问题
How can I efficiently calculate the sum of all pixels in an image, by using a HSLS pixel shader? I'm interested in Pixel Shader 2.0, that I could invoke as a WPF shader effect.
回答1:
There is a much simpler solution that doesn't use shaders: load the image as a texture, create a mipmap chain and read back the value of the last mipmap (1x1 pixel). This trick is used in games extensively to calculate, for example, the average brigthness of a scene (in order to apply HDR tonemapping). It's a great trick if you value simplicity and efficiency over accuracy.
回答2:
Assuming that you want to sum r/g/b channel values in image (and assuming that you can destroy original pixels) - here is my partial solution:
- Each 10th pixel would calculate average color of 10 neighboring pixels and put this averaged color on image.
- Then in CPU side sum
Pix_Value*10
of each 10th pixel.
In this case we get about 10x speed-up compared to summing pixel values only in CPU side.
回答3:
Effects in WPF are much more suited to producing visual result. It sounds like you want to use them for a different type of calculation, to do this you would need to treat the image result as data this would need RenderTargetBitmap which would be done in software anyway. You might want to look at these projects designed for GPGPU.
Accelerator
Brahma
OpenTK I dont know the state of the OpenCL bindings in OpenTK. Other technologies such as DirectCompute and CUDA might be worth a look too.
来源：https://stackoverflow.com/questions/2944290/efficient-pixel-shader-sum-of-all-pixels