Greater

逐元素比较两个输入张量，判断 input1 的元素是否大于 input2 的对应元素。支持广播。

\[\text{output}_i = (\text{input1}_i > \text{input2}_i)\]

输出一个布尔张量，如果比较结果为真，则对应元素为 true，否则为 false。

输入：

  • input1 - 第一个输入张量的数据地址。

  • input2 - 第二个输入张量的数据地址。

  • params - 参数打包数组，格式如下：

    • input0_dims - 第一个输入张量 input1 的维度。

    • input1_dims - 第二个输入张量 input2 的维度。

    • output_dims - 输出张量的维度。

    • strides0 - 第一个输入张量 input1 的步长。

    • strides1 - 第二个输入张量 input2 的步长。

    • strides_output - 输出张量的步长。

    • num_dims - 维度数量。用于广播判断。

  • core_mask - 核掩码。

输出：

  • output - 写入比较结果的布尔张量。

支持平台：

FT78NE MT7004

备注

  • FT78NE 支持fp32, fp64, int8, int16, int32

  • MT7004 支持fp16, fp32, int16, int32

共享存储版本:

void fp_greater_s(float *input1, float *input2, bool *output, long long *params, int core_mask)
void hp_greater_s(half *input1, half *input2, bool *output, long long *params, int core_mask)
void dp_greater_s(double *input1, double *input2, bool *output, long long *params, int core_mask)
void i32_greater_s(int32_t *input1, int32_t *input2, bool *output, long long *params, int core_mask)
void i16_greater_s(int16_t *input1, int16_t *input2, bool *output, long long *params, int core_mask)
void i8_greater_s(int8_t *input1, int8_t *input2, bool *output, long long *params, int core_mask)

C调用示例：

 1//FT78NE示例
 2#include <stdio.h>
 3#include <greater.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*)0x85000000;
 9    int *strides1 = (int*)0x86000000;
10    int *strides_output = (int*)0x87000000;
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[] = {64, 16, 63};  // 2x2
21    int input1_dims[] = {1, 1, 1};  // 2x2
22    int output_dims[] = {64, 16, 63};      // 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    int core_id = get_core_id();
39    int logic_core_id = GetLogicCoreId(core_mask, core_id);
40    int core_num = GetCoreNum(core_mask);
41
42    for(i = 0;i < total_input0;++i) {
43        input0[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
44    }
45
46    for(i = 0;i < total_input1;++i) {
47        input1[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
48    }
49
50    fp_greater_s(input0, input1, output, params, core_mask);//调用汇编
51    return 0;
52}

私有存储版本:

void fp_greater_p(float *input1, float *input2, bool *output, long long *params)
void hp_greater_p(half *input1, half *input2, bool *output, long long *params)
void i32_greater_p(int32_t *input1, int32_t *input2, bool *output, long long *params)
void i16_greater_p(int16_t *input1, int16_t *input2, bool *output, long long *params)
void i8_greater_p(int8_t *input1, int8_t *input2, bool *output, long long *params)

C调用示例：

 1//FT78NE示例
 2#include <stdio.h>
 3#include <greater.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[] = {1, 1, 1};  // 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
39    for(i = 0;i < total_input0;++i) {
40        input0[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
41    }
42
43    for(i = 0;i < total_input1;++i) {
44        input1[i] = f_min + ((float)rand() / (float)RAND_MAX) * (f_max - f_min);
45    }
46
47    fp_greater_p(input0, input1, output, params);//调用汇编
48    return 0;
49}