Addfusion
带参数alpha的逐元素求和
\[output_i = in0_i + in1_i*alpha\]
- 输入：
in0 - 第一个输入数据地址。
in1 - 第二个输入数据地址。
alpha - 乘法因子。
size - 计算长度。
core_mask - 核掩码。
- 输出：
out - 计算结果地址。
- 支持平台：
FT78NEMT7004
备注
FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
MT7004 支持fp16, fp32, int16, int32, cplx64
共享存储版本:
-
void i8_addext_s(int8_t *in0, int8_t *in1, int8_t alpha, int8_t *out, int size, int core_mask)
-
void i16_addext_s(int16_t *in0, int16_t *in1, int16_t alpha, int16_t *out, int size, int core_mask)
-
void i32_addext_s(int32_t *in0, int32_t *in1, int32_t alpha, int32_t *out, int size, int core_mask)
-
void hp_addext_s(half *in0, half *in1, half alpha, half *out, int size, int core_mask)
-
void fp_addext_s(float *in0, float *in1, float alpha, float *out, int size, int core_mask)
-
void dp_addext_s(double *in0, double *in1, double alpha, double *out, int size, int core_mask)
-
void c64_addext_s(float *in0, float *in1, float alpha, float *out, int size, int core_mask)
-
void c128_addext_s(double *in0, double *in1, double alpha, double *out, int size, int core_mask)
C调用示例：
1//FT78NE示例
2#include <stdio.h>
3#include <addfusion.h>
4
5int main(int argc, char* argv[]) {
6 float *in0 = (float *)0xA0000000; //input在DDR空间
7 float *in1 = (float *)0xB0000000;
8 bool *out = (bool *)0xC0000000;
9 float alpha = 0.1;
10 int size = 1000;
11 int core_mask = 0xff;
12 fp_addext_s(in0, in1, alpha, out, size, core_mask);
13 return 0;
14}
私有存储版本:
-
void i8_addext_p(int8_t *in0, int8_t *in1, int8_t alpha, int8_t *out, int size)
-
void i16_addext_p(int16_t *in0, int16_t *in1, int16_t alpha, int16_t *out, int size)
-
void i32_addext_p(int32_t *in0, int32_t *in1, int32_t alpha, int32_t *out, int size)
-
void hp_addext_p(half *in0, half *in1, half alpha, half *out, int size)
-
void fp_addext_p(float *in0, float *in1, float alpha, float *out, int size)
-
void dp_addext_p(double *in0, double *in1, double alpha, double *out, int size)
-
void c64_addext_p(float *in0, float *in1, float alpha, float *out, int size)
-
void c128_addext_p(double *in0, double *in1, double alpha, double *out, int size)
C调用示例：
1//FT78NE示例
2#include <stdio.h>
3#include <addfusion.h>
4
5int main(int argc, char* argv[]) {
6 float *in0 = (float *)0x10000000; //input在L2空间
7 float *in1 = (float *)0x10001000;
8 bool *out = (bool *)0xC0000000;
9 int length = 1000;
10 fp_addext_s(in0, in1, alpha, out, size);
11 return 0;
12}
逐元素求和并计算relu激活函数
\[out_i = max(0, in0_i + in1_i)\]
- 输入：
in0 - 第一个输入数据地址。
in1 - 第二个输入数据地址。
size - 计算长度。
core_mask - 核掩码。
- 输出：
out - 计算结果地址。
- 支持平台：
FT78NEMT7004
备注
FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
MT7004 支持fp16, fp32, int16, int32, cplx64
共享存储版本:
-
void i8_addrelu_s(int8_t *in0, int8_t *in1, int8_t *out, int size, int core_mask)
-
void i16_addrelu_s(int16_t *in0, int16_t *in1, int16_t *out, int size, int core_mask)
-
void i32_addrelu_s(int32_t *in0, int32_t *in1, int32_t *out, int size, int core_mask)
-
void hp_addrelu_s(half *in0, half *in1, half *out, int size, int core_mask)
-
void fp_addrelu_s(float *in0, float *in1, float *out, int size, int core_mask)
-
void dp_addrelu_s(double *in0, double *in1, double *out, int size, int core_mask)
-
void c64_addrelu_s(float *in0, float *in1, float *out, int size, int core_mask)
-
void c128_addrelu_s(double *in0, double *in1, double *out, int size, int core_mask)
C调用示例：
1//FT78NE示例
2#include <stdio.h>
3#include <addfusion.h>
4
5int main(int argc, char* argv[]) {
6 float *in0 = (float *)0xA0000000; //input在DDR空间
7 float *in1 = (float *)0xB0000000;
8 bool *out = (bool *)0xC0000000;
9 int size = 1000;
10 int core_mask = 0xff;
11 fp_addrelu_s(in0, in1, out, size, core_mask);
12 return 0;
13}
私有存储版本:
-
void i8_addrelu_p(int8_t *in0, int8_t *in1, int8_t *out, int size)
-
void i16_addrelu_p(int16_t *in0, int16_t *in1, int16_t *out, int size)
-
void i32_addrelu_p(int32_t *in0, int32_t *in1, int32_t *out, int size)
-
void hp_addrelu_p(half *in0, half *in1, half *out, int size)
-
void fp_addrelu_p(float *in0, float *in1, float *out, int size)
-
void dp_addrelu_p(double *in0, double *in1, double *out, int size)
-
void c64_addrelu_p(float *in0, float *in1, float *out, int size)
-
void c128_addrelu_p(double *in0, double *in1, double *out, int size)
C调用示例：
1//FT78NE示例
2#include <stdio.h>
3#include <addfusion.h>
4
5int main(int argc, char* argv[]) {
6 float *in0 = (float *)0x10000000; //input在L2空间
7 float *in1 = (float *)0x10001000;
8 bool *out = (bool *)0xC0000000;
9 int size = 1000;
10 fp_addrelu_s(in0, in1, out, size);
11 return 0;
12}
逐元素求和并计算relu6激活函数
\[out_i = min(max(0, in0_i + in1_i), 6)\]
- 输入：
in0 - 第一个输入数据地址。
in1 - 第二个输入数据地址。
size - 计算长度。
core_mask - 核掩码。
- 输出：
out - 计算结果地址。
- 支持平台：
FT78NEMT7004
备注
FT78NE 支持int8, int16, int32, fp32, fp64, cplx64, cplx128
MT7004 支持fp16, fp32, int16, int32, cplx64
共享存储版本:
-
void i8_addrelu6_s(int8_t *in0, int8_t *in1, int8_t *out, int size, int core_mask)
-
void i16_addrelu6_s(int16_t *in0, int16_t *in1, int16_t *out, int size, int core_mask)
-
void i32_addrelu6_s(int32_t *in0, int32_t *in1, int32_t *out, int size, int core_mask)
-
void hp_addrelu6_s(half *in0, half *in1, half *out, int size, int core_mask)
-
void fp_addrelu6_s(float *in0, float *in1, float *out, int size, int core_mask)
-
void dp_addrelu6_s(double *in0, double *in1, double *out, int size, int core_mask)
-
void c64_addrelu6_s(float *in0, float *in1, float *out, int size, int core_mask)
-
void c128_addrelu6_s(double *in0, double *in1, double *out, int size, int core_mask)
C调用示例：
1//FT78NE示例
2#include <stdio.h>
3#include <addfusion.h>
4
5int main(int argc, char* argv[]) {
6 float *in0 = (float *)0xA0000000; //input在DDR空间
7 float *in1 = (float *)0xB0000000;
8 bool *out = (bool *)0xC0000000;
9 int size = 1000;
10 int core_mask = 0xff;
11 fp_addrelu6_s(in0, in1, out, size, core_mask);
12 return 0;
13}
私有存储版本:
-
void i8_addrelu6_p(int8_t *in0, int8_t *in1, int8_t *out, int size)
-
void i16_addrelu6_p(int16_t *in0, int16_t *in1, int16_t *out, int size)
-
void i32_addrelu6_p(int32_t *in0, int32_t *in1, int32_t *out, int size)
-
void hp_addrelu6_p(half *in0, half *in1, half *out, int size)
-
void fp_addrelu6_p(float *in0, float *in1, float *out, int size)
-
void dp_addrelu6_p(double *in0, double *in1, double *out, int size)
-
void c64_addrelu6_p(float *in0, float *in1, float *out, int size)
-
void c128_addrelu6_p(double *in0, double *in1, double *out, int size)
C调用示例：
1//FT78NE示例
2#include <stdio.h>
3#include <addfusion.h>
4
5int main(int argc, char* argv[]) {
6 float *in0 = (float *)0x10000000; //input在L2空间
7 float *in1 = (float *)0x10001000;
8 bool *out = (bool *)0xC0000000;
9 int size = 1000;
10 fp_addrelu6_s(in0, in1, out, size);
11 return 0;
12}