When I use rayQueryProceedEXT(query), the work register value is very high.
Work registers are used to store the traversal state and traversal results, so higher work register usage is expected due to how the current design functions.
HTH,
Pete
higher work register lead to half thread.If it can be optimized, could improve the performance of ray tracing?
More threads help with latency hiding, but half threads is enough to keep the core busy, so it's not guaranteed to be beneficial. A lot of complex gaming content has shaders using more than 32 registers, so I wouldn't worry too much.