[llvm] [Offload] Allow CUDA Kernels to use arbitrarily large shared memory (PR #145963)

[Joseph Huber via llvm-commits]

jhuber6 wrote:

> I wonder what the attribute this corresponds? Is it`CU_FUNC_ATTRIBUTE_MAX_DYNAMIC_SHARED_SIZE_BYTES` or `CU_DEVICE_ATTRIBUTE_MAX_SHARED_MEMORY_PER_BLOCK_OPTIN`?

I believe the former is the standard maximum and the latter is the maximum to can opt-in to. I think it's 48 KiB and 64 KiB respectively right now.

https://github.com/llvm/llvm-project/pull/145963