FloorDiv

传入两个等长的浮点型数组，对于数组中对位元素做除法，再对除法所得结果执行向下取整操作。

\[dst_i = floorf(\frac{input0_i}{input1_i})\]
输入：

  • input0 - 被除数数据地址。

  • input1 - 除数数据地址。

  • params - 其他参数打包成数组：

    • input0_dims - input0 的维度。

    • input1_dims - input1的维度。

    • output_dims - output的维度。

    • strides0 - Input0 的步长，大小为8*sizeof(int)

    • strides1 - Input1 的步长，大小为8*sizeof(int)

    • strides_output - Output 的步长，大小为8*sizeof(int)

    • num_dims - 维度数量。

  • core_mask - 核掩码（仅适用于共享存储版本）。

输出：

  • output - 计算结果地址。

支持平台：

FT78NE MT7004

备注

  • FT78NE 支持fp32, fp64

  • MT7004 支持fp16, fp32

共享存储版本:

void hp_floor_div_s(half *src_data0, half *src_data1, half *dst_data, unsigned long long *params, int core_mask)
void fp_floor_div_s(float *src_data0, float *src_data1, float *dst_data, unsigned long long *params, int core_mask)
void dp_floor_div_s(double *src_data0, double *src_data1, double *dst_data, unsigned long long *params, int core_mask)

C调用示例：

 1//FT78NE示例
 2#include <stdio.h>
 3#include <floordiv.h>
 4
 5int main(int argc, char* argv[]) {
 6    float *input0 = (float *)0xA0000000;   //input在DDR空间
 7    float *input1 = (float *)0xB0000000;
 8    float *output = (float *)0xC0000000;
 9    int *strides0 = (int*)0x84000000;
10    int *strides1 = (int*)0x85000000;
11    int *strides_output = (int*)0x86000000;
12
13    int core_mask = 0xff;
14    int input0_dims[] = {4, 8, 17};  // 2x2
15    int input1_dims[] = {4, 8, 17};  // 2x2
16    int output_dims[] = {4, 8, 17};      // 2x2
17    int num_dims = 3;
18
19    unsigned long long params[9];
20    params[0] = (unsigned long long)input0_dims;
21    params[1] = (unsigned long long)input1_dims;
22    params[2] = (unsigned long long)output_dims;
23    params[3] = (unsigned long long)strides0;
24    params[4] = (unsigned long long)strides1;
25    params[5] = (unsigned long long)strides_output;
26    params[6] = (unsigned long long)num_dims;
27    fp_floor_div_s( input0,input1, output,params,core_mask);
28    return 0;
29}

私有存储版本:

void hp_floor_div_p(half *src_data0, half *src_data1, half *dst_data, unsigned long long *params)
void fp_floor_div_p(float *src_data0, float *src_data1, float *dst_data, unsigned long long *params)
void dp_floor_div_p(double *src_data0, double *src_data1, double *dst_data, unsigned long long *params)

C调用示例：

 1//FT78NE示例
 2#include <stdio.h>
 3#include <floordiv.h>
 4int main(int argc, char* argv[]) {
 5    float *input0 = (float *)0x10810000;   //input在L2空间
 6    float *input1 = (float *)0x10820000;
 7    float *output = (float *)0x10830000;
 8    int *strides0 = (int*)0x10841000;
 9    int *strides1 = (int*)0x10842000;
10    int *strides_output = (int*)0x10843000;
11    int input0_dims[] = {4, 8, 17};  // 2x2
12    int input1_dims[] = {4, 8, 17};  // 2x2
13    int output_dims[] = {4, 8, 17};      // 2x2
14    int num_dims = 3;
15
16    unsigned long long params[9];
17    params[0] = (unsigned long long)input0_dims;
18    params[1] = (unsigned long long)input1_dims;
19    params[2] = (unsigned long long)output_dims;
20    params[3] = (unsigned long long)strides0;
21    params[4] = (unsigned long long)strides1;
22    params[5] = (unsigned long long)strides_output;
23    params[6] = (unsigned long long)num_dims;
24    fp_floor_div_p( input0, input1, output, params);
25    return 0;
26}