Instruction hit count is same within kernel asm?
hgtsoi opened this issue · comments
How the RGP profiles instruction hit count?
All the assembly code lines got the same hit count. The num of hit count might vary for different kernels.
Does it mean how many times the whole kernel being executed? Or it just equals to that then.
When you're looking at the Instruction timing view, you are looking at a single event -- a single draw call or dispatch. The hit counts shown indicate how many times instructions were executed for that particular draw call or event. If a kernel is used by more than one event, you would need to select each event (using the "Event" drop down combo box at the top of the view) to view the instruction timing data.
If a kernel has no branching/looping, then I would expect each instruction to show the same count for a given event, however if there is branching/looping within the kernel, then it is expected that different instruction blocks would have different hit counts.
Also, as mentioned in the RGP documentation: "Instruction Timing information is only sampled from some of the compute units on a single shader engine of the GPU.", so the hit counts shown only reflect the wavefronts which run on the sampled compute units.
Closing this issue, and there doesn't seem to be a bug here. If I am mistaken and you are seeing a bug, please reopen with more details and a test case. Thanks!