FloorMod ================= 逐元素计算两个输入张量的 floor-modulus。 .. math:: \text{output}_i = \text{input0}_i - \lfloor \frac{\text{input0}_i}{\text{input1}_i} \rfloor \cdot \text{input1}_i 其中 :math:`\lfloor \cdot \rfloor` 表示向下取整 (floor) 操作。 输入: - **input0** - 第一个输入张量(被除数)的数据地址。 - **input1** - 第二个输入张量(除数)的数据地址。 - **params** - 参数打包,格式如下。 - **input0_dims** - input0的维度信息。 - **input1_dims** - input1的维度信息。 - **output_dims** - output的维度信息。 - **strides0** - 输入张量0的步长信息。 - **strides1** - 输入张量1的步长信息。 - **strides_output** - 输出张量的步长信息。 - **num_dims** - 张量的维度数。 - **core_mask** - 核掩码。(仅共享存储版本)。 输出: - **output** - 输出张量的数据地址,其大小与输入张量相同。 支持平台: ``FT78NE`` ``MT7004`` .. note:: - FT78NE 支持fp32,fp64 - MT7004 支持fp16,fp32 **共享存储版本:** .. c:function:: void fp_floor_mod_s(float* input0, float* input1, float* output, unsigned long long *params, int core_mask) .. c:function:: void hp_floor_mod_s(half* input0, half* input1, half* output, unsigned long long *params, int core_mask) .. c:function:: void dp_floor_mod_s(double* input0, double* input1, double* output, unsigned long long *params, int core_mask) **C调用示例:** .. code-block:: c :linenos: :emphasize-lines: 44 //FT78NE示例 #include #include int main(int argc, char* argv[]) { float* input0 = (float*)0x81000000; float* input1 = (float*)0x82000000; float* output = (float*)0x83000000; int *strides0 = (int*)0x84000000; int *strides1 = (int*)0x85000000; int *strides_output = (int*)0x86000000; int core_mask = 0b1111; // same shape int input0_dims[] = {4, 1, 16}; // 2x2 int input1_dims[] = {4, 8, 16}; // 2x2 int output_dims[] = {4, 8, 16}; // 2x2 int num_dims = 3; unsigned long long params[9]; params[0] = (unsigned long long)input0_dims; params[1] = (unsigned long long)input1_dims; params[2] = (unsigned long long)output_dims; params[3] = (unsigned long long)strides0; params[4] = (unsigned long long)strides1; params[5] = (unsigned long long)strides_output; params[6] = (unsigned long long)num_dims; int total_input0 = get_total_elements(num_dims, input0_dims); int total_input1 = get_total_elements(num_dims, input1_dims); int total_output = get_total_elements(num_dims, output_dims); srand(time(0)); int i; for (i = 0; i < total_input0; ++i) { input0[i] = (float)(rand() % 100) / 10.0f; } for (i = 0; i < total_input1; ++i) { input1[i] = (float)(rand() % 100) / 10.0f + 0.01f; } fp_floor_mod_s(input0, input1, output, params, core_mask); return 0; } **私有存储版本:** .. c:function:: void fp_floor_mod_p(float* input0, float* input1, float* output, unsigned long long *params) .. c:function:: void hp_floor_mod_p(half* input0, half* input1, half* output, unsigned long long *params) .. c:function:: void dp_floor_mod_p(double* input0, double* input1, double* output, unsigned long long *params) **C调用示例:** .. code-block:: c :linenos: :emphasize-lines: 42 //FT04示例 #include #include int main(int argc, char* argv[]) { float* input0 = (float*)0x10010000; float* input1 = (float*)0x10020000; float* output = (float*)0x10030000; int *strides0 = (int*)0x10050000; int *strides1 = (int*)0x10053000; int *strides_output = (int*)0x10056000; // same shape int input0_dims[] = {4, 1, 16}; // 2x2 int input1_dims[] = {4, 8, 16}; // 2x2 int output_dims[] = {4, 8, 16}; // 2x2 int num_dims = 3; unsigned long long params[9]; params[0] = (unsigned long long)input0_dims; params[1] = (unsigned long long)input1_dims; params[2] = (unsigned long long)output_dims; params[3] = (unsigned long long)strides0; params[4] = (unsigned long long)strides1; params[5] = (unsigned long long)strides_output; params[6] = (unsigned long long)num_dims; int total_input0 = get_total_elements(num_dims, input0_dims); int total_input1 = get_total_elements(num_dims, input1_dims); int total_output = get_total_elements(num_dims, output_dims); srand(time(0)); int i; for (i = 0; i < total_input0; ++i) { input0[i] = (float)(rand() % 100) / 10.0f; } for (i = 0; i < total_input1; ++i) { input1[i] = (float)(rand() % 100) / 10.0f + 0.01f; } fp_floor_mod_p(input0, input1, output, params); return 0; }