xref: /aosp_15_r20/external/ComputeLibrary/cl_kernels/common/gemm.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_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

#if defined(M0) && defined(N0) && defined(K0) && defined(H0) && defined(DATA_TYPE)

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

#define ARM_DOT1(a, b, c) \
    ({                    \
        c = fma(a, b, c); \
    })
#define ARM_DOT2(a, b, c)       \
    ({                          \
        c = fma(a.s0, b.s0, c); \
        c = fma(a.s1, b.s1, c); \
    })
#define ARM_DOT3(a, b, c)           \
    ({                              \
        ARM_DOT2(a, b, c);          \
        c = fma((a.s2), (b.s2), c); \
    })
#define ARM_DOT4(a, b, c)           \
    ({                              \
        ARM_DOT3(a, b, c);          \
        c = fma((a.s3), (b.s3), 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); \
    })

#if N0 == 2
#define ARM_DOT_K0XN0(k0, a, b, c) \
    ({                             \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##0), (c.s0));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##1), (c.s1));     \
    })
#elif N0 == 3
#define ARM_DOT_K0XN0(k0, a, b, c) \
    ({                             \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##0), (c.s0));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##1), (c.s1));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##2), (c.s2));     \
    })
#elif N0 == 4
#define ARM_DOT_K0XN0(k0, a, b, c) \
    ({                             \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##0), (c.s0));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##1), (c.s1));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##2), (c.s2));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##3), (c.s3));     \
    })
#elif N0 == 8
#define ARM_DOT_K0XN0(k0, a, b, c) \
    ({                             \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##0), (c.s0));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##1), (c.s1));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##2), (c.s2));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##3), (c.s3));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##4), (c.s4));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##5), (c.s5));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##6), (c.s6));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##7), (c.s7));     \
    })
#elif N0 == 16
#define ARM_DOT_K0XN0(k0, a, b, c) \
    ({                             \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##0), (c.s0));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##1), (c.s1));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##2), (c.s2));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##3), (c.s3));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##4), (c.s4));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##5), (c.s5));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##6), (c.s6));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##7), (c.s7));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##8), (c.s8));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##9), (c.s9));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##A), (c.sA));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##B), (c.sB));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##C), (c.sC));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##D), (c.sD));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##E), (c.sE));     \
        CONCAT(ARM_DOT, k0)        \
        ((a), (b##F), (c.sF));     \
    })
#else
#error "N0 value not supported"
#endif

#if defined(GEMM_MM_RESHAPED_ONLY_RHS_T)

__kernel void gemm_mm_reshaped_only_rhs_t(IMAGE_DECLARATION(lhs),
                                          IMAGE_DECLARATION(rhs),
#if defined(BETA)
                                          IMAGE_DECLARATION(bias),
#endif
                                          IMAGE_DECLARATION(dst),
                                          uint lhs_stride_z,
                                          uint rhs_stride_z,
#if defined(BETA)
                                          uint bias_stride_z,
#endif
                                          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
                                          ,
                                          const int M,
                                          const int N,
                                          const int K)
{

#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);

    const bool cond_y = y == 0;
    const bool cond_x = ((x + 1) * N0 >= N);

#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 % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (x / (uint)H0) * rhs_stride_y;

#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, zero, 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(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(N0, K0, DATA_TYPE, b, rhs_ptr, rhs_offset, RHS_STEP_X * sizeof(DATA_TYPE), zero);


        ARM_DOT_K0XN0(K0, a0, b, c0);
#if M0 > 1
        ARM_DOT_K0XN0(K0, a1, b, c1);
#endif
#if M0 > 2
        ARM_DOT_K0XN0(K0, a2, b, c2);
#endif
#if M0 > 3
        ARM_DOT_K0XN0(K0, a3, b, c3);
#endif
#if M0 > 4
        ARM_DOT_K0XN0(K0, a4, b, c4);
#endif
#if M0 > 5
        ARM_DOT_K0XN0(K0, a5, b, c5);
#endif
#if M0 > 6
        ARM_DOT_K0XN0(K0, a6, b, c6);
#endif
#if M0 > 7
        ARM_DOT_K0XN0(K0, a7, b, c7);
#endif

        lhs_offset += K0 * sizeof(DATA_TYPE);
        rhs_offset += (N0 * RHS_STEP_X * RHS_STEP_LOOP) * sizeof(DATA_TYPE);
    }


    for(; i < K; ++i)
    {

        LOAD_BLOCK(M0, 1, DATA_TYPE, a, lhs_ptr, lhs_offset, lhs_stride_y, zlhs);


        LOAD_BLOCK(N0, 1, DATA_TYPE, b, rhs_ptr, rhs_offset, RHS_STEP_X * sizeof(DATA_TYPE), zero);


        ARM_DOT_K0XN0(1, a0, b, c0);
#if M0 > 1
        ARM_DOT_K0XN0(1, a1, b, c1);
#endif
#if M0 > 2
        ARM_DOT_K0XN0(1, a2, b, c2);
#endif
#if M0 > 3
        ARM_DOT_K0XN0(1, a3, b, c3);
#endif
#if M0 > 4
        ARM_DOT_K0XN0(1, a4, b, c4);
#endif
#if M0 > 5
        ARM_DOT_K0XN0(1, a5, b, c5);
#endif
#if M0 > 6
        ARM_DOT_K0XN0(1, a6, b, c6);
#endif
#if M0 > 7
        ARM_DOT_K0XN0(1, a7, b, c7);
#endif

        lhs_offset += sizeof(DATA_TYPE);
        rhs_offset += sizeof(DATA_TYPE);
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_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, 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


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK_BROADCAST(M0, c, bias0);

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) * bias_stride_y) + z * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK(M0, c, bias);

#endif
#endif

#if defined(ACTIVATION_TYPE)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif


    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef RHS_STEP_LOOP
}
#endif

#if defined(OPENCL_IMAGE_SUPPORT) && defined(GEMM_MM_RESHAPED_ONLY_RHS_T_TEXTURE)

__kernel void gemm_mm_reshaped_only_rhs_t_texture(IMAGE_DECLARATION(lhs),
                                                  __read_only image2d_t rhs_img,
#if defined(BETA)
                                                  IMAGE_DECLARATION(bias),
#endif
                                                  IMAGE_DECLARATION(dst),
                                                  uint lhs_stride_z,
                                                  uint rhs_stride_z,
#if defined(BETA)
                                                  uint bias_stride_z,
#endif
                                                  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
                                                  ,
                                                  const int M,
                                                  const int N,
                                                  const int K)
{

#define PIXEL_UNIT CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(K0)

    const uint LEFTOVER_K = K % K0;


#define RHS_BLOCK_SIZE (PIXEL_UNIT * (N0))


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (PIXEL_UNIT)
#define RHS_STEP_X (PIXEL_UNIT * (H0))
#define RHS_STEP_LOOP (1)
#else
#define RHS_OFFSET_X (RHS_BLOCK_SIZE)
#define RHS_STEP_X PIXEL_UNIT
#define RHS_STEP_LOOP (H0)
#endif

    uint x = get_global_id(0);
    uint y = get_global_id(1);
    uint z = get_global_id(2);

    const bool cond_y = y == 0;
    const bool cond_x = ((x + 1) * N0 >= N);

#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;

#if defined(MATRIX_B_DEPTH)

    const uint z_rhs = (get_global_id(2) % MATRIX_B_DEPTH);
#else
    const uint z_rhs = get_global_id(2);
#endif


    uint       x_rhs = (get_global_id(0) % H0) * (uint)RHS_OFFSET_X;
    const uint y_rhs = (get_global_id(0) / (uint)H0) + z_rhs * RHS_HEIGHT;

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zlhs, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 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(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);


        REPEAT_VAR_INIT_TO_CONST(N0, VEC_DATA_TYPE(DATA_TYPE, K0), b, 0);
        LOAD_TEXTURE2D(N0, PIXEL_UNIT, DATA_TYPE, b, rhs_img, x_rhs, y_rhs, RHS_STEP_X, 0);


        ARM_DOT_K0XN0(K0, a0, b, c0);
#if M0 > 1
        ARM_DOT_K0XN0(K0, a1, b, c1);
#endif
#if M0 > 2
        ARM_DOT_K0XN0(K0, a2, b, c2);
#endif
#if M0 > 3
        ARM_DOT_K0XN0(K0, a3, b, c3);
#endif
#if M0 > 4
        ARM_DOT_K0XN0(K0, a4, b, c4);
#endif
#if M0 > 5
        ARM_DOT_K0XN0(K0, a5, b, c5);
#endif
#if M0 > 6
        ARM_DOT_K0XN0(K0, a6, b, c6);
#endif
#if M0 > 7
        ARM_DOT_K0XN0(K0, a7, b, c7);
#endif

        lhs_offset += K0 * sizeof(DATA_TYPE);
        x_rhs += N0 * RHS_STEP_X * RHS_STEP_LOOP;
    }

    if(LEFTOVER_K != 0)
    {




        union UNION_VEC_TYPE
        {
            DATA_TYPE s[K0];
            VEC_DATA_TYPE(DATA_TYPE, K0)
            v;
        };

        union UNION_VEC_TYPE a0 = {.v = 0 };
#if M0 > 1
        union UNION_VEC_TYPE a1 = {.v = 0 };
#endif
#if M0 > 2
        union UNION_VEC_TYPE a2 = {.v = 0 };
#endif
#if M0 > 3
        union UNION_VEC_TYPE a3 = {.v = 0 };
#endif
#if M0 > 4
        union UNION_VEC_TYPE a4 = {.v = 0 };
#endif
#if M0 > 5
        union UNION_VEC_TYPE a5 = {.v = 0 };
#endif
#if M0 > 6
        union UNION_VEC_TYPE a6 = {.v = 0 };
#endif
#if M0 > 7
        union UNION_VEC_TYPE a7 = {.v = 0 };
#endif

        REPEAT_VAR_INIT_TO_CONST(N0, VEC_DATA_TYPE(DATA_TYPE, K0), b, 0);


        LOAD_TEXTURE2D(N0, PIXEL_UNIT, DATA_TYPE, b, rhs_img, x_rhs, y_rhs, RHS_STEP_X, 0);


        for(int k = 0; k < LEFTOVER_K; ++k)
        {
            a0.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 0 * lhs_stride_y + zlhs0);
#if M0 > 1
            a1.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 1 * lhs_stride_y + zlhs1);
#endif
#if M0 > 2
            a2.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 2 * lhs_stride_y + zlhs2);
#endif
#if M0 > 3
            a3.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 3 * lhs_stride_y + zlhs3);
#endif
#if M0 > 4
            a4.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 4 * lhs_stride_y + zlhs4);
#endif
#if M0 > 5
            a5.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 5 * lhs_stride_y + zlhs5);
#endif
#if M0 > 6
            a6.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 6 * lhs_stride_y + zlhs6);
#endif
#if M0 > 7
            a7.s[k] = *(__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zlhs7);
#endif

            lhs_offset += sizeof(DATA_TYPE);
        }


        ARM_DOT_K0XN0(K0, a0.v, b, c0);
#if M0 > 1
        ARM_DOT_K0XN0(K0, a1.v, b, c1);
#endif
#if M0 > 2
        ARM_DOT_K0XN0(K0, a2.v, b, c2);
#endif
#if M0 > 3
        ARM_DOT_K0XN0(K0, a3.v, b, c3);
#endif
#if M0 > 4
        ARM_DOT_K0XN0(K0, a4.v, b, c4);
#endif
#if M0 > 5
        ARM_DOT_K0XN0(K0, a5.v, b, c5);
#endif
#if M0 > 6
        ARM_DOT_K0XN0(K0, a6.v, b, c6);
#endif
#if M0 > 7
        ARM_DOT_K0XN0(K0, a7.v, b, c7);
#endif
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (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


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK_BROADCAST(M0, c, bias0);

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) * bias_stride_y) + z * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK(M0, c, bias);

#endif
#endif

#if defined(ACTIVATION_TYPE)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif


    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef RHS_STEP_LOOP
#undef PIXEL_UNIT
}
#endif

#define VFMA(a, b, c)     \
    ({                    \
        c = fma(a, b, c); \
    })

#if M0 == 1
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
    })
#elif M0 == 2
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
    })
#elif M0 == 3
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
    })
#elif M0 == 4
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
    })
#elif M0 == 5
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
    })
#elif M0 == 6
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \
    })
#elif M0 == 7
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \
    })
#elif M0 == 8
#define VFMA_M0xN0(i, a, b, c)                                        \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##7).s##i), b, (c##7)); \
    })
#else
#error "M0 not supported"
#endif

#if defined(GEMM_MM_RESHAPED_ONLY_RHS_NT)

__kernel void gemm_mm_reshaped_only_rhs_nt(IMAGE_DECLARATION(lhs),
                                           IMAGE_DECLARATION(rhs),
#if defined(BETA)
                                           IMAGE_DECLARATION(bias),
#endif
                                           IMAGE_DECLARATION(dst),
                                           uint lhs_stride_z,
                                           uint rhs_stride_z,
#if defined(BETA)
                                           uint bias_stride_z,
#endif
                                           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
                                           ,
                                           const int M,
                                           const int N,
                                           const int K)
{

#define RHS_BLOCK_SIZE ((K0) * (N0))


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (N0)
#define RHS_STEP_X ((N0) * (H0))
#define RHS_STEP_LOOP (1)
#else
#define RHS_OFFSET_X (RHS_BLOCK_SIZE)
#define RHS_STEP_X (N0)
#define RHS_STEP_LOOP (H0)
#endif

    uint x = get_global_id(0);
    uint y = get_global_id(1);
    uint z = get_global_id(2);

    const bool cond_y = y == 0;
    const bool cond_x = ((x + 1) * N0 >= N);

#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 % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (x / (uint)H0) * rhs_stride_y;

#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, zin, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0);

#if defined(REINTERPRET_INPUT_AS_3D)


    CALCULATE_Z_OFFSET(M0, uint, zin, 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(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, zin);

        VEC_DATA_TYPE(DATA_TYPE, N0)
        b0;

        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(0, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 1 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(1, a, b0, c);
#if K0 > 2
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 2 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(2, a, b0, c);
#endif
#if K0 > 3
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 3 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(3, a, b0, c);
#endif
#if K0 > 4
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 4 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(4, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 5 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(5, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 6 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(6, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 7 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(7, a, b0, c);
#endif
#if K0 > 8
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 8 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(8, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 9 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(9, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 10 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(A, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 11 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(B, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 12 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(C, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 13 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(D, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 14 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(E, a, b0, c);
        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 15 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(F, a, b0, c);
#endif

        lhs_offset += K0 * sizeof(DATA_TYPE);
        rhs_offset += K0 * RHS_STEP_X * RHS_STEP_LOOP * sizeof(DATA_TYPE);
    }


    for(; i < K; ++i)
    {

        VEC_DATA_TYPE(DATA_TYPE, 2)
        a0 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 0 * lhs_stride_y + zin0));
#if M0 > 1
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a1 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 1 * lhs_stride_y + zin1));
#endif
#if M0 > 2
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a2 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 2 * lhs_stride_y + zin2));
#endif
#if M0 > 3
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a3 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 3 * lhs_stride_y + zin3));
#endif
#if M0 > 4
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a4 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 4 * lhs_stride_y + zin4));
#endif
#if M0 > 5
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a5 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 5 * lhs_stride_y + zin5));
#endif
#if M0 > 6
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a6 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 6 * lhs_stride_y + zin6));
#endif
#if M0 > 7
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a7 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zin7));
#endif

        VEC_DATA_TYPE(DATA_TYPE, N0)
        b0;

        b0 = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0 * RHS_STEP_X * sizeof(DATA_TYPE)));
        VFMA_M0xN0(0, a, b0, c);

        lhs_offset += sizeof(DATA_TYPE);
        rhs_offset += RHS_STEP_X * sizeof(DATA_TYPE);
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_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, 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


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK_BROADCAST(M0, c, bias0);

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) * bias_stride_y) + z * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK(M0, c, bias);

#endif
#endif

#if defined(ACTIVATION_TYPE)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif


    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef RHS_STEP_LOOP
}
#endif

#if defined(OPENCL_IMAGE_SUPPORT) && defined(GEMM_MM_RESHAPED_ONLY_RHS_NT_TEXTURE)

__kernel void gemm_mm_reshaped_only_rhs_nt_texture(IMAGE_DECLARATION(lhs),
                                                   __read_only image2d_t rhs_img,
#if defined(BETA)
                                                   IMAGE_DECLARATION(bias),
#endif
                                                   IMAGE_DECLARATION(dst),
                                                   uint lhs_stride_z,
                                                   uint rhs_stride_z,
#if defined(BETA)
                                                   uint bias_stride_z,
#endif
                                                   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
                                                   ,
                                                   const int M,
                                                   const int N,
                                                   const int K)
{

#define PIXEL_UNIT CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(N0)


#define RHS_BLOCK_SIZE ((K0) * (PIXEL_UNIT))


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (PIXEL_UNIT)
#define RHS_STEP_X ((PIXEL_UNIT) * (H0))
#define RHS_STEP_LOOP 1
#else
#define RHS_OFFSET_X (RHS_BLOCK_SIZE)
#define RHS_STEP_X (PIXEL_UNIT)
#define RHS_STEP_LOOP (H0)
#endif

    uint x = get_global_id(0);
    uint y = get_global_id(1);
    uint z = get_global_id(2);

    const bool cond_y = y == 0;
    const bool cond_x = ((x + 1) * N0 >= N);

#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;

#if defined(MATRIX_B_DEPTH)

    const uint z_rhs = (z % MATRIX_B_DEPTH);
#else
    const uint z_rhs = z;
#endif


    uint       x_rhs = (x % H0) * (uint)RHS_OFFSET_X;
    const uint y_rhs = (x / (uint)H0) + z_rhs * RHS_HEIGHT;

    REPEAT_VAR_INIT_TO_CONST(8, uint, zin, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0);

#if defined(REINTERPRET_INPUT_AS_3D)


    CALCULATE_Z_OFFSET(M0, uint, zin, 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(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, zin);

        VEC_DATA_TYPE(DATA_TYPE, N0)
        b0;

        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 0 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(0, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 1 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(1, a, b0, c);
#if K0 > 2
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 2 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(2, a, b0, c);
#endif
#if K0 > 3
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 3 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(3, a, b0, c);
#endif
#if K0 > 4
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 4 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(4, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 5 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(5, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 6 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(6, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 7 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(7, a, b0, c);
#endif
#if K0 > 8
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 8 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(8, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 9 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(9, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 10 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(A, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 11 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(B, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 12 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(C, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 13 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(D, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 14 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(E, a, b0, c);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 15 * RHS_STEP_X), (y_rhs));
        VFMA_M0xN0(F, a, b0, c);
#endif

        lhs_offset += K0 * sizeof(DATA_TYPE);
        x_rhs += K0 * RHS_STEP_X * RHS_STEP_LOOP;
    }


    for(; i < K; ++i)
    {

        VEC_DATA_TYPE(DATA_TYPE, 2)
        a0 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 0 * lhs_stride_y + zin0));
#if M0 > 1
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a1 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 1 * lhs_stride_y + zin1));
#endif
#if M0 > 2
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a2 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 2 * lhs_stride_y + zin2));
#endif
#if M0 > 3
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a3 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 3 * lhs_stride_y + zin3));
#endif
#if M0 > 4
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a4 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 4 * lhs_stride_y + zin4));
#endif
#if M0 > 5
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a5 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 5 * lhs_stride_y + zin5));
#endif
#if M0 > 6
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a6 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 6 * lhs_stride_y + zin6));
#endif
#if M0 > 7
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a7 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zin7));
#endif

        VEC_DATA_TYPE(DATA_TYPE, N0)
        b0;
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 0 * RHS_STEP_X), (y_rhs));

        VFMA_M0xN0(0, a, b0, c);

        lhs_offset += sizeof(DATA_TYPE);
        x_rhs += RHS_STEP_X;
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_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, 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


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK_BROADCAST(M0, c, bias0);

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) * bias_stride_y) + z * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK(M0, c, bias);

#endif
#endif

#if defined(ACTIVATION_TYPE)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif


    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef RHS_STEP_LOOP
}
#endif
#endif

#if defined(M0) && defined(N0) && defined(K0) && defined(V0) && defined(H0) && defined(DATA_TYPE) && defined(DATA_TYPE_ACCUMULATOR)

#if defined(MIXED_PRECISION)
#if K0 == 2
#define ARM_DOT_K0(a, b, c) \
    ({                      \
        c += a.s0 * b.s0;   \
        c += a.s1 * b.s1;   \
    })
#elif K0 == 3
#define ARM_DOT_K0(a, b, c) \
    ({                      \
        c += a.s0 * b.s0;   \
        c += a.s1 * b.s1;   \
        c += a.s2 * b.s2;   \
    })
#elif K0 == 4
#define ARM_DOT_K0(a, b, c) \
    ({                      \
        c += a.s0 * b.s0;   \
        c += a.s1 * b.s1;   \
        c += a.s2 * b.s2;   \
        c += a.s3 * b.s3;   \
    })
#elif K0 == 8
#define ARM_DOT_K0(a, b, c) \
    ({                      \
        c += a.s0 * b.s0;   \
        c += a.s1 * b.s1;   \
        c += a.s2 * b.s2;   \
        c += a.s3 * b.s3;   \
        c += a.s4 * b.s4;   \
        c += a.s5 * b.s5;   \
        c += a.s6 * b.s6;   \
        c += a.s7 * b.s7;   \
    })
#elif K0 == 16
#define ARM_DOT_K0(a, b, c) \
    ({                      \
        c += a.s0 * b.s0;   \
        c += a.s1 * b.s1;   \
        c += a.s2 * b.s2;   \
        c += a.s3 * b.s3;   \
        c += a.s4 * b.s4;   \
        c += a.s5 * b.s5;   \
        c += a.s6 * b.s6;   \
        c += a.s7 * b.s7;   \
        c += a.s8 * b.s8;   \
        c += a.s9 * b.s9;   \
        c += a.sA * b.sA;   \
        c += a.sB * b.sB;   \
        c += a.sC * b.sC;   \
        c += a.sD * b.sD;   \
        c += a.sE * b.sE;   \
        c += a.sF * b.sF;   \
    })
#else
#error "K0 value not supported"
#endif
#else
#if K0 == 2
#define ARM_DOT_K0(a, b, c)     \
    ({                          \
        c = fma(a.s0, b.s0, c); \
        c = fma(a.s1, b.s1, c); \
    })
#elif K0 == 3
#define ARM_DOT_K0(a, b, c)     \
    ({                          \
        c = fma(a.s0, b.s0, c); \
        c = fma(a.s1, b.s1, c); \
        c = fma(a.s2, b.s2, c); \
    })
#elif K0 == 4
#define ARM_DOT_K0(a, b, c)     \
    ({                          \
        c = fma(a.s0, b.s0, c); \
        c = fma(a.s1, b.s1, c); \
        c = fma(a.s2, b.s2, c); \
        c = fma(a.s3, b.s3, c); \
    })
#elif K0 == 8
#define ARM_DOT_K0(a, b, c)     \
    ({                          \
        c = fma(a.s0, b.s0, c); \
        c = fma(a.s1, b.s1, c); \
        c = fma(a.s2, b.s2, c); \
        c = fma(a.s3, b.s3, c); \
        c = fma(a.s4, b.s4, c); \
        c = fma(a.s5, b.s5, c); \
        c = fma(a.s6, b.s6, c); \
        c = fma(a.s7, b.s7, c); \
    })
#elif K0 == 16
#define ARM_DOT_K0(a, b, c)     \
    ({                          \
        c = fma(a.s0, b.s0, c); \
        c = fma(a.s1, b.s1, c); \
        c = fma(a.s2, b.s2, c); \
        c = fma(a.s3, b.s3, c); \
        c = fma(a.s4, b.s4, c); \
        c = fma(a.s5, b.s5, c); \
        c = fma(a.s6, b.s6, c); \
        c = fma(a.s7, b.s7, c); \
        c = fma(a.s8, b.s8, c); \
        c = fma(a.s9, b.s9, c); \
        c = fma(a.sA, b.sA, c); \
        c = fma(a.sB, b.sB, c); \
        c = fma(a.sC, b.sC, c); \
        c = fma(a.sD, b.sD, c); \
        c = fma(a.sE, b.sE, c); \
        c = fma(a.sF, b.sF, c); \
    })
#else
#error "K0 value not supported"
#endif
#endif

#if defined(ARM_DOT_K0XN0)
#undef ARM_DOT_K0XN0
#endif

#if N0 == 2
#define ARM_DOT_K0XN0(a, b, c)           \
    ({                                   \
        ARM_DOT_K0((a), (b##0), (c.s0)); \
        ARM_DOT_K0((a), (b##1), (c.s1)); \
    })
#elif N0 == 3
#define ARM_DOT_K0XN0(a, b, c)           \
    ({                                   \
        ARM_DOT_K0((a), (b##0), (c.s0)); \
        ARM_DOT_K0((a), (b##1), (c.s1)); \
        ARM_DOT_K0((a), (b##2), (c.s2)); \
    })
#elif N0 == 4
#define ARM_DOT_K0XN0(a, b, c)           \
    ({                                   \
        ARM_DOT_K0((a), (b##0), (c.s0)); \
        ARM_DOT_K0((a), (b##1), (c.s1)); \
        ARM_DOT_K0((a), (b##2), (c.s2)); \
        ARM_DOT_K0((a), (b##3), (c.s3)); \
    })
#elif N0 == 8
#define ARM_DOT_K0XN0(a, b, c)           \
    ({                                   \
        ARM_DOT_K0((a), (b##0), (c.s0)); \
        ARM_DOT_K0((a), (b##1), (c.s1)); \
        ARM_DOT_K0((a), (b##2), (c.s2)); \
        ARM_DOT_K0((a), (b##3), (c.s3)); \
        ARM_DOT_K0((a), (b##4), (c.s4)); \
        ARM_DOT_K0((a), (b##5), (c.s5)); \
        ARM_DOT_K0((a), (b##6), (c.s6)); \
        ARM_DOT_K0((a), (b##7), (c.s7)); \
    })
#elif N0 == 16
#define ARM_DOT_K0XN0(a, b, c)           \
    ({                                   \
        ARM_DOT_K0((a), (b##0), (c.s0)); \
        ARM_DOT_K0((a), (b##1), (c.s1)); \
        ARM_DOT_K0((a), (b##2), (c.s2)); \
        ARM_DOT_K0((a), (b##3), (c.s3)); \
        ARM_DOT_K0((a), (b##4), (c.s4)); \
        ARM_DOT_K0((a), (b##5), (c.s5)); \
        ARM_DOT_K0((a), (b##6), (c.s6)); \
        ARM_DOT_K0((a), (b##7), (c.s7)); \
        ARM_DOT_K0((a), (b##8), (c.s8)); \
        ARM_DOT_K0((a), (b##9), (c.s9)); \
        ARM_DOT_K0((a), (b##A), (c.sA)); \
        ARM_DOT_K0((a), (b##B), (c.sB)); \
        ARM_DOT_K0((a), (b##C), (c.sC)); \
        ARM_DOT_K0((a), (b##D), (c.sD)); \
        ARM_DOT_K0((a), (b##E), (c.sE)); \
        ARM_DOT_K0((a), (b##F), (c.sF)); \
    })
#else
#error "N0 value not supported"
#endif

#if defined(GEMM_MM_RESHAPED_LHS_NT_RHS_T)

__kernel void gemm_mm_reshaped_lhs_nt_rhs_t(IMAGE_DECLARATION(lhs),
                                            IMAGE_DECLARATION(rhs),
#if defined(BETA)
                                            IMAGE_DECLARATION(bias),
#endif
                                            IMAGE_DECLARATION(dst),
                                            uint lhs_stride_z,
                                            uint rhs_stride_z,
#if defined(BETA)
                                            uint bias_stride_z,
#endif
                                            uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
                                            ,
                                            uint dst_cross_plane_pad
#endif
                                            ,
                                            const int M,
                                            const int N,
                                            const int K)
{

#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

#if defined(DUMMY_WORK_ITEMS)
    if((get_global_id(0) * N0 >= N) || (get_global_id(1) * M0 >= M))
    {
        return;
    }
#endif


    __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (get_global_id(1) % V0) * (uint)LHS_OFFSET_X * sizeof(DATA_TYPE) + (get_global_id(1) / V0) * (uint)lhs_stride_y +
                               (get_global_id(2) * lhs_stride_z);


    __global uchar *rhs_addr = rhs_ptr + rhs_offset_first_element_in_bytes + (get_global_id(0) % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (get_global_id(0) / (uint)H0) * rhs_stride_y;

#if defined(MATRIX_B_DEPTH)

    rhs_addr += (get_global_id(2) % MATRIX_B_DEPTH) * rhs_stride_z;
#else
    rhs_addr += get_global_id(2) * rhs_stride_z;
#endif


    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0), c, 0);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zlhs, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0);

    for(int i = 0; i < K; i += K0)
    {










        LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_addr, 0, LHS_STEP_X * sizeof(DATA_TYPE), zlhs);


        LOAD_BLOCK(N0, K0, DATA_TYPE, b, rhs_addr, 0, RHS_STEP_X * sizeof(DATA_TYPE), zero);


        ARM_DOT_K0XN0(a0, b, c0);
#if M0 > 1
        ARM_DOT_K0XN0(a1, b, c1);
#endif
#if M0 > 2
        ARM_DOT_K0XN0(a2, b, c2);
#endif
#if M0 > 3
        ARM_DOT_K0XN0(a3, b, c3);
#endif
#if M0 > 4
        ARM_DOT_K0XN0(a4, b, c4);
#endif
#if M0 > 5
        ARM_DOT_K0XN0(a5, b, c5);
#endif
#if M0 > 6
        ARM_DOT_K0XN0(a6, b, c6);
#endif
#if M0 > 7
        ARM_DOT_K0XN0(a7, b, c7);
#endif

        lhs_addr += (M0 * LHS_STEP_X * LHS_STEP_LOOP) * sizeof(DATA_TYPE);
        rhs_addr += (N0 * RHS_STEP_X * RHS_STEP_LOOP) * sizeof(DATA_TYPE);
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * dst_stride_y);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zout, 0);

    const bool cond_y = ((get_global_id(1) + 1) * M0 >= M);
    const bool cond_x = ((get_global_id(0) + 1) * N0 >= N);

#if defined(REINTERPRET_OUTPUT_AS_3D)


    CALCULATE_Z_OFFSET(M0, uint, zout, get_global_id(1) * (uint)M0, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y);


    dst_addr += get_global_id(2) * dst_stride_z * DEPTH_GEMM3D;

#else


    dst_addr += get_global_id(2) * dst_stride_z;

#endif


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(1, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK_BROADCAST(M0, c, bias_hp0);
#else
    ADD_BLOCK_BROADCAST(M0, c, bias0);
#endif

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * bias_stride_y) + get_global_id(
                                    2) * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK(M0, c, bias_hp);
#else
    ADD_BLOCK(M0, c, bias);
#endif

#endif
#endif

#if defined(ACTIVATION_TYPE)
#if defined(MIXED_PRECISION)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE_ACCUMULATOR, N0, c, A_VAL, B_VAL);
#else
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#else
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#endif

#undef LHS_BLOCK_SIZE
#undef LHS_OFFSET_X
#undef LHS_STEP_X
#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef LHS_STEP_LOOP
#undef RHS_STEP_LOOP
}
#endif

#if defined(OPENCL_IMAGE_SUPPORT) && defined(GEMM_MM_RESHAPED_LHS_NT_RHS_T_TEXTURE)

__kernel void gemm_mm_reshaped_lhs_nt_rhs_t_texture(IMAGE_DECLARATION(lhs),
                                                    __read_only image2d_t rhs_img,
#if defined(BETA)
                                                    IMAGE_DECLARATION(bias),
#endif
                                                    IMAGE_DECLARATION(dst),
                                                    uint lhs_stride_z,
                                                    uint rhs_stride_z,
#if defined(BETA)
                                                    uint bias_stride_z,
#endif
                                                    uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
                                                    ,
                                                    uint dst_cross_plane_pad
#endif
                                                    ,
                                                    const int M,
                                                    const int N,
                                                    const int K)
{

#define PIXEL_UNIT CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(K0)


#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 (PIXEL_UNIT * (N0))


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (PIXEL_UNIT)
#define RHS_STEP_X (PIXEL_UNIT * (H0))
#define RHS_STEP_LOOP (1)
#else
#define RHS_OFFSET_X (RHS_BLOCK_SIZE)
#define RHS_STEP_X PIXEL_UNIT
#define RHS_STEP_LOOP (H0)
#endif

#if defined(DUMMY_WORK_ITEMS)
    if((get_global_id(0) * N0 >= N) || (get_global_id(1) * M0 >= M))
    {
        return;
    }
#endif


    __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (get_global_id(1) % V0) * (uint)LHS_OFFSET_X * sizeof(DATA_TYPE) + (get_global_id(1) / V0) * (uint)lhs_stride_y +
                               (get_global_id(2) * lhs_stride_z);

#if defined(MATRIX_B_DEPTH)

    const uint z_rhs = (get_global_id(2) % MATRIX_B_DEPTH);
#else
    const uint z_rhs = get_global_id(2);
#endif


    uint       x_rhs = (get_global_id(0) % H0) * (uint)RHS_OFFSET_X;
    const uint y_rhs = (get_global_id(0) / (uint)H0) + z_rhs * RHS_HEIGHT;


    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0), c, 0);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zlhs, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 0);

    for(int i = 0; i < K; i += K0)
    {

        LOAD_BLOCK(M0, K0, DATA_TYPE, a, lhs_addr, 0, LHS_STEP_X * sizeof(DATA_TYPE), zlhs);


        REPEAT_VAR_INIT_TO_CONST(N0, VEC_DATA_TYPE(DATA_TYPE, K0), b, 0);
        LOAD_TEXTURE2D(N0, PIXEL_UNIT, DATA_TYPE, b, rhs_img, x_rhs, y_rhs, RHS_STEP_X, 0);


        ARM_DOT_K0XN0(a0, b, c0);
#if M0 > 1
        ARM_DOT_K0XN0(a1, b, c1);
#endif
#if M0 > 2
        ARM_DOT_K0XN0(a2, b, c2);
#endif
#if M0 > 3
        ARM_DOT_K0XN0(a3, b, c3);
#endif
#if M0 > 4
        ARM_DOT_K0XN0(a4, b, c4);
#endif
#if M0 > 5
        ARM_DOT_K0XN0(a5, b, c5);
#endif
#if M0 > 6
        ARM_DOT_K0XN0(a6, b, c6);
#endif
#if M0 > 7
        ARM_DOT_K0XN0(a7, b, c7);
#endif

        lhs_addr += (M0 * LHS_STEP_X * LHS_STEP_LOOP) * sizeof(DATA_TYPE);

        x_rhs += N0 * RHS_STEP_X * RHS_STEP_LOOP;
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * dst_stride_y);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zout, 0);

    const bool cond_y = ((get_global_id(1) + 1) * M0 >= M);
    const bool cond_x = ((get_global_id(0) + 1) * N0 >= N);

#if defined(REINTERPRET_OUTPUT_AS_3D)


    CALCULATE_Z_OFFSET(M0, uint, zout, get_global_id(1) * (uint)M0, HEIGHT_GEMM3D, DEPTH_GEMM3D, dst_cross_plane_pad, dst_stride_y);


    dst_addr += get_global_id(2) * dst_stride_z * DEPTH_GEMM3D;

#else


    dst_addr += get_global_id(2) * dst_stride_z;

#endif


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(1, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK_BROADCAST(M0, c, bias_hp0);
#else
    ADD_BLOCK_BROADCAST(M0, c, bias0);
#endif

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * bias_stride_y) + get_global_id(
                                    2) * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK(M0, c, bias_hp);
#else
    ADD_BLOCK(M0, c, bias);
#endif

#endif
#endif

#if defined(ACTIVATION_TYPE)
#if defined(MIXED_PRECISION)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE_ACCUMULATOR, N0, c, A_VAL, B_VAL);
#else
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#else
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#endif

#undef LHS_BLOCK_SIZE
#undef LHS_OFFSET_X
#undef LHS_STEP_X
#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef PIXEL_UNIT
#undef LHS_STEP_LOOP
#undef RHS_STEP_LOOP
}
#endif

#if defined(LHS_TRANSPOSE)

#define VTYPE(TYPE, SIZE) VEC_DATA_TYPE(TYPE, SIZE)

#if defined(MIXED_PRECISION)

#if(GPU_ARCH == GPU_ARCH_MIDGARD)
#define ARM_VFMA(N0, a, b, c) c += (CONVERT(a, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0))) * (CONVERT(b, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0)));
#else
#define ARM_VFMA(N0, a, b, c) c = fma((CONVERT(a, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0))), (CONVERT(b, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0))), (c));
#endif

#else

#if(GPU_ARCH == GPU_ARCH_MIDGARD)
#define ARM_VFMA(N0, a, b, c) c += (a) * (b);
#else
#define ARM_VFMA(N0, a, b, c) c = fma((a), (b), (c));
#endif

#endif

#define ARM_VVM_T_NT_1xN0x1(N0, TYPE, a, b, C)         \
    ({                                                 \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a), b, (C##0)); \
    })
#define ARM_VVM_T_NT_2xN0x1(N0, TYPE, a, b, C)            \
    ({                                                    \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s0), b, (C##0)); \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s1), b, (C##1)); \
    })
#define ARM_VVM_T_NT_3xN0x1(N0, TYPE, a, b, C)            \
    ({                                                    \
        ARM_VVM_T_NT_2xN0x1(N0, TYPE, a, b, C);           \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s2), b, (C##2)); \
    })
#define ARM_VVM_T_NT_4xN0x1(N0, TYPE, a, b, C)            \
    ({                                                    \
        ARM_VVM_T_NT_3xN0x1(N0, TYPE, a, b, C);           \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s3), b, (C##3)); \
    })
#define ARM_VVM_T_NT_8xN0x1(N0, TYPE, a, b, C)            \
    ({                                                    \
        ARM_VVM_T_NT_4xN0x1(N0, TYPE, a, b, C);           \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s4), b, (C##4)); \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s5), b, (C##5)); \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s6), b, (C##6)); \
        ARM_VFMA(N0, (VTYPE(TYPE, N0))(a.s7), b, (C##7)); \
    })







#define ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, a, b, C) ARM_VVM_T_NT_##M0##xN0x1(N0, TYPE, a, b, C)

#define ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, A, B, C)             \
    ({                                                         \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##0), (B##0), C); \
    })
#define ARM_MM_T_NT_M0xN0x2(M0, N0, TYPE, A, B, C)             \
    ({                                                         \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, A, B, C);            \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##1), (B##1), C); \
    })
#define ARM_MM_T_NT_M0xN0x3(M0, N0, TYPE, A, B, C)             \
    ({                                                         \
        ARM_MM_T_NT_M0xN0x2(M0, N0, TYPE, A, B, C);            \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##2), (B##2), C); \
    })
#define ARM_MM_T_NT_M0xN0x4(M0, N0, TYPE, A, B, C)             \
    ({                                                         \
        ARM_MM_T_NT_M0xN0x3(M0, N0, TYPE, A, B, C);            \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##3), (B##3), C); \
    })
#define ARM_MM_T_NT_M0xN0x8(M0, N0, TYPE, A, B, C)             \
    ({                                                         \
        ARM_MM_T_NT_M0xN0x4(M0, N0, TYPE, A, B, C);            \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##4), (B##4), C); \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##5), (B##5), C); \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##6), (B##6), C); \
        ARM_VVM_T_NT_M0xN0x1(M0, N0, TYPE, (A##7), (B##7), C); \
    })
#define ARM_MM_T_NT_M0xN0x16(M0, N0, TYPE, A, B, C)           \
    ({                                                        \
        ARM_MM_T_NT_M0xN0x8(M0, N0, TYPE, A, B, C);           \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##8), (B##8), C); \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##9), (B##9), C); \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##A), (B##A), C); \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##B), (B##B), C); \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##C), (B##C), C); \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##D), (B##D), C); \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##E), (B##E), C); \
        ARM_MM_T_NT_M0xN0x1(M0, N0, TYPE, (A##F), (B##F), C); \
    })









#define ARM_MM_T_NT(M0, N0, K0, TYPE, A, B, C) \
    CONCAT(ARM_MM_T_NT_M0xN0x, K0)             \
    (M0, N0, TYPE, A, B, C)

#if defined(GEMM_MM_RESHAPED_LHS_T_RHS_NT)

__kernel void gemm_mm_reshaped_lhs_t_rhs_nt(IMAGE_DECLARATION(lhs),
                                            IMAGE_DECLARATION(rhs),
#if defined(BETA)
                                            IMAGE_DECLARATION(bias),
#endif
                                            IMAGE_DECLARATION(dst),
                                            uint lhs_stride_z,
                                            uint rhs_stride_z,
#if defined(BETA)
                                            uint bias_stride_z,
#endif
                                            uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
                                            ,
                                            uint dst_cross_plane_pad
#endif
                                            ,
                                            const int M,
                                            const int N,
                                            const int K)
{

#define LHS_BLOCK_SIZE ((K0) * (M0))

#if defined(LHS_INTERLEAVE)
#define LHS_OFFSET_X (M0)
#define LHS_STEP_X ((M0) * (V0))
#define LHS_STEP_LOOP (1)
#else
#define LHS_OFFSET_X (LHS_BLOCK_SIZE)
#define LHS_STEP_X (M0)
#define LHS_STEP_LOOP (V0)
#endif


#define RHS_BLOCK_SIZE ((K0) * (N0))


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (N0)
#define RHS_STEP_X ((N0) * (H0))
#else
#define RHS_OFFSET_X (RHS_BLOCK_SIZE)
#define RHS_STEP_X (N0)
#endif

    const uint x = get_global_id(0);
    const uint y = get_global_id(1);
    const uint z = get_global_id(2);

    const bool cond_y = ((get_global_id(1) + 1) * M0 >= M);
    const bool cond_x = ((get_global_id(0) + 1) * N0 >= N);

#if defined(DUMMY_WORK_ITEMS)
    if((x * N0 >= N) || (y * M0 >= M))
    {
        return;
    }
#endif


    __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (y % V0) * (uint)LHS_OFFSET_X * sizeof(DATA_TYPE) + (y / V0) * (uint)lhs_stride_y + (z * lhs_stride_z);


    __global uchar *rhs_addr = rhs_ptr + rhs_offset_first_element_in_bytes + (x % H0) * (uint)RHS_OFFSET_X * sizeof(DATA_TYPE) + (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(M0, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0), c, 0);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zero, 0);

    __global DATA_TYPE *lhs = (__global DATA_TYPE *)(lhs_addr);
    __global DATA_TYPE *rhs = (__global DATA_TYPE *)(rhs_addr);

    for(int i = 0; i < K; i += K0)
    {
        VEC_DATA_TYPE(DATA_TYPE, M0)
        a0;
        VEC_DATA_TYPE(DATA_TYPE, N0)
        b0;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

#if K0 > 1
        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;
#endif

#if K0 > 2
        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;
#endif

#if K0 > 3
        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;
#endif

#if K0 > 4
        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;
#endif

#if K0 > 8
        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = VLOAD(N0)(0, rhs);

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
        rhs += RHS_STEP_X;
#endif

#ifndef LHS_INTERLEAVE
        lhs += (M0 * K0 * (V0 - 1));
#endif

#ifndef RHS_INTERLEAVE
        rhs += (N0 * K0 * (H0 - 1));
#endif
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)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, y * (uint)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


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(1, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK_BROADCAST(M0, c, bias_hp0);
#else
    ADD_BLOCK_BROADCAST(M0, c, bias0);
#endif

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE)) + (get_global_id(1) * (uint)M0 * bias_stride_y) + get_global_id(
                                    2) * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif

#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK(M0, c, bias_hp);
#else
    ADD_BLOCK(M0, c, bias);
#endif

#endif
#endif

#if defined(ACTIVATION_TYPE)
#if defined(MIXED_PRECISION)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE_ACCUMULATOR, N0, c, A_VAL, B_VAL);
#else
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#else
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#endif

#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(OPENCL_IMAGE_SUPPORT) && defined(GEMM_MM_RESHAPED_LHS_T_RHS_NT_TEXTURE)

__kernel void gemm_mm_reshaped_lhs_t_rhs_nt_texture(IMAGE_DECLARATION(lhs),
                                                    __read_only image2d_t rhs_img,
#if defined(BETA)
                                                    IMAGE_DECLARATION(bias),
#endif
                                                    IMAGE_DECLARATION(dst),
                                                    uint lhs_stride_z,
                                                    uint rhs_stride_z,
#if defined(BETA)
                                                    uint bias_stride_z,
#endif
                                                    uint dst_stride_z
#if defined(REINTERPRET_OUTPUT_AS_3D)
                                                    ,
                                                    uint dst_cross_plane_pad
#endif
                                                    ,
                                                    const int M,
                                                    const int N,
                                                    const int K)
{

#define PIXEL_UNIT CONVERT_VECTOR_SIZE_TO_PIXEL_UNIT(N0)


#define LHS_BLOCK_SIZE ((K0) * (M0))

#if defined(LHS_INTERLEAVE)
#define LHS_OFFSET_X (M0)
#define LHS_STEP_X ((M0) * (V0))
#define LHS_STEP_LOOP (1)
#else
#define LHS_OFFSET_X (LHS_BLOCK_SIZE)
#define LHS_STEP_X (M0)
#define LHS_STEP_LOOP (V0)
#endif


#define RHS_BLOCK_SIZE ((K0) * (PIXEL_UNIT))


#if defined(RHS_INTERLEAVE)
#define RHS_OFFSET_X (PIXEL_UNIT)
#define RHS_STEP_X ((PIXEL_UNIT) * (H0))
#else
#define RHS_OFFSET_X (RHS_BLOCK_SIZE)
#define RHS_STEP_X (PIXEL_UNIT)
#endif

    const uint x = get_global_id(0);
    const uint y = get_global_id(1);
    const uint z = get_global_id(2);

#if defined(DUMMY_WORK_ITEMS)
    if((x * N0 >= N) || (y * M0 >= M))
    {
        return;
    }
#endif


    __global uchar *lhs_addr = lhs_ptr + lhs_offset_first_element_in_bytes + (y % V0) * (uint)LHS_OFFSET_X * sizeof(DATA_TYPE) + (y / V0) * (uint)lhs_stride_y + (z * lhs_stride_z);

#if defined(MATRIX_B_DEPTH)

    const uint z_rhs = (z % MATRIX_B_DEPTH);
#else
    const uint z_rhs = z;
#endif


    uint       x_rhs = (x % H0) * (uint)RHS_OFFSET_X;
    const uint y_rhs = (x / (uint)H0) + z_rhs * RHS_HEIGHT;


    REPEAT_VAR_INIT_TO_CONST(M0, VEC_DATA_TYPE(DATA_TYPE_ACCUMULATOR, N0), c, 0);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zero, 0);

    __global DATA_TYPE *lhs = (__global DATA_TYPE *)(lhs_addr);

    for(int i = 0; i < K; i += K0)
    {
        VEC_DATA_TYPE(DATA_TYPE, M0)
        a0;
        VEC_DATA_TYPE(DATA_TYPE, N0)
        b0;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 0 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

#if K0 > 1
        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 1 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
#endif

#if K0 > 2
        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 2 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
#endif

#if K0 > 3
        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 3 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
#endif

#if K0 > 4
        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 4 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 5 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 6 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 7 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
#endif

#if K0 > 8
        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 8 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 9 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 10 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 11 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 12 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 13 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 14 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;

        a0 = VLOAD(M0)(0, lhs);
        b0 = READ_IMAGE2D(DATA_TYPE, PIXEL_UNIT, rhs_img, (x_rhs + 15 * RHS_STEP_X), (y_rhs));

        ARM_MM_T_NT(M0, N0, 1, DATA_TYPE, a, b, c);

        lhs += LHS_STEP_X;
#endif

#ifndef LHS_INTERLEAVE
        lhs += (M0 * K0 * (V0 - 1));
#endif

        x_rhs += K0 * RHS_STEP_X;
#ifndef RHS_INTERLEAVE
        x_rhs += (PIXEL_UNIT * K0 * (H0 - 1));
#endif
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)M0 * dst_stride_y);

    REPEAT_VAR_INIT_TO_CONST(M0, uint, zout, 0);

    const bool cond_y = ((get_global_id(1) + 1) * M0 >= M);
    const bool cond_x = ((get_global_id(0) + 1) * N0 >= N);

#if defined(REINTERPRET_OUTPUT_AS_3D)


    CALCULATE_Z_OFFSET(M0, uint, zout, y * (uint)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


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK_BOUNDARY_AWARE(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, 1, PARTIAL_STORE_N0, false, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(1, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK_BROADCAST(M0, c, bias_hp0);
#else
    ADD_BLOCK_BROADCAST(M0, c, bias0);
#endif

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (y * (uint)M0 * bias_stride_y) + z * bias_stride_z;

    LOAD_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif

#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE_ACCUMULATOR, bias, bias_hp);
    ADD_BLOCK(M0, c, bias_hp);
#else
    ADD_BLOCK(M0, c, bias);
#endif

#endif
#endif

#if defined(ACTIVATION_TYPE)
#if defined(MIXED_PRECISION)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE_ACCUMULATOR, N0, c, A_VAL, B_VAL);
#else
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif
#endif


#if defined(MIXED_PRECISION)
    CONVERT_BLOCK(M0, N0, DATA_TYPE, c, c_lp);
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c_lp, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#else
    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
#endif

#undef LHS_BLOCK_SIZE
#undef LHS_OFFSET_X
#undef LHS_STEP_X
#undef RHS_BLOCK_SIZE
#undef RHS_OFFSET_X
#undef RHS_STEP_X
#undef PIXEL_UNIT
#undef LHS_STEP_LOOP
#undef RHS_STEP_LOOP
}
#endif

#endif

#endif

#if defined(M0) && defined(N0) && defined(K0) && defined(DATA_TYPE)

#define VFMA(a, b, c)     \
    ({                    \
        c = fma(a, b, c); \
    })

#if M0 == 1
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
    })
#elif M0 == 2
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
    })
#elif M0 == 3
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
    })
#elif M0 == 4
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
    })
#elif M0 == 5
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
    })
#elif M0 == 6
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \
    })
#elif M0 == 7
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \
    })
#elif M0 == 8
#define RHS_VFMA_M0xN0(i, a, b, c)                                    \
    ({                                                                \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##0).s##i), b, (c##0)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##1).s##i), b, (c##1)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##2).s##i), b, (c##2)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##3).s##i), b, (c##3)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##4).s##i), b, (c##4)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##5).s##i), b, (c##5)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##6).s##i), b, (c##6)); \
        VFMA((VEC_DATA_TYPE(DATA_TYPE, N0))((a##7).s##i), b, (c##7)); \
    })
#else
#error "M0 not supported"
#endif

#if defined(GEMM_MM_NATIVE)

__kernel void gemm_mm_native(IMAGE_DECLARATION(lhs),
                             IMAGE_DECLARATION(rhs),
#if defined(BETA)
                             IMAGE_DECLARATION(bias),
#endif
                             IMAGE_DECLARATION(dst),
                             uint lhs_stride_z,
                             uint rhs_stride_z,
#if defined(BETA)
                             uint bias_stride_z,
#endif
                             uint      dst_stride_z,
                             const int M,
                             const int N,
                             const int K
#if defined(REINTERPRET_INPUT_AS_3D)
                             ,
                             uint lhs_cross_plane_pad
#endif
#if defined(REINTERPRET_OUTPUT_AS_3D)
                             ,
                             uint dst_cross_plane_pad
#endif
                            )
{

#define RHS_BLOCK_SIZE ((K0) * (N0))


#define RHS_OFFSET_X (RHS_BLOCK_SIZE)

    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(M0, uint, zlhs, 0);
    REPEAT_VAR_INIT_TO_CONST(16, uint, zero, 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(DATA_TYPE, N0), c, 0);

    int i = 0;
#if K0 > 1
    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, zero);

        RHS_VFMA_M0xN0(0, a, b0, c);
        RHS_VFMA_M0xN0(1, a, b1, c);
#if K0 > 2
        RHS_VFMA_M0xN0(2, a, b2, c);
#endif
#if K0 > 3
        RHS_VFMA_M0xN0(3, a, b3, c);
#endif
#if K0 > 4
        RHS_VFMA_M0xN0(4, a, b4, c);
        RHS_VFMA_M0xN0(5, a, b5, c);
        RHS_VFMA_M0xN0(6, a, b6, c);
        RHS_VFMA_M0xN0(7, a, b7, c);
#endif
#if K0 > 8
        RHS_VFMA_M0xN0(8, a, b8, c);
        RHS_VFMA_M0xN0(9, a, b9, c);
        RHS_VFMA_M0xN0(A, a, bA, c);
        RHS_VFMA_M0xN0(B, a, bB, c);
        RHS_VFMA_M0xN0(C, a, bC, c);
        RHS_VFMA_M0xN0(D, a, bD, c);
        RHS_VFMA_M0xN0(E, a, bE, c);
        RHS_VFMA_M0xN0(F, a, bF, c);
#endif

        lhs_offset += K0 * sizeof(DATA_TYPE);
        rhs_offset += K0 * rhs_stride_y;
    }
#endif

    for(; i < K; ++i)
    {

        VEC_DATA_TYPE(DATA_TYPE, 2)
        a0 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 0 * lhs_stride_y + zlhs0));
#if M0 > 1
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a1 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 1 * lhs_stride_y + zlhs1));
#endif
#if M0 > 2
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a2 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 2 * lhs_stride_y + zlhs2));
#endif
#if M0 > 3
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a3 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 3 * lhs_stride_y + zlhs3));
#endif
#if M0 > 4
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a4 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 4 * lhs_stride_y + zlhs4));
#endif
#if M0 > 5
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a5 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 5 * lhs_stride_y + zlhs5));
#endif
#if M0 > 6
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a6 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 6 * lhs_stride_y + zlhs6));
#endif
#if M0 > 7
        VEC_DATA_TYPE(DATA_TYPE, 2)
        a7 = *((__global DATA_TYPE *)(lhs_ptr + lhs_offset + 7 * lhs_stride_y + zlhs7));
#endif

        VEC_DATA_TYPE(DATA_TYPE, N0)
        b = VLOAD(N0)(0, (__global DATA_TYPE *)(rhs_ptr + rhs_offset + 0 * rhs_stride_y));
        RHS_VFMA_M0xN0(0, a, b, c);

        lhs_offset += sizeof(DATA_TYPE);
        rhs_offset += rhs_stride_y;
    }

    __global uchar *dst_addr = dst_ptr + dst_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (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


#if defined(ALPHA)
    SCALE_BLOCK(M0, DATA_TYPE, c, ALPHA);
#endif


#if defined(BETA)
#if defined(BROADCAST_BIAS)
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (get_global_id(0) * (uint)N0 * sizeof(DATA_TYPE));

    LOAD_BLOCK(1, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero);

#ifndef UNIT_BETA
    SCALE_BLOCK(1, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK_BROADCAST(M0, c, bias0);

#else
    __global uchar *bias_addr = bias_ptr + bias_offset_first_element_in_bytes + (x * (uint)N0 * sizeof(DATA_TYPE)) + (COMPUTE_M0_START_ROW(y, M0, PARTIAL_STORE_M0) * bias_stride_y) + z * bias_stride_z;

    LOAD_BLOCK(M0, N0, DATA_TYPE, bias, bias_addr, 0, bias_stride_y, zero);

#ifndef UNIT_BETA
    SCALE_BLOCK(M0, DATA_TYPE, bias, BETA);
#endif


    ADD_BLOCK(M0, c, bias);

#endif
#endif

#if defined(ACTIVATION_TYPE)
    ACTIVATION_BLOCK(M0, ACTIVATION_TYPE, DATA_TYPE, N0, c, A_VAL, B_VAL);
#endif

    const bool cond_y = y == 0;
    const bool cond_x = ((x + 1) * N0 >= N);


    STORE_BLOCK_BOUNDARY_AWARE(M0, N0, DATA_TYPE, c, dst_addr, dst_stride_y, zout, PARTIAL_STORE_M0, PARTIAL_STORE_N0, cond_y, cond_x);
}
#endif
#endif

#if defined(BETA)

__kernel void gemm_ma_f32(TENSOR3D_DECLARATION(src),
                          TENSOR3D_DECLARATION(dst))
{

    Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
    Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst);


    float4 alpha_ab = vload4(0, (__global float *)dst.ptr);


    float4 c = vload4(0, (__global float *)src.ptr);


    float4 out = alpha_ab + (float4)BETA * c;


    vstore4(out, 0, (__global float *)dst.ptr);
}

#if defined(ARM_COMPUTE_OPENCL_FP16_ENABLED)

__kernel void gemm_ma_f16(TENSOR3D_DECLARATION(src),
                          TENSOR3D_DECLARATION(dst))
{

    Tensor3D src = CONVERT_TO_TENSOR3D_STRUCT(src);
    Tensor3D dst = CONVERT_TO_TENSOR3D_STRUCT(dst);


    half8 alpha_ab = vload8(0, (__global half *)dst.ptr);


    half8 c = vload8(0, (__global half *)src.ptr);


    half8 out = alpha_ab + (half8)BETA * c;


    vstore8(out, 0, (__global half *)dst.ptr);
}
#endif
#endif  )"