BroadcastTo
=================

将输入张量按 NumPy / MindSpore 广播规则扩展到目标形状 ``output_shape``。
输入维数可少于输出维数（左侧隐式补 1）；每一维上，输入长度必须为 1（可广播）或与输出长度相等。
形状维数不超过 8。

设输出维数为 :math:`M`，输入维数为 :math:`K`，且满足 :math:`K \le M`。先将输入形状左侧补 1，
得到对齐后的输入形状 :math:`\tilde{I}`：

.. math::

    \tilde{I}_i =
    \begin{cases}
    1, & 0 \le i < M - K \\
    I_{i - (M - K)}, & M - K \le i < M
    \end{cases}

对每个维度 :math:`i`，要求 :math:`\tilde{I}_i = 1` 或 :math:`\tilde{I}_i = O_i`，
其中 :math:`O` 为输出形状。

**索引映射**

对输出线性下标 :math:`idx`，将其按输出 stride 分解为多维坐标 :math:`(p_0,\ldots,p_{M-1})`，
再映射到输入偏移：

.. math::

    \begin{aligned}
    S^{out}_{M-1} &= 1,\quad
    S^{out}_{i} = S^{out}_{i+1} \cdot O_{i+1} \\
    S^{in}_{M-1} &=
    \begin{cases}
    0, & \tilde{I}_{M-1} = 1 \\
    1, & \text{otherwise}
    \end{cases},\quad
    S^{in}_{i} =
    \begin{cases}
    0, & \tilde{I}_{i} = 1 \\
    S^{in}_{i+1} \cdot \tilde{I}_{i+1}, & \text{otherwise}
    \end{cases} \\
    \mathrm{in\_off} &= \sum_{i=0}^{M-1} p_i \cdot S^{in}_{i}
    \end{aligned}

即 :math:`O[idx] = I[\mathrm{in\_off}]`。当某维 :math:`\tilde{I}_i = 1` 时，
该维 stride 为 0，对应元素沿该维重复。

输入：
        - **input** - 输入数据地址
        - **input_shape** - 输入形状数组
        - **input_shape_size** - 输入形状维数 :math:`K`
        - **output_shape** - 目标输出形状数组
        - **output_shape_size** - 输出形状维数 :math:`M`
        - **core_mask** - 核掩码（仅共享存储版本使用）

输出：
        - **output** - 输出数据地址

支持平台：
        ``FT78NE``
        ``MT7004``

.. note::
        - FT78NE 支持 int8, int16, int32, fp32, fp64, cplx64, cplx128
        - MT7004 支持 int16, int32, fp16, fp32, cplx64

**共享存储版本:**

.. c:function:: void i8_broadcast_to_s(int8_t *input, int8_t *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)
.. c:function:: void i16_broadcast_to_s(int16_t *input, int16_t *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)
.. c:function:: void i32_broadcast_to_s(int *input, int *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)
.. c:function:: void hp_broadcast_to_s(float16 *input, float16 *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)
.. c:function:: void fp_broadcast_to_s(float *input, float *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)
.. c:function:: void dp_broadcast_to_s(double *input, double *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)
.. c:function:: void c64_broadcast_to_s(float *input, float *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)
.. c:function:: void c128_broadcast_to_s(double *input, double *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size, int core_mask)

**C调用示例：**

.. code-block:: c
    :linenos:
    :emphasize-lines: 12

    // MT7004 示例（共享存储多核，DDR 地址）
    // 例：input {1,N} -> output {N,N}
    void TestBroadcastToSMCFp32(int n, int core_mask) {
        int core_id = get_core_id();
        int logic_core_id = GetLogicCoreId(core_mask, core_id);
        int core_num = GetCoreNum(core_mask);
        float *input = (float *)0x81000000;
        float *output = (float *)0x82000000;
        int input_shape[2] = {1, n};
        int output_shape[2] = {n, n};
        sys_bar(0, core_num);
        fp_broadcast_to_s(input, output, input_shape, 2, output_shape, 2, core_mask);
    }

    void main() {
        int core_mask = 0b1111;
        TestBroadcastToSMCFp32(32, core_mask);
    }

**私有存储版本:**

.. c:function:: void i8_broadcast_to_p(int8_t *input, int8_t *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)
.. c:function:: void i16_broadcast_to_p(int16_t *input, int16_t *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)
.. c:function:: void i32_broadcast_to_p(int *input, int *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)
.. c:function:: void hp_broadcast_to_p(float16 *input, float16 *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)
.. c:function:: void fp_broadcast_to_p(float *input, float *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)
.. c:function:: void dp_broadcast_to_p(double *input, double *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)
.. c:function:: void c64_broadcast_to_p(float *input, float *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)
.. c:function:: void c128_broadcast_to_p(double *input, double *output, int *input_shape, int input_shape_size, int *output_shape, int output_shape_size)

**C调用示例：**

.. code-block:: c
    :linenos:
    :emphasize-lines: 7

    // MT7004 示例（私有存储单核，AM 地址）
    void TestBroadcastToAMFp32(int n) {
        float *input = (float *)0x10000000;
        float *output = (float *)0x10020000;
        int input_shape[2] = {1, n};
        int output_shape[2] = {n, n};
        fp_broadcast_to_p(input, output, input_shape, 2, output_shape, 2);
    }

    void main() {
        TestBroadcastToAMFp32(32);
    }