To get ALU/invocation:ALU Count: CVT + SFU + FMA instructions * 11 (11 cores)
invocations: $MaliCoreNonFragmentWarps * 16
Is the above calculation Right?
Don't need the x 11 - it cancels out with the same x11 you need for the warp count. This should do it:
((CVT + SFU + FMA) * 4) / ($MaliCoreNonFragmentWarps * 16)
Thanks Peter for replying.
1. Why we are multiplying by 4 the total instructions?
2. Are the CVT instructions count for all warps or all threads executed over the cores?
Yash Agrawal said:1. Why we are multiplying by 4 the total instructions?
Mali-G715 contains 4 processing units per core, the performance counters only count instructions in PU0.
Yash Agrawal said:2. Are the CVT instructions count for all warps or all threads executed over the cores?
All the instruction issues counters count per warp.