i'm tring to make reduction work on CAL and check its performance.
Is there a way of doing gather and scatter in the same kernel written in IL?
There are samples of gather and scattering in the SDK. But both of them use g as parameter name, how to seperate them if i want to use both of them?
On RV8xx I believe you can use UAVs to seperate input and output. However RV7xx supports only 1 UAV.
If you care about backward compatibility, you need to seperate the input and output area of the global buffer g, like a ping-pong buffer.
You can also read data from g, manipulate it and store the results back to g, overriding the input.
Besides, you do not necessarily "gather" from g, samplers can also access arbitrary locations.