BiasAdd
=================

将偏置向量 input_bias 沿通道维广播，加到输入张量 input_x 上。

.. math::

   \mathrm{output} = \mathrm{input\_x} + \mathrm{input\_bias}

- NCHW：:math:`\mathrm{output}(n,c,h,w) = \mathrm{input\_x}(n,c,h,w) + \mathrm{input\_bias}(c)`
- NHWC：:math:`\mathrm{output}(n,h,w,c) = \mathrm{input\_x}(n,h,w,c) + \mathrm{input\_bias}(c)`

输入：
        - **input_x** - 输入张量地址
        - **input_bias** - 偏置地址，长度等于通道数 C；c64 时为交错复数，长度 2C 个 float
        - **dims** - 维度数组；NCHW 为 [N, C, H, W]，NHWC 为 [N, H, W, C]
        - **shape_size** - 维数（测试常用 4）
        - **data_format** - 布局：0 为 NCHW，1 为 NHWC（整型，不是字符串）
        - **length** - input_x 元素个数（复数时为复数个数）
        - **core_mask** - 核掩码（仅共享存储版本）

输出：
        - **output** - 输出地址，形状与 input_x 相同

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
        - data_format 为 int：0 = NCHW，非 0（测试用 1）= NHWC

**共享存储版本:**

.. c:function:: void i8_bias_add_s(int8_t *input_x, int8_t *input_bias, int8_t *output, int *dims, int shape_size, int data_format, int length, int core_mask)
.. c:function:: void i16_bias_add_s(int16_t *input_x, int16_t *input_bias, int16_t *output, int *dims, int shape_size, int data_format, int length, int core_mask)
.. c:function:: void i32_bias_add_s(int32_t *input_x, int32_t *input_bias, int32_t *output, int *dims, int shape_size, int data_format, int length, int core_mask)
.. c:function:: void hp_bias_add_s(float16 *input_x, float16 *input_bias, float16 *output, int *dims, int shape_size, int data_format, int length, int core_mask)
.. c:function:: void fp_bias_add_s(float *input_x, float *input_bias, float *output, int *dims, int shape_size, int data_format, int length, int core_mask)
.. c:function:: void dp_bias_add_s(double *input_x, double *input_bias, double *output, int *dims, int shape_size, int data_format, int length, int core_mask)
.. c:function:: void c64_bias_add_s(float *input_x, float *input_bias, float *output, int *dims, int shape_size, int data_format, int length, int core_mask)
.. c:function:: void c128_bias_add_s(double *input_x, double *input_bias, double *output, int *dims, int shape_size, int data_format, int length, int core_mask)

**C调用示例：**

.. code-block:: c
    :linenos:
    :emphasize-lines: 14

    // MT7004 示例（共享存储多核，DDR 地址，NHWC）
    void TestBiasAddSMCFp32(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_x = (float *)0x81000000;
        float *input_bias = (float *)0x82000000;
        float *output = (float *)0x83000000;
        int *dims = (int *)0x85000000;
        if (logic_core_id == 0) {
            dims[0] = 1; dims[1] = 4; dims[2] = 4; dims[3] = 16; /* N,H,W,C */
        }
        sys_bar(0, core_num);
        fp_bias_add_s(input_x, input_bias, output, dims, 4, 1 /* NHWC */, 256, core_mask);
    }

    void main() {
        int core_mask = 0b1111;
        TestBiasAddSMCFp32(core_mask);
    }

**私有存储版本:**

.. c:function:: void i8_bias_add_p(int8_t *input_x, int8_t *input_bias, int8_t *output, int *dims, int shape_size, int data_format, int length)
.. c:function:: void i16_bias_add_p(int16_t *input_x, int16_t *input_bias, int16_t *output, int *dims, int shape_size, int data_format, int length)
.. c:function:: void i32_bias_add_p(int32_t *input_x, int32_t *input_bias, int32_t *output, int *dims, int shape_size, int data_format, int length)
.. c:function:: void hp_bias_add_p(float16 *input_x, float16 *input_bias, float16 *output, int *dims, int shape_size, int data_format, int length)
.. c:function:: void fp_bias_add_p(float *input_x, float *input_bias, float *output, int *dims, int shape_size, int data_format, int length)
.. c:function:: void dp_bias_add_p(double *input_x, double *input_bias, double *output, int *dims, int shape_size, int data_format, int length)
.. c:function:: void c64_bias_add_p(float *input_x, float *input_bias, float *output, int *dims, int shape_size, int data_format, int length)
.. c:function:: void c128_bias_add_p(double *input_x, double *input_bias, double *output, int *dims, int shape_size, int data_format, int length)

**C调用示例：**

.. code-block:: c
    :linenos:
    :emphasize-lines: 8

    // MT7004 示例（私有存储单核，AM 地址，NHWC）
    void TestBiasAddAMFp32(void) {
        float *input_x = (float *)0x10010000;
        float *input_bias = (float *)0x10020000;
        float *output = (float *)0x10030000;
        int *dims = (int *)0x10050000;
        dims[0] = 1; dims[1] = 4; dims[2] = 4; dims[3] = 16; /* N,H,W,C */
        fp_bias_add_p(input_x, input_bias, output, dims, 4, 1 /* NHWC */, 256);
    }

    void main() {
        TestBiasAddAMFp32();
    }