DCPerf v2-beta : `Rebatch` benchmark memcpy may be optimized out with Clang

[Dhruvaraj-007]

Hi,

We observed that the memcpy operations in packages/ai_wdl/rebatch/rebatchBench.cpp appear to be optimized out when the benchmark is built with Clang instead of GCC.

What would Meta/DCPerf like to do?

Thanks

[LDong-Arm]

@YifanYuan3 @ahmadelyoussef
Could you please have a look? rebatchBench.cpp mentions that it uses checksums to prevent work from being optimized out, but the code doesn't seem to do any checksums? It does cause work to be optimized out when we modify the install script to use clang instead of gcc.