[compiler-rt] [scudo] Calling initCache() in init() of SizeClassAllocatorLocalCache (PR #71427)

[via llvm-commits]

ChiaHungDuan wrote:

> IIRC the only reason we had it this way was to avoid dirtying the Cache for new threads that don't do allocations, which might be only really relevant for the Exclusive TSD model on heavy threaded applications. I don't think we ever gathered data about this, but might want to run a few tests (I usually ran the RPC benchmarks in g3 with Scudo compiled in).

Thanks for the context! Let me try that benchmark first and if it does have some impact, I'll try to adopt a different layout to minimize the size of unused dirty memory

https://github.com/llvm/llvm-project/pull/71427