Reference \:
> * **Agner Fog's VCL (Vector Class Library Installation and Overview)**
https://www.youtube.com/watch?v=TKjYdLIMTrI&list=PLKK11LigqithMn_3ipTSSTZdbLHSh5Iy3&index=3

*This is an old note from the early days 2022\/06\/11. Please refer to other posts if you want to use intrinsics directly.*

## Vector Class Library

Compiler intrinsic can be used to take advantage on these registers. However, the syntax is difficult to read and maintain. VCL provide SIMD vectors of 128 bits, 256 bits, or 512 bits for int from 8 bits to 64 bits and 32/64 bits floating point operation. The instruction sets can emulate AVX512 using two 256 bits registers.

:::info
:bulb: **Compiler Settings**

Check the default option selected by the gcc compiler
***gcc -march=native -Q --help=target | grep march***

> https://stackoverflow.com/questions/52653025/why-is-march-native-used-so-rarely

Add default include path to C++ compiler
***export CPLUS_INCLUDE_PATH="/home/erebus/VCL"***

Additionally, the program needs to be compiled in 64 bits and the C++ standard should be set as c++17

***g++ -std=c++17 -m64 -march=native -o binaryname filename***

:::

## Examples

This program does a simple reduction to calculate the sum of all elements in an array which has size is not a multiple of the vector size.

Set the data size and calculate the part that is the multiple of vector size.

/*
Array is not always with data size that is a multiple of the vector size.
Therefore, increase the array size to fit into a vector register or handling the remaining data is required.
*/

#include <iostream>
#include <vectorclass.h>

const int datasize = 134;
const int vectorsize = 8;
const int regularpart = datasize & (-vectorsize); // = 128
// (AND - ing with -vectorsize will round down to the nearest
// lower multiple of vectorsize . This works only if vectorsize
// is a power of 2)

### Method \#1

Handling the remaining data with smaller vector size.

if (method_select == '0')
{
    std::cout << "method 0" << std::endl;

    int i;
    int mydata[datasize];
    for (i = 0; i < datasize; ++i) // initialize mydata
        mydata[i] = i;

    Vec8i sum1(0), temp;
    int sum = 0;

    // loop for 8 numbers at a time
    for (i = 0; i < regularpart; i += vectorsize)
    {
        temp.load(mydata + i); // load 8 elements
        sum1 += temp;          // add 8 elements
    }

    sum = horizontal_add(sum1); // sum of first 128 numbers

    if (datasize - i >= 4)
    {
        // get four more numbers
        Vec4i sum2;
        sum2.load(mydata + i);
        i += 4;
        sum += horizontal_add(sum2);
    }
    // loop for the remaining 2 numbers
    for (; i < datasize; i++)
    {
        sum += mydata[i];
    }
}

### Method \#2

Use partial load for the last vector

// Use partial load for the last vector

else if (method_select == '1')
{
    std::cout << "method 1" << std::endl;

    int i;
    int mydata[datasize];
    for (i = 0; i < datasize; ++i) // initialize mydata
        mydata[i] = i;

    Vec8i sum1(0), temp;

    // loop for 8 numbers at a time
    for (int i = 0; i < regularpart; i += vectorsize)
    {
        temp.load(mydata + i); // load 8 elements
        sum1 += temp;          // add 8 elements
    }
    // load the last 6 elements
    temp.load_partial(datasize - regularpart, mydata + regularpart);
    sum1 += temp;                   // add last 6 elements
    int sum = horizontal_add(sum1); // vector sum
}

### Method \#3

Read past the end of the array and ignore excess data

// method 3
// Read past the end of the array and ignore excess data

else if (method_select == '2') // potential hazard exist
{
    std::cout << "method 2" << std::endl;

    int i;
    int mydata[datasize];
    for (i = 0; i < datasize; ++i) // initialize mydata
        mydata[i] = i;

    Vec8i sum1(0), temp;

    // loop for 8 numbers at a time , reading 136 numbers
    for (int i = 0; i < datasize; i += vectorsize)
    {
        temp.load(mydata + i); // load 8 elements
        if (datasize - i < vectorsize)
        {
            // set excess data to zero
            // ( this may be faster than load_partial )
            temp.cutoff(datasize - i);
        }
        sum1 += temp; // add 8 elements
    }
    int sum = horizontal_add(sum1); // vector sum
}

### Method \#4

Make array bigger and set excess data to zero

// Make array bigger and set excess data to zero

else if (method_select == '3')
{
    std::cout << "method 3" << std::endl;

    const int arraysize = (datasize + vectorsize - 1) & (-vectorsize); // = 136
    int i;
    int mydata[arraysize];

    for (i = 0; i < datasize; ++i) // initialize mydata
        mydata[i] = i;

    // set excess data to zero
    for (i = datasize; i < arraysize; ++i)
        mydata[i] = 0;

    Vec8i sum1(0), temp;

    // loop for 8 numbers at a time , reading 136 numbers
    for (i = 0; i < arraysize; i += vectorsize)
    {
        temp.load(mydata + i); // load 8 elements
        sum1 += temp;          // add 8 elements
    }
    int sum = horizontal_add(sum1); // vector sum
}

## Matrix Multiplication Implementation

*2024\/08\/01*

### Implementation with Intrinsics

Please refer to the other post for this version

Compile with `-g`

./main 
Compute Time: 0.031000

Compile with `-O3`

./main 
Compute Time: 0.035000

https://hackmd.io/@Erebustsai/HkdXPx-rh

### Implementation with VLC2

Compile with `-g`

./main 
Compute Time: 0.096000

Compile with `-O3`

./main 
Compute Time: 0.036000

Using the `.load_partial()` might be what cause the performance difference since intrinsic version use mask. Further investigate required