french toast
Veteran
Running completely out of L1 icache/dcache is one problem, probably the most major one, but Dhrystone has a lot of other problems.. Note that the small footprint also makes it fit unrealistically well into other CPU buffers like the TLB and BTB.
But it has other problems. It's a mess of pretty much random code rather than some adaptation of real programs working on real input, it interacts badly with compiler optimizations, and it spends too much time in library code that tends to get tweaked to perform well with it. It's also poorly documented or standardized, and has various different versions floating around. On the flip side, it's very small and portable so it's easy for a project to grab the C code and slap it in their program, calling it a CPU integer performance test (and Whetstone for FPU). BOINC for instance does exactly this. So I don't know for sure what AnTuTu does but I doubt they wrote their own benchmark.
Whetstone actually came first.. I don't really know anything about it but the interesting criteria for an FP specific program are different so it might not be as useless.
Thanks for breakdown.
Hopefully that is not the foundation for antutu cpu benchmark then..it is odd that they dont specify just what their benchmark measures. .its been round long enough now.