如何并行化for循环

Question

I try to parallelize this for in for loop because I want to use in code with parallel stream in java . The problem is.. Every time I've try to do this I don't get the entire result. I mean... this code should rotate an image with some angle, but if I'm doing a parallelize I will receive just half of image rotated.

Answer 1

One issue could be the a, b, xx, yy variables being declared outside the for loop. It could be that different (interleaved) iterations overwrite the values of these variables. For instance,

1. iteration 0 writes to a and b
2. Iteration 0 stop and iteration 1 starts and write to a and b
3. Scheduler goes back to iteration 0 but the values of a and b are now those from iteration 1.

Answer 2

I think there are some problems with this implementation.

1) each of your thread from parallelstream is going to override the ImageIo.write the other thread did.

2) Also, forEach is unordered so you don't know which half will happen first.

I think your better bet is to use collect method from the Stream API. I can give some code if you need help. However if you give me a testable code in your question, I will write the parallel collect operation. Else, all the best.

---

Edit

I did some checking and found that your rotateImage is the culprit. The problem has nothing to do with parallelStream . There is lot of transformation required for each point (x1,y1) and it is bit too much for me to code right now. Your algorithm should be taking care of following:

1. you need to calculate new height and width of the image as when you rotate by 90 degree clockwise your destHeight = srcWidth and vice versa.
2. First you have to rotate your point around origin (x0,y0)
3. Not only that you need to translate your point (x1,y1) to the new coordinate system
   1. This new coordinate system has origin at top-left corner of image , hence you have to not only rotate but translate your point to this new coordinate system.
   2. Also, do note that in the rotation equations you wrote , you were probably assuming that x increases towards right and y increases towards top, however when you are writing the image to a pixel your xx increases towards right , but yy increases towards bottom (instead of top).

Once you have taken care of all this translations and rotations for each pixel along with the buffer's height and width, in rotateImage function, I think you can just call the parallelStream function on your locations stream and you won't have issues.

Some libraries like OpenGL can make this very easier for you. They take care of rotating and translating each pixel vector around the center with rotation vector and translation vector.