[llvm] [SROA] Allow canonicalizing homogeneous array allocas to vectors (PR #216456)

[Eli Friedman via llvm-commits]

efriedma-quic wrote:

>  I am not sure if it is fine under the security model of LLVM.

I don't think security is relevant here?  A compiler error on valid code isn't a security issue under our model.

> At least we allow memory usage to grow linearly with the integer bit width (e.g., during the legalization of BitInt).

We try to support very large arrays the best we can... we haven't done similar work for vectors.

------

> Fixes https://github.com/llvm/llvm-project/issues/164308.

This is still using the IR type (we pass `AI.getAllocatedType()` to compute TypePartitionTy); can we remove that dependency completely?

https://github.com/llvm/llvm-project/pull/216456