xref: /aosp_15_r20/external/ComputeLibrary/cl_kernels/common/fft.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 defined(DATA_TYPE)

#define TWIDDLE_FACTOR_MULTIPLICATION(phi, input)  \
    {                                              \
        VEC_DATA_TYPE(DATA_TYPE, 2)                \
        w, tmp;                                    \
        w.x   = cos(phi);                          \
        w.y   = sin(phi);                          \
        tmp.x = (w.x * input.x) - (w.y * input.y); \
        tmp.y = (w.x * input.y) + (w.y * input.x); \
        input = tmp;                               \
    }


#define DFT_2(c0, c1)               \
    {                               \
        VEC_DATA_TYPE(DATA_TYPE, 2) \
        v0;                         \
        v0 = c0;                    \
        c0 = v0 + c1;               \
        c1 = v0 - c1;               \
    }


#define SQRT3DIV2 0.86602540378443f


#define DFT_3(c0, c1, c2)                             \
    {                                                 \
        VEC_DATA_TYPE(DATA_TYPE, 2)                   \
        v0 = c1 + c2;                                 \
        VEC_DATA_TYPE(DATA_TYPE, 2)                   \
        v1   = c1 - c2;                               \
        c1.x = c0.x - 0.5f * v0.x + v1.y * SQRT3DIV2; \
        c1.y = c0.y - 0.5f * v0.y - v1.x * SQRT3DIV2; \
        c2.x = c0.x - 0.5f * v0.x - v1.y * SQRT3DIV2; \
        c2.y = c0.y - 0.5f * v0.y + v1.x * SQRT3DIV2; \
        c0   = c0 + v0;                               \
    }


#define DFT_4(c0, c1, c2, c3)       \
    {                               \
        VEC_DATA_TYPE(DATA_TYPE, 2) \
        v0, v1, v2, v3;             \
        v0   = c0 + c2;             \
        v1   = c1 + c3;             \
        v2   = c0 - c2;             \
        v3.x = c1.y - c3.y;         \
        v3.y = c3.x - c1.x;         \
        c0   = v0 + v1;             \
        c2   = v0 - v1;             \
        c1   = v2 + v3;             \
        c3   = v2 - v3;             \
    }


#define W5_A (DATA_TYPE)0.30901699437494f
#define W5_B (DATA_TYPE)0.95105651629515f
#define W5_C (DATA_TYPE)0.80901699437494f
#define W5_D (DATA_TYPE)0.58778525229247f


#define DFT_5(c0, c1, c2, c3, c4)                                  \
    {                                                              \
        VEC_DATA_TYPE(DATA_TYPE, 2)                                \
        v0, v1, v2, v3, v4;                                        \
        v0 = c0;                                                   \
        v1 = W5_A * (c1 + c4) - W5_C * (c2 + c3);                  \
        v2 = W5_C * (c1 + c4) - W5_A * (c2 + c3);                  \
        v3 = W5_D * (c1 - c4) - W5_B * (c2 - c3);                  \
        v4 = W5_B * (c1 - c4) + W5_D * (c2 - c3);                  \
        c0 = v0 + c1 + c2 + c3 + c4;                               \
        c1 = v0 + v1 + (VEC_DATA_TYPE(DATA_TYPE, 2))(v4.y, -v4.x); \
        c2 = v0 - v2 + (VEC_DATA_TYPE(DATA_TYPE, 2))(v3.y, -v3.x); \
        c3 = v0 - v2 + (VEC_DATA_TYPE(DATA_TYPE, 2))(-v3.y, v3.x); \
        c4 = v0 + v1 + (VEC_DATA_TYPE(DATA_TYPE, 2))(-v4.y, v4.x); \
    }


#define W7_A (DATA_TYPE)0.62348980185873f
#define W7_B (DATA_TYPE)0.78183148246802f
#define W7_C (DATA_TYPE)0.22252093395631f
#define W7_D (DATA_TYPE)0.97492791218182f
#define W7_E (DATA_TYPE)0.90096886790241f
#define W7_F (DATA_TYPE)0.43388373911755f


#define DFT_7(c0, c1, c2, c3, c4, c5, c6)                            \
    {                                                                \
        VEC_DATA_TYPE(DATA_TYPE, 2)                                  \
        v0, v1, v2, v3, v4, v5, v6;                                  \
        v0 = c0;                                                     \
        v1 = W7_A * (c1 + c6) - W7_C * (c2 + c5) - W7_E * (c3 + c4); \
        v2 = W7_C * (c1 + c6) + W7_E * (c2 + c5) - W7_A * (c3 + c4); \
        v3 = W7_E * (c1 + c6) - W7_A * (c2 + c5) + W7_C * (c3 + c4); \
        v4 = W7_B * (c1 - c6) + W7_D * (c2 - c5) + W7_F * (c3 - c4); \
        v5 = W7_D * (c1 - c6) - W7_F * (c2 - c5) - W7_B * (c3 - c4); \
        v6 = W7_F * (c1 - c6) - W7_B * (c2 - c5) + W7_D * (c3 - c4); \
        c0 = v0 + c1 + c2 + c3 + c4 + c5 + c6;                       \
        c1 = v0 + v1 + (VEC_DATA_TYPE(DATA_TYPE, 2))(v4.y, -v4.x);   \
        c2 = v0 - v2 + (VEC_DATA_TYPE(DATA_TYPE, 2))(v5.y, -v5.x);   \
        c3 = v0 - v3 + (VEC_DATA_TYPE(DATA_TYPE, 2))(v6.y, -v6.x);   \
        c4 = v0 - v3 + (VEC_DATA_TYPE(DATA_TYPE, 2))(-v6.y, v6.x);   \
        c5 = v0 - v2 + (VEC_DATA_TYPE(DATA_TYPE, 2))(-v5.y, v5.x);   \
        c6 = v0 + v1 + (VEC_DATA_TYPE(DATA_TYPE, 2))(-v4.y, v4.x);   \
    }


#define DFT_8(c0, c1, c2, c3, c4, c5, c6, c7) \
    {                                         \
        VEC_DATA_TYPE(DATA_TYPE, 2)           \
        v0, v1, v2, v3, v4, v5, v6, v7;       \
        VEC_DATA_TYPE(DATA_TYPE, 2)           \
        s0, s1, s2, s3, s4, s5, s6, s7;       \
        VEC_DATA_TYPE(DATA_TYPE, 2)           \
        t0, t1, t2;                           \
        v0   = c0 + c4;                       \
        v1   = c1 + c5;                       \
        v2   = c2 + c6;                       \
        v3   = c3 + c7;                       \
        v4   = c0 - c4;                       \
        v5   = c1 - c5;                       \
        v6   = c2 - c6;                       \
        v7   = c3 - c7;                       \
        s0   = v0 + v2;                       \
        s1   = v1 + v3;                       \
        s2   = v0 - v2;                       \
        s3   = v1 - v3;                       \
        s4.x = v4.x - v6.y;                   \
        s4.y = v4.y + v6.x;                   \
        s5.x = v5.x - v7.y;                   \
        s5.y = v5.y + v7.x;                   \
        s6.x = v4.x + v6.y;                   \
        s6.y = v4.y - v6.x;                   \
        s7.x = v5.x + v7.y;                   \
        s7.y = v5.y - v7.x;                   \
        t0.x = -s3.y;                         \
        t0.y = s3.x;                          \
        t1.x = M_SQRT1_2_F * (s5.x - s5.y);   \
        t1.y = M_SQRT1_2_F * (s5.x + s5.y);   \
        t2.x = -M_SQRT1_2_F * (s7.x + s7.y);  \
        t2.y = M_SQRT1_2_F * (s7.x - s7.y);   \
        c0   = s0 + s1;                       \
        c1   = s6 - t2;                       \
        c2   = s2 - t0;                       \
        c3   = s4 - t1;                       \
        c4   = s0 - s1;                       \
        c5   = s6 + t2;                       \
        c6   = s2 + t0;                       \
        c7   = s4 + t1;                       \
    }


__kernel void fft_radix_2_first_stage_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 4)
    data = vload4(0, (__global DATA_TYPE *)input.ptr);


    DFT_2(data.s01, data.s23);


    vstore4(data, 0, (__global DATA_TYPE *)output.ptr);
}


__kernel void fft_radix_2_first_stage_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0));


    DFT_2(data1, data2);


    vstore2(data1, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 1, 0));
}


__kernel void fft_radix_3_first_stage_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 4)
    data0 = vload4(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 2, 0, 0));


    DFT_3(data0.s01, data0.s23, data1.s01);


    vstore4(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 2, 0, 0));
}


__kernel void fft_radix_3_first_stage_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    data0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0));


    DFT_3(data0, data1, data2);


    vstore2(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 1, 0));
    vstore2(data2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2, 0));
}


__kernel void fft_radix_4_first_stage_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 8)
    data = vload8(0, (__global DATA_TYPE *)input.ptr);


    DFT_4(data.s01, data.s23, data.s45, data.s67);


    vstore8(data, 0, (__global DATA_TYPE *)output.ptr);
}


__kernel void fft_radix_4_first_stage_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    data0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3, 0));


    DFT_4(data0, data1, data2, data3);


    vstore2(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 1, 0));
    vstore2(data2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2, 0));
    vstore2(data3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3, 0));
}


__kernel void fft_radix_5_first_stage_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 8)
    data0 = vload8(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 4, 0, 0));


    DFT_5(data0.s01, data0.s23, data0.s45, data0.s67, data1.s01);


    vstore8(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 4, 0, 0));
}


__kernel void fft_radix_5_first_stage_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    data0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 4, 0));


    DFT_5(data0, data1, data2, data3, data4);


    vstore2(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 1, 0));
    vstore2(data2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2, 0));
    vstore2(data3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3, 0));
    vstore2(data4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 4, 0));
}


__kernel void fft_radix_7_first_stage_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 8)
    data0 = vload8(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 4)
    data1 = vload4(0, (__global DATA_TYPE *)tensor3D_offset(&input, 4, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 6, 0, 0));


    DFT_7(data0.s01, data0.s23, data0.s45, data0.s67, data1.s01, data1.s23, data2.s01);


    vstore8(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore4(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 4, 0, 0));
    vstore2(data2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 6, 0, 0));
}


__kernel void fft_radix_7_first_stage_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    data0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 4, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data5 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 5, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data6 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 6, 0));


    DFT_7(data0, data1, data2, data3, data4, data5, data6);


    vstore2(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 1, 0));
    vstore2(data2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2, 0));
    vstore2(data3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3, 0));
    vstore2(data4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 4, 0));
    vstore2(data5, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 5, 0));
    vstore2(data6, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 6, 0));
}


__kernel void fft_radix_8_first_stage_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 16)
    data = vload16(0, (__global DATA_TYPE *)input.ptr);


    DFT_8(data.s01, data.s23, data.s45, data.s67, data.s89, data.sAB, data.sCD, data.sEF);


    vstore16(data, 0, (__global DATA_TYPE *)output.ptr);
}


__kernel void fft_radix_8_first_stage_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
)
{

    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT(input);
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT(output);
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    data0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 1, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 4, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data5 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 5, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data6 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 6, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    data7 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 7, 0));


    DFT_8(data0, data1, data2, data3, data4, data5, data6, data7);


    vstore2(data0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(data1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 1, 0));
    vstore2(data2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2, 0));
    vstore2(data3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3, 0));
    vstore2(data4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 4, 0));
    vstore2(data5, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 5, 0));
    vstore2(data6, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 6, 0));
    vstore2(data7, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 7, 0));
}


__kernel void fft_radix_2_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(0);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += n * input.stride_x + get_global_id(1) * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += n * output.stride_x + get_global_id(1) * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, Nx, 0, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);


    DFT_2(c0, c1);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, Nx, 0, 0));
}


__kernel void fft_radix_2_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(1);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += get_global_id(0) * input.stride_x + n * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += get_global_id(0) * output.stride_x + n * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, Nx, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);


    DFT_2(c0, c1);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, Nx, 0));
}


__kernel void fft_radix_3_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(0);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += n * input.stride_x + get_global_id(1) * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += n * output.stride_x + get_global_id(1) * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 2 * Nx, 0, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);


    DFT_3(c0, c1, c2);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, Nx, 0, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 2 * Nx, 0, 0));
}


__kernel void fft_radix_3_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(1);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += get_global_id(0) * input.stride_x + n * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += get_global_id(0) * output.stride_x + n * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2 * Nx, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);


    DFT_3(c0, c1, c2);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, Nx, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2 * Nx, 0));
}


__kernel void fft_radix_4_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(0);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += n * input.stride_x + get_global_id(1) * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += n * output.stride_x + get_global_id(1) * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 2 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 3 * Nx, 0, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);


    DFT_4(c0, c1, c2, c3);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, Nx, 0, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 2 * Nx, 0, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 3 * Nx, 0, 0));
}


__kernel void fft_radix_4_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(1);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += get_global_id(0) * input.stride_x + n * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += get_global_id(0) * output.stride_x + n * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3 * Nx, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);


    DFT_4(c0, c1, c2, c3);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, Nx, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2 * Nx, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3 * Nx, 0));
}


__kernel void fft_radix_5_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(0);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += n * input.stride_x + get_global_id(1) * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += n * output.stride_x + get_global_id(1) * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 2 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 3 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 4 * Nx, 0, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);
    TWIDDLE_FACTOR_MULTIPLICATION(4 * phi, c4);


    DFT_5(c0, c1, c2, c3, c4);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, Nx, 0, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 2 * Nx, 0, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 3 * Nx, 0, 0));
    vstore2(c4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 4 * Nx, 0, 0));
}


__kernel void fft_radix_5_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(1);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += get_global_id(0) * input.stride_x + n * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += get_global_id(0) * output.stride_x + n * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 4 * Nx, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);
    TWIDDLE_FACTOR_MULTIPLICATION(4 * phi, c4);


    DFT_5(c0, c1, c2, c3, c4);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, Nx, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2 * Nx, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3 * Nx, 0));
    vstore2(c4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 4 * Nx, 0));
}


__kernel void fft_radix_7_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(0);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += n * input.stride_x + get_global_id(1) * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += n * output.stride_x + get_global_id(1) * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 2 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 3 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 4 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c5 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 5 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c6 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 6 * Nx, 0, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);
    TWIDDLE_FACTOR_MULTIPLICATION(4 * phi, c4);
    TWIDDLE_FACTOR_MULTIPLICATION(5 * phi, c5);
    TWIDDLE_FACTOR_MULTIPLICATION(6 * phi, c6);


    DFT_7(c0, c1, c2, c3, c4, c5, c6);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, Nx, 0, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 2 * Nx, 0, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 3 * Nx, 0, 0));
    vstore2(c4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 4 * Nx, 0, 0));
    vstore2(c5, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 5 * Nx, 0, 0));
    vstore2(c6, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 6 * Nx, 0, 0));
}


__kernel void fft_radix_7_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(1);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += get_global_id(0) * input.stride_x + n * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += get_global_id(0) * output.stride_x + n * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 4 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c5 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 5 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c6 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 6 * Nx, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);
    TWIDDLE_FACTOR_MULTIPLICATION(4 * phi, c4);
    TWIDDLE_FACTOR_MULTIPLICATION(5 * phi, c5);
    TWIDDLE_FACTOR_MULTIPLICATION(6 * phi, c6);


    DFT_7(c0, c1, c2, c3, c4, c5, c6);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, Nx, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2 * Nx, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3 * Nx, 0));
    vstore2(c4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 4 * Nx, 0));
    vstore2(c5, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 5 * Nx, 0));
    vstore2(c6, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 6 * Nx, 0));
}


__kernel void fft_radix_8_axis_0(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(0);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += n * input.stride_x + get_global_id(1) * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += n * output.stride_x + get_global_id(1) * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 2 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 3 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 4 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c5 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 5 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c6 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 6 * Nx, 0, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c7 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 7 * Nx, 0, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);
    TWIDDLE_FACTOR_MULTIPLICATION(4 * phi, c4);
    TWIDDLE_FACTOR_MULTIPLICATION(5 * phi, c5);
    TWIDDLE_FACTOR_MULTIPLICATION(6 * phi, c6);
    TWIDDLE_FACTOR_MULTIPLICATION(7 * phi, c7);


    DFT_8(c0, c1, c2, c3, c4, c5, c6, c7);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, Nx, 0, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 2 * Nx, 0, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 3 * Nx, 0, 0));
    vstore2(c4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 4 * Nx, 0, 0));
    vstore2(c5, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 5 * Nx, 0, 0));
    vstore2(c6, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 6 * Nx, 0, 0));
    vstore2(c7, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 7 * Nx, 0, 0));
}


__kernel void fft_radix_8_axis_1(
    TENSOR3D_DECLARATION(input)
#ifndef IN_PLACE
    ,
    TENSOR3D_DECLARATION(output)
#endif
    ,
    uint Nx, uint Ni, float exp_const)
{

    uint kx = get_global_id(1);


    uint nx = kx % Nx;


    uint n = nx + (kx / Nx) * Ni;


    Tensor3D input = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(input);
    input.ptr += get_global_id(0) * input.stride_x + n * input.stride_y + get_global_id(2) * input.stride_z;
#ifdef IN_PLACE
    Tensor3D output = input;
#else
    Tensor3D output = CONVERT_TO_TENSOR3D_STRUCT_NO_STEP(output);
    output.ptr += get_global_id(0) * output.stride_x + n * output.stride_y + get_global_id(2) * output.stride_z;
#endif


    VEC_DATA_TYPE(DATA_TYPE, 2)
    c0 = vload2(0, (__global DATA_TYPE *)input.ptr);
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c1 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c2 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 2 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c3 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 3 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c4 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 4 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c5 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 5 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c6 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 6 * Nx, 0));
    VEC_DATA_TYPE(DATA_TYPE, 2)
    c7 = vload2(0, (__global DATA_TYPE *)tensor3D_offset(&input, 0, 7 * Nx, 0));


    DATA_TYPE phi = (DATA_TYPE)nx * (DATA_TYPE)exp_const;


    TWIDDLE_FACTOR_MULTIPLICATION(phi, c1);
    TWIDDLE_FACTOR_MULTIPLICATION(2 * phi, c2);
    TWIDDLE_FACTOR_MULTIPLICATION(3 * phi, c3);
    TWIDDLE_FACTOR_MULTIPLICATION(4 * phi, c4);
    TWIDDLE_FACTOR_MULTIPLICATION(5 * phi, c5);
    TWIDDLE_FACTOR_MULTIPLICATION(6 * phi, c6);
    TWIDDLE_FACTOR_MULTIPLICATION(7 * phi, c7);


    DFT_8(c0, c1, c2, c3, c4, c5, c6, c7);


    vstore2(c0, 0, (__global DATA_TYPE *)output.ptr);
    vstore2(c1, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, Nx, 0));
    vstore2(c2, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 2 * Nx, 0));
    vstore2(c3, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 3 * Nx, 0));
    vstore2(c4, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 4 * Nx, 0));
    vstore2(c5, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 5 * Nx, 0));
    vstore2(c6, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 6 * Nx, 0));
    vstore2(c7, 0, (__global DATA_TYPE *)tensor3D_offset(&output, 0, 7 * Nx, 0));
}
#endif  )"