LessEqual
逐元素计算第一个输入 (Input0) 的元素是否小于等于第二个输入 (Input1) 的对应元素。该算子支持广播机制。
\[\begin{split}output_i = \begin{cases}
\text{True}, & \text{if } Input0_0 <= Input1_i \text{ (Input0 is broadcasted)} \\
\text{True}, & \text{if } Input0_i <= Input1_0 \text{ (Input1 is broadcasted)} \\
\text{True}, & \text{if } Input0_i <= Input1_i \text{ (No broadcast)} \\
\text{False}, & \text{otherwise}
\end{cases}\end{split}\]
- 输入:
Input0 - 第一个输入数据地址。
Input1 - 第二个输入数据地址。
- 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(仅共享存储版本需要)。
- 输出:
Output - 计算结果地址。
- 支持平台:
FT78NEMT7004
备注
FT78NE 支持int8, int16, int32, fp32, fp64
MT7004 支持fp16, fp32, int16, int32
共享存储版本:
-
void i8_less_equal_s(int8_t *Input0, int8_t *Input1, bool *output, long long *params, int core_mask)
-
void i16_less_equal_s(int16_t *Input0, int16_t *Input1, bool *output, long long *params, int core_mask)
-
void i32_less_equal_s(int32_t *Input0, int32_t *Input1, bool *output, long long *params, int core_mask)
-
void hp_less_equal_s(half *Input0, half *Input1, bool *output, long long *params, int core_mask)
-
void fp_less_equal_s(float *Input0, float *Input1, bool *output, long long *params, int core_mask)
-
void dp_less_equal_s(double *Input0, double *Input1, bool *output, long long *params, int core_mask)
C调用示例:
1//FT78NE示例
2#include <stdio.h>
3#include <less.h> // 假设头文件名为 less.h
4int main(int argc, char* argv[]) {
5 float *input0 = (float *)0x81000000;
6 float *input1 = (float *)0x82000000;
7 bool *output = (bool *)0x83000000;
8 int *strides0 = (int*)0x84000000;
9 int *strides1 = (int*)0x85000000;
10 int *strides_output = (int*)0x86000000;
11
12 int i = 0;
13
14 srand(seed++);
15
16 float f_min = -100.0;
17 float f_max = 100.0;
18
19 // same shape
20 int input0_dims[] = {4, 8, 17}; // 2x2
21 int input1_dims[] = {4, 8, 17}; // 2x2
22 int output_dims[] = {4, 8, 17}; // 2x2
23 int num_dims = 3;
24
25 unsigned long long params[9];
26 params[0] = (unsigned long long)input0_dims;
27 params[1] = (unsigned long long)input1_dims;
28 params[2] = (unsigned long long)output_dims;
29 params[3] = (unsigned long long)strides0;
30 params[4] = (unsigned long long)strides1;
31 params[5] = (unsigned long long)strides_output;
32 params[6] = (unsigned long long)num_dims;
33
34 int total_input0 = get_total_elements(num_dims, input0_dims);
35 int total_input1 = get_total_elements(num_dims, input1_dims);
36 int total_output = get_total_elements(num_dims, output_dims);
37
38 for(i = 0;i < total_input0;++i) {
39 input0[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
40 }
41
42 for(i = 0;i < total_input1;++i) {
43 input1[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
44 }
45 int core_mask = 0b1111;
46 fp_less_equal_s(input0, input1, output, params, core_mask);//调用汇编
47 return 0;
48}
私有存储版本:
-
void i8_less_equal_p(int8_t *Input0, int8_t *Input1, bool *output, long long *params)
-
void i16_less_equal_p(int16_t *Input0, int16_t *Input1, bool *output, long long *params)
-
void i32_less_equal_p(int32_t *Input0, int32_t *Input1, bool *output, long long *params)
-
void hp_less_equal_p(half *Input0, half *Input1, bool *output, long long *params)
-
void fp_less_equal_p(float *Input0, float *Input1, bool *output, long long *params)
-
void dp_less_equal_p(double *Input0, double *Input1, bool *output, long long *params)
C调用示例:
1//FT78NE示例
2#include <stdio.h>
3#include <less.h> // 假设头文件名为 less.h
4int main(int argc, char* argv[]) {
5 float *input0 = (float *)0x10010000;
6 float *input1 = (float *)0x10020000;
7 bool *output = (bool *)0x10030000;
8 int *strides0 = (int*)0x10050000;
9 int *strides1 = (int*)0x10060000;
10 int *strides_output = (int*)0x10070000;
11
12 int i = 0;
13
14 srand(seed++);
15
16 float f_min = -100.0;
17 float f_max = 100.0;
18
19 // same shape
20 int input0_dims[] = {4, 8, 17}; // 2x2
21 int input1_dims[] = {4, 8, 17}; // 2x2
22 int output_dims[] = {4, 8, 17}; // 2x2
23 int num_dims = 3;
24
25 unsigned long long params[9];
26 params[0] = (unsigned long long)input0_dims;
27 params[1] = (unsigned long long)input1_dims;
28 params[2] = (unsigned long long)output_dims;
29 params[3] = (unsigned long long)strides0;
30 params[4] = (unsigned long long)strides1;
31 params[5] = (unsigned long long)strides_output;
32 params[6] = (unsigned long long)num_dims;
33
34 int total_input0 = get_total_elements(num_dims, input0_dims);
35 int total_input1 = get_total_elements(num_dims, input1_dims);
36 int total_output = get_total_elements(num_dims, output_dims);
37
38 for(i = 0;i < total_input0;++i) {
39 input0[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
40 }
41
42 for(i = 0;i < total_input1;++i) {
43 input1[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
44 }
45
46 fp_less_equal_p(input0, input1, output, params);//调用汇编
47 return 0;
48}