Loop Reversal in C# Speeds Up app - c#

We are working on a video processing application using EmguCV and recently had to do some pixel level operation. I initially wrote the loops to go across all the pixels in the image as follows:
for (int j = 0; j < Img.Width; j++ )
{
for (int i = 0; i < Img.Height; i++)
{
// Pixel operation code
}
}
The time to execute the loops was pretty bad. Then I posted on the EmguCV forum and got a suggestion to switch the loops like this:
for (int j = Img.Width; j-- > 0; )
{
for (int i = Img.Height; i-- > 0; )
{
// Pixel operation code
}
}
I was very surprised to find that the code executed in half the time!
The only thing I can think of is the comparison that takes place in the loops each time accesses a property, which it no longer has to. Is this the reason for the speed up? Or is there something else? I was thrilled to see this improvement. And would love it if someone could clarify the reason for this.

The difference isn't the cost of branching, it's the fact that you are fetching an object property Img.Width and Img.Height in the inner loop. The optimizer has no way of knowing that these are constants for purposes of that loop.
You should get the same performance speedup by doing this.
const int Width = Img.Width;
const int Height = Img.Height;
for (int j = 0; j < Width; j++ )
{
for (int i = 0; i < Height; i++)
{
// Pixel operation code
}
}
Edit:
As Joshua Suggests, putting Width in the inner loop will have you walking through the memory sequentially, which will be better cache coherency, and might be faster. (depends on how big your bitmap is).
const int Width = Img.Width;
const int Height = Img.Height;
for (int i = 0; i < Height; i++)
{
for (int j = 0; j < Width; j++ )
{
// Pixel operation code
}
}

I assume you are using the System.Drawing.Image class? Looking at the implementation of .Width and .Height I see they do a function call into GDI+ (GdipGetImageHeight and GdipGetImageWidth in gdiplus.dll), which seems to be rather expensive.
By going backwards you make that call once, rather than in every iteration.

It's not the loop reversal that speeds things up -- it's the fact that you're accessing the Width and Height properties far fewer times.

It's because the CPUs are like hockey players, they go faster when going backward ;-)
More seriously:
This is not related in the direction of the loop in any way, but rather to the fact that the in the original construct, the loop control conditions implied dereferencing the Img object to index to its Width or Height property (for each and single iteration in the loops), whereby the second construct evaluates these properties only once.
Also, the fact that the new condition tests against the value 0, saves even the loading of an immediate value.
This probably explains the difference (assuming the work done inside the inner was relatively minimal, i.e. +/- the same as work to test an Object.Property, since you indicate a roughly 50% gain).
Edit:
see Michael Stum's answer, which indicates that the Img.Width/Height reference is even more costly than thought. As it sometimes happens with properties, the implementation of the object may run a significant amount of code to produce the value (for example it may do a bunch of math to get to the width, each time, rather than somehow caching it etc..). This seems to be the case with this Img object, hence the interest to do this only once (if you are sure that the value will remain constant for the duration of the loop logic).

Related

time complexity of Method

I want to learn about big-o, I hope someone can help me count operators in Method and tell me what the time complexity of this method is and teach me how to count. I tried to study on Youtube and I was a bit confused.
static void SelectionSort(int[] data)
{
int temp, min;
for (int i = 0; i < data.Length - 1 ; i++)
{
min = i;
for (int j = 0; j < data.Length; j++)
{
if (data[j] < data[min])
{
min = j;
}
temp = data[min];
data[min] = data[i];
data[i] = temp;
}
}
}
first of all this is a function not a method because a method is simply a function inside a class.
the time complexity of this algorithm is O(n^2) because of the double for loop that means that this algorithm will take around n^2 operations to be done.
for example if you input an array of length 10 it will make 100 steps that's not an exact number but it means a lot if you try an array of length 100 it will make 10000 steps that means that if would take 100 more time to finish.
so the less the time complexity the faster the algorithm is.
to learn about time complexity check this video it will help a lot-->
https://www.youtube.com/watch?v=6aDHWSNKlVw&t=6s
Time Complexity will be O(n^2) for this one. Reason is you have nested loop inside another loop.
The outer loop iterates n times giving an element to the inner loop which again loops n times, per one loop of the outer array.
https://adrianmejia.com/most-popular-algorithms-time-complexity-every-programmer-should-know-free-online-tutorial-course/#Bubble-sort

Fill sparse matrix parallel

I have a sparse-matrix from Extreme.Mathematics.LinearAlgebra like:
SparseMatrix<double> J = Matrix.CreateSparse<double>(amountI, amountJ);
Now I want to fill it in a parallel loop, since filling it in parallel shound be way faster.
Parallel.For(0, amountI, i =>
{
for (int j = 0; j < amountJ; j++)
J[i, j] = random.Next();
});
This gives me the error: out of range exception.
However, for a normal for loop, it works pretty fine.
for (int i = 0; i < amountI; i++)
{
for (int j = 0; j < amountJ; j++)
J[i, j] = random.Next();
}
Also, if I use a 2D array instead of a sparse matrix it works fine.
double[,] M = new double[amountI, amountJ];
Parallel.For(0, amountI, i =>
{
for (int j = 0; j < amountJ; j++)
M[i, j] = random.Next();
});
How do I achieve, to fill a sparse matrix in parallel without running into out of range exceptions?
I know this is a little bit late but better than nothing.
The Sparse matrix is something completely different from a normal array. It uses indices of rows and columns, to store only the non-zero values of the matrix. For further information I guess you should read the docs of Extreme.
In general: The backend of a sparse matrix has to reallocate memory every time you modify the underlying data. So given the case you loop normally, the allocations will happen in sync and you are fine. As soon as multiple threads try to modify the matrix, you will get into trouble, because on one thread you allocated new memory, which is immediately overwritten by another thread, giving you an exception on the original thread.
So: Parallel filling of sparse matrices wont work. I didn't find a proper library which gives the possibility of writing to a sparse matrix in parallel yet.

2D Array vs Array of Arrays in tight loop performance C#

I had a look and couldn't see anything quite answering my question.
I'm not exactly the best at creating accurate 'real life' tests, so i'm not sure if that's the problem here. Basically I want to create a few simple neural networks to create something to the effect of Gridworld. Performance of these neural networks will be critical and i dont want the hidden layer to be a bottleneck as much as possible.
I would rather use more memory and be faster, so I opted to use arrays instead of lists (due to lists having an extra bounds check over arrays). The arrays aren't always full, but because the if statement (check if the element is null) is the same until the end, it can be predicted and there is no performance drop from that at all.
My question comes from how I store the data for the network to process. I figured due to 2D arrays storing all the data together it would be better cache wise and would run faster. But from my mock up test that an array of arrays performs much better in this scenario.
Some Code:
private void RunArrayOfArrayTest(float[][] testArray, Data[] data)
{
for (int i = 0; i < testArray.Length; i++) {
for (int j = 0; j < testArray[i].Length; j++) {
var inputTotal = data[i].bias;
for (int k = 0; k < data[i].weights.Length; k++) {
inputTotal += testArray[i][k];
}
}
}
}
private void Run2DArrayTest(float[,] testArray, Data[] data, int maxI, int maxJ)
{
for (int i = 0; i < maxI; i++) {
for (int j = 0; j < maxJ; j++) {
var inputTotal = data[i].bias;
for (int k = 0; k < maxJ; k++) {
inputTotal += testArray[i, k];
}
}
}
}
These are the two functions that are timed. Each 'creature' has its own network (The first for loop), each network has hidden nodes (The second for loop) and i need to find the sum of the weights for each input (The third loop). In my test i stripped it so that it's not really what i am doing in my actual code, but the same amount of loops happen (The data variable would have it's own 2D array, but i didn't want to possibly skew the results). From this i was trying to get a feel for which one is faster, and to my surprise the array of arrays was.
Code to start the tests:
// Array of Array test
Stopwatch timer = Stopwatch.StartNew();
RunArrayOfArrayTest(arrayOfArrays, dataArrays);
timer.Stop();
Console.WriteLine("Array of Arrays finished in: " + timer.ElapsedTicks);
// 2D Array test
timer = Stopwatch.StartNew();
Run2DArrayTest(array2D, dataArrays, NumberOfNetworks, NumberOfInputNeurons);
timer.Stop();
Console.WriteLine("2D Array finished in: " + timer.ElapsedTicks);
Just wanted to show how i was testing it. The results from this in release mode give me values like:
Array of Arrays finished in: 8972
2D Array finished in: 16376
Can someone explain to me what i'm doing wrong? Why is an array of arrays faster in this situation by so much? Isn't a 2D array all stored together, meaning it would be more cache friendly?
Note i really do need this to be fast as it needs to sum up hundreds of thousands - millions of numbers per frame, and like i said i don't want this is be a problem. I know this can be multi threaded in the future quite easily because each network is completely separate and even each node is completely separate.
Last question i suppose, would something like this be possible to run on the GPU instead? I figure a GPU would not struggle to have much larger amounts of networks with much larger numbers of input/hidden neurons.
In the CLR, there are two different types of array:
Vectors, which are zero-based, single-dimensional arrays
Arrays, which can have non-zero bases and multiple dimensions
Your "array of arrays" is a "vector of vectors" in CLR terms.
Vectors are significantly faster than arrays, basically. It's possible that arrays could be optimized further in later CLR versions, but I doubt that there'll get the same amount of love as vectors, as they're so relatively rarely used. There's not a lot you can do to make CLR arrays faster. As you say, they'll be more cache friendly, but they have this CLR penalty.
You can improve your array-of-arrays code already, however, by only performing the first indexing operation once per row:
private void RunArrayOfArrayTest(float[][] testArray, Data[] data)
{
for (int i = 0; i < testArray.Length; i++) {
// These don't change in the loop below, so extract them
var row = testArray[i];
var inputTotal = data[i].bias;
var weightLength = data[i].weights.Length;
for (int j = 0; j < row.Length; j++) {
for (int k = 0; k < weightLength; k++) {
inputTotal += row[k];
}
}
}
}
If you want to get the cache friendliness and still use a vector, you could have a single float[] and perform the indexing yourself... but I'd probably start off with the array-of-arrays approach.

Accessing Height, Width properties of a control is slower in C#

I found some surprising results while drawing some shapes in .Net Compact Framework.
Method1 and Method2 draws some rectangles but Method1 is faster then Method2, here is the code:
Method1:
int height = Height;
for (int i = 0; i < data.Length; i++)
{
barYPos = Helper.GetPixelValue(Point1, Point2, data[i]);
barRect.X = barXPos;
barRect.Y = barYPos;
barRect.Height = height - barYPos;
//
//rects.Add(barRect);
_gBmp.FillRectangle(_barBrush, barRect);
//
barXPos += (WidthOfBar + DistanceBetweenBars);
}
Method2:
for (int i = 0; i < data.Length; i++)
{
barYPos = Helper.GetPixelValue(Point1, Point2, data[i]);
barRect.X = barXPos;
barRect.Y = barYPos;
barRect.Height = Height - barYPos;
//
//rects.Add(barRect);
_gBmp.FillRectangle(_barBrush, barRect);
//
barXPos += (WidthOfBar + DistanceBetweenBars);
}
The only difference between two is in Method1 I am storing the Height of the control in a local variable.
Can anyone please explain the reason and some guidelines for drawings in .Net Compact Framework?
Method 2 is slower because you're accessing the Height property at each iteration of your for loop. This property may cause some time consuming calculations, and putting it in a local variable outside the loop acts as a cache.
A call to a property in C# has more associated cost than it would directly accessing a variable in memory; as properties are generated as methods with a backing field in the background (and/or worse.. perhaps it queries something else!)
If your application is indeed single threaded and you can afford to cache it, do so. Avoid properties in tight loops.
I believe thats because you access Height- data.Lenght amoutns of times. and in the first methood you only initalize it once.

Can this code be optimised?

I have some image processing code that loops through 2 multi-dimensional byte arrays (of the same size). It takes a value from the source array, performs a calculation on it and then stores the result in another array.
int xSize = ResultImageData.GetLength(0);
int ySize = ResultImageData.GetLength(1);
for (int x = 0; x < xSize; x++)
{
for (int y = 0; y < ySize; y++)
{
ResultImageData[x, y] = (byte)((CurrentImageData[x, y] * AlphaValue) +
(AlphaImageData[x, y] * OneMinusAlphaValue));
}
}
The loop currently takes ~11ms, which I assume is mostly due to accessing the byte arrays values as the calculation is pretty simple (2 multiplications and 1 addition).
Is there anything I can do to speed this up? It is a time critical part of my program and this code gets called 80-100 times per second, so any speed gains, however small will make a difference. Also at the moment xSize = 768 and ySize = 576, but this will increase in the future.
Update: Thanks to Guffa (see answer below), the following code saves me 4-5ms per loop. Although it is unsafe code.
int size = ResultImageData.Length;
int counter = 0;
unsafe
{
fixed (byte* r = ResultImageData, c = CurrentImageData, a = AlphaImageData)
{
while (size > 0)
{
*(r + counter) = (byte)(*(c + counter) * AlphaValue +
*(a + counter) * OneMinusAlphaValue);
counter++;
size--;
}
}
}
To get any real speadup for this code you would need to use pointers to access the arrays, that removes all the index calculations and bounds checking.
int size = ResultImageData.Length;
unsafe
{
fixed(byte* rp = ResultImageData, cp = CurrentImageData, ap = AlphaImageData)
{
byte* r = rp;
byte* c = cp;
byte* a = ap;
while (size > 0)
{
*r = (byte)(*c * AlphaValue + *a * OneMinusAlphaValue);
r++;
c++;
a++;
size--;
}
}
}
Edit:
Fixed variables can't be changed, so I added code to copy the pointers to new pointers that can be changed.
These are all independent calculations so if you have a multicore CPU you should be able to gain some benefit by parallelizing the calculation. Note that you'd need to keep the threads around and just hand them work to do since the overhead of thread creation would probably make this slower rather than faster if the threads are recreated each time.
The other thing that may work is farming the work off to the graphics processor. Look at this question for some ideas, for example, using Accelerator.
An option would be to use unsafe code: fixing the array in memory and use pointer operations. I doubt the speed increase will be that dramatic though.
One note: how are you timing? If you are using DateTime then be aware that this class has poor resolution. You should add an outer loop and repeat the operation say ten times -- I bet the result is less than 110ms.
for (int outer = 0; outer < 10; ++outer)
{
for (int x = 0; x < xSize; x++)
{
for (int y = 0; y < ySize; y++)
{
ResultImageData[x, y] = (byte)((CurrentImageData[x, y] * AlphaValue) +
(AlphaImageData[x, y] * OneMinusAlphaValue));
}
}
}
Since it appears that each cell in the matrix is calculated entirely independent of the others. You may want to look into having more than one thread handle this. To avoid the cost of creating threads you could have a thread pool.
If the matrix is of sufficient size, it could be a very nice speed gain. On the other hand, if it is too small, it may not help (even hurt). Worth a try though.
An example (pseudo code) could be like this:
void process(int x, int y) {
ResultImageData[x, y] = (byte)((CurrentImageData[x, y] * AlphaValue) +
(AlphaImageData[x, y] * OneMinusAlphaValue));
}
ThreadPool pool(3); // 3 threads big
int xSize = ResultImageData.GetLength(0);
int ySize = ResultImageData.GetLength(1);
for (int x = 0; x < xSize; x++) {
for (int y = 0; y < ySize; y++) {
pool.schedule(x, y); // this will add all tasks to the pool's work queue
}
}
pool.waitTilFinished(); // wait until all scheduled tasks are complete
EDIT: Michael Meadows mentioned in a comment that plinq may be a suitable alternative: http://msdn.microsoft.com/en-us/magazine/cc163329.aspx
I'd recommend running a few empty tests to figure out what your theoretical bounds are. For example, take out the calculation from inside the loop and see how much time is saved. Try replacing the double loop with a single loop that runs the same number of times and see how much time that saves. Then you can be sure you are going down the right path for optimization (the two paths I see are flattening the double loop into a single loop and working with the multiplication [maybe using a lookup table would be faster]).
Just real quick, you can get an optimization by looping in reverse and comparing against 0. Most CPUs have a fast op for comparison to 0.
E.g.
int xSize = ResultImageData.GetLength(0) -1;
int ySize = ResultImageData.GetLength(1) -1; //minor optimization suggested by commenter
for (int x = xSize; x >= 0; --x)
{
for (int y = ySize; y >=0; --y)
{
ResultImageData[x, y] = (byte)((CurrentImageData[x, y] * AlphaValue) +
(AlphaImageData[x, y] * OneMinusAlphaValue));
}
}
See http://dotnetperls.com/Content/Decrement-Optimization.aspx
You are probably suffering from Boundschecking. Like Jon Skeet states, a jagged array instead of a multidimensional (that is data[][] instead of data[,]) will be faster, strange as that may seem.
The compiler will optimize
for (int i = 0; i < data.Length; i++)
by eliminating the per-element range check. But it's some kind of special case, it won't do the same for Getlength().
For the same reason, caching or hoisting the Length property (putting it in a variable like xSize) also used to be a bad thing though I haven't been able to verify that with Framework 3.5
Try swapping the x and y for loops for a more linear memory access pattern and (thus) less cache misses, like so.
int xSize = ResultImageData.GetLength(0);
int ySize = ResultImageData.GetLength(1);
for (int y = 0; y < ySize; y++)
{
for (int x = 0; x < xSize; x++)
{
ResultImageData[x, y] = (byte)((CurrentImageData[x, y] * AlphaValue) +
(AlphaImageData[x, y] * OneMinusAlphaValue));
}
}
If you are using LockBits to get at the image buffer, you should loop through y in the outer loop and x in the inner loop as that is how it is stored in memory (by row, not column). I would say that 11ms is pretty darn fast though...
Does the image data have to be stored in a multi-dimensional (rectangular) array? If you use jagged arrays instead, you may well find the JIT has more optimizations available (including removing the bounds checking).
If CurrentImageData and/or AlphaImageData don't change every time you run your code snippet, you could store the product prior to running the code snippet you show and avoid that multiplication in your loops.
Edit: Another thing I just thought of: Sometimes int operations are quicker than byte operations. Offset this with your processor cache utilization (you'll increase the data size considerably and stand a greater risk of a cache miss).
442,368 additions and 884,736 multiplications for the calculation i would think 11ms is actually extremely slow on a modern CPU.
while i don't know much about the specifics of .net i do know high speed calculation is not its strong suit. In the past i've built java apps with similar problems, i've always used C libraries to do the image / audio processing.
coming from a hardware perspective you want to make sure the memory accesses are sequential, that is step through the buffer in the order it exists in memory. you also may need to reorder this such that the compiler takes advantage of available instructions such as SIMD. How to approach this will end up being dependent on your compiler and i can't help on vs.net.
on an embedded DSP i would break out
(AlphaImageData[x, y] * OneMinusAlphaValue) and (CurrentImageData[x, y] * AlphaValue) and use SIMD instructions to calculate buffers, possibly in parallel before performing the addition. perhaps doing small enough chunks to keep the buffers in cache on the cpu.
i believe anything you do will require more direct access to the memory/cpu than .net allows.
You may also want to take a look at the Mono runtime and its Simd extensions. Perhaps some of your calculations can make use of the SSE acceleration as I gather that you basically do vector calculations (I don't know up to which vector size there is acceleration for multiplication but there is for some sizes)
(Blog post announcing Mono.Simd: http://tirania.org/blog/archive/2008/Nov-03.html)
Of course, that wouldn't work on Microsoft .NET but maybe you are interested in some experimentation.
Interestingly, image data is frequently pretty similar, meaning that the calculations are likely very repetitive. Have you explored doing a lookup table for the calculations? So any time 0.8 was multiplied by 128 - value[80,128] which you've precalculated to 102.4, you simply looked that up? You're basically trading memory space for CPU speed, but it could work for you.
Of course, if your image data has too high a resolution (and goes to too significant a digit), this may not be practical.

Categories

Resources