API Reference

When applications that use Intel intrinsic instructions are ported from the x86 architecture to the Kunpeng architecture, the instructions need to be further developed because the names and functions of AArch64 instructions are different from those of x86 instructions. Therefore, a huge porting workload is generated. In this project, common AVX instructions are encapsulated as independent modules to reduce repeated development workload. Even when AVX instructions are replaced by the corresponding NEON or SVE SIMD instructions, the instruction names and functions remain unchanged. You can import related header files to application software to call corresponding instructions.

AVX Instruction Class

AVX instructions are Intel intrinsic instructions applicable to the x86 architecture only. The instruction names and functions implemented by the AVX2KI library remain identical to those of the AVX instructions.

Example

Create a testavx.cpp file.

Press i to enter the insert mode and add the following content to the file:

#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include "avx2ki.h"

void AvxExample()
{
    int32_t a[4] = {-5, 13, 4, -20};
    int32_t b[4] = {12, 3, 0, 7};
    int32_t c[4] = {0};
    __m128i t1 = _mm_load_epi32(a);
    __m128i t2 = _mm_load_epi32(b);
    __m128i dst = _mm_add_epi32(t1, t2);
    _mm_store_epi32(c, dst);
    printf("dst: %d %d %d %d\n", c[0], c[1], c[2], c[3]);
}

int main(void) {
    AvxExample();
    return 0;
}

Press Esc, type :wq!, and press Enter to save the settings and exit.
Compile the testavx.cpp file and specify the name of the output executable file as testavx.
```
g++ testavx.cpp -o testavx -I/usr/local/ksl/include -L/usr/local/ksl/lib -lavx2ki
```
Run the testavx executable file.
```
./testavx
```
The execution result is as follows:
```
dst: 7 16 4 -13
```

AVX Operator Overloading Class

AVX operator overloading performs overloaded operations between AVX vectors. These operations include addition, subtraction, multiplication, division, addition assignment, subtraction assignment, multiplication assignment, and division assignment.

Interface Definition

Addition:

__m128i operator + (const __m128i &a, const __m128i &b);

__m256i operator + (const __m256i &a, const __m256i &b);

__m256 operator + (const __m256 &a, const __m256 &b);

__m256d operator + (const __m256d &a, const __m256d &b);

__m512i operator + (const __m512i &a, const __m512i &b);

__m512 operator + (const __m512 &a, const __m512 &b);

__m512d operator + (const __m512d &a, const __m512d &b);

Subtraction:

__m128i operator - (const __m128i &a, const __m128i &b);

__m256i operator - (const __m256i &a, const __m256i &b);

__m256 operator - (const __m256 &a, const __m256 &b);

__m256d operator - (const __m256d &a, const __m256d &b);

__m512i operator - (const __m512i &a, const __m512i &b);

__m512 operator - (const __m512 &a, const __m512 &b);

__m512d operator - (const __m512d &a, const __m512d &b);

Multiplication:

__m128i operator * (const __m128i &a, const __m128i &b);

__m256i operator * (const __m256i &a, const __m256i &b);

__m256 operator * (const __m256 &a, const __m256 &b);

__m256d operator * (const __m256d &a, const __m256d &b);

__m512i operator * (const __m512i &a, const __m512i &b);

__m512 operator * (const __m512 &a, const __m512 &b);

__m512d operator * (const __m512d &a, const __m512d &b);

Division:

__m128i operator / (const __m128i &a, const __m128i &b);

__m256i operator / (const __m256i &a, const __m256i &b);

__m256 operator / (const __m256 &a, const __m256 &b);

__m256d operator / (const __m256d &a, const __m256d &b);

__m512i operator / (const __m512i &a, const __m512i &b);

__m512 operator / (const __m512 &a, const __m512 &b);

__m512d operator / (const __m512d &a, const __m512d &b);

AND:

__m128i operator & (const __m128i &a, const __m128i &b);

__m256i operator & (const __m256i &a, const __m256i &b);

__m256 operator & (const __m256 &a, const __m256 &b);

__m256d operator & (const __m256d &a, const __m256d &b);

__m512i operator & (const __m512i &a, const __m512i &b);

__m512 operator & (const __m512 &a, const __m512 &b);

__m512d operator & (const __m512d &a, const __m512d &b);

OR:

__m128i operator | (const __m128i &a, const __m128i &b);

__m256i operator | (const __m256i &a, const __m256i &b);

__m256 operator | (const __m256 &a, const __m256 &b);

__m256d operator | (const __m256d &a, const __m256d &b);

__m512i operator | (const __m512i &a, const __m512i &b);

__m512 operator | (const __m512 &a, const __m512 &b);

__m512d operator | (const __m512d &a, const __m512d &b);

XOR:

__m128i operator ^ (const __m128i &a, const __m128i &b);

__m256i operator ^ (const __m256i &a, const __m256i &b);

__m256 operator ^ (const __m256 &a, const __m256 &b);

__m256d operator ^ (const __m256d &a, const __m256d &b);

__m512i operator ^ (const __m512i &a, const __m512i &b);

__m512 operator ^ (const __m512 &a, const __m512 &b);

__m512d operator ^ (const __m512d &a, const __m512d &b);

Addition assignment:

__m128i &operator += (__m128i &a, const __m128i &b);

__m256i &operator += (__m256i &a, const __m256i &b);

__m256 &operator += (__m256 &a, const __m256 &b);

__m256d &operator += (__m256d &a, const __m256d &b);

__m512i &operator += (__m512i &a, const __m512i &b);

__m512 &operator += (__m512 &a, const __m512 &b);

__m512d &operator += (__m512d &a, const __m512d &b);

Subtraction assignment:

__m128i &operator -= (__m128i &a, const __m128i &b);

__m256i &operator -= (__m256i &a, const __m256i &b);

__m256 &operator -= (__m256 &a, const __m256 &b);

__m256d &operator -= (__m256d &a, const __m256d &b);

__m512i &operator -= (__m512i &a, const __m512i &b);

__m512 &operator -= (__m512 &a, const __m512 &b);

__m512d &operator -= (__m512d &a, const __m512d &b);

Multiplication assignment:

__m128i &operator *= (__m128i &a, const __m128i &b);

__m256i &operator *= (__m256i &a, const __m256i &b);

__m256 &operator *= (__m256 &a, const __m256 &b);

__m256d &operator *= (__m256d &a, const __m256d &b);

__m512i &operator *= (__m512i &a, const __m512i &b);

__m512 &operator *= (__m512 &a, const __m512 &b);

__m512d &operator *= (__m512d &a, const __m512d &b);

Division assignment:

__m128i &operator /= (__m128i &a, const __m128i &b);

__m256i &operator /= (__m256i &a, const __m256i &b);

__m256 &operator /= (__m256 &a, const __m256 &b);

__m256d &operator /= (__m256d &a, const __m256d &b);

__m512i &operator /= (__m512i &a, const __m512i &b);

__m512 &operator /= (__m512 &a, const __m512 &b);

__m512d &operator /= (__m512d &a, const __m512d &b);

AND assignment:

__m128i &operator &= (__m128i &a, const __m128i &b);

__m256i &operator &= (__m256i &a, const __m256i &b);

__m256 &operator &= (__m256 &a, const __m256 &b);

__m256d &operator &= (__m256d &a, const __m256d &b);

__m512i &operator &= (__m512i &a, const __m512i &b);

__m512 &operator &= (__m512 &a, const __m512 &b);

__m512d &operator &= (__m512d &a, const __m512d &b);

OR assignment:

__m128i &operator |= (__m128i &a, const __m128i &b);

__m256i &operator |= (__m256i &a, const __m256i &b);

__m256 &operator |= (__m256 &a, const __m256 &b);

__m256d &operator |= (__m256d &a, const __m256d &b);

__m512i &operator |= (__m512i &a, const __m512i &b);

__m512 &operator |= (__m512 &a, const __m512 &b);

__m512d &operator |= (__m512d &a, const __m512d &b);

XOR assignment:

__m128i &operator ^= (__m128i &a, const __m128i &b);

__m256i &operator ^= (__m256i &a, const __m256i &b);

__m256 &operator ^= (__m256 &a, const __m256 &b);

__m256d &operator ^= (__m256d &a, const __m256d &b);

__m512i &operator ^= (__m512i &a, const __m512i &b);

__m512 &operator ^= (__m512 &a, const __m512 &b);

__m512d &operator ^= (__m512d &a, const __m512d &b);

Parameters

Parameter	Description	Value Range	Input/Output
a	Reference to the instruction vector data type.	Unrestricted	Input/Output
b	Reference to the instruction vector data type.	Unrestricted	Input

Output Result

Instruction vector data of the same type as the input parameters.

The header file operatoroverload.h must be included to use the operator overloading functions.
Operator overloading functions support C++, while instruction functions support both C and C++.

Example

Create a testavx.cpp file with the following content.

#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include "avx2ki.h"
#include "operatoroverload.h"

void OperatorExample()
{
    int64_t a[2] = {-5, 13};
    int64_t b[2] = {12, 3};
    int64_t c[2] = {0};
    __m128i t1 = _mm_load_epi64(a);
    __m128i t2 = _mm_load_epi64(b);
    __m128i dst = t1 + t2;
    _mm_store_epi64(c, dst);
    printf("dst: %ld %ld\n", c[0], c[1]);
}

int main(void) {
    OperatorExample();
    return 0;
}

Perform compilation.

g++ testavx.cpp -o testavx -I/usr/local/ksl/include -L/usr/local/ksl/lib -lavx2ki

Output:

dst: 7 16

Kunpeng System Class

Obtains the AVX2KI version information.

Avx2kiResult AVX2KI_GetProductVersion(Avx2kiProVersion *packageInfo);

Parameters

Parameter	Description	Value Range	Input/Output
packageInfo	Pointer to the product information structure.	Not null	Input/Output

Return Value

Success: Returns AVX2KI_STS_NO_ERR.
Failure: Returns an error code.

Error Codes

Error Code	Description
AVX2KI_STS_NULL_PTR_ERR	`Avx2kiProVersion` is a null pointer.

Example

Create a testavx.cpp file with the following content.

#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include "avx2ki.h"
#include "operatoroverload.h"

void GetProductVersionExample()
{
    Avx2kiProVersion versionGet;
    Avx2kiResult result = AVX2KI_GetProductVersion(&versionGet);
    if (result == AVX2KI_STS_NO_ERR) {
        printf("Product Name: %s\n", versionGet.productName);
        printf("Product Version: %s\n", versionGet.productVersion);
        printf("Component Name: %s\n", versionGet.componentName);
        printf("Component Version: %s\n", versionGet.componentVersion);
        printf("Component AppendInfo: %s\n", versionGet.componentAppendInfo);
        printf("Software Name: %s\n", versionGet.softwareName);
        printf("Software Version: %s\n", versionGet.softwareVersion);
    }
}

int main(void) {
    GetProductVersionExample();
    return 0;
}

Perform compilation.

g++ testavx.cpp -o testavx -I/usr/local/ksl/include -L/usr/local/ksl/lib -lavx2ki

Output:

Product Name: Kunpeng Boostkit
Product Version: 23.0.RC2
Component Name: BoostKit-ksl
Component Version: 1.0.0
Component AppendInfo: gcc
Software Name: boostkit-ksl-avx2neon
Software Version: 1.0.0

The version number and build time in the example above are for reference only.