If we look at the LLVM MCA Documentation, for the sample example that they’ve run, “Bottleneck Analysis” says that IPC is limited by resource pressure and not by data dependencies. Here, they run the same block of code for 500 iterations.
But for the same block of code, when it is run for 300 iterations, they say the IPC is mostly limited by data dependencies and not resource pressure. This can be seen at the end of “Extra Statistics to Further Diagnose Performance Issues”.
How are they arriving at 2 different conclusions for different iterations ? Shouldn’t the result be the same? Please help me make sense of the statistics presented.
Also, if you try to run LLVM-MCA for the same example code, for 300 iterations, it presents a similar number.
Result for 300 iterations
Hemant Kumar is a new contributor to this site. Take care in asking for clarification, commenting, and answering.
Check out our Code of Conduct.
3