I was testing the new job system and noticed that calculations with NativeArrays are running slower than the same calculations with regular arrays. For tests I used three cases for velocity calculations:
1 - jobified IJobParallelFor
2 - the same as 1 but running single-threaded with positions and velocities stored in NativeArray
3 - the same as 2 but with regular arrays instead of native arrays.
The results on my computer for 100k points are:
1 - 60 FPS; 2 - 34 FPS; 3 - 79 FPS.
This shows that single-threaded calculations with regular arrays are about 2.3x faster than the same calculations using NativeArray. IJobParallelFor speeds up calculations by 1.8x compared to the single-threaded NativeArray usage. However, the IJobParallelFor speedup seems to be not enough to reach FPS when using regular arrays.
Do I miss something here?
The full code is bellow:
using System.Collections;
using System.Collections.Generic;
using UnityEngine;
using Unity.Jobs;
using Unity.Collections;
public class JobDemo1 : MonoBehaviour {
struct VelocityJob1 : IJobParallelFor
{
[ReadOnly]
public NativeArray<Vector3> velocity;
public NativeArray<Vector3> position;
public float deltaTime;
public void Execute(int i)
{
position[i] = position[i] + velocity[i] * deltaTime;
}
}
struct VelocityJob2
{
[ReadOnly]
public NativeArray<Vector3> velocity;
public NativeArray<Vector3> position;
public float deltaTime;
public void Execute(int i)
{
position[i] = position[i] + velocity[i] * deltaTime;
}
}
struct VelocityJob3
{
[ReadOnly]
public Vector3[] velocity;
public Vector3[] position;
public float deltaTime;
public void Execute(int i)
{
position[i] = position[i] + velocity[i] * deltaTime;
}
}
NativeArray<Vector3> position1;
NativeArray<Vector3> velocity1;
Vector3[] position3;
Vector3[] velocity3;
void Start(){
int n = 100000;
position1 = new NativeArray<Vector3>(n, Allocator.Persistent);
velocity1 = new NativeArray<Vector3>(n, Allocator.Persistent);
position3 = new Vector3[n];
velocity3 = new Vector3[n];
for (var i = 0; i < velocity1.Length; i++){
velocity1[i] = new Vector3(0, 10, 0);
velocity3[i] = new Vector3(0, 10, 0);
}
}
int jobMode = 1;
void Update(){
if(jobMode == 1){
Update1();
}
else if(jobMode == 2){
Update2();
}
else if(jobMode == 3){
Update3();
}
if(Input.GetKeyDown(KeyCode.A)){
jobMode = 1;
Debug.Log("jobMode "+jobMode);
}
else if(Input.GetKeyDown(KeyCode.B)){
jobMode = 2;
Debug.Log("jobMode "+jobMode);
}
else if(Input.GetKeyDown(KeyCode.C)){
jobMode = 3;
Debug.Log("jobMode "+jobMode);
}
}
void Update1(){
int processorCount = System.Environment.ProcessorCount;
var job = new VelocityJob1()
{
deltaTime = Time.deltaTime,
position = position1,
velocity = velocity1
};
JobHandle jobHandle = job.Schedule(position1.Length, processorCount);
jobHandle.Complete();
}
void Update2(){
var job = new VelocityJob2(){
deltaTime = Time.deltaTime,
position = position1,
velocity = velocity1
};
for(int i=0; i<position1.Length; i++){
job.Execute(i);
}
}
void Update3(){
var job = new VelocityJob3(){
deltaTime = Time.deltaTime,
position = position3,
velocity = velocity3
};
for(int i=0; i<position3.Length; i++){
job.Execute(i);
}
}
void OnApplicationQuit(){
position1.Dispose();
velocity1.Dispose();
}
}
. I am curious isn’t there possible to disable ENABLE_UNITY_COLLECTIONS_CHECKS while in Editor? Couldn’t find it so far. Though it would be a bit weird that in order to check performance each time would be needed to build the project.