(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##C), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_13(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##D), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_14(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##E), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define CONVERT_STORE_ROW_16(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    CONVERT_STORE_ROW_15(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE(N0)                                                          \
    (CONVERT_SAT((BASENAME##F), VEC_DATA_TYPE(DATA_TYPE, N0)), 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));




#define STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_ROW_##M0(N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define CONVERT_STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) CONVERT_STORE_BLOCK_STR(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)



#define STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##0, 0, (__global DATA_TYPE *)(PTR + 0 * STRIDE_Y + Z##0));

#define STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_1(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##1, 0, (__global DATA_TYPE *)(PTR + 1 * STRIDE_Y + Z##1));

#define STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_2(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##2, 0, (__global DATA_TYPE *)(PTR + 2 * STRIDE_Y + Z##2));

#define STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_3(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##3, 0, (__global DATA_TYPE *)(PTR + 3 * STRIDE_Y + Z##3));

#define STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_4(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##4, 0, (__global DATA_TYPE *)(PTR + 4 * STRIDE_Y + Z##4));

#define STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_5(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##5, 0, (__global DATA_TYPE *)(PTR + 5 * STRIDE_Y + Z##5));

#define STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_6(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##6, 0, (__global DATA_TYPE *)(PTR + 6 * STRIDE_Y + Z##6));

#define STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_7(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##7, 0, (__global DATA_TYPE *)(PTR + 7 * STRIDE_Y + Z##7));

#define STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_8(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                 \
    (BASENAME##8, 0, (__global DATA_TYPE *)(PTR + 8 * STRIDE_Y + Z##8));

#define STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_9(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)      \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##9, 0, (__global DATA_TYPE *)(PTR + 9 * STRIDE_Y + Z##9));

#define STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_10(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##A, 0, (__global DATA_TYPE *)(PTR + 10 * STRIDE_Y + Z##A));

#define STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_11(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##B, 0, (__global DATA_TYPE *)(PTR + 11 * STRIDE_Y + Z##B));

#define STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_12(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##C, 0, (__global DATA_TYPE *)(PTR + 12 * STRIDE_Y + Z##C));

#define STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_13(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##D, 0, (__global DATA_TYPE *)(PTR + 13 * STRIDE_Y + Z##D));

#define STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_14(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##E, 0, (__global DATA_TYPE *)(PTR + 14 * STRIDE_Y + Z##E));

#define STORE_ROW_PARTIAL_16(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) \
    STORE_ROW_PARTIAL_15(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)     \
    VSTORE_PARTIAL(N0, STORE_N0)                                                  \
    (BASENAME##F, 0, (__global DATA_TYPE *)(PTR + 15 * STRIDE_Y + Z##F));



#define STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_ROW_PARTIAL_##STORE_M0(N0, STORE_N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)
#define STORE_BLOCK_PARTIAL(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z) STORE_BLOCK_PARTIAL_STR(STORE_M0, STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#define STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X) && !(PARTIAL_COND_Y))                                                                                                            \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                                           \
    }                                                                                                                                                     \
    else if((PARTIAL_COND_Y) && !(PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else if(!(PARTIAL_COND_Y) && (PARTIAL_COND_X))                                                                                                        \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                                             \
    }                                                                                                                                                     \
    else                                                                                                                                                  \
    {                                                                                                                                                     \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                               \
    }

#define STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X) \
    if(!(PARTIAL_COND_X))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, PARTIAL_STORE_N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }

#define STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y) \
    if(!(PARTIAL_COND_Y))                                                                                         \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                                   \
    }                                                                                                             \
    else                                                                                                          \
    {                                                                                                             \
        STORE_BLOCK_PARTIAL(PARTIAL_STORE_M0, N0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z);                     \
    }


#if defined(PARTIAL_STORE_M0) && defined(PARTIAL_STORE_N0)


#if PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z)

#elif PARTIAL_STORE_M0 > 0 && PARTIAL_STORE_N0 == 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_COND_Y)

#elif PARTIAL_STORE_M0 == 0 && PARTIAL_STORE_N0 > 0

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_N0, PARTIAL_COND_X)

#else

#define STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X) \
    STORE_BLOCK_PARTIAL_IN_X_AND_Y(M0, N0, DATA_TYPE, BASENAME, PTR, STRIDE_Y, Z, PARTIAL_STORE_M0, PARTIAL_STORE_N0, PARTIAL_COND_Y, PARTIAL_COND_X)

#endif

#endif


#if defined(PARTIAL_STORE_M0)

#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(max(0, (int)(y * M0) - (int)((M0 - PARTIAL_STORE_M0) % M0))))
#else
#define COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) \
    ((uint)(y * M0))
#endif



#define STORE_VECTOR_SELECT(basename, data_type, ptr, vec_size, leftover, cond) \
    STORE_BLOCK_PARTIAL_IN_X(1, vec_size, data_type, basename, ptr, 0, 0, leftover, cond)


#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#pragma OPENCL EXTENSION cl_khr_fp16 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_int8 : enable
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8)
#pragma OPENCL EXTENSION cl_arm_integer_dot_product_accumulate_int8 : enable
#endif

#if defined(ARM_COMPUTE_DEBUG_ENABLED) && defined(cl_arm_printf)
#pragma OPENCL EXTENSION cl_arm_printf : enable
#endif

#define GPU_ARCH_MIDGARD 0x100
#define GPU_ARCH_BIFROST 0x200
#define GPU_ARCH_VALHALL 0x300


#define CONCAT(a, b) a##b


#define EXPAND(x) x


#define CLAMP(x, min_val, max_val) min(max(x, min_val), max_val)


#define REV1(x) ((x))
#define REV2(x) ((x).s10)
#define REV3(x) ((x).s210)
#define REV4(x) ((x).s3210)
#define REV8(x) ((x).s76543210)
#define REV16(x) ((x).sFEDCBA9876543210)



#define REVERSE_STR(x, s) REV##s((x))
#define REVERSE(x, s) REVERSE_STR(x, s)



#define ROT1_0(x) ((x))
#define ROT1_1(x) ((x))

#define ROT2_0(x) ((x))
#define ROT2_1(x) ((x).s10)
#define ROT2_2(x) ((x))

#define ROT3_0(x) ((x))
#define ROT3_1(x) ((x).s201)
#define ROT3_2(x) ((x).s120)
#define ROT3_3(x) ((x))

#define ROT4_0(x) ((x))
#define ROT4_1(x) ((x).s3012)
#define ROT4_2(x) ((x).s2301)
#define ROT4_3(x) ((x).s1230)
#define ROT4_4(x) ((x))

#define ROT8_0(x) ((x))
#define ROT8_1(x) ((x).s70123456)
#define ROT8_2(x) ((x).s67012345)
#define ROT8_3(x) ((x).s56701234)
#define ROT8_4(x) ((x).s45670123)
#define ROT8_5(x) ((x).s34567012)
#define ROT8_6(x) ((x).s23456701)
#define ROT8_7(x) ((x).s12345670)
#define ROT8_8(x) ((x))

#define ROT16_0(x) ((x))
#define ROT16_1(x) ((x).sF0123456789ABCDE)
#define ROT16_2(x) ((x).sEF0123456789ABCD)
#define ROT16_3(x) ((x).sDEF0123456789ABC)
#define ROT16_4(x) ((x).sCDEF0123456789AB)
#define ROT16_5(x) ((x).sBCDEF0123456789A)
#define ROT16_6(x) ((x).sABCDEF0123456789)
#define ROT16_7(x) ((x).s9ABCDEF012345678)
#define ROT16_8(x) ((x).s89ABCDEF01234567)
#define ROT16_9(x) ((x).s789ABCDEF0123456)
#define ROT16_10(x) ((x).s6789ABCDEF012345)
#define ROT16_11(x) ((x).s56789ABCDEF01234)
#define ROT16_12(x) ((x).s456789ABCDEF0123)
#define ROT16_13(x) ((x).s3456789ABCDEF012)
#define ROT16_14(x) ((x).s23456789ABCDEF01)
#define ROT16_15(x) ((x).s123456789ABCDEF0)
#define ROT16_16(x) ((x))



#define ROTATE_STR(x, s, n) ROT##s##_##n(x)
#define ROTATE(x, s, n) ROTATE_STR(x, s, n)



#define V_OFFS1(dt) (dt##1)(0)
#define V_OFFS2(dt) (dt##2)(0, 1)
#define V_OFFS3(dt) (dt##3)(0, 1, 2)
#define V_OFFS4(dt) (dt##4)(0, 1, 2, 3)
#define V_OFFS8(dt) (dt##8)(0, 1, 2, 3, 4, 5, 6, 7)
#define V_OFFS16(dt) (dt##16)(0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15)



#define VEC_OFFS_STR(dt, s) V_OFFS##s(dt)
#define VEC_OFFS(dt, s) VEC_OFFS_STR(dt, s)


#define VLOAD_STR(size) vload##size
#define VLOAD(size) VLOAD_STR(size)


#define VLOAD_PARTIAL_STR(size, load_size) vload_partial_##size##_##load_size
#define VLOAD_PARTIAL(size, load_size) VLOAD_PARTIAL_STR(size, load_size)

#define NO_LOAD(data, offs, ptr) \
    {                            \
    }


#define vload_partial_1_0 NO_LOAD
#define vload_partial_1_1 vload1
#define vload_partial_1_2 NO_LOAD
#define vload_partial_1_3 NO_LOAD
#define vload_partial_1_4 NO_LOAD
#define vload_partial_1_5 NO_LOAD
#define vload_partial_1_6 NO_LOAD
#define vload_partial_1_7 NO_LOAD
#define vload_partial_1_8 NO_LOAD
#define vload_partial_1_9 NO_LOAD
#define vload_partial_1_10 NO_LOAD
#define vload_partial_1_11 NO_LOAD
#define vload_partial_1_12 NO_LOAD
#define vload_partial_1_13 NO_LOAD
#define vload_partial_1_14 NO_LOAD
#define vload_partial_1_15 NO_LOAD
#define vload_partial_1_16 NO_LOAD

#define vload_partial_2_0 NO_LOAD
#define vload_partial_2_1 vload_partial_1
#define vload_partial_2_2 vload_partial_2
#define vload_partial_2_3 NO_LOAD
#define vload_partial_2_4 NO_LOAD
#define vload_partial_2_5 NO_LOAD
#define vload_partial_2_6 NO_LOAD
#define vload_partial_2_7 NO_LOAD
#define vload_partial_2_8 NO_LOAD
#define vload_partial_2_9 NO_LOAD
#define vload_partial_2_10 NO_LOAD
#define vload_partial_2_11 NO_LOAD
#define vload_partial_2_12 NO_LOAD
#define vload_partial_2_13 NO_LOAD
#define vload_partial_2_14 NO_LOAD
#define vload_partial_2_15 NO_LOAD
#define vload_partial_2_16 NO_LOAD

#define vload_partial_3_0 NO_LOAD
#define vload_partial_3_1 vload_partial_1
#define vload_partial_3_2 vload_partial_2
#define vload_partial_3_3 vload_partial_3
#define vload_partial_3_4 NO_LOAD
#define vload_partial_3_5 NO_LOAD
#define vload_partial_3_6 NO_LOAD
#define vload_partial_3_7 NO_LOAD
#define vload_partial_3_8 NO_LOAD
#define vload_partial_3_9 NO_LOAD
#define vload_partial_3_10 NO_LOAD
#define vload_partial_3_11 NO_LOAD
#define vload_partial_3_12 NO_LOAD
#define vload_partial_3_13 NO_LOAD
#define vload_partial_3_14 NO_LOAD
#define vload_partial_3_15 NO_LOAD
#define vload_partial_3_16 NO_LOAD

#define vload_partial_4_0 NO_LOAD
#define vload_partial_4_1 vload_partial_1
#define vload_partial_4_2 vload_partial_2
#define vload_partial_4_3 vload_partial_3
#define vload_partial_4_4 vload_partial_4
#define vload_partial_4_5 NO_LOAD
#define vload_partial_4_6 NO_LOAD
#define vload_partial_4_7 NO_LOAD
#define vload_partial_4_8 NO_LOAD
#define vload_partial_4_9 NO_LOAD
#define vload_partial_4_10 NO_LOAD
#define vload_partial_4_11 NO_LOAD
#define vload_partial_4_12 NO_LOAD
#define vload_partial_4_13 NO_LOAD
#define vload_partial_4_14 NO_LOAD
#define vload_partial_4_15 NO_LOAD
#define vload_partial_4_16 NO_LOAD

#define vload_partial_8_0 NO_LOAD
#define vload_partial_8_1 vload_partial_1
#define vload_partial_8_2 vload_partial_2
#define vload_partial_8_3 vload_partial_3
#define vload_partial_8_4 vload_partial_4
#define vload_partial_8_5 vload_partial_5
#define vload_partial_8_6 vload_partial_6
#define vload_partial_8_7 vload_partial_7
#define vload_partial_8_8 vload_partial_8
#define vload_partial_8_9 NO_LOAD
#define vload_partial_8_10 NO_LOAD
#define vload_partial_8_11 NO_LOAD
#define vload_partial_8_12 NO_LOAD
#define vload_partial_8_13 NO_LOAD
#define vload_partial_8_14 NO_LOAD
#define vload_partial_8_15 NO_LOAD
#define vload_partial_8_16 NO_LOAD

#define vload_partial_16_0 NO_LOAD
#define vload_partial_16_1 vload_partial_1
#define vload_partial_16_2 vload_partial_2
#define vload_partial_16_3 vload_partial_3
#define vload_partial_16_4 vload_partial_4
#define vload_partial_16_5 vload_partial_5
#define vload_partial_16_6 vload_partial_6
#define vload_partial_16_7 vload_partial_7
#define vload_partial_16_8 vload_partial_8
#define vload_partial_16_9 vload_partial_9
#define vload_partial_16_10 vload_partial_10
#define vload_partial_16_11 vload_partial_11
#define vload_partial_16_12 vload_partial_12
#define vload_partial_16_13 vload_partial_13
#define vload_partial_16_14 vload_partial_14
#define vload_partial_16_15 vload_partial_15
#define vload_partial_16_16 vload_partial_16


#define vload_partial_1(DATA, OFFSET, PTR) \
    DATA.s0 = vload1(OFFSET, PTR);

#define vload_partial_2(DATA, OFFSET, PTR) \
    DATA.s01 = vload2(OFFSET, PTR);

#define vload_partial_3(DATA, OFFSET, PTR) \
    DATA.s012 = vload3(OFFSET, PTR);

#define vload_partial_4(DATA, OFFSET, PTR) \
    DATA.s0123 = vload4(OFFSET, PTR);

#define vload_partial_5(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    DATA.s4 = vload1(OFFSET, PTR + 4);

#define vload_partial_6(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vload_partial_7(DATA, OFFSET, PTR)    \
    vload_partial_4(DATA.s0123, OFFSET, PTR); \
    vload_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vload_partial_8(DATA, OFFSET, PTR) \
    DATA.s01234567 = vload8(OFFSET, PTR);

#define vload_partial_9(DATA, OFFSET, PTR)        \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    DATA.s8 = vload1(OFFSET, PTR + 8);

#define vload_partial_10(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vload_partial_11(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_3(DATA.s89A, OFFSET, PTR + 8);

#define vload_partial_12(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_4(DATA.s89AB, OFFSET, PTR + 8);

#define vload_partial_13(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_5(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_14(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_6(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_15(DATA, OFFSET, PTR)       \
    vload_partial_8(DATA.s01234567, OFFSET, PTR); \
    vload_partial_7(DATA.s89ABCDEF, OFFSET, PTR + 8);

#define vload_partial_16(DATA, OFFSET, PTR) \
    DATA = vload16(OFFSET, PTR);



#define PIXEL_UNIT4 1
#define PIXEL_UNIT8 2
#define PIXEL_UNIT16 4


#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size) PIXEL_UNIT##vec_size
#define CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(vec_size) CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT_STR(vec_size)


#define read_image2d_floatx1(img, x_coord, y_coord) (float4)(read_imagef(img, (int2)(x_coord, y_coord)));
#define read_image2d_floatx2(img, x_coord, y_coord) (float8)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_floatx4(img, x_coord, y_coord) (float16)(read_imagef(img, (int2)(x_coord, y_coord)), read_imagef(img, (int2)(x_coord + 1, y_coord)), read_imagef(img, (int2)(x_coord + 2, y_coord)), read_imagef(img, (int2)(x_coord + 3, y_coord)));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define read_image2d_halfx1(img, x_coord, y_coord) (half4)(read_imageh(img, (int2)(x_coord, y_coord)));
#define read_image2d_halfx2(img, x_coord, y_coord) (half8)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)));
#define read_image2d_halfx4(img, x_coord, y_coord) (half16)(read_imageh(img, (int2)(x_coord, y_coord)), read_imageh(img, (int2)(x_coord + 1, y_coord)), read_imageh(img, (int2)(x_coord + 2, y_coord)), read_imageh(img, (int2)(x_coord + 3, y_coord)));
#endif

#define write_image2d_floatx1(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values));
#define write_image2d_floatx2(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_floatx4(img, x_coord, y_coord, values) (write_imagef(img, (int2)(x_coord, y_coord), values.s0123), write_imagef(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imagef(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imagef(img, (int2)(x_coord + 3, y_coord), values.sCDEF));

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED) && defined(cl_khr_fp16)
#define write_image2d_halfx1(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values));
#define write_image2d_halfx2(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567));
#define write_image2d_halfx4(img, x_coord, y_coord, values) (write_imageh(img, (int2)(x_coord, y_coord), values.s0123), write_imageh(img, (int2)(x_coord + 1, y_coord), values.s4567), write_imageh(img, (int2)(x_coord + 2, y_coord), values.s89AB), write_imageh(img, (int2)(x_coord + 3, y_coord), values.sCDEF));
#endif


#define READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord) read_image2d_##data_type##x##n0(img, x_coord, y_coord)
#define READ_IMAGE2D(data_type, n0, img, x_coord, y_coord) READ_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord)


#define WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values) write_image2d_##data_type##x##n0(img, x_coord, y_coord, values)
#define WRITE_IMAGE2D(data_type, n0, img, x_coord, y_coord, values) WRITE_IMAGE2D_STR(data_type, n0, img, x_coord, y_coord, values)

#define VSTORE_STR(size) vstore##size
#define VSTORE(size) VSTORE_STR(size)

#define float1 float
#define half1 half
#define char1 char
#define uchar1 uchar
#define short1 short
#define ushort1 ushort
#define int1 int
#define uint1 uint
#define long1 long
#define ulong1 ulong
#define double1 double

#define vload1(OFFSET, PTR) *(OFFSET + PTR)
#define vstore1(DATA, OFFSET, PTR) *(OFFSET + PTR) = DATA


#define VSTORE_PARTIAL_STR(size, store_size) vstore_partial_##size##_##store_size
#define VSTORE_PARTIAL(size, store_size) VSTORE_PARTIAL_STR(size, store_size)

#define NO_STORE(data, offs, ptr) \
    {                             \
    }


#define vstore_partial_1_0 NO_STORE
#define vstore_partial_1_1 vstore1
#define vstore_partial_1_2 NO_STORE
#define vstore_partial_1_3 NO_STORE
#define vstore_partial_1_4 NO_STORE
#define vstore_partial_1_5 NO_STORE
#define vstore_partial_1_6 NO_STORE
#define vstore_partial_1_7 NO_STORE
#define vstore_partial_1_8 NO_STORE
#define vstore_partial_1_9 NO_STORE
#define vstore_partial_1_10 NO_STORE
#define vstore_partial_1_11 NO_STORE
#define vstore_partial_1_12 NO_STORE
#define vstore_partial_1_13 NO_STORE
#define vstore_partial_1_14 NO_STORE
#define vstore_partial_1_15 NO_STORE
#define vstore_partial_1_16 NO_STORE

#define vstore_partial_2_0 NO_STORE
#define vstore_partial_2_1 vstore_partial_1
#define vstore_partial_2_2 vstore_partial_2
#define vstore_partial_2_3 NO_STORE
#define vstore_partial_2_4 NO_STORE
#define vstore_partial_2_5 NO_STORE
#define vstore_partial_2_6 NO_STORE
#define vstore_partial_2_7 NO_STORE
#define vstore_partial_2_8 NO_STORE
#define vstore_partial_2_9 NO_STORE
#define vstore_partial_2_10 NO_STORE
#define vstore_partial_2_11 NO_STORE
#define vstore_partial_2_12 NO_STORE
#define vstore_partial_2_13 NO_STORE
#define vstore_partial_2_14 NO_STORE
#define vstore_partial_2_15 NO_STORE
#define vstore_partial_2_16 NO_STORE

#define vstore_partial_3_0 NO_STORE
#define vstore_partial_3_1 vstore_partial_1
#define vstore_partial_3_2 vstore_partial_2
#define vstore_partial_3_3 vstore_partial_3
#define vstore_partial_3_4 NO_STORE
#define vstore_partial_3_5 NO_STORE
#define vstore_partial_3_6 NO_STORE
#define vstore_partial_3_7 NO_STORE
#define vstore_partial_3_8 NO_STORE
#define vstore_partial_3_9 NO_STORE
#define vstore_partial_3_10 NO_STORE
#define vstore_partial_3_11 NO_STORE
#define vstore_partial_3_12 NO_STORE
#define vstore_partial_3_13 NO_STORE
#define vstore_partial_3_14 NO_STORE
#define vstore_partial_3_15 NO_STORE
#define vstore_partial_3_16 NO_STORE

#define vstore_partial_4_0 NO_STORE
#define vstore_partial_4_1 vstore_partial_1
#define vstore_partial_4_2 vstore_partial_2
#define vstore_partial_4_3 vstore_partial_3
#define vstore_partial_4_4 vstore_partial_4
#define vstore_partial_4_5 NO_STORE
#define vstore_partial_4_6 NO_STORE
#define vstore_partial_4_7 NO_STORE
#define vstore_partial_4_8 NO_STORE
#define vstore_partial_4_9 NO_STORE
#define vstore_partial_4_10 NO_STORE
#define vstore_partial_4_11 NO_STORE
#define vstore_partial_4_12 NO_STORE
#define vstore_partial_4_13 NO_STORE
#define vstore_partial_4_14 NO_STORE
#define vstore_partial_4_15 NO_STORE
#define vstore_partial_4_16 NO_STORE

#define vstore_partial_8_0 NO_STORE
#define vstore_partial_8_1 vstore_partial_1
#define vstore_partial_8_2 vstore_partial_2
#define vstore_partial_8_3 vstore_partial_3
#define vstore_partial_8_4 vstore_partial_4
#define vstore_partial_8_5 vstore_partial_5
#define vstore_partial_8_6 vstore_partial_6
#define vstore_partial_8_7 vstore_partial_7
#define vstore_partial_8_8 vstore_partial_8
#define vstore_partial_8_9 NO_STORE
#define vstore_partial_8_10 NO_STORE
#define vstore_partial_8_11 NO_STORE
#define vstore_partial_8_12 NO_STORE
#define vstore_partial_8_13 NO_STORE
#define vstore_partial_8_14 NO_STORE
#define vstore_partial_8_15 NO_STORE
#define vstore_partial_8_16 NO_STORE

#define vstore_partial_16_0 NO_STORE
#define vstore_partial_16_1 vstore_partial_1
#define vstore_partial_16_2 vstore_partial_2
#define vstore_partial_16_3 vstore_partial_3
#define vstore_partial_16_4 vstore_partial_4
#define vstore_partial_16_5 vstore_partial_5
#define vstore_partial_16_6 vstore_partial_6
#define vstore_partial_16_7 vstore_partial_7
#define vstore_partial_16_8 vstore_partial_8
#define vstore_partial_16_9 vstore_partial_9
#define vstore_partial_16_10 vstore_partial_10
#define vstore_partial_16_11 vstore_partial_11
#define vstore_partial_16_12 vstore_partial_12
#define vstore_partial_16_13 vstore_partial_13
#define vstore_partial_16_14 vstore_partial_14
#define vstore_partial_16_15 vstore_partial_15
#define vstore_partial_16_16 vstore_partial_16


#define vstore_partial_1(DATA, OFFSET, PTR) \
    vstore1(DATA.s0, OFFSET, PTR);

#define vstore_partial_2(DATA, OFFSET, PTR) \
    vstore2(DATA.s01, OFFSET, PTR);

#define vstore_partial_3(DATA, OFFSET, PTR) \
    vstore3(DATA.s012, OFFSET, PTR);

#define vstore_partial_4(DATA, OFFSET, PTR) \
    vstore4(DATA.s0123, OFFSET, PTR);

#define vstore_partial_5(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore1(DATA.s4, OFFSET, PTR + 4);

#define vstore_partial_6(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_2(DATA.s45, OFFSET, PTR + 4);

#define vstore_partial_7(DATA, OFFSET, PTR)    \
    vstore_partial_4(DATA.s0123, OFFSET, PTR); \
    vstore_partial_3(DATA.s456, OFFSET, PTR + 4);

#define vstore_partial_8(DATA, OFFSET, PTR) \
    vstore8(DATA.s01234567, OFFSET, PTR);

#define vstore_partial_9(DATA, OFFSET, PTR)        \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore1(DATA.s8, OFFSET, PTR + 8);

#define vstore_partial_10(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_2(DATA.s89, OFFSET, PTR + 8);

#define vstore_partial_11(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_3(DATA.s89a, OFFSET, PTR + 8);

#define vstore_partial_12(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_4(DATA.s89ab, OFFSET, PTR + 8);

#define vstore_partial_13(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_5(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_14(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_6(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_15(DATA, OFFSET, PTR)       \
    vstore_partial_8(DATA.s01234567, OFFSET, PTR); \
    vstore_partial_7(DATA.s89abcdef, OFFSET, PTR + 8);

#define vstore_partial_16(DATA, OFFSET, PTR) \
    vstore16(DATA, OFFSET, PTR);





#define convert_float_sat convert_float
#define convert_float1_sat convert_float
#define convert_float2_sat convert_float2
#define convert_float3_sat convert_float3
#define convert_float4_sat convert_float4
#define convert_float8_sat convert_float8
#define convert_float16_sat convert_float16
#define convert_half_sat convert_float
#define convert_half1_sat convert_half
#define convert_half2_sat convert_half2
#define convert_half3_sat convert_half3
#define convert_half4_sat convert_half4
#define convert_half8_sat convert_half8
#define convert_half16_sat convert_half16

#define convert_float1 convert_float
#define convert_half1 convert_half
#define convert_char1 convert_char
#define convert_uchar1 convert_uchar
#define convert_short1 convert_short
#define convert_ushort1 convert_ushort
#define convert_int1 convert_int
#define convert_uint1 convert_uint
#define convert_long1 convert_long
#define convert_ulong1 convert_ulong
#define convert_double1 convert_double

#define convert_char1_sat convert_char_sat
#define convert_uchar1_sat convert_uchar_sat
#define convert_uchar2_sat convert_uchar2_sat
#define convert_uchar3_sat convert_uchar3_sat
#define convert_uchar4_sat convert_uchar4_sat
#define convert_uchar8_sat convert_uchar8_sat
#define convert_uchar16_sat convert_uchar16_sat
#define convert_short1_sat convert_short_sat
#define convert_ushort1_sat convert_ushort_sat
#define convert_int1_sat convert_int_sat
#define convert_uint1_sat convert_uint_sat
#define convert_long1_sat convert_long_sat
#define convert_ulong1_sat convert_ulong_sat
#define convert_double1_sat convert_double_sat

#define VEC_DATA_TYPE_STR(type, size) type##size
#define VEC_DATA_TYPE(type, size) VEC_DATA_TYPE_STR(type, size)

#define CONVERT_STR(x, type) (convert_##type((x)))
#define CONVERT(x, type) CONVERT_STR(x, type)

#define CONVERT_SAT_STR(x, type) (convert_##type##_sat((x)))
#define CONVERT_SAT(x, type) CONVERT_SAT_STR(x, type)

#define CONVERT_SAT_ROUND_STR(x, type, round) (convert_##type##_sat_##round((x)))
#define CONVERT_SAT_ROUND(x, type, round) CONVERT_SAT_ROUND_STR(x, type, round)

#define select_vec_dt_uchar(size) uchar##size
#define select_vec_dt_char(size) char##size
#define select_vec_dt_ushort(size) ushort##size
#define select_vec_dt_short(size) short##size
#define select_vec_dt_half(size) short##size
#define select_vec_dt_uint(size) uint##size
#define select_vec_dt_int(size) int##size
#define select_vec_dt_float(size) int##size
#define select_vec_dt_ulong(size) ulong##size
#define select_vec_dt_long(size) long##size

#define SELECT_VEC_DATA_TYPE_STR(type, size) select_vec_dt_##type(size)
#define SELECT_VEC_DATA_TYPE(type, size) SELECT_VEC_DATA_TYPE_STR(type, size)
#define SELECT_DATA_TYPE(type) SELECT_VEC_DATA_TYPE_STR(type, 1)

#define signed_int_vec_dt_uchar(size) char##size
#define signed_int_vec_dt_char(size) char##size
#define signed_int_vec_dt_ushort(size) short##size
#define signed_int_vec_dt_short(size) short##size
#define signed_int_vec_dt_half(size) short##size
#define signed_int_vec_dt_uint(size) int##size
#define signed_int_vec_dt_int(size) int##size
#define signed_int_vec_dt_float(size) int##size
#define signed_int_vec_dt_ulong(size) long##size
#define signed_int_vec_dt_long(size) long##size

#define SIGNED_INT_VEC_DATA_TYPE_STR(type, size) signed_int_vec_dt_##type(size)
#define SIGNED_INT_VEC_DATA_TYPE(type, size) SIGNED_INT_VEC_DATA_TYPE_STR(type, size)
#define SIGNED_INT_DATA_TYPE(type) SIGNED_INT_VEC_DATA_TYPE_STR(type, 1)

#define sum_reduce_1(x) (x)
#define sum_reduce_2(x) ((x).s0) + ((x).s1)
#define sum_reduce_3(x) sum_reduce_2((x).s01) + ((x).s2)
#define sum_reduce_4(x) sum_reduce_2((x).s01) + sum_reduce_2((x).s23)
#define sum_reduce_8(x) sum_reduce_4((x).s0123) + sum_reduce_4((x).s4567)
#define sum_reduce_16(x) sum_reduce_8((x).s01234567) + sum_reduce_8((x).s89ABCDEF)

#define SUM_REDUCE_STR(x, size) sum_reduce_##size(x)
#define SUM_REDUCE(x, size) SUM_REDUCE_STR(x, size)

#define prod_reduce_1(x) (x)
#define prod_reduce_2(x) ((x).s0) * ((x).s1)
#define prod_reduce_3(x) prod_reduce_2((x).s01) * ((x).s2)
#define prod_reduce_4(x) prod_reduce_2((x).s01) * prod_reduce_2((x).s23)
#define prod_reduce_8(x) prod_reduce_4((x).s0123) * prod_reduce_4((x).s4567)
#define prod_reduce_16(x) prod_reduce_8((x).s01234567) * prod_reduce_8((x).s89ABCDEF)

#define PROD_REDUCE_STR(x, size) prod_reduce_##size(x)
#define PROD_REDUCE(x, size) PROD_REDUCE_STR(x, size)

#define max_reduce_1(x) (x)
#define max_reduce_2(x) max(((x).s0), ((x).s1))
#define max_reduce_3(x) max(max_reduce_2((x).s01), ((x).s2))
#define max_reduce_4(x) max(max_reduce_2((x).s01), max_reduce_2((x).s23))
#define max_reduce_8(x) max(max_reduce_4((x).s0123), max_reduce_4((x).s4567))
#define max_reduce_16(x) max(max_reduce_8((x).s01234567), max_reduce_8((x).s89ABCDEF))

#define MAX_REDUCE_STR(x, size) max_reduce_##size(x)
#define MAX_REDUCE(x, size) MAX_REDUCE_STR(x, size)

#define VECTOR_DECLARATION(name)     \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_offset_first_element_in_bytes

#define IMAGE_DECLARATION(name)      \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR3D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR4D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR5D_DECLARATION(name)   \
    __global uchar *name##_ptr,      \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_stride_v, \
    uint        name##_step_v,   \
    uint        name##_offset_first_element_in_bytes

#define CONVERT_TO_VECTOR_STRUCT(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x)

#define CONVERT_TO_VECTOR_STRUCT_NO_STEP(name) \
    update_vector_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0)

#define CONVERT_TO_IMAGE_STRUCT(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y)

#define CONVERT_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT_NO_STEP(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, name##_step_z)

#define CONVERT_TENSOR3D_TO_IMAGE_STRUCT(name) \
    update_image_from_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT(name)                                                                                                           \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(name) \
    update_tensor3D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0)

#define CONVERT_TO_TENSOR4D_STRUCT(name, mod_size)                                                                                                 \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                                 name##_stride_z, name##_step_z, name##_stride_w, name##_step_w, mod_size)

#define CONVERT_TO_TENSOR4D_STRUCT_NO_STEP(name, mod_size) \
    update_tensor4D_workitem_ptr(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, 0, name##_stride_y, 0, name##_stride_z, 0, name##_stride_w, 0, mod_size)

#define CONVERT_TO_TENSOR3D_STRUCT_NO_UPDATE_PTR(name)                                                                                       \
    tensor3D_ptr_no_update(name##_ptr, name##_offset_first_element_in_bytes, name##_stride_x, name##_step_x, name##_stride_y, name##_step_y, \
                           name##_stride_z, name##_step_z)


typedef struct Vector
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
} Vector;


typedef struct Image
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
} Image;


typedef struct Tensor3D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
} Tensor3D;


typedef struct Tensor4D
{
    __global uchar *ptr;
    int             offset_first_element_in_bytes;
    int             stride_x;
    int             stride_y;
    int             stride_z;
    int             stride_w;
} Tensor4D;


inline Vector update_vector_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x)
{
    Vector vector =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
    };
    vector.ptr += vector.offset_first_element_in_bytes + get_global_id(0) * step_x;
    return vector;
}


inline Image update_image_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y;
    return img;
}


inline Image update_image_from_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Image img =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y
    };
    img.ptr += img.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return img;
}


inline Tensor3D update_tensor3D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + get_global_id(2) * step_z;
    return tensor;
}


inline Tensor3D tensor3D_ptr_no_update(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z)
{
    Tensor3D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z
    };
    return tensor;
}

inline Tensor4D update_tensor4D_workitem_ptr(__global uchar *ptr, uint offset_first_element_in_bytes, uint stride_x, uint step_x, uint stride_y, uint step_y, uint stride_z, uint step_z, uint stride_w,
                                             uint step_w,
                                             uint mod_size)
{
    Tensor4D tensor =
    {
        .ptr                           = ptr,
        .offset_first_element_in_bytes = offset_first_element_in_bytes,
        .stride_x                      = stride_x,
        .stride_y                      = stride_y,
        .stride_z                      = stride_z,
        .stride_w                      = stride_w
    };

    tensor.ptr += tensor.offset_first_element_in_bytes + get_global_id(0) * step_x + get_global_id(1) * step_y + (get_global_id(2) % mod_size) * step_z + (get_global_id(2) / mod_size) * step_w;
    return tensor;
}


inline __global const uchar *vector_offset(const Vector *vec, int x)
{
    return vec->ptr + x * vec->stride_x;
}


inline __global uchar *offset(const Image *img, int x, int y)
{
    return img->ptr + x * img->stride_x + y * img->stride_y;
}


inline __global const uchar *tensor3D_offset(const Tensor3D *tensor, int x, int y, int z)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z;
}


inline __global const uchar *tensor4D_offset(const Tensor4D *tensor, int x, int y, int z, int w)
{
    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + w * tensor->stride_w;
}


inline __global const uchar *tensor3D_index2ptr(const Tensor3D *tensor, uint width, uint height, uint depth, uint index)
{
    uint num_elements = width * height;

    const uint z = index / num_elements;

    index %= num_elements;

    const uint y = index / width;

    index %= width;

    const uint x = index;

    return tensor->ptr + x * tensor->stride_x + y * tensor->stride_y + z * tensor->stride_z + tensor->offset_first_element_in_bytes;
}

#endif



#define REPEAT_3_1(P_X, P_A, P_B, P_C) P_X##_DEF(0, P_A, P_B, P_C)
#define REPEAT_3_2(P_X, P_A, P_B, P_C) \
    P_X##_DEF(1, P_A, P_B, P_C);       \
    REPEAT_3_1(P_X, P_A, P_B, P_C)
#define REPEAT_3_3(P_X, P_A, P_B, P_C) \
    P_X##_DEF(2, P_A, P_B, P_C);       \
    REPEAT_3_2(P_X, P_A, P_B, P_C)
#define REPEAT_3_4(P_X, P_A, P_B, P_C) \
    P_X##_DEF(3, P_A, P_B, P_C);       \
    REPEAT_3_3(P_X, P_A, P_B, P_C)
#define REPEAT_3_5(P_X, P_A, P_B, P_C) \
    P_X##_DEF(4, P_A, P_B, P_C);       \
    REPEAT_3_4(P_X, P_A, P_B, P_C)
#define REPEAT_3_6(P_X, P_A, P_B, P_C) \
    P_X##_DEF(5, P_A, P_B, P_C);       \
    REPEAT_3_5(P_X, P_A, P_B, P_C)
#define REPEAT_3_7(P_X, P_A, P_B, P_C) \
    P_X##_DEF(6, P_A, P_B, P_C);       \
    REPEAT_3_6(P_X, P_A, P_B, P_C)
#define REPEAT_3_8(P_X, P_A, P_B, P_C) \
    P_X##_DEF(7, P_A, P_B, P_C);       \
    REPEAT_3_7(P_X, P_A, P_B, P_C)
#define REPEAT_3_9(P_X, P_A, P_B, P_C) \
    P_X##_DEF(8, P_A, P_B, P_C);       \
    REPEAT_3_8(P_X, P_A, P_B, P_C)
#define REPEAT_3_10(P_X, P_A, P_B, P_C) \
    P_X##_DEF(9, P_A, P_B, P_C);        \
    REPEAT_3_9(P_X, P_A, P_B, P_C)
#define REPEAT_3_11(P_X, P_A, P_B, P_C) \
    P_X##_DEF(A, P_A, P_B, P_C);        \
    REPEAT_3_10(P_X, P_A, P_B, P_C)
#define REPEAT_3_12(P_X, P_A, P_B, P_C) \
    P_X##_DEF(B, P_A, P_B, P_C);        \
    REPEAT_3_11(P_X, P_A, P_B, P_C)
#define REPEAT_3_13(P_X, P_A, P_B, P_C) \
    P_X##_DEF(C, P_A, P_B, P_C);        \
    REPEAT_3_12(P_X, P_A, P_B, P_C)
#define REPEAT_3_14(P_X, P_A, P_B, P_C) \
    P_X##_DEF(D, P_A, P_B, P_C);        \
    REPEAT_3_13(P_X, P_A, P_B, P_C)
#define REPEAT_3_15(P_X, P_A, P_B, P_C) \
    P_X##_DEF(E, P_A, P_B, P_C);        \
    REPEAT_3_14(P_X, P_A, P_B, P_C)
#define REPEAT_3_16(P_X, P_A, P_B, P_C) \
    P_X##_DEF(F, P_A, P_B, P_C);        \
    REPEAT_3_15(P_X, P_A, P_B, P_C)

#define REPEAT_DEF_3_N(P_NUM, P_OP, P_A, P_B, P_C) REPEAT_3_##P_NUM(P_OP, P_A, P_B, P_C)
#define REPEAT_3_N(P_NUM, P_OP, P_A, P_B, P_C) REPEAT_DEF_3_N(P_NUM, P_OP, P_A, P_B, P_C)


#define REPEAT_4_1(P_X, P_A, P_B, P_C, P_D) P_X##_DEF(0, P_A, P_B, P_C, P_D)
#define REPEAT_4_2(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(1, P_A, P_B, P_C, P_D);       \
    REPEAT_4_1(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_3(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(2, P_A, P_B, P_C, P_D);       \
    REPEAT_4_2(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_4(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(3, P_A, P_B, P_C, P_D);       \
    REPEAT_4_3(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_5(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(4, P_A, P_B, P_C, P_D);       \
    REPEAT_4_4(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_6(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(5, P_A, P_B, P_C, P_D);       \
    REPEAT_4_5(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_7(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(6, P_A, P_B, P_C, P_D);       \
    REPEAT_4_6(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_8(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(7, P_A, P_B, P_C, P_D);       \
    REPEAT_4_7(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_9(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(8, P_A, P_B, P_C, P_D);       \
    REPEAT_4_8(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_10(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(9, P_A, P_B, P_C, P_D);        \
    REPEAT_4_9(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_11(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(A, P_A, P_B, P_C, P_D);        \
    REPEAT_4_10(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_12(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(B, P_A, P_B, P_C, P_D);        \
    REPEAT_4_11(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_13(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(C, P_A, P_B, P_C, P_D);        \
    REPEAT_4_12(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_14(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(D, P_A, P_B, P_C, P_D);        \
    REPEAT_4_13(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_15(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(E, P_A, P_B, P_C, P_D);        \
    REPEAT_4_14(P_X, P_A, P_B, P_C, P_D)
#define REPEAT_4_16(P_X, P_A, P_B, P_C, P_D) \
    P_X##_DEF(F, P_A, P_B, P_C, P_D);        \
    REPEAT_4_15(P_X, P_A, P_B, P_C, P_D)

#define REPEAT_DEF_4_N(P_NUM, P_OP, P_A, P_B, P_C, P_D) REPEAT_4_##P_NUM(P_OP, P_A, P_B, P_C, P_D)
#define REPEAT_4_N(P_NUM, P_OP, P_A, P_B, P_C, P_D) REPEAT_DEF_4_N(P_NUM, P_OP, P_A, P_B, P_C, P_D)


#define VAR_INIT_TO_CONST_DEF(ID, TYPE, VAR, VAL) TYPE VAR##ID = VAL
#define REPEAT_VAR_INIT_TO_CONST(N, TYPE, VAR, VAL) REPEAT_3_N(N, VAR_INIT_TO_CONST, TYPE, VAR, VAL)


#define VAR_INIT_CONVERT_DEF(ID, TYPE_OUT, VAR_IN, VAR_OUT) TYPE_OUT VAR_OUT##ID = CONVERT(VAR_IN##ID, TYPE_OUT)
#define REPEAT_VAR_INIT_CONVERT(N, TYPE_OUT, VAR_IN, VAR_OUT) REPEAT_3_N(N, VAR_INIT_CONVERT, TYPE_OUT, VAR_IN, VAR_OUT)


#define VAR_INIT_CONVERT_SAT_DEF(ID, TYPE_OUT, VAR_IN, VAR_OUT) TYPE_OUT VAR_OUT##ID = CONVERT_SAT(VAR_IN##ID, TYPE_OUT)
#define REPEAT_VAR_INIT_CONVERT_SAT(N, TYPE_OUT, VAR_IN, VAR_OUT) REPEAT_3_N(N, VAR_INIT_CONVERT_SAT, TYPE_OUT, VAR_IN, VAR_OUT)


#define ADD_CONST_TO_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID += (TYPE)VAL
#define REPEAT_ADD_CONST_TO_VAR(N, TYPE, VAR, VAL) REPEAT_3_N(N, ADD_CONST_TO_VAR, TYPE, VAR, VAL)


#define MLA_VAR_WITH_CONST_VEC_DEF(ID, VAR_A, VAR_B, VAL) VAR_A##ID += VAR_B##ID * VAL
#define REPEAT_MLA_VAR_WITH_CONST_VEC(N, VAR_A, VAR_B, VAL) REPEAT_3_N(N, MLA_VAR_WITH_CONST_VEC, VAR_A, VAR_B, VAL)


#define ADD_VECTOR_TO_VAR_DEF(ID, TYPE, VAR, VEC) VAR##ID += VEC
#define REPEAT_ADD_VECTOR_TO_VAR(N, VAR, VEC) REPEAT_3_N(N, ADD_VECTOR_TO_VAR, "", VAR, VEC)


#define ADD_TWO_VARS_DEF(ID, TYPE, VAR_A, VAR_B) VAR_A##ID += VAR_B##ID
#define REPEAT_ADD_TWO_VARS(N, VAR_A, VAR_B) REPEAT_3_N(N, ADD_TWO_VARS, "", VAR_A, VAR_B)


#define MAX_CONST_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID = max(VAR##ID, (TYPE)VAL)
#define REPEAT_MAX_CONST_VAR(N, TYPE, VAR, VAL) REPEAT_3_N(N, MAX_CONST_VAR, TYPE, VAR, VAL)


#define MIN_CONST_VAR_DEF(ID, TYPE, VAR, VAL) VAR##ID = min(VAR##ID, (TYPE)VAL)
#define REPEAT_MIN_CONST_VAR(N, TYPE, VAR, VAL) REPEAT_3_N(N, MIN_CONST_VAR, TYPE, VAR, VAL)


#define ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT) VAR##ID = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, SIZE)
#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(N, SIZE, VAR, RES_MUL, RES_SHIFT) REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE, SIZE, VAR, RES_MUL, RES_SHIFT)


#define ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT) VAR##ID = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, SIZE)
#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(N, SIZE, VAR, RES_MUL, RES_SHIFT) REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE, SIZE, VAR, RES_MUL, RES_SHIFT)


#define ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL_DEF(ID, SIZE, VAR, RES_MUL, RES_SHIFT)                     \
    ({                                                                                                        \
        VEC_DATA_TYPE(int, N0)                                                                                \
        VAR##ID_shift_lt0 = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, N0); \
        VEC_DATA_TYPE(int, N0)                                                                                \
        VAR##ID_shift_gt0 = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(VAR##ID, RES_MUL, RES_SHIFT, N0);    \
        VAR##ID           = select(VAR##ID_shift_lt0, VAR##ID_shift_gt0, RES_SHIFT >= 0);                     \
    })
#define REPEAT_ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL(N, SIZE, VAR, RES_MUL, RES_SHIFT) REPEAT_4_N(N, ASYMM_MULT_BY_QUANT_MULTIPLIER_PER_CHANNEL, SIZE, VAR, RES_MUL, RES_SHIFT)

#endif

#ifndef SRC_CORE_CL_CL_KERNELS_TILE_HELPERS
#define SRC_CORE_CL_CL_KERNELS_TILE_HELPERS




#define TILE_VECTOR_SIZE1 1
#define TILE_VECTOR_SIZE2 2
#define TILE_VECTOR_SIZE3 3
#define TILE_VECTOR_SIZE4 4
#define TILE_VECTOR_SIZE5 8
#define TILE_VECTOR_SIZE6 8
#define TILE_VECTOR_SIZE7 8
#define TILE_VECTOR_SIZE8 8
#define TILE_VECTOR_SIZE9 16
#define TILE_VECTOR_SIZE10 16
#define TILE_VECTOR_SIZE11 16
#define TILE_VECTOR_SIZE12 16
#define TILE_VECTOR_SIZE13 16
#define TILE_VECTOR_SIZE14 16
#define TILE_VECTOR_SIZE15 16
#define TILE_VECTOR_SIZE16 16

#define TILE_VECTOR_TYPE1(DATA_TYPE) DATA_TYPE##1
#define TILE_VECTOR_TYPE2(DATA_TYPE) DATA_TYPE##2
#define TILE_VECTOR_TYPE3(DATA_TYPE) DATA_TYPE##3
#define TILE_VECTOR_TYPE4(DATA_TYPE) DATA_TYPE##4
#define TILE_VECTOR_TYPE5(DATA_TYPE) DATA_TYPE##8
#define TILE_VECTOR_TYPE6(DATA_TYPE) DATA_TYPE##8
#define TILE_VECTOR_TYPE7(DATA_TYPE) DATA_TYPE##8
#define TILE_VECTOR_TYPE8(DATA_TYPE) DATA_TYPE##8
#define TILE_VECTOR_TYPE9(DATA_TYPE) DATA_TYPE##16
#define TILE_VECTOR_TYPE10(DATA_TYPE) DATA_TYPE##16
#define TILE_VECTOR_TYPE11(DATA_TYPE) DATA_TYPE##16
#define TILE_VECTOR_TYPE12(DATA_TYPE) DATA_TYPE##16
#define TILE_VECTOR_TYPE13(DATA_TYPE) DATA_TYPE##16
#define TILE_VECTOR_TYPE14(DATA_TYPE) DATA_TYPE##16
#define TILE_VECTOR_TYPE15(DATA_TYPE) DATA_TYPE##16
#define TILE_VECTOR_TYPE16(DATA_TYPE) DATA_TYPE##16


#define TILE(DATA_TYPE, H, W, BASENAME) TILE_STR(DATA_TYPE, H, W, BASENAME)
#define TILE_STR(DATA_TYPE, H, W, BASENAME) \
    union {                                 \
        DATA_TYPE                      s[TILE_VECTOR_SIZE##W];                  \
        TILE_VECTOR_TYPE##W(DATA_TYPE) v;                     \
    } BASENAME[H]

#define TENSOR4D_IMAGE(name)          \
    __read_only image2d_t name##_img, \
    __global uchar *name##_ptr,       \
    uint            name##_stride_x,  \
    uint            name##_step_x,    \
    uint            name##_stride_y,  \
    uint            name##_step_y,    \
    uint            name##_stride_z,  \
    uint            name##_step_z,    \
    uint            name##_stride_w,  \
    uint            name##_step_w,    \
    uint            name##_offset_first_element_in_bytes

#define TENSOR4D_BUFFER(name)    \
    __global uchar *name##_ptr,  \
    uint        name##_stride_x, \
    uint        name##_step_x,   \
    uint        name##_stride_y, \
    uint        name##_step_y,   \
    uint        name##_stride_z, \
    uint        name##_step_z,   \
    uint        name##_stride_w, \
    uint        name##_step_w,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR4D_STR(name, type) TENSOR4D_##type(name)
#define TENSOR4D(name, type) TENSOR4D_STR(name, type)

#define TENSOR4D_T_IMAGE(name)          \
    __read_only image2d_t name##_img, \
    __global uchar *name##_ptr,       \
    uint        name##_stride_y, \
    uint        name##_stride_z, \
    uint        name##_stride_w, \
    uint        name##_c,   \
    uint        name##_w,   \
    uint        name##_h,   \
    uint        name##_n,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR4D_T_BUFFER(name)    \
    __global uchar *name##_ptr,  \
    uint        name##_stride_y, \
    uint        name##_stride_z, \
    uint        name##_stride_w, \
    uint        name##_c,   \
    uint        name##_w,   \
    uint        name##_h,   \
    uint        name##_n,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR4D_T_STR(name, type) TENSOR4D_T_##type(name)


#define TENSOR4D_T(name, type) TENSOR4D_T_STR(name, type)

#define TENSOR4D_RO_T_IMAGE(name)          \
    __read_only image2d_t name##_img, \
    TENSOR4D_T_BUFFER(name)

#define TENSOR4D_RO_T_BUFFER(name) TENSOR4D_T_BUFFER(name)

#define TENSOR4D_RO_T_STR(name, type) TENSOR4D_RO_T_##type(name)


#define TENSOR4D_RO_T(name, type) TENSOR4D_RO_T_STR(name, type)

#define TENSOR4D_WO_T_IMAGE(name)          \
    __write_only image2d_t name##_img, \
    TENSOR4D_T_BUFFER(name)

#define TENSOR4D_WO_T_BUFFER(name) TENSOR4D_T_BUFFER(name)

#define TENSOR4D_WO_T_STR(name, type) TENSOR4D_WO_T_##type(name)


#define TENSOR4D_WO_T(name, type) TENSOR4D_WO_T_STR(name, type)

#define TENSOR3D_T_IMAGE(name)          \
    __read_only image2d_t name##_img, \
    __global uchar *name##_ptr,       \
    uint        name##_stride_y, \
    uint        name##_stride_z, \
    uint        name##_w,   \
    uint        name##_h,   \
    uint        name##_n,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR3D_T_BUFFER(name)    \
    __global uchar *name##_ptr,  \
    uint        name##_stride_y, \
    uint        name##_stride_z, \
    uint        name##_w,   \
    uint        name##_h,   \
    uint        name##_n,   \
    uint        name##_offset_first_element_in_bytes

#define TENSOR3D_T_STR(name, type) TENSOR3D_T_##type(name)
#define TENSOR3D_T(name, type) TENSOR3D_T_STR(name, type)

#if !defined(UNROLL_WITH_PRAGMA)
#define UNROLL_INCR(idx, step, macro) idx += (step); (macro)

#define LOOP_UNROLLING_1(idx, step, macro) (macro)
#define LOOP_UNROLLING_2(idx, step, macro) LOOP_UNROLLING_1(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_3(idx, step, macro) LOOP_UNROLLING_2(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_4(idx, step, macro) LOOP_UNROLLING_3(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_5(idx, step, macro) LOOP_UNROLLING_4(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_6(idx, step, macro) LOOP_UNROLLING_5(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_7(idx, step, macro) LOOP_UNROLLING_6(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_8(idx, step, macro) LOOP_UNROLLING_7(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_9(idx, step, macro) LOOP_UNROLLING_8(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_10(idx, step, macro) LOOP_UNROLLING_9(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_11(idx, step, macro) LOOP_UNROLLING_10(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_12(idx, step, macro) LOOP_UNROLLING_11(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_13(idx, step, macro) LOOP_UNROLLING_12(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_14(idx, step, macro) LOOP_UNROLLING_13(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_15(idx, step, macro) LOOP_UNROLLING_14(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_16(idx, step, macro) LOOP_UNROLLING_15(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_17(idx, step, macro) LOOP_UNROLLING_16(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_18(idx, step, macro) LOOP_UNROLLING_17(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_19(idx, step, macro) LOOP_UNROLLING_18(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_20(idx, step, macro) LOOP_UNROLLING_19(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_21(idx, step, macro) LOOP_UNROLLING_20(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_22(idx, step, macro) LOOP_UNROLLING_21(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_23(idx, step, macro) LOOP_UNROLLING_22(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_24(idx, step, macro) LOOP_UNROLLING_23(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_25(idx, step, macro) LOOP_UNROLLING_24(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_26(idx, step, macro) LOOP_UNROLLING_25(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_27(idx, step, macro) LOOP_UNROLLING_26(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_28(idx, step, macro) LOOP_UNROLLING_27(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_29(idx, step, macro) LOOP_UNROLLING_28(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_30(idx, step, macro) LOOP_UNROLLING_29(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_31(idx, step, macro) LOOP_UNROLLING_30(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_32(idx, step, macro) LOOP_UNROLLING_31(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_33(idx, step, macro) LOOP_UNROLLING_32(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_34(idx, step, macro) LOOP_UNROLLING_33(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_35(idx, step, macro) LOOP_UNROLLING_34(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_36(idx, step, macro) LOOP_UNROLLING_35(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_37(idx, step, macro) LOOP_UNROLLING_36(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_38(idx, step, macro) LOOP_UNROLLING_37(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_39(idx, step, macro) LOOP_UNROLLING_38(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_40(idx, step, macro) LOOP_UNROLLING_39(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_41(idx, step, macro) LOOP_UNROLLING_40(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_42(idx, step, macro) LOOP_UNROLLING_41(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_43(idx, step, macro) LOOP_UNROLLING_42(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_44(idx, step, macro) LOOP_UNROLLING_43(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_45(idx, step, macro) LOOP_UNROLLING_44(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_46(idx, step, macro) LOOP_UNROLLING_45(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_47(idx, step, macro) LOOP_UNROLLING_46(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_48(idx, step, macro) LOOP_UNROLLING_47(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_49(idx, step, macro) LOOP_UNROLLING_48(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_50(idx, step, macro) LOOP_UNROLLING_49(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_51(idx, step, macro) LOOP_UNROLLING_50(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_52(idx, step, macro) LOOP_UNROLLING_51(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_53(idx, step, macro) LOOP_UNROLLING_52(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_54(idx, step, macro) LOOP_UNROLLING_53(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_55(idx, step, macro) LOOP_UNROLLING_54(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_56(idx, step, macro) LOOP_UNROLLING_55(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_57(idx, step, macro) LOOP_UNROLLING_56(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_58(idx, step, macro) LOOP_UNROLLING_57(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_59(idx, step, macro) LOOP_UNROLLING_58(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_60(idx, step, macro) LOOP_UNROLLING_59(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_61(idx, step, macro) LOOP_UNROLLING_60(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_62(idx, step, macro) LOOP_UNROLLING_61(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_63(idx, step, macro) LOOP_UNROLLING_62(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_64(idx, step, macro) LOOP_UNROLLING_63(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_65(idx, step, macro) LOOP_UNROLLING_64(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_66(idx, step, macro) LOOP_UNROLLING_65(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_67(idx, step, macro) LOOP_UNROLLING_66(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_68(idx, step, macro) LOOP_UNROLLING_67(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_69(idx, step, macro) LOOP_UNROLLING_68(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_70(idx, step, macro) LOOP_UNROLLING_69(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_71(idx, step, macro) LOOP_UNROLLING_70(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_72(idx, step, macro) LOOP_UNROLLING_71(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_73(idx, step, macro) LOOP_UNROLLING_72(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_74(idx, step, macro) LOOP_UNROLLING_73(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_75(idx, step, macro) LOOP_UNROLLING_74(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_76(idx, step, macro) LOOP_UNROLLING_75(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_77(idx, step, macro) LOOP_UNROLLING_76(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_78(idx, step, macro) LOOP_UNROLLING_77(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_79(idx, step, macro) LOOP_UNROLLING_78(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_80(idx, step, macro) LOOP_UNROLLING_79(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_81(idx, step, macro) LOOP_UNROLLING_80(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_82(idx, step, macro) LOOP_UNROLLING_81(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_83(idx, step, macro) LOOP_UNROLLING_82(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_84(idx, step, macro) LOOP_UNROLLING_83(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_85(idx, step, macro) LOOP_UNROLLING_84(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_86(idx, step, macro) LOOP_UNROLLING_85(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_87(idx, step, macro) LOOP_UNROLLING_86(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_88(idx, step, macro) LOOP_UNROLLING_87(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_89(idx, step, macro) LOOP_UNROLLING_88(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_90(idx, step, macro) LOOP_UNROLLING_89(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_91(idx, step, macro) LOOP_UNROLLING_90(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_92(idx, step, macro) LOOP_UNROLLING_91(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_93(idx, step, macro) LOOP_UNROLLING_92(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_94(idx, step, macro) LOOP_UNROLLING_93(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_95(idx, step, macro) LOOP_UNROLLING_94(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_96(idx, step, macro) LOOP_UNROLLING_95(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_97(idx, step, macro) LOOP_UNROLLING_96(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_98(idx, step, macro) LOOP_UNROLLING_97(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_99(idx, step, macro) LOOP_UNROLLING_98(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_100(idx, step, macro) LOOP_UNROLLING_99(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_101(idx, step, macro) LOOP_UNROLLING_100(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_102(idx, step, macro) LOOP_UNROLLING_101(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_103(idx, step, macro) LOOP_UNROLLING_102(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_104(idx, step, macro) LOOP_UNROLLING_103(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_105(idx, step, macro) LOOP_UNROLLING_104(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_106(idx, step, macro) LOOP_UNROLLING_105(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_107(idx, step, macro) LOOP_UNROLLING_106(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_108(idx, step, macro) LOOP_UNROLLING_107(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_109(idx, step, macro) LOOP_UNROLLING_108(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_110(idx, step, macro) LOOP_UNROLLING_109(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_111(idx, step, macro) LOOP_UNROLLING_110(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_112(idx, step, macro) LOOP_UNROLLING_111(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_113(idx, step, macro) LOOP_UNROLLING_112(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_114(idx, step, macro) LOOP_UNROLLING_113(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_115(idx, step, macro) LOOP_UNROLLING_114(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_116(idx, step, macro) LOOP_UNROLLING_115(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_117(idx, step, macro) LOOP_UNROLLING_116(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_118(idx, step, macro) LOOP_UNROLLING_117(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_119(idx, step, macro) LOOP_UNROLLING_118(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_120(idx, step, macro) LOOP_UNROLLING_119(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_121(idx, step, macro) LOOP_UNROLLING_120(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_122(idx, step, macro) LOOP_UNROLLING_121(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_123(idx, step, macro) LOOP_UNROLLING_122(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_124(idx, step, macro) LOOP_UNROLLING_123(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_125(idx, step, macro) LOOP_UNROLLING_124(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_126(idx, step, macro) LOOP_UNROLLING_125(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_127(idx, step, macro) LOOP_UNROLLING_126(idx, step, macro); UNROLL_INCR(idx, step, macro)
#define LOOP_UNROLLING_128(idx, step, macro) LOOP_UNROLLING_127(idx, step, macro); UNROLL_INCR(idx, step, macro)

#define LOOP_UNROLLING_STR(type, idx, start, step, num, macro) \
    {                                                          \
        type idx = start;                                      \
        LOOP_UNROLLING_##num(idx, step, macro);                \
    }
#else
#define LOOP_UNROLLING_STR(type, idx, start, step, num, macro) \
    {                                                          \
        _Pragma("unroll")                                      \
        for(type idx = start; idx < (num * step); idx += step) \
        {                                                      \
            (macro);                                           \
        }                                                      \
    }
#endif
#define LOOP_UNROLLING(type, idx, start, step, num, macro) LOOP_UNROLLING_STR(type, idx, start, step, num, macro)


#define GET_SPATIAL_IDX(IDX, N0, PARTIAL_N0) (max((int)(get_global_id(IDX) * N0 - (N0 - PARTIAL_N0) % N0), 0))


#define DOT_PRODUCT_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, K0, a, b, c) DOT_PRODUCT_INTEGER8_STR(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, K0, a, b, c)
#define DOT_PRODUCT_INTEGER8_STR(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, K0, a, b, c) DOT_PRODUCT##K0##_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c)
#define DOT_PRODUCT1_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        c += (C_DATA_TYPE)(a) * (C_DATA_TYPE)(b);     \
    })
#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_khr_integer_dot_product)
#define DOT_PRODUCT2_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c += dot((A_DATA_TYPE##4)((a).s01, (A_DATA_TYPE##2)(0)), (B_DATA_TYPE##4)(((b).s01), (B_DATA_TYPE##2)(0)));
#define DOT_PRODUCT3_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c += dot((A_DATA_TYPE##4)((a).s012, (A_DATA_TYPE)0), (B_DATA_TYPE##4)(((b).s012), (B_DATA_TYPE)0));
#define DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c += dot((a), (b));
#elif defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8)
#define DOT_PRODUCT2_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c = arm_dot_acc((A_DATA_TYPE##4)((a).s01, (A_DATA_TYPE##2)(0)), (B_DATA_TYPE##4)(((b).s01), (B_DATA_TYPE##2)(0)), (c));
#define DOT_PRODUCT3_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c = arm_dot_acc((A_DATA_TYPE##4)((a).s012, (A_DATA_TYPE)0), (B_DATA_TYPE##4)(((b).s012), (B_DATA_TYPE)0), (c));
#define DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c = arm_dot_acc((a), (b), (c));
#elif defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)
#define DOT_PRODUCT2_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c += arm_dot((A_DATA_TYPE##4)((a).s01, (A_DATA_TYPE##2)(0)), (B_DATA_TYPE##4)(((b).s01), (B_DATA_TYPE##2)(0)));
#define DOT_PRODUCT3_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c += arm_dot((A_DATA_TYPE##4)((a).s012, (A_DATA_TYPE)0), (B_DATA_TYPE##4)(((b).s012), (B_DATA_TYPE)0));
#define DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) c += arm_dot((a), (b));
#else
#define DOT_PRODUCT2_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c)   \
    ({                                                  \
        c += (C_DATA_TYPE)(a).s0 * (C_DATA_TYPE)(b).s0; \
        c += (C_DATA_TYPE)(a).s1 * (C_DATA_TYPE)(b).s1; \
    })
#define DOT_PRODUCT3_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c)   \
    ({                                                  \
        DOT_PRODUCT2_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c);  \
        c += (C_DATA_TYPE)(a).s2 * (C_DATA_TYPE)(b).s2; \
    })
#define DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, x, y, val)   \
    ({                                                    \
        val += (C_DATA_TYPE)(x).s0 * (C_DATA_TYPE)(y).s0; \
        val += (C_DATA_TYPE)(x).s1 * (C_DATA_TYPE)(y).s1; \
        val += (C_DATA_TYPE)(x).s2 * (C_DATA_TYPE)(y).s2; \
        val += (C_DATA_TYPE)(x).s3 * (C_DATA_TYPE)(y).s3; \
    })
#endif
#define DOT_PRODUCT5_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s0123), ((b).s0123), c);     \
        DOT_PRODUCT1_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s4), ((b).s4), c);     \
    })
#define DOT_PRODUCT6_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s0123), ((b).s0123), c);     \
        DOT_PRODUCT2_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s45), ((b).s45), c);     \
    })
#define DOT_PRODUCT7_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s0123), ((b).s0123), c);     \
        DOT_PRODUCT3_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s456), ((b).s456), c);     \
    })
#define DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).lo), ((b).lo), c);     \
        DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).hi), ((b).hi), c);     \
    })
#define DOT_PRODUCT9_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s01234567), ((b).s01234567), c);     \
        DOT_PRODUCT1_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s8), ((b).s8), c);     \
    })
#define DOT_PRODUCT10_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s01234567), ((b).s01234567), c);     \
        DOT_PRODUCT2_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s89), ((b).s89), c);     \
    })
#define DOT_PRODUCT11_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s01234567), ((b).s01234567), c);     \
        DOT_PRODUCT3_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s89A), ((b).s89A), c);     \
    })
#define DOT_PRODUCT12_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s01234567), ((b).s01234567), c);     \
        DOT_PRODUCT4_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s89AB), ((b).s89AB), c);     \
    })
#define DOT_PRODUCT13_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s01234567), ((b).s01234567), c);     \
        DOT_PRODUCT5_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s89ABC), ((b).s89ABC), c);     \
    })
#define DOT_PRODUCT14_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s01234567), ((b).s01234567), c);     \
        DOT_PRODUCT6_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s89ABCD), ((b).s89ABCD), c);     \
    })
#define DOT_PRODUCT15_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s01234567), ((b).s01234567), c);     \
        DOT_PRODUCT7_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).s89ABCDE), ((b).s89ABCDE), c);     \
    })
#define DOT_PRODUCT16_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, a, b, c) \
    ({                                                 \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).lo), ((b).lo), c);      \
        DOT_PRODUCT8_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, ((a).hi), ((b).hi), c);      \
    })


#define REDUCE_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, K0, a, c) REDUCE_INTEGER8_STR(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, K0, a, c)
#define REDUCE_INTEGER8_STR(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, K0, a, c) DOT_PRODUCT_INTEGER8(A_DATA_TYPE, B_DATA_TYPE, C_DATA_TYPE, K0, a, (TILE_VECTOR_TYPE##K0(B_DATA_TYPE))1, c)


#define V_LOAD(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, Y, STRIDE_Y) V_LOAD_STR(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, Y, STRIDE_Y)
#define V_LOAD_STR(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, Y, STRIDE_Y) V_LOAD_##TENSOR_TYPE(DATA_TYPE, WIDTH, TENSOR, X, Y, STRIDE_Y)
#define V_LOAD_BUFFER(DATA_TYPE, WIDTH, TENSOR, X, Y, STRIDE_Y) \
    VLOAD(WIDTH)                                                \
    (0, (__global DATA_TYPE *)(TENSOR##_ptr + TENSOR##_offset_first_element_in_bytes + (X) * sizeof(DATA_TYPE) + (Y) * (STRIDE_Y)))
#define V_LOAD_IMAGE(DATA_TYPE, WIDTH, TENSOR, X, Y, STRIDE_Y) READ_IMAGE2D(DATA_TYPE, CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(WIDTH), TENSOR##_img, (X) / 4, (Y))


#define V_STORE(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, Y, STRIDE_Y, VALUES) V_STORE_STR(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, Y, STRIDE_Y, VALUES)
#define V_STORE_STR(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, Y, STRIDE_Y, VALUES) V_STORE_##TENSOR_TYPE(DATA_TYPE, WIDTH, TENSOR, X, Y, STRIDE_Y, VALUES)
#define V_STORE_BUFFER(DATA_TYPE, WIDTH, TENSOR, X, Y, STRIDE_Y, VALUES) \
    VSTORE(WIDTH)                                                \
    (VALUES, 0, (__global DATA_TYPE *)(TENSOR##_ptr + TENSOR##_offset_first_element_in_bytes + (X) * sizeof(DATA_TYPE) + (Y) * (STRIDE_Y)))
#define V_STORE_IMAGE(DATA_TYPE, WIDTH, TENSOR, X, Y, STRIDE_Y, VALUES) WRITE_IMAGE2D(DATA_TYPE, CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(WIDTH), TENSOR##_img, (X) / 4, (Y), VALUES)


#define T_LOAD(DATA_TYPE, HEIGHT, WIDTH, TENSOR_TYPE, TENSOR, X, Y, YI_MULTIPLIER, STRIDE_Y, dst)                      \
    ({                                                                                                                 \
        LOOP_UNROLLING(int, _i, 0, 1, HEIGHT,                                                                          \
        {                                                                                                              \
            dst[_i].v = V_LOAD(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, ((Y) + _i * (int)(YI_MULTIPLIER)), STRIDE_Y); \
        })                                                                                                             \
    })


#define T_LOAD_INDIRECT(DATA_TYPE, HEIGHT, WIDTH, TENSOR_TYPE, TENSOR, X, STRIDE_Y, indirect_y, dst)    \
    ({                                                                                                  \
        LOOP_UNROLLING(int, _i, 0, 1, HEIGHT,                                                           \
        {                                                                                               \
            dst[_i].v = V_LOAD(DATA_TYPE, WIDTH, TENSOR_TYPE, TENSOR, X, (indirect_y[_i].v), STRIDE_Y); \
        })                                                                                              \
    })


#define T_LOAD_INDIRECT_WIDTH_SELECT(DATA_TYPE, HEIGHT, WIDTH0, WIDTH1, TENSOR_TYPE, TENSOR, X, STRIDE_Y, WIDTH1_CONDITION, dst, indirect_y)                                                      \
    ({                                                                                                                                                                                             \
        if(WIDTH1_CONDITION)                                                                                                                                                                       \
        {                                                                                                                                                                                          \
            LOOP_UNROLLING(int, _i, 0, 1, HEIGHT,                                                                                                                                                  \
            {                                                                                                                                                                                      \
                VLOAD_PARTIAL(WIDTH0, WIDTH1)                                                         \
                (dst[HEIGHT - 1 - _i].v, 0, (__global DATA_TYPE *)(TENSOR##_ptr + TENSOR##_offset_first_element_in_bytes + (X) * sizeof(DATA_TYPE) + (indirect_y[HEIGHT - 1 - _i].v) * STRIDE_Y));               \
            })                                                                                                                                                                                     \
        }                                                                                                                                                                                          \
        else                                                                                                                                                                                       \
        {                                                                                                                                                                                          \
            LOOP_UNROLLING(int, _i, 0, 1, HEIGHT,                                                                                                                                                  \
            {                                                                                                                                                                                      \
                dst[HEIGHT - 1 - _i].v = V_LOAD(DATA_TYPE, WIDTH0, TENSOR_TYPE, TENSOR, X, (indirect_y[HEIGHT - 1 - _i].v), STRIDE_Y); \
            })                                                                                                                                                                                     \
        }                                                                                                                                                                                          \
    })

#define T_LOAD_NHWC(DATA_TYPE, TILE_HEIGHT, TILE_WIDTH, TILE_CHANNELS, TENSOR_TYPE, TENSOR, B, Y, X, C, TENSOR_WIDTH, TENSOR_HEIGHT, STRIDE_Y, dst)   \
    ({                                                                                                                                                \
        LOOP_UNROLLING(int, _yk, 0, 1, TILE_HEIGHT,                                                                                                   \
        {                                                                                                                                             \
            LOOP_UNROLLING(int, _xk, 0, 1, TILE_WIDTH,                                                                                                \
            {                                                                                                                                         \
                int _src_y = (X) + _xk + ((Y) + _yk) * (TENSOR_WIDTH);                                                                                \
                _src_y    += (B) * (int)(TENSOR_WIDTH) * (int)(TENSOR_HEIGHT);                                                                        \
                int _src_valid_y = (((X) + _xk) >= 0 && ((X) + _xk) < (int)(TENSOR_WIDTH) && ((Y) + _yk) >= 0 && ((Y) + _yk) < (int)(TENSOR_HEIGHT)); \
                if(_src_valid_y != 0)                                                                                                                 \
                {                                                                                                                                     \
                    dst[_xk + _yk * (TILE_WIDTH)].v = V_LOAD(DATA_TYPE, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, _src_y, STRIDE_Y);                     \
                }                                                                                                                                     \
            })                                                                                                                                        \
        })                                                                                                                                            \
    })


#define T_LOAD_NHWC_WITH_DILATION(DATA_TYPE, TILE_HEIGHT, TILE_WIDTH, TILE_CHANNELS, TENSOR_TYPE, TENSOR, B, Y, X, C, TENSOR_WIDTH, TENSOR_HEIGHT, DILATION_X, DILATION_Y, BOUNDARY_CHECK, dst)         \
    ({ \
        LOOP_UNROLLING(int, _yk, 0, 1, TILE_HEIGHT, \
        { \
            LOOP_UNROLLING(int, _xk, 0, 1, TILE_WIDTH, \
            { \
                int _src_y = (X) + _xk * (DILATION_X); \
                int _src_z = ((Y) + _yk * (DILATION_Y)); \
                int _src_w    = (B); \
                bool _src_valid_y = (((X) + _xk * (DILATION_X)) >= 0) && (((X) + _xk * (DILATION_X)) < (int)(TENSOR_WIDTH)) && (((Y) + _yk * (DILATION_Y)) >= 0) && (((Y) + _yk * (DILATION_Y)) < (int)(TENSOR_HEIGHT)); \
                if(!(BOUNDARY_CHECK)) \
                { \
                    dst[_xk + _yk * (TILE_WIDTH)].v = VLOAD(TILE_CHANNELS)                                                \
                    (0, (__global DATA_TYPE *)(TENSOR##_ptr + TENSOR##_offset_first_element_in_bytes + (C) * sizeof(DATA_TYPE) + (_src_y) * (TENSOR##_stride_y) + (_src_z) * (TENSOR##_stride_z) + (_src_w) * (TENSOR##_stride_w))); \
                } \
                else \
                { \
                    if(_src_valid_y) \
                    { \
                        dst[_xk + _yk * (TILE_WIDTH)].v = VLOAD(TILE_CHANNELS)                                                \
                    (0, (__global DATA_TYPE *)(TENSOR##_ptr + TENSOR##_offset_first_element_in_bytes + (C) * sizeof(DATA_TYPE) + (_src_y) * (TENSOR##_stride_y) + (_src_z) * (TENSOR##_stride_z) + (_src_w) * (TENSOR##_stride_w))); \
                    }                                                                                                                                                                                                 \
                } \
            })                                                                                                                                                                                                             \
        })                                                                                                                                                                                                             \
    })


#define T_LOAD_NHWC_INDIRECT(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, B, Y, X, C, TENSOR_WIDTH, TENSOR_HEIGHT, STRIDE_Y, xi, yi, dst)                \
    ({                                                                                                                                                                \
        LOOP_UNROLLING(int, _i, 0, 1, TILE_AREA,                                                                                                                      \
        {                                                                                                                                                             \
            int _src_y = (X) + xi[_i].v + ((Y) + yi[_i].v) * (TENSOR_WIDTH);                                                                                          \
            _src_y += (B) * (int)(TENSOR_WIDTH) * (int)(TENSOR_HEIGHT);                                                                                               \
            int _src_valid_y = (((X) + xi[_i].v) >= 0 && ((X) + xi[_i].v) < (int)(TENSOR_WIDTH) && ((Y) + yi[_i].v) >= 0 && ((Y) + yi[_i].v) < (int)(TENSOR_HEIGHT)); \
            if(_src_valid_y != 0)                                                                                                                                     \
            {                                                                                                                                                         \
                dst[_i].v = V_LOAD(DATA_TYPE, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, _src_y, STRIDE_Y);                                                               \
            }                                                                                                                                                         \
        })                                                                                                                                                            \
    })


#define T_LOAD2D_INDIRECT(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, STRIDE_Y, yi, dst) T_LOAD2D_INDIRECT_STR(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, STRIDE_Y, yi, dst)
#define T_LOAD2D_INDIRECT_STR(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, STRIDE_Y, yi, dst) T_LOAD2D_INDIRECT_##TENSOR_TYPE(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, STRIDE_Y, yi, dst)
#define T_LOAD2D_INDIRECT_BUFFER(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, STRIDE_Y, yi, dst) \
    ({ \
        LOOP_UNROLLING(int, _i, 0, 1, TILE_AREA, \
        { \
            if(yi[0].s[_i] >= 0) \
            { \
                dst[_i].v = V_LOAD(DATA_TYPE, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, yi[0].s[_i], STRIDE_Y); \
            } \
        }) \
    })

#define T_LOAD2D_INDIRECT_IMAGE(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, STRIDE_Y, yi, dst) \
    ({ \
        LOOP_UNROLLING(int, _i, 0, 1, TILE_AREA, \
        { \
            dst[_i].v = V_LOAD(DATA_TYPE, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, yi[0].s[_i], STRIDE_Y); \
        }) \
    })


#define T_LOAD_NDHWC_INDIRECT(DATA_TYPE, TILE_AREA, TILE_CHANNELS, TENSOR_TYPE, TENSOR, B, Z, Y, X, C, TENSOR_WIDTH, TENSOR_HEIGHT, TENSOR_DEPTH, STRIDE_Y, xi, yi, zi, dst) \
    ({                                                                                                                                                                \
        LOOP_UNROLLING(int, _i, 0, 1, TILE_AREA,                                                                                                                      \
        {                                                                                                                                                             \
            int _src_y = (X) + xi[_i].v + ((Y) + yi[_i].v) * (TENSOR_WIDTH) + ((Z) + zi[_i].v) * (TENSOR_WIDTH * TENSOR_HEIGHT);                                      \
            _src_y += (B) * (int)(TENSOR_WIDTH) * (int)(TENSOR_HEIGHT) * (int)(TENSOR_DEPTH);                                                                         \
            int _src_valid_y = (((X) + xi[_i].v) >= 0 && ((X) + xi[_i].v) < (int)(TENSOR_WIDTH) && ((Y) + yi[_i].v) >= 0 && ((Y) + yi[_i].v) < (int)(TENSOR_HEIGHT)   \
                             && ((Z) + zi[_i].v) >= 0 && ((Z) + zi[_i].v) < (int)(TENSOR_DEPTH));                                                                     \
            if(_src_valid_y != 0)                                                                                                                                     \
            {                                                                                                                                                         \
                dst[_i].v = V_LOAD(DATA_TYPE, TILE_CHANNELS, TENSOR_TYPE, TENSOR, C, _src_y, STRIDE_Y);                                                               \
            }                                                                                                                                                         \
        })                                                                                                                                                            \
    })


#define T_STORE_INDIRECT_WIDTH_SELECT(DATA_TYPE, HEIGHT, WIDTH0, WIDTH1, TENSOR_TYPE, TENSOR, X, STRIDE_Y, WIDTH1_CONDITION, src, indirect_y)                                                      \
    ({                                                                                                                                                                                             \
        if(WIDTH1_CONDITION)                                                                                                                                                                       \
        {                                                                                                                                                                                          \
            LOOP_UNROLLING(int, _i, 0, 1, HEIGHT,                                                                                                                                                  \
            {                                                                                                                                                                                      \
                VSTORE_PARTIAL(WIDTH0, WIDTH1)                                                                                                                                                     \
                (CONVERT(src[HEIGHT - 1 - _i].v, VEC_DATA_TYPE(DATA_TYPE, WIDTH0)), 0, (__global DATA_TYPE *)(TENSOR##_ptr + TENSOR##_offset_first_element_in_bytes + (X) * sizeof(DATA_TYPE) + (indirect_y[HEIGHT - 1 - _i].v) * STRIDE_Y)); \
            })                                                                                                                                                                                     \
        }                                                                                                                                                                                          \
        else                                                                                                                                                                                       \
        {                                                                                                                                                                                          \
            LOOP_UNROLLING(int, _i, 0, 1, HEIGHT,                                                                                                                                                  \
            {                                                                                                                                                                                      \
                VSTORE(WIDTH0)                                                                                                                                                                     \
                (CONVERT(src[HEIGHT - 1 - _i].v, VEC_DATA_TYPE(DATA_TYPE, WIDTH0)), 0, (__global DATA_TYPE *)(TENSOR##_ptr + TENSOR##_offset_first_element_in_bytes + (X) * sizeof(DATA_TYPE) + (indirect_y[HEIGHT - 1 - _i].v) * STRIDE_Y)); \
            })                                                                                                                                                                                     \
        }                                                                                                                                                                                          \
    })


#define T_OFFSET_CORRECTION(ACC_DATA_TYPE, M0, N0, K0, SRC_OFFSET, WEI_OFFSET, lhs, rhs, dst)        \
    ({                                                                                               \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                                                           \
        {                                                                                            \
            ACC_DATA_TYPE _tm = 0;                                                                   \
            LOOP_UNROLLING(int, _k0, 0, 1, K0,                                                       \
            {                                                                                        \
                _tm += ((ACC_DATA_TYPE)lhs[_m0].s[_k0] * (ACC_DATA_TYPE)WEI_OFFSET);                 \
            })                                                                                       \
            LOOP_UNROLLING(int, _n0, 0, 1, N0,                                                       \
            {                                                                                        \
                dst[_m0].s[_n0] += _tm;                                                              \
                LOOP_UNROLLING(int, _k0, 0, 1, K0,                                                   \
                {                                                                                    \
                    dst[_m0].s[_n0] += ((ACC_DATA_TYPE)rhs[_n0].s[_k0] * (ACC_DATA_TYPE)SRC_OFFSET); \
                })                                                                                   \
            })                                                                                       \
        })                                                                                          \
    })


#define T_QUANTIZE8(SRC_DATA_TYPE, DST_DATA_TYPE, QUANTIZATION_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, src, dst_multipliers, dst_shifts, dst) T_QUANTIZE8_STR(SRC_DATA_TYPE, DST_DATA_TYPE, QUANTIZATION_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, src, dst_multipliers, dst_shifts, dst)
#define T_QUANTIZE8_STR(SRC_DATA_TYPE, DST_DATA_TYPE, QUANTIZATION_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, src, dst_multipliers, dst_shifts, dst) T_QUANTIZE8_##QUANTIZATION_TYPE(SRC_DATA_TYPE, DST_DATA_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, src, dst_multipliers, dst_shifts, dst)


#define T_QUANTIZE8_PER_TENSOR(SRC_DATA_TYPE, DST_DATA_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, src, dst_multipliers, dst_shifts, dst)                          \
    ({ \
        LOOP_UNROLLING(int, _m0, 0, 1, M0, \
        { \
            LOOP_UNROLLING(int, _n0, 0, 1, N0, \
            { \
                SRC_DATA_TYPE _tmp = 0; \
                SRC_DATA_TYPE _src = src[_m0].s[_n0]; \
                _src *= select((SRC_DATA_TYPE)1, ((SRC_DATA_TYPE)1 << (SRC_DATA_TYPE)(-DST_SHIFT)), ((SRC_DATA_TYPE)DST_SHIFT < (SRC_DATA_TYPE)0)); \
                SRC_DATA_TYPE overflow = _src == DST_MULTIPLIER && _src == INT_MIN; \
                long a_64 = (long)(_src); \
                long b_64 = (long)(DST_MULTIPLIER); \
                long ab_64 = a_64 * b_64; \
                long mask1 = 1 << 30; \
                long mask2 = 1 - (1 << 30); \
                long is_positive_or_zero = ab_64 >= 0; \
                long nudge = select(mask2, mask1, is_positive_or_zero); \
                SRC_DATA_TYPE ab_x2_high32 = CONVERT((ab_64 + nudge) / (long)(1ll << 31), SRC_DATA_TYPE); \
                _tmp = select(ab_x2_high32, (SRC_DATA_TYPE)INT_MAX, overflow); \
                if(DST_SHIFT >= 0) \
                { \
                    long mask = ((((int)1) << DST_SHIFT) - (long)1); \
                    long threshold = _tmp < (int)0 ? (mask >> 1) + (long)1 : (mask >> 1) + 0; \
                    _tmp = (_tmp & mask) > threshold ? (_tmp >> DST_SHIFT) + (int)1 : (_tmp >> DST_SHIFT); \
                } \
                _tmp += DST_OFFSET; \
                dst[_m0].s[_n0] = CONVERT_SAT(_tmp, DST_DATA_TYPE);                                                                            \
            })                                                                                                                                          \
        })                                                                                                                                          \
    })


#define T_QUANTIZE8_PER_CHANNEL(SRC_DATA_TYPE, DST_DATA_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, src, dst_multipliers, dst_shifts, dst)                          \
    ({ \
        LOOP_UNROLLING(int, _m0, 0, 1, M0, \
        { \
            LOOP_UNROLLING(int, _n0, 0, 1, N0, \
            { \
                SRC_DATA_TYPE _tmp = 0; \
                SRC_DATA_TYPE _tmp2 = 0; \
                SRC_DATA_TYPE _src = src[_m0].s[_n0]; \
                SRC_DATA_TYPE _dst_multiplier = dst_multipliers[0].s[_n0]; \
                SRC_DATA_TYPE _dst_shift = dst_shifts[0].s[_n0]; \
                _src *= select((SRC_DATA_TYPE)1, ((SRC_DATA_TYPE)1 << (SRC_DATA_TYPE)(-_dst_shift)), ((SRC_DATA_TYPE)_dst_shift < (SRC_DATA_TYPE)0)); \
                SRC_DATA_TYPE overflow = _src == _dst_multiplier && _src == INT_MIN; \
                long a_64 = (long)(_src); \
                long b_64 = (long)(_dst_multiplier); \
                long ab_64 = a_64 * b_64; \
                long mask1 = 1 << 30; \
                long mask2 = 1 - (1 << 30); \
                long is_positive_or_zero = ab_64 >= 0; \
                long nudge = select(mask2, mask1, is_positive_or_zero); \
                SRC_DATA_TYPE ab_x2_high32 = CONVERT((ab_64 + nudge) / (long)(1ll << 31), SRC_DATA_TYPE); \
                _tmp = select(ab_x2_high32, (SRC_DATA_TYPE)INT_MAX, overflow); \
                long mask = ((((int)1) << _dst_shift) - (int)1); \
                long threshold = (mask >> 1) + any(_tmp); \
                _tmp2 = _tmp >> _dst_shift; \
                _tmp2 += select(0, 1, (_tmp & mask) > threshold); \
                _tmp = select(_tmp, _tmp2, _dst_shift >= 0); \
                _tmp += DST_OFFSET; \
                dst[_m0].s[_n0] = CONVERT_SAT(_tmp, DST_DATA_TYPE);                                                                            \
            })                                                                                                                                          \
        })                                                                                                                                         \
    })


#define T_QUANTIZE8_ASYMMETRIC(SRC_DATA_TYPE, DST_DATA_TYPE, M0, N0, DST_OFFSET, DST_SHIFT, DST_MULTIPLIER, src, dst)                          \
    ({ \
        LOOP_UNROLLING(int, _m0, 0, 1, M0, \
        { \
            LOOP_UNROLLING(int, _n0, 0, 1, N0, \
            { \
                SRC_DATA_TYPE _tmp = 0; \
                SRC_DATA_TYPE _src = src[_m0].s[_n0]; \
                _src *= select((SRC_DATA_TYPE)1, ((SRC_DATA_TYPE)1 << (SRC_DATA_TYPE)(-DST_SHIFT)), ((SRC_DATA_TYPE)DST_SHIFT < (SRC_DATA_TYPE)0)); \
                SRC_DATA_TYPE overflow = _src == DST_MULTIPLIER && _src == INT_MIN; \
                long a_64 = (long)(_src); \
                long b_64 = (long)(DST_MULTIPLIER); \
                long ab_64 = a_64 * b_64; \
                long mask1 = 1 << 30; \
                long mask2 = 1 - (1 << 30); \
                long is_positive_or_zero = ab_64 >= 0; \
                long nudge = select(mask2, mask1, is_positive_or_zero); \
                SRC_DATA_TYPE ab_x2_high32 = CONVERT((ab_64 + nudge) / (long)(1ll << 31), SRC_DATA_TYPE); \
                _tmp = select(ab_x2_high32, (SRC_DATA_TYPE)INT_MAX, overflow); \
                if(DST_SHIFT >= 0) \
                { \
                    long mask = ((((int)1) << DST_SHIFT) - (int)1); \
                    long threshold = _tmp < (int)0 ? (mask >> 1) + (long)1 : (mask >> 1) + 0; \
                    _tmp = (_tmp & mask) > threshold ? (_tmp >> DST_SHIFT) + (int)1 : (_tmp >> DST_SHIFT); \
                } \
                _tmp += DST_OFFSET; \
                dst[_m0].s[_n0] = CONVERT_SAT(_tmp, DST_DATA_TYPE);                                                                            \
            })                                                                                                                                          \
        })                                                                                                                                          \
    })


#define T_ROWSET_MASK(DATA_TYPE, M0, N0, VALUE_TO_SET, a, mask)                                                                                            \
    ({                                                                                                                                                     \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                                                                                                                 \
        {                                                                                                                                                  \
            LOOP_UNROLLING(int, _n0, 0, 1, N0,                                                                                                             \
            {                                                                                                                                              \
                a[_m0].s[_n0] = select((DATA_TYPE)(a[_m0].s[_n0]), (DATA_TYPE)(VALUE_TO_SET), (SELECT_DATA_TYPE(DATA_TYPE))(mask[_m0].v == (DATA_TYPE)0)); \
            })                                                                                                                                             \
        })                                                                                                                                                 \
    })


#define T_ACTIVATION(DATA_TYPE, M0, N0, ACTIVATION_TYPE, A_VAL, B_VAL, src, dst)               \
    ({                                                                                         \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                                                     \
        {                                                                                      \
            dst[_m0].v = ACTIVATION(ACTIVATION_TYPE, DATA_TYPE, N0, src[_m0].v, A_VAL, B_VAL); \
        })                                                                                     \
    })


#define relu_op_quantized(DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) (max((DATA_TYPE)ZERO_VALUE, x))

#define brelu_op_quantized(DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) (min((DATA_TYPE)A_VAL, max((DATA_TYPE)ZERO_VALUE, x)))

#define lu_brelu_op_quantized(DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) (min(max(x, (DATA_TYPE)B_VAL), (DATA_TYPE)A_VAL))

#define hard_swish_op_quantized(DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) (x * ((min(max((DATA_TYPE)(x + (DATA_TYPE)3.f), (DATA_TYPE)0.f), (DATA_TYPE)6.f)) * (DATA_TYPE)0.166666667f))

#define identity_op_quantized(DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) (x)

#define ACT_OP_QUANTIZED(op, DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) op##_op_quantized(DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x)
#define ACTIVATION_QUANTIZED(op, DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x) ACT_OP_QUANTIZED(op, DATA_TYPE, VEC_SIZE, ZERO_VALUE, A_VAL, B_VAL, x)

#define V_ADD(A_VAL, B_VAL) ((A_VAL) + (B_VAL))
#define V_SUB(A_VAL, B_VAL) ((A_VAL) - (B_VAL))
#define V_DIV(A_VAL, B_VAL) ((A_VAL) / (B_VAL))
#define V_MUL(A_VAL, B_VAL) ((A_VAL) * (B_VAL))


#define T_ACTIVATION_QUANTIZED(DATA_TYPE, M0, N0, ACTIVATION_TYPE, ZERO_VALUE, A_VAL, B_VAL, src, dst)               \
    ({ \
        LOOP_UNROLLING(int, _m0, 0, 1, M0, \
        { \
            dst[_m0].v = ACTIVATION_QUANTIZED(ACTIVATION_TYPE, DATA_TYPE, N0, ZERO_VALUE, A_VAL, B_VAL, src[_m0].v); \
        })                                                                                          \
    })


#define T_ADD(DATA_TYPE, M0, N0, lhs, rhs, dst) \
    ({                                                            \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                        \
        {                                                         \
            dst[_m0].v = lhs[_m0].v + rhs[_m0].v; \
        })                                                        \
    })


#define T_ADD_CONSTANT(DATA_TYPE, M0, N0, lhs, rhs_constant, dst) \
    ({                                                            \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                        \
        {                                                         \
            dst[_m0].v = lhs[_m0].v + (DATA_TYPE)rhs_constant;               \
        })                                                        \
    })

#define T_ELTWISE_BROADCAST_ADD_X(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
#define T_ELTWISE_BROADCAST_LHS_X_ADD(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_LHS_X(V_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
#define T_ELTWISE_BROADCAST_RHS_X_ADD(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)

#define T_ELTWISE_BROADCAST_LHS_X_SUB(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_LHS_X(V_SUB, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
#define T_ELTWISE_BROADCAST_RHS_X_SUB(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_SUB, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)

#define T_ELTWISE_BROADCAST_DIV_X(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)

#define T_ELTWISE_BROADCAST_LHS_X_MUL(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_LHS_X(V_MUL, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
#define T_ELTWISE_BROADCAST_RHS_X_MUL(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE_BROADCAST_X(V_MUL, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)


#define T_SCALE_CONSTANT(DATA_TYPE, M0, N0, lhs, rhs_constant, dst) \
    ({                                                            \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                        \
        {                                                         \
            dst[_m0].v = lhs[_m0].v * (DATA_TYPE)rhs_constant; \
        })                                                        \
    })


#define T_ELTWISE_BROADCAST_X(T_ELWISE_OP, DST_DATA_TYPE, M0, N0, lhs, rhs, dst) \
    ({                                                      \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                  \
        {                                                   \
            dst[_m0].v = T_ELWISE_OP(CONVERT(lhs[_m0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)), CONVERT(rhs[0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)));             \
        })                                                  \
    })


#define T_ELTWISE_BROADCAST_LHS_X(T_ELWISE_OP, DST_DATA_TYPE, M0, N0, lhs, rhs, dst) \
    ({                                                      \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                  \
        {                                                   \
            dst[_m0].v = T_ELWISE_OP(CONVERT(lhs[0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)), CONVERT(rhs[_m0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)));             \
        })                                                  \
    })

#define T_ELTWISE_ADD(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_ADD, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
#define T_ELTWISE_SUB(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_SUB, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
#define T_ELTWISE_DIV(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_DIV, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)
#define T_ELTWISE_MUL(DST_DATA_TYPE, M0, N0, lhs, rhs, dst) T_ELTWISE(V_MUL, DST_DATA_TYPE, M0, N0, lhs, rhs, dst)


#define T_ELTWISE(T_ELWISE_OP, DST_DATA_TYPE, M0, N0, lhs, rhs, dst) \
    ({                                                      \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                  \
        {                                                   \
            dst[_m0].v = T_ELWISE_OP(CONVERT(lhs[_m0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)), CONVERT(rhs[_m0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)));             \
        })                                                  \
    })


#define T_FLOOR(DST_DATA_TYPE, M0, N0, src, dst) \
    ({                                                      \
        LOOP_UNROLLING(int, _m0, 0, 1, M0,                  \
        {                                                   \
            dst[_m0].v = floor(CONVERT(src[_m0].v, VEC_DATA_TYPE(DST_DATA_TYPE, N0)));             \
        })                                                  \
    })


#define T_MMUL(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, LHS_LAYOUT, RHS_LAYOUT, lhs, rhs, dst) T_MMUL_##LHS_LAYOUT##_##RHS_LAYOUT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_##LHS_DATA_TYPE##_##RHS_DATA_TYPE##_##DST_DATA_TYPE(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T_float_float_float(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T_half_half_float(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T_half_half_half(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T_char_char_int(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_INTEGER8(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T_uchar_uchar_uint(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_INTEGER8(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T_uchar_uchar_int(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst) T_MMUL_NT_T_INTEGER8(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)
#define T_MMUL_NT_T_FLOAT(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)                       \
    {                                                                                     \
        LOOP_UNROLLING(int, _m, 0, 1, M0,                                                 \
        {                                                                                 \
            LOOP_UNROLLING(int, _n, 0, 1, N0,                                             \
            {                                                                             \
                LOOP_UNROLLING(int, _k, 0, 1, K0,                                         \
                {                                                                         \
                    dst[_m].s[_n] = fma((DST_DATA_TYPE)(lhs[_m].s[_k]), (DST_DATA_TYPE)(rhs[_n].s[_k]), dst[_m].s[_n]); \
                })                                                                        \
            })                                                                            \
        })                                                                                \
    }

#define T_MMUL_NT_T_INTEGER8(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, M0, N0, K0, lhs, rhs, dst)                            \
    ({ \
        LOOP_UNROLLING(int, _m, 0, 1, M0, \
        { \
            LOOP_UNROLLING(int, _n, 0, 1, N0, \
            { \
                DOT_PRODUCT_INTEGER8(LHS_DATA_TYPE, RHS_DATA_TYPE, DST_DATA_TYPE, K0, (lhs[_m].v), (rhs[_n].v), dst[_m].s[_n]); \
            })                                                                                             \
        })                                                                                             \
    })

#endif

#if defined(DATA_TYPE) && defined(ACC_DATA_TYPE)

#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)
#if defined(ARM_COMPUTE_OPENCL_DOT8_ACC_ENABLED) && defined(cl_arm_integer_dot_product_accumulate_int8)
#define ARM_DOT(x, y, val) val = arm_dot_acc((x), (y), (val));
#else
#define ARM_DOT(x, y, val) val += arm_dot((x), (y));
#endif
#endif

#if defined(ARM_COMPUTE_OPENCL_DOT8_ENABLED) && defined(cl_arm_integer_dot_product_int8)

#define ARM_DOT1(a, b, c)                                                                                                                               \
    ({                                                                                                                                                  \
        ARM_DOT((VEC_DATA_TYPE(DATA_TYPE, 4))(a, (VEC_DATA_TYPE(DATA_TYPE, 3))0), (VEC_DATA_TYPE(DATA_TYPE, 4))(b, (VEC_DATA_TYPE(DATA_TYPE, 3))0), c); \
    })
#define ARM_DOT2(a, b, c)                                                                                                                               \
    ({                                                                                                                                                  \
        ARM_DOT((VEC_DATA_TYPE(DATA_TYPE, 4))(a, (VEC_DATA_TYPE(DATA_TYPE, 2))0), (VEC_DATA_TYPE(DATA_TYPE, 4))(b, (VEC_DATA_TYPE(DATA_TYPE, 2))0), c); \
    })
#define ARM_DOT3(a, b, c)                                                                                           \
    ({                                                                                                              \
        ARM_DOT((VEC_DATA_TYPE(DATA_TYPE, 4))(a, (DATA_TYPE)0), (VEC_DATA_TYPE(DATA_TYPE, 4))(b, (DATA_TYPE)0), c); \
    })
#define ARM_DOT4(a, b, c) \
    ({                    \
        ARM_DOT(a, b, c); \
    })
#define ARM_DOT8(a, b, c)            \
    ({                               \
        ARM_DOT4((a.lo), (b.lo), c); \
        ARM_DOT4((a.hi), (b.hi), c); \
    })
#define ARM_DOT16(a, b, c)           \
    ({                               \
        ARM_DOT8((a.lo), (b.lo), c); \
        ARM_DOT8((a.hi), (b.hi), c); \
    })

#else


#define ARM_DOT1(a, b, c)          \
    ({                             \
        c += (ACC_DATA_TYPE)a * b; \
    })
#define ARM_DOT2(a, b, c)                \
    ({                                   \
        c += (ACC_DATA_TYPE)a.s0 * b.s0; \
        c += (ACC_DATA_TYPE)a.s1 * b.s1; \
    })
#define ARM_DOT3(a, b, c)                \
    ({                                   \
        ARM_DOT2(a, b, c);               \
        c += (ACC_DATA_TYPE)a.s2 * b.s2; \
    })
#define ARM_DOT4(a, b, c)                \
    ({                                   \
        ARM_DOT3(a, b, c);               \
        c += (ACC_DATA_TYPE)a.s3 * b.s3; \
    })
#define ARM_DOT8(a, b, c)            \
    ({                               \
        ARM_DOT4((a.lo), (b.lo), c); \
        ARM_DOT4((a.hi), (b.hi), c); \
    })
#define ARM_DOT16(a, b, c)           \
    ({                               \
        ARM_DOT8((a.lo), (b.lo), c); \
        ARM_DOT8((a.hi), (b.hi), c); \
    })
#endif


#define ARM_DOT_K0X1(k0, a, b, c)         \
    ({                                    \
        ARM_DOT_K0(k0, (a), (b##0), (c)); \
    })
#define ARM_DOT_K0X2(k0, a, b, c)            \
    ({                                       \
        ARM_DOT_K0(k0, (a), (b##0), (c.s0)); \
        ARM_DOT_K0(k0, (a), (b##1), (c.s1)); \
    })
#define ARM_DOT_K0X3(k0, a, b, c)            \
    ({                                       \
        ARM_DOT_K0X2(k0, a, b, c);           \
        ARM_DOT_K0(k0, (a), (b##2), (c.s2)); \
    })
#define ARM_DOT_K0X4(k0, a, b, c)            \
    ({                                       \
        ARM_DOT_K0X3(k0, a, b, c);           \
        ARM_DOT_K0(k0, (a), (b##3), (c.s3)); \
    })
#define ARM_DOT_K0X8(k0, a, b, c)            \
    ({                                       \
        ARM_DOT_K0X4(k0, a, b, c);           \
        ARM_DOT_K0(k0, (a), (b##4), (c.s4)); \
        ARM_DOT_K0(k0, (a), (b##5), (c.s5)); \
        ARM_DOT_K0(k0, (a), (b##6), (c.s6)); \
        ARM_DOT_K0(k0, (a), (b##7), (c.s7)); \
    })
#define ARM_DOT_K0X16(k0, a, b, c)           \
    ({                                       \
        ARM_DOT_K0X8(k0, a, b, c);           \
        ARM_DOT_K0(k0, (a), (b##8), (c.s8)); \
        ARM_DOT_K0(k0, (a), (b##9), (c.s9)); \
        ARM_DOT_K0(k0, (a), (b##A), (c.sA)); \
        ARM_DOT_K0(k0, (a), (b##B), (c.sB)); \
        ARM_DOT_K0(k0, (a), (b##C), (c.sC)); \
        ARM_DOT_K0(k0, (a), (b##D), (c.sD)); \
        ARM_DOT_K0(k0, (a), (b##E), (c.sE)); \
        ARM_DOT_K0(k0, (a), (b##F), (c.sF)); \
    })


#define ARM_MM_K0XN0X1(n0, k0, a, b, c)           \
    ({                                            \
        ARM_DOT_K0XN0(n0, k0, (a##0), b, (c##0)); \
    })
#define ARM_MM_K0XN0X2(n0, k0, a, b, c)           \
    ({                                            \
        ARM_MM_K0XN0X1(n0, k0, a, b, c);          \
        ARM_DOT_K0XN0(n0, k0, (a##1), b, (c##1)); \
    })
#define ARM_MM_K0XN0X3(n0, k0, a, b, c)           \
    ({                                            \
        ARM_MM_K0XN0X2(n0, k0, a, b, c);          \
        ARM_DOT_K0XN0(n0, k0, (a##2), b, (c##2)); \
    })
#define ARM_MM_K0XN0X4(n0, k0, a, b, c)           \
    ({                                            \
        ARM_MM_K0XN0X3(n0, k0, a, b, c);          \
        ARM_DOT_K0XN0(n0, k0, (a##3), b, (c##3)); \
    })
#define ARM_MM_K0XN0X5(n0, k0, a, b, c)           \
    ({                                            \
        ARM_MM_K0XN0X4(n0, k0, a, b, c);          \
        ARM_DOT_K0XN0(n0, k0, (a##4), b, (c##4)); \
    })
#define ARM_MM_K0XN0X6(n0, k0, a, b, c)           \
    ({                                            \
        ARM_MM_K0XN0X5(n0, k0, a, b, c);          \
        ARM_DOT_K0XN0(n0, k0, (a##5), b, (c##5)); \
    })
#define ARM_MM_K0XN0X7(n0, k0, a, b, c)           \
    ({                                            \
        ARM_MM_K0XN0X6(n0, k0, a, b, c);          \
        ARM_DOT_K0XN0(n0, k0, (a##6), b, (c##6)); \
    })
#define ARM_MM_K0XN0X8(n0, k0, a, b, c)           \
    ({                                            \
        ARM_MM_K0XN0X7(n0, k0, a, b, c);          \
        ARM_DOT_K0XN0(n0, k0, (a##7), b, (c##7)); \
    })

#define ARM_DOT_K0(k0, a, b, c) \
    ({                          \
        CONCAT(ARM_DOT, k0)     \
        ((a), (b), (c));        \
    })

#define ARM_DOT_K0XN0(n0, k0, a, b, c) \
    ({                                 \
        CONCAT(ARM_DOT_K0X, n0)        \
        (k0, (a), b, (c));             \
    })

#define ARM_MM_K0XN0XM0(m0, n0, k0, a, b, c) \
    ({                                       \
        CONCAT(ARM_MM_K0XN0X, m0)            \
        (n0, k0, a, b, c);                   \
    })


#define ARM_MUL_N0X1(VECTOR_ACC_TYPE, a, b, c)   \
    ({                                           \
        c += CONVERT(b##0, VECTOR_ACC_TYPE) * a; \
    })
#define ARM_MUL_N0X2(VECTOR_ACC_TYPE, a, b, c)        \
    ({                                                \
        c += CONVERT(b##0, VECTOR_ACC_TYPE) * a.s##0; \
        c += CONVERT(b##1, VECTOR_ACC_TYPE) * a.s##1; \
    })
#define ARM_MUL_N0X3(VECTOR_ACC_TYPE, a, b, c)        \
    ({                                                \
        ARM_MUL_N0X2(VECTOR_ACC_TYPE, a, b, c);       \
        c += CONVERT(b##2, VECTOR_ACC_TYPE) * a.s##2; \
    })
#define ARM_MUL_N0X4(VECTOR_ACC_TYPE, a, b, c)        \
    ({                                                \
        ARM_MUL_N0X3(VECTOR_ACC_TYPE, a, b, c);       \
        c += CONVERT(b##3, VECTOR_ACC_TYPE) * a.s##3; \
    })
#define ARM_MUL_N0X8(VECTOR_ACC_TYPE, a, b, c)        \
    ({                                                \
        ARM_MUL_N0X4(VECTOR_ACC_TYPE, a, b, c);       \
        c += CONVERT(b##4, VECTOR_ACC_TYPE) * a.s##4; \
        c += CONVERT(b##5, VECTOR_ACC_TYPE) * a.s##5; \
        c += CONVERT(b##6, VECTOR_ACC_TYPE) * a.s##6; \
        c += CONVERT(b##7, VECTOR_ACC_TYPE) * a.s##7; \
    })
#define ARM_MUL_N0X16(VECTOR_ACC_TYPE, a, b, c)       \
    ({                                                \
        ARM_MUL_N0X8(VECTOR_ACC_TYPE, a, b, c);       \
        c += CONVERT(b##8, VECTOR_ACC_TYPE) * a.s##8; \
        c += CONVERT(b##9, VECTOR_ACC_TYPE) * a.s##9; \
        c += CONVERT(b##A, VECTOR_ACC_TYPE) * a.s##A; \
        c += CONVERT(b##B, VECTOR_ACC_TYPE) * a.s##B; \
        c += CONVERT(b##C, VECTOR_ACC_TYPE) * a.s##C; \
        c += CONVERT(b##D, VECTOR_ACC_TYPE) * a.s##D; \
        c += CONVERT(b##E, VECTOR_ACC_TYPE) * a.s##E; \
        c += CONVERT(b##F, VECTOR_ACC_TYPE) * a.s##F; \
    })

#define ARM_MM_NATIVE_N0XK0X1(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##0), b, (c##0)); \
    })
#define ARM_MM_NATIVE_N0XK0X2(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MM_NATIVE_N0XK0X1(VECTOR_ACC_TYPE, k0, a, b, c);   \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##1), b, (c##1)); \
    })
#define ARM_MM_NATIVE_N0XK0X3(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MM_NATIVE_N0XK0X2(VECTOR_ACC_TYPE, k0, a, b, c);   \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##2), b, (c##2)); \
    })
#define ARM_MM_NATIVE_N0XK0X4(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MM_NATIVE_N0XK0X3(VECTOR_ACC_TYPE, k0, a, b, c);   \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##3), b, (c##3)); \
    })
#define ARM_MM_NATIVE_N0XK0X5(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MM_NATIVE_N0XK0X4(VECTOR_ACC_TYPE, k0, a, b, c);   \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##4), b, (c##4)); \
    })
#define ARM_MM_NATIVE_N0XK0X6(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MM_NATIVE_N0XK0X5(VECTOR_ACC_TYPE, k0, a, b, c);   \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##5), b, (c##5)); \
    })
#define ARM_MM_NATIVE_N0XK0X7(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MM_NATIVE_N0XK0X6(VECTOR_ACC_TYPE, k0, a, b, c);   \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##6), b, (c##6)); \
    })
#define ARM_MM_NATIVE_N0XK0X8(VECTOR_ACC_TYPE, k0, a, b, c)    \
    ({                                                         \
        ARM_MM_NATIVE_N0XK0X7(VECTOR_ACC_TYPE, k0, a, b, c);   \
        ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, (a##7), b, (c##7)); \
    })
#define ARM_MUL_N0XK0(VECTOR_ACC_TYPE, k0, a, b, c) \
    ({                                              \
        CONCAT(ARM_MUL_N0X, k0)                     \
        (VECTOR_ACC_TYPE, (a), b, (c));             \
    })
#define ARM_MM_NATIVE_N0XK0XM0(VECTOR_ACC_TYPE, m0, k0, a, b, c) \
    ({                                                           \
        CONCAT(ARM_MM_NATIVE_N0XK0X, m0)                         \
        (VECTOR_ACC_TYPE, k0, a, b, c);                          \
    })

#if defined(GEMMLOWP_MM_RESHAPED_LHS_NT_RHS_T)

__kernel void gemmlowp_mm_reshaped_lhs_nt_rhs_t(IMAGE_DECLARATION(lhs),
                                                IMAGE_DECLARATION(rhs),
                                                IMAGE_DECLARATION(dst),
                                                uint k,
                                                uint lhs_stride_z,
                                                uint rhs_stride_z,
                                                uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
                                                ,
                                                uint dst_cross_plane_pad
#endif
                                               )
{

#define LHS_BLOCK_SIZE ((K0) * (M0))

#if defined(LHS_INTERLEAVE)
#define LHS_OFFSET_X (K0)
#define LHS_STEP_X ((K0) * (V0))
#define LHS_STEP_LOOP (1)
#else
#define LHS_OFFSET_X (LHS_BLOCK_SIZE)
#define LHS_STEP_X (K0)
#define LHS_STEP_LOOP (V0)
#endif


#define RHS_BLOCK_SIZE ((K0) * (N0))


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (K0)
#define RHS_STEP_X ((K0) * (H0))
#define RHS_STEP_LOOP (1)
#else
#define RHS_OFFSET_X (RHS_BLOCK_SIZE)
#define RHS_STEP_X (K0)
#define RHS_STEP_LOOP (H0)
#endif

    uint x = get_global_id(0);
    uint y = get_global_id(1);
    uint z = get_global_id(2);

#if defined(DUMMY_WORK_ITEMS)
    if((x * N0 >= N) || (y * M0 >= M))
    {
        return;
    }
#endif


    __global DATA_TYPE *lhs_addr = (__global DATA_TYPE *)(lhs_ptr + lhs_offset_first_element_in_bytes + (y % V0) * (uint)LHS_OFFSET_X + (y / V0) * (uint)lhs_stride_y + (z * lhs_stride_z));


    __global DATA_TYPE *rhs_addr = (__global DATA_TYPE *)(rhs_ptr + rhs_offset_first_element_in_bytes + (x % H0) * (uint)RHS_OFFSET_X + (x / (uint)H0) * rhs_stride_y);

#if defined(MATRIX_B_DEPTH)

    rhs_addr += (z % MATRIX_B_DEPTH) * rhs_stride_z;
#else
    rhs_addr += z * rhs_stride_z;
#endif

    REPEAT_VAR_INIT_TO_CONST(8, uint, zlhs, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zrhs, 0);


    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(ACC_DATA_TYPE, N0), c, 0);

    for(int i = 0; i < k; i += K0)
    {

        LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_addr, 0, LHS_STEP_X, zlhs);


        LOAD_BLOCK(N0, K0, DATA_TYPE, b, rhs_addr, 0, RHS_STEP_X, zrhs);


        ARM_MM_K0XN0XM0(M0, N0, K0, a, b, c);


        lhs_addr += (M0 * LHS_STEP_X * LHS_STEP_LOOP);
        rhs_addr += (N0 * RHS_STEP_X * RHS_STEP_LOOP);
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(int)) + (y * (uint)M0 * dst_stride_y);

    REPEAT_VAR_INIT_TO_CONST(8, uint, zout, 0);

#if defined(REINTERPRET_OUTPUT_AS_3D)

    CALCULATE_Z_OFFSET(M0, uint, zout, y * M0, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y);



    dst_addr += z * dst_stride_z * DEPTH_GEMM3D;

#else


    dst_addr += z * dst_stride_z;

#endif


    const bool cond_y = ((get_global_id(1) + 1) * M0 >= M);
    const bool cond_x = ((get_global_id(0) + 1) * N0 >= N);


    REPEAT_VAR_INIT_CONVERT_SAT(M0, VEC_DATA_TYPE(int, N0), c, c_lp);
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, int, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#undef LHS_BLOCK_SIZE
#undef LHS_OFFSET_X
#undef LHS_STEP_X
#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
}
#endif

#if defined(GEMMLOWP_MM_RESHAPED_ONLY_RHS_T_FUSED_OUTPUT_STAGE_FIXEDPOINT) || defined(GEMMLOWP_MM_RESHAPED_ONLY_RHS_T)
#if defined(RESULT_OFFSET) && defined(RESULT_MULTIPLIER) && defined(RESULT_SHIFT)
#define FUSED_OUTPUT_STAGE_FIXED_POINT
#endif


#if defined(GEMMLOWP_MM_RESHAPED_ONLY_RHS_T_FUSED_OUTPUT_STAGE_FIXEDPOINT)
__kernel void gemmlowp_mm_reshaped_only_rhs_t_fused_output_stage_fixedpoint
#elif defined(GEMMLOWP_MM_RESHAPED_ONLY_RHS_T)
__kernel void gemmlowp_mm_reshaped_only_rhs_t
#endif
(IMAGE_DECLARATION(lhs),
 IMAGE_DECLARATION(rhs),
 IMAGE_DECLARATION(dst),
 uint lhs_stride_z,
 uint rhs_stride_z,
 uint dst_stride_z
#if defined(REINTERPRET_INPUT_AS_3D)
 ,
 uint lhs_cross_plane_pad
#endif
#if defined(REINTERPRET_OUTPUT_AS_3D)
 ,
 uint dst_cross_plane_pad
#endif
#if defined(A_OFFSET)
 ,
 IMAGE_DECLARATION(sum_col)
#endif
#if defined(B_OFFSET)
 ,
 IMAGE_DECLARATION(sum_row)
#endif
#if defined(ADD_BIAS)
 ,
 VECTOR_DECLARATION(biases)
#endif
#if defined(PER_CHANNEL_QUANTIZATION)
 ,
 VECTOR_DECLARATION(result_multipliers),
 VECTOR_DECLARATION(result_shifts)
#endif
)
{

#define FULL_LHS_HEIGHT (lhs_stride_z / lhs_stride_y)
#define FULL_DST_HEIGHT (dst_stride_z / dst_stride_y)


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (K0)
#define RHS_STEP_X (K0 * H0)
#else
#define RHS_OFFSET_X (K0 * N0)
#define RHS_STEP_X (K0)
#endif
#define RHS_STEP_LOOP (N0 * K0 * H0)

    uint x  = GET_SPATIAL_IDX(0, 1, 1);
    uint y  = GET_SPATIAL_IDX(1, M0, PARTIAL_STORE_M0);
    uint z  = GET_SPATIAL_IDX(2, 1, 1);
    int  xo = (x * N0);

#if defined(DUMMY_WORK_ITEMS)
    if((xo >= N) || (y >= M))
    {
        return;
    }
#endif


    uint lhs_y = y + z * FULL_LHS_HEIGHT;


    uint rhs_offset_x = (x % H0) * RHS_OFFSET_X;
    uint rhs_offset_y = (x / H0) * rhs_stride_y;

#if defined(MATRIX_B_DEPTH)

    rhs_offset_y += (z % MATRIX_B_DEPTH) * rhs_stride_z;
#else
    rhs_offset_y += z * rhs_stride_z;
#endif


    TILE(ACC_DATA_TYPE, M0, N0, c);
    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        c[i].v = 0;
    })

    int i = 0;
    for(; i <= (K - K0); i += K0)
    {
        TILE(DATA_TYPE, M0, K0, a);
        TILE(DATA_TYPE, N0, K0, b);


        T_LOAD(DATA_TYPE, M0, K0, BUFFER, lhs, i, lhs_y, 1, lhs_stride_y, a);


        LOOP_UNROLLING(int, _i, 0, 1, N0,
        {
            b[_i].v = VLOAD(K0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset_first_element_in_bytes + rhs_offset_x + rhs_offset_y + _i * RHS_STEP_X));
        })


        T_MMUL(DATA_TYPE, DATA_TYPE, ACC_DATA_TYPE, M0, N0, K0, NT, T, a, b, c);

        rhs_offset_x += RHS_STEP_LOOP;
    }

#if((K % K0) != 0)


    for(; i < K; ++i)
    {
        TILE(DATA_TYPE, M0, 1, a);
        TILE(DATA_TYPE, N0, 1, b);


        T_LOAD(DATA_TYPE, M0, 1, BUFFER, lhs, i, lhs_y, 1, lhs_stride_y, a);

        LOOP_UNROLLING(int, _i, 0, 1, N0,
        {
            b[_i].v = *(__global DATA_TYPE *)(rhs_ptr + rhs_offset_first_element_in_bytes + rhs_offset_x + rhs_offset_y + _i * RHS_STEP_X);
        })

        T_MMUL(DATA_TYPE, DATA_TYPE, ACC_DATA_TYPE, M0, N0, 1, NT, T, a, b, c);

        rhs_offset_x += 1;
    }
#endif

#if defined(FUSED_OUTPUT_STAGE_FIXED_POINT)

    TILE(int, M0, N0, c_int);
    TILE(int, M0, N0, offset_s32);
    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        offset_s32[i].v = (VEC_DATA_TYPE(int, N0))K_OFFSET;
    })

    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        c_int[i].v = CONVERT_SAT(c[i].v, VEC_DATA_TYPE(int, N0));
    })

#if defined(A_OFFSET)

#if defined(SUM_COL_HAS_BATCHES)
    int sum_col_y = z;
#else
    int sum_col_y = 0;
#endif
    TILE(int, 1, N0, a_offset_s32);

    T_LOAD(int, 1, N0, BUFFER, sum_col, xo, sum_col_y, 1, sum_col_stride_y, a_offset_s32);

    a_offset_s32[0].v *= A_OFFSET;

    T_ELTWISE_BROADCAST_ADD_X(int, M0, N0, offset_s32, a_offset_s32, offset_s32);
#endif

#if defined(B_OFFSET)




    TILE(int, M0, N0, b_offset_s32);

    T_LOAD(int, M0, 1, BUFFER, sum_row, y + z * (sum_row_stride_y / sizeof(int)), 0, 1, sum_row_stride_x, b_offset_s32);

    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        offset_s32[i].v += b_offset_s32[i].v *B_OFFSET;
    })

#endif

#if defined(ADD_BIAS)

    TILE(int, 1, N0, bias);

    T_LOAD(int, 1, N0, BUFFER, biases, xo, 0, 1, 0, bias);

    T_ELTWISE_BROADCAST_ADD_X(int, M0, N0, offset_s32, bias, offset_s32);
#endif

    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        c_int[i].v += offset_s32[i].v;
    })

    TILE(DATA_TYPE, M0, N0, c_lp);


#if defined(PER_CHANNEL_QUANTIZATION)
    TILE(int, 1, N0, res_mul);
    TILE(int, 1, N0, res_shift);

    T_LOAD(int, 1, N0, BUFFER, result_multipliers, xo, 0, 0, 0, res_mul);
    T_LOAD(int, 1, N0, BUFFER, result_shifts, xo, 0, 0, 0, res_shift);

    T_QUANTIZE8(int, DATA_TYPE, PER_CHANNEL, M0, N0, RESULT_OFFSET, RESULT_SHIFT, RESULT_MULTIPLIER, c_int, res_mul, res_shift, c_lp);
#else
    T_QUANTIZE8(int, DATA_TYPE, PER_TENSOR, M0, N0, RESULT_OFFSET, RESULT_SHIFT, RESULT_MULTIPLIER, c_int, 0, 0, c_lp);
#endif

#if defined(MIN_BOUND)
    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        c_lp[i].v = max(c_lp[i].v, (VEC_DATA_TYPE(DATA_TYPE, N0))MIN_BOUND);
    })
#endif
#if defined(MAX_BOUND)
    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        c_lp[i].v = min(c_lp[i].v, (VEC_DATA_TYPE(DATA_TYPE, N0))MAX_BOUND);
    })
#endif

#else
    TILE(int, M0, N0, c_lp);

    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
        c_lp[i].v = CONVERT_SAT(c[i].v, VEC_DATA_TYPE(int, N0));
    })
#endif

    TILE(uint, M0, 1, dst_indirect_y);

    LOOP_UNROLLING(int, i, 0, 1, M0,
    {
#if defined(REINTERPRET_OUTPUT_AS_3D)
        dst_indirect_y[i].v = (uint)min((int)((y + i) % HEIGHT_GEMM3D), (int)HEIGHT_GEMM3D - 1);
        dst_indirect_y[i].v += (uint)min((int)((y + i) / HEIGHT_GEMM3D), (int)DEPTH_GEMM3D - 1) * FULL_DST_HEIGHT;
        dst_indirect_y[i].v += z *FULL_DST_HEIGHT *DEPTH_GEMM3D;
#else
        dst_indirect_y[i].v = (uint)min((int)y + i, (int)M - 1) + z *FULL_DST_HEIGHT;
#endif
    })

    const bool cond_x = (xo > (N - N0)) & (PARTIAL_STORE_N0 != 0);

#if defined(FUSED_OUTPUT_STAGE_FIXED_POINT)
    T_STORE_INDIRECT_WIDTH_SELECT(DATA_TYPE, M0, N0, PARTIAL_STORE_N0, BUFFER, dst, xo, dst_stride_y, cond_x, c_lp, dst_indirect_y);
#else
    T_STORE_INDIRECT_WIDTH_SELECT(int, M0, N0, PARTIAL_STORE_N0, BUFFER, dst, xo, dst_stride_y, cond_x, c_lp, dst_indirect_y);
#endif

#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef RHS_STEP_LOOP
}
#endif

#if defined(GEMMLOWP_MM_NATIVE)


__kernel void gemmlowp_mm_native(IMAGE_DECLARATION(lhs),
                                 IMAGE_DECLARATION(rhs),
                                 IMAGE_DECLARATION(dst),
                                 uint lhs_stride_z,
                                 uint rhs_stride_z,
                                 uint dst_stride_z
#if defined(REINTERPRET_INPUT_AS_3D)
                                 ,
                                 uint lhs_cross_plane_pad
#endif
#if defined(REINTERPRET_OUTPUT_AS_3D)
                                 ,
                                 uint dst_cross_plane_pad
#endif
                                )
{
    uint x = get_global_id(0);
    uint y = get_global_id(1);
    uint z = get_global_id(2);

#if defined(DUMMY_WORK_ITEMS)
    if((x * N0 >= N) || (y * M0 >= M))
    {
        return;
    }
#endif


    uint lhs_offset = lhs_offset_first_element_in_bytes + COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) * (uint)lhs_stride_y;


    uint rhs_offset = rhs_offset_first_element_in_bytes + x * N0 * sizeof(DATA_TYPE);

#if defined(MATRIX_B_DEPTH)

    rhs_offset += (z % MATRIX_B_DEPTH) * rhs_stride_z;
#else
    rhs_offset += z * rhs_stride_z;
#endif

    REPEAT_VAR_INIT_TO_CONST(8, uint, zlhs, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zrhs, 0);

#if defined(REINTERPRET_INPUT_AS_3D)

    CALCULATE_Z_OFFSET(M0, uint, zlhs, COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0), HEIGHT_GEMM3D, DEPTH_GEMM3D, lhs_cross_plane_pad, lhs_stride_y);



    lhs_offset += z * lhs_stride_z * DEPTH_GEMM3D;

#else


    lhs_offset += z * lhs_stride_z;

#endif


    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(ACC_DATA_TYPE, N0), c, 0);

    int i = 0;

    for(; i <= (K - K0); i += K0)
    {

        LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs);


        LOAD_BLOCK(K0, N0, DATA_TYPE, b, rhs_ptr, rhs_offset, rhs_stride_y, zrhs);


#if(GPU_ARCH == GPU_ARCH_MIDGARD)
        ARM_MM_NATIVE_N0XK0XM0(VEC_DATA_TYPE(ACC_DATA_TYPE, N0), M0, K0, a, b, c);
#else

        TRANSPOSE_K0XN0(K0, N0, b_t, b, DATA_TYPE);

        ARM_MM_K0XN0XM0(M0, N0, K0, a, b_t, c);
#endif


        lhs_offset += K0;
        rhs_offset += K0 * rhs_stride_y;
    }


    for(; i < K; ++i)
    {

        LOAD_BLOCK(M0, 1, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs);


        LOAD_BLOCK(1, N0, DATA_TYPE, b, rhs_ptr, rhs_offset, rhs_stride_y, zrhs);


#if(GPU_ARCH == GPU_ARCH_MIDGARD)
        ARM_MM_NATIVE_N0XK0XM0(VEC_DATA_TYPE(ACC_DATA_TYPE, N0), M0, 1, a, b, c);
#else

        TRANSPOSE_K0XN0(1, N0, b_t, b, DATA_TYPE);

        ARM_MM_K0XN0XM0(M0, N0, 1, a, b_t, c);
#endif


        lhs_offset += 1;
        rhs_offset += rhs_stride_y;
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(int)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) * dst_stride_y);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zout, 0);

#if defined(REINTERPRET_OUTPUT_AS_3D)

    CALCULATE_Z_OFFSET(M0, uint, zout, COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0), HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y);



    dst_addr += z * dst_stride_z * DEPTH_GEMM3D;

#else


    dst_addr += z * dst_stride_z;

#endif
    const bool cond_y = y == 0;
    const bool cond_x = ((x + 1) * N0 >= N);


    REPEAT_VAR_INIT_CONVERT(M0, VEC_DATA_TYPE(int, N0), c, res);
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, int, res, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
}
#endif

#if defined(GEMMLOWP_MATRIX_A_REDUCTION)

__kernel void gemmlowp_matrix_a_reduction(TENSOR3D_DECLARATION(src),
                                          IMAGE_DECLARATION(dst))
{

    Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
    Image    dst = CONVERT_TO_IMAGE_STRUCT(dst);

    VEC_DATA_TYPE(ACC_DATA_TYPE, 4)
    sum_row_32            = (VEC_DATA_TYPE(ACC_DATA_TYPE, 4))0;
    ACC_DATA_TYPE sum_row = 0;

    __global const DATA_TYPE *matrix_a = (__global const DATA_TYPE *)(src.ptr + get_global_id(0) * src_stride_y + get_global_id(1) * src_stride_z);

    int i = 0;


    for(; i <= ((int)COLS_A - 16); i += 16)
    {
        const VEC_DATA_TYPE(DATA_TYPE, 16) a0 = vload16(0, matrix_a + i);

        sum_row_32 += CONVERT(a0.s0123, VEC_DATA_TYPE(ACC_DATA_TYPE, 4)) + CONVERT(a0.s4567, VEC_DATA_TYPE(ACC_DATA_TYPE, 4)) + CONVERT(a0.s89AB, VEC_DATA_TYPE(ACC_DATA_TYPE, 4)) + CONVERT(a0.sCDEF,
                      VEC_DATA_TYPE(ACC_DATA_TYPE, 4));
    }


    for(; i < COLS_A; ++i)
    {
        sum_row += (ACC_DATA_TYPE)matrix_a[i];
    }

    sum_row += sum_row_32.s0 + sum_row_32.s1 + sum_row_32.s2 + sum_row_32.s3;

#if defined(SCALAR)
    sum_row *= (int)SCALAR;
#endif
    *((__global int *)dst.ptr) = (int)sum_row;
}
#endif

#if defined(GEMMLOWP_MATRIX_A_REDUCTION_DOT8)

__kernel void gemmlowp_matrix_a_reduction_dot8(TENSOR3D_DECLARATION(src),
                                               IMAGE_DECLARATION(dst))
{

    Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
    Image    dst = CONVERT_TO_IMAGE_STRUCT(dst);

    ACC_DATA_TYPE sum_row = 0;

    __global const DATA_TYPE *matrix_a = (__global const DATA_TYPE *)(src.ptr + get_global_id(0) * src_stride_y + get_global_id(1) * src_stride_z);

    int i = 0;


    for(; i <= ((int)COLS_A - 32); i += 32)
    {
        VEC_DATA_TYPE(DATA_TYPE, 16)
        a0 = vload16(0, matrix_a + i);

        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.s0123, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);
        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.s4567, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);
        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.s89AB, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);
        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.sCDEF, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);

        a0 = vload16(1, matrix_a + i);

        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.s0123, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);
        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.s4567, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);
        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.s89AB, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);
        DOT_PRODUCT4_INTEGER8(DATA_TYPE, DATA_TYPE, DATA_TYPE, a0.sCDEF, (VEC_DATA_TYPE(DATA_TYPE, 4))(1), sum_row);
    }


    for(; i < COLS_A; ++i)
    {
        sum_row += (ACC_DATA_TYPE)matrix_a[i];
    }

#if defined(SCALAR)
    sum_row *= (int)SCALAR;
#endif
    *((__global int *)dst.ptr) = (int)sum_row;
}
#endif

#if defined(GEMMLOWP_MATRIX_B_REDUCTION)

__kernel void gemmlowp_matrix_b_reduction(TENSOR3D_DECLARATION(src),
                                          IMAGE_DECLARATION(dst))
{

    const uint x_offs = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    const uint y      = get_global_id(1);

    __global const DATA_TYPE *matrix_b = (__global const DATA_TYPE *)(src_ptr + src_offset_first_element_in_bytes + x_offs * sizeof(DATA_TYPE) + y * src_step_y + y * src_stride_z);
    __global uchar *dst_addr           = dst_ptr + dst_offset_first_element_in_bytes + x_offs * sizeof(int) + y * dst_stride_y;

    VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE)
    sum_col_32 = (VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE))0;

    int i = 0;

    for(; i <= ((int)ROWS_B - 4); i += 4)
    {
        const VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
        b0 = VLOAD(VEC_SIZE)(0, matrix_b + 0 * src_stride_y);
        const VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
        b1 = VLOAD(VEC_SIZE)(0, matrix_b + 1 * src_stride_y);
        const VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
        b2 = VLOAD(VEC_SIZE)(0, matrix_b + 2 * src_stride_y);
        const VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
        b3 = VLOAD(VEC_SIZE)(0, matrix_b + 3 * src_stride_y);

        sum_col_32 += CONVERT(b0, VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE)) + CONVERT(b1, VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE)) + CONVERT(b2, VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE)) + CONVERT(b3,
                      VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE));

        matrix_b += 4 * src_stride_y;
    }


    for(; i < (int)ROWS_B; ++i)
    {
        const VEC_DATA_TYPE(DATA_TYPE, VEC_SIZE)
        b0 = VLOAD(VEC_SIZE)(0, matrix_b);

        sum_col_32 += CONVERT(b0, VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE));

        matrix_b += src_stride_y;
    }

#if defined(SCALAR)
    sum_col_32 *= (VEC_DATA_TYPE(ACC_DATA_TYPE, VEC_SIZE))SCALAR;
#endif
    VEC_DATA_TYPE(int, VEC_SIZE)
    res0 = CONVERT(sum_col_32, VEC_DATA_TYPE(int, VEC_SIZE));

    STORE_VECTOR_SELECT(res, int, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif

#endif

#if defined(K_OFFSET) && defined(VEC_SIZE) && defined(VEC_SIZE_LEFTOVER)

#define VEC_INT VEC_DATA_TYPE(int, VEC_SIZE)


inline VEC_INT offset_contribution(
    int x,
    int y,
    int z
#if defined(A_OFFSET)
    ,
    IMAGE_DECLARATION(sum_col)
#endif
#if defined(B_OFFSET)
    ,
    IMAGE_DECLARATION(sum_row)
#endif
#if defined(ADD_BIAS)
    ,
    VECTOR_DECLARATION(biases)
#endif
)
{
    VEC_INT a_offset_s32 = (VEC_INT)0;
    VEC_INT b_offset_s32 = (VEC_INT)0;

    int batch_id = z;
#if defined(DEPTH_INPUT3D)
    batch_id /= (int)DEPTH_INPUT3D;
#endif

#if defined(A_OFFSET)

    __global uchar *sum_col_addr = sum_col_ptr + sum_col_offset_first_element_in_bytes + x * sizeof(int);


#if defined(SUM_COL_HAS_BATCHES)
    a_offset_s32 = VLOAD(VEC_SIZE)(0, (__global int *)(sum_col_addr + batch_id * sum_col_stride_y));
#else
    a_offset_s32 = VLOAD(VEC_SIZE)(0, (__global int *)sum_col_addr);
#endif

    a_offset_s32 *= (VEC_INT)A_OFFSET;
#endif

#if defined(B_OFFSET)

    __global uchar *sum_row_addr = sum_row_ptr + sum_row_offset_first_element_in_bytes + y * sizeof(int);


#if defined(HEIGHT_INPUT3D) && defined(DEPTH_INPUT3D)
    b_offset_s32 = (VEC_INT) * (((__global int *)(sum_row_addr + batch_id * sum_row_stride_y)) + (z % (int)DEPTH_INPUT3D) * (int)HEIGHT_INPUT3D);
#else
    b_offset_s32 = (VEC_INT) * (((__global int *)(sum_row_addr + batch_id * sum_row_stride_y)));
#endif
    b_offset_s32 *= (VEC_INT)B_OFFSET;
#endif

#if defined(ADD_BIAS)

    __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * sizeof(int);

    VEC_INT biases_values = VLOAD(VEC_SIZE)(0, (__global int *)bias_addr);
    b_offset_s32 += (VEC_INT)biases_values;
#endif

    return (VEC_INT)K_OFFSET + a_offset_s32 + b_offset_s32;
}

#if defined(GEMMLOWP_OFFSET_CONTRIBUTION)

__kernel void gemmlowp_offset_contribution(TENSOR3D_DECLARATION(mm_result)
#if defined(A_OFFSET)
                                           ,
                                           IMAGE_DECLARATION(sum_col)
#endif
#if defined(B_OFFSET)
                                           ,
                                           IMAGE_DECLARATION(sum_row)
#endif
#if defined(ADD_BIAS)
                                           ,
                                           VECTOR_DECLARATION(biases)
#endif
                                          )
{
    const int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    const int y = get_global_id(1);
    const int z = get_global_id(2);


    VEC_INT offset_term_s32 = offset_contribution(
                                  x, y, z
#if defined(A_OFFSET)
                                  ,
                                  sum_col_ptr,
                                  sum_col_stride_x,
                                  sum_col_step_x,
                                  sum_col_stride_y,
                                  sum_col_step_y,
                                  sum_col_offset_first_element_in_bytes
#endif
#if defined(B_OFFSET)
                                  ,
                                  sum_row_ptr,
                                  sum_row_stride_x,
                                  sum_row_step_x,
                                  sum_row_stride_y,
                                  sum_row_step_y,
                                  sum_row_offset_first_element_in_bytes
#endif
#if defined(ADD_BIAS)
                                  ,
                                  biases_ptr,
                                  biases_stride_x,
                                  biases_step_x,
                                  biases_offset_first_element_in_bytes
#endif
                              );

    __global uchar *mm_result_addr = mm_result_ptr + mm_result_offset_first_element_in_bytes + x * sizeof(int) + y * mm_result_stride_y + z * mm_result_stride_z;

    VEC_INT in_s32_0 = VLOAD(VEC_SIZE)(0, (__global int *)mm_result_addr);


    in_s32_0 += offset_term_s32;


    STORE_VECTOR_SELECT(in_s32_, int, mm_result_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif

#if defined(GEMMLOWP_OFFSET_CONTRIBUTION_QUANTIZE_DOWN)

__kernel void gemmlowp_offset_contribution_quantize_down(TENSOR3D_DECLARATION(mm_result)
#if defined(A_OFFSET)
                                                         ,
                                                         IMAGE_DECLARATION(sum_col)
#endif
#if defined(B_OFFSET)
                                                         ,
                                                         IMAGE_DECLARATION(sum_row)
#endif
                                                         ,
#if defined(ADD_BIAS)
                                                         VECTOR_DECLARATION(biases),
#endif
                                                         TENSOR3D_DECLARATION(dst)
#if defined(PER_CHANNEL_QUANTIZATION)
                                                         ,
                                                         VECTOR_DECLARATION(result_multipliers),
                                                         VECTOR_DECLARATION(result_shifts)
#endif
                                                        )
{
    const int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    const int y = get_global_id(1);
    const int z = get_global_id(2);

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;


    VEC_INT offset_term_s32 = offset_contribution(
                                  x, y, z
#if defined(A_OFFSET)
                                  ,
                                  sum_col_ptr,
                                  sum_col_stride_x,
                                  sum_col_step_x,
                                  sum_col_stride_y,
                                  sum_col_step_y,
                                  sum_col_offset_first_element_in_bytes
#endif
#if defined(B_OFFSET)
                                  ,
                                  sum_row_ptr,
                                  sum_row_stride_x,
                                  sum_row_step_x,
                                  sum_row_stride_y,
                                  sum_row_step_y,
                                  sum_row_offset_first_element_in_bytes
#endif
#if defined(ADD_BIAS)
                                  ,
                                  biases_ptr,
                                  biases_stride_x,
                                  biases_step_x,
                                  biases_offset_first_element_in_bytes
#endif
                              );

    __global uchar *mm_result_addr = mm_result_ptr + mm_result_offset_first_element_in_bytes + x * sizeof(int) + y * mm_result_stride_y + z * mm_result_stride_z;

    VEC_INT in_s32 = VLOAD(VEC_SIZE)(0, (__global int *)mm_result_addr);


    in_s32 += offset_term_s32;




    in_s32 += (VEC_INT)RESULT_OFFSET;


#if defined(PER_CHANNEL_QUANTIZATION)
    __global uchar *result_multipliers_addr   = result_multipliers_ptr + result_multipliers_offset_first_element_in_bytes + x * sizeof(int);
    __global uchar *result_shifts_addr        = result_shifts_ptr + result_shifts_offset_first_element_in_bytes + x * sizeof(int);
    VEC_INT         result_multipliers_values = VLOAD(VEC_SIZE)(0, (__global int *)result_multipliers_addr);
    VEC_INT         result_shifts_values      = VLOAD(VEC_SIZE)(0, (__global int *)result_shifts_addr);

    in_s32 *= result_multipliers_values;
    in_s32 >>= result_shifts_values;
#else
    in_s32 *= RESULT_MULTIPLIER;

    in_s32 >>= RESULT_SHIFT;
#endif

    VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE)
    res0 = CONVERT_SAT(in_s32, VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE));

#if defined(MIN_BOUND)
    res0 = max(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MIN_BOUND);
#endif
#if defined(MAX_BOUND)
    res0 = min(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MAX_BOUND);
#endif


    STORE_VECTOR_SELECT(res, OUTPUT_DATA_TYPE, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif

#if defined(GEMMLOWP_OFFSET_CONTRIBUTION_QUANTIZE_DOWN_FIXEDPOINT)

__kernel void gemmlowp_offset_contribution_quantize_down_fixedpoint(TENSOR3D_DECLARATION(mm_result)
#if defined(A_OFFSET)
                                                                    ,
                                                                    IMAGE_DECLARATION(sum_col)
#endif
#if defined(B_OFFSET)
                                                                    ,
                                                                    IMAGE_DECLARATION(sum_row)
#endif
                                                                    ,
#if defined(ADD_BIAS)
                                                                    VECTOR_DECLARATION(biases),
#endif
                                                                    TENSOR3D_DECLARATION(dst)
#if defined(PER_CHANNEL_QUANTIZATION)
                                                                    ,
                                                                    VECTOR_DECLARATION(result_multipliers),
                                                                    VECTOR_DECLARATION(result_shifts)
#endif
                                                                   )
{
    const int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    const int y = get_global_id(1);
    const int z = get_global_id(2);


    VEC_INT offset_term_s32 = offset_contribution(
                                  x, y, z
#if defined(A_OFFSET)
                                  ,
                                  sum_col_ptr,
                                  sum_col_stride_x,
                                  sum_col_step_x,
                                  sum_col_stride_y,
                                  sum_col_step_y,
                                  sum_col_offset_first_element_in_bytes
#endif
#if defined(B_OFFSET)
                                  ,
                                  sum_row_ptr,
                                  sum_row_stride_x,
                                  sum_row_step_x,
                                  sum_row_stride_y,
                                  sum_row_step_y,
                                  sum_row_offset_first_element_in_bytes
#endif
#if defined(ADD_BIAS)
                                  ,
                                  biases_ptr,
                                  biases_stride_x,
                                  biases_step_x,
                                  biases_offset_first_element_in_bytes
#endif
                              );

    __global uchar *mm_result_addr = mm_result_ptr + mm_result_offset_first_element_in_bytes + x * sizeof(int) + y * mm_result_stride_y + z * mm_result_stride_z;

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;

    VEC_INT in_s32 = VLOAD(VEC_SIZE)(0, (__global int *)mm_result_addr);


    in_s32 += offset_term_s32;




#if defined(PER_CHANNEL_QUANTIZATION)
    __global uchar *result_multipliers_addr   = result_multipliers_ptr + result_multipliers_offset_first_element_in_bytes + x * sizeof(int);
    __global uchar *result_shifts_addr        = result_shifts_ptr + result_shifts_offset_first_element_in_bytes + x * sizeof(int);
    VEC_INT         result_multipliers_values = VLOAD(VEC_SIZE)(0, (__global int *)result_multipliers_addr);
    VEC_INT         result_shifts_values      = VLOAD(VEC_SIZE)(0, (__global int *)result_shifts_addr);

    VEC_INT in_s32_shift_lt0 = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(in_s32, result_multipliers_values, result_shifts_values, VEC_SIZE);
    VEC_INT in_s32_shift_gt0 = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(in_s32, result_multipliers_values, result_shifts_values, VEC_SIZE);
    in_s32                   = select(in_s32_shift_lt0, in_s32_shift_gt0, result_shifts_values >= 0);
#else

#if RESULT_SHIFT < 0
    in_s32 = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(in_s32, RESULT_MULTIPLIER, RESULT_SHIFT, VEC_SIZE);
#else
    in_s32 = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(in_s32, RESULT_MULTIPLIER, RESULT_SHIFT, VEC_SIZE);
#endif

#endif


    in_s32 += (VEC_INT)RESULT_OFFSET;

    VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE)
    res0 = CONVERT_SAT(in_s32, VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE));

#if defined(MIN_BOUND)
    res0 = max(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MIN_BOUND);
#endif
#if defined(MAX_BOUND)
    res0 = min(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MAX_BOUND);
#endif


    STORE_VECTOR_SELECT(res, OUTPUT_DATA_TYPE, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif

#undef VEC_INT

#endif

#if defined(GEMMLOWP_OUTPUT_STAGE_QUANTIZE_DOWN)

__kernel void gemmlowp_output_stage_quantize_down(TENSOR3D_DECLARATION(src),
#if defined(ADD_BIAS)
                                                  VECTOR_DECLARATION(biases),
#endif
                                                  TENSOR3D_DECLARATION(dst))
{

    int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    int y = get_global_id(1);
    int z = get_global_id(2);

    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(int) + y * src_stride_y + z * src_stride_z;

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;

    VEC_DATA_TYPE(int, VEC_SIZE)
    input_values = VLOAD(VEC_SIZE)(0, (__global int *)src_addr);

#if defined(ADD_BIAS)

    __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * sizeof(int);

    VEC_DATA_TYPE(int, VEC_SIZE)
    biases_values = VLOAD(VEC_SIZE)(0, (__global int *)bias_addr);
    input_values += biases_values;
#endif


    input_values += (VEC_DATA_TYPE(int, VEC_SIZE))RESULT_OFFSET;


    input_values *= RESULT_MULT_INT;

#if RESULT_SHIFT < 0
    input_values >>= -RESULT_SHIFT;
#else
    input_values >>= RESULT_SHIFT;
#endif

    VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE)
    res0 = CONVERT_SAT(input_values, VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE));

#if defined(MIN_BOUND)
    res0 = max(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MIN_BOUND);
#endif
#if defined(MAX_BOUND)
    res0 = min(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MAX_BOUND);
#endif


    STORE_VECTOR_SELECT(res, OUTPUT_DATA_TYPE, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif

#if defined(GEMMLOWP_OUTPUT_STAGE_QUANTIZE_DOWN_FIXEDPOINT)

__kernel void gemmlowp_output_stage_quantize_down_fixedpoint(TENSOR3D_DECLARATION(src),
#if defined(ADD_BIAS)
                                                             VECTOR_DECLARATION(biases),
#endif
                                                             TENSOR3D_DECLARATION(dst))
{

    int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    int y = get_global_id(1);
    int z = get_global_id(2);

    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(int) + y * src_stride_y + z * src_stride_z;

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;

    VEC_DATA_TYPE(int, VEC_SIZE)
    input_values = VLOAD(VEC_SIZE)(0, (__global int *)src_addr);

#if defined(ADD_BIAS)

    __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * sizeof(int);

    VEC_DATA_TYPE(int, VEC_SIZE)
    biases_values = VLOAD(VEC_SIZE)(0, (__global int *)bias_addr);
    input_values += biases_values;
#endif


#if RESULT_SHIFT < 0
    input_values = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(input_values, RESULT_FIXEDPOINT_MULTIPLIER, RESULT_SHIFT, VEC_SIZE);
#else
    input_values = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(input_values, RESULT_FIXEDPOINT_MULTIPLIER, RESULT_SHIFT, VEC_SIZE);
#endif


    input_values += (VEC_DATA_TYPE(int, VEC_SIZE))RESULT_OFFSET_AFTER_SHIFT;

    VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE)
    res0 = CONVERT_SAT(input_values, VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE));

#if defined(MIN_BOUND)
    res0 = max(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MIN_BOUND);
#endif
#if defined(MAX_BOUND)
    res0 = min(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MAX_BOUND);
#endif


    STORE_VECTOR_SELECT(res, OUTPUT_DATA_TYPE, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif

#if defined(GEMMLOWP_OUTPUT_STAGE_QUANTIZE_DOWN_FIXEDPOINT_QSYMM16)

__kernel void gemmlowp_output_stage_quantize_down_fixedpoint_qsymm16(TENSOR3D_DECLARATION(src),
#if defined(ADD_BIAS)
                                                                     VECTOR_DECLARATION(biases),
#endif
                                                                     TENSOR3D_DECLARATION(dst))
{

    int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    int y = get_global_id(1);
    int z = get_global_id(2);

    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(int) + y * src_stride_y + z * src_stride_z;

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x * sizeof(short) + y * dst_stride_y + z * dst_stride_z;

    VEC_DATA_TYPE(int, VEC_SIZE)
    input_values = VLOAD(VEC_SIZE)(0, (__global int *)src_addr);

#if defined(ADD_BIAS)

    __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * sizeof(int);

    VEC_DATA_TYPE(int, VEC_SIZE)
    biases_values = VLOAD(VEC_SIZE)(0, (__global int *)bias_addr);
    input_values += biases_values;
#endif


#if RESULT_SHIFT < 0
    input_values = ASYMM_MULT_BY_QUANT_MULTIPLIER_GREATER_THAN_ONE(input_values, RESULT_FIXEDPOINT_MULTIPLIER, RESULT_SHIFT, VEC_SIZE);
#else
    input_values = ASYMM_MULT_BY_QUANT_MULTIPLIER_LESS_THAN_ONE(input_values, RESULT_FIXEDPOINT_MULTIPLIER, RESULT_SHIFT, VEC_SIZE);
#endif

    VEC_DATA_TYPE(short, VEC_SIZE)
    res0 = CONVERT_SAT(input_values, VEC_DATA_TYPE(short, VEC_SIZE));

#if defined(MIN_BOUND)
    res0 = max(res0, (VEC_DATA_TYPE(short, VEC_SIZE))MIN_BOUND);
#endif
#if defined(MAX_BOUND)
    res0 = min(res0, (VEC_DATA_TYPE(short, VEC_SIZE))MAX_BOUND);
#endif


    STORE_VECTOR_SELECT(res, short, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif

#if defined(GEMMLOWP_OUTPUT_STAGE_QUANTIZE_DOWN_FLOAT)

__kernel void gemmlowp_output_stage_quantize_down_float(TENSOR3D_DECLARATION(src),
#if defined(ADD_BIAS)
                                                        VECTOR_DECLARATION(biases),
#endif
#if defined(DST_HEIGHT)
                                                        TENSOR4D_DECLARATION(dst))
#else
                                                        TENSOR3D_DECLARATION(dst))
#endif
{

    int x = max((int)(get_global_id(0) * VEC_SIZE - (VEC_SIZE - VEC_SIZE_LEFTOVER) % VEC_SIZE), 0);
    int y = get_global_id(1);
    int z = get_global_id(2);

    __global uchar *src_addr = src_ptr + src_offset_first_element_in_bytes + x * sizeof(int) + y * src_stride_y + z * src_stride_z;

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + x + y * dst_stride_y + z * dst_stride_z;

    VEC_DATA_TYPE(int, VEC_SIZE)
    input_values = VLOAD(VEC_SIZE)(0, (__global int *)src_addr);

#if defined(ADD_BIAS)

    __global uchar *bias_addr = biases_ptr + biases_offset_first_element_in_bytes + x * sizeof(int);

    VEC_DATA_TYPE(int, VEC_SIZE)
    biases_values = VLOAD(VEC_SIZE)(0, (__global int *)bias_addr);
    input_values += (VEC_DATA_TYPE(int, VEC_SIZE))biases_values;
#endif


    VEC_DATA_TYPE(float, VEC_SIZE)
    input_values_f = CONVERT(input_values, VEC_DATA_TYPE(float, VEC_SIZE));
    input_values_f = round(input_values_f * (float)REAL_MULTIPLIER + (float)OUTPUT_OFFSET);

    VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE)
    res0 = CONVERT_SAT(input_values_f, VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE));

#if defined(MIN_BOUND)
    res0 = max(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MIN_BOUND);
#endif
#if defined(MAX_BOUND)
    res0 = min(res0, (VEC_DATA_TYPE(OUTPUT_DATA_TYPE, VEC_SIZE))MAX_BOUND);
#endif


    STORE_VECTOR_SELECT(res, OUTPUT_DATA_TYPE, dst_addr, VEC_SIZE, VEC_SIZE_LEFTOVER, VEC_SIZE_LEFTOVER != 0 && get_global_id(0) == 0)
}
#endif  )"