[WIP] Interface rework

[mzient]

trafficstars

This is a proposal for the interface rework that would satisfy the requirements outlined in #660

The changes are:

• add notion of memory_kind and allocation_order
• add a base class that mimics pmr::memory_resource, but is templated with respect to memory_kind and allocation_order
• add a base class stream_aware_memory_resource which covers the functionality previously implemented by device_memory_resource, but with configurable memory_kind
• make host_memory_resource an alias for memory_resource<memory_kind::host, allocation_order::host>
• make device_memory_resource derive from stream_aware_resource<memory_kind::device>
• add synchronous calls to stream_aware_memory_resource by inheriting from memory_resource<kind, allocation_order::host>
• add default implementation for synchronous allocations
• remove default stream argument
• add default_alignment<memory_kind> constant, which resolves to alignof(max_align_t) for host and 256B for others

Behavior changes:

• a call to stream_aware_memory_resource::allocate without a stream will cause synchronization (to run on default stream, user needs to explicitly specify the default stream)

Breaking changes:

• do_allocate and do_deallocate in stream_aware_memory_resource have been renamed to avoid an unfixable warning when compiling with NVCC (see below)
• do_alloc_async / do_dealloc_async now accept an alignment

NVCC warning problem:

Overriding a subset of virtual function's overloads causes a warning about hiding a virtual function. It's impossible to disable this warning or to mitigate it in the code, because the functions involved are private and therefore not accessible in the derived class - so we can't add dummy overrides or add using base::do_allocate.

[GPUtester]

Can one of the admins verify this patch?

[GPUtester]

Can one of the admins verify this patch?

[GPUtester]

Can one of the admins verify this patch?

[github-actions[bot]]

This PR has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this PR if it is no longer required. Otherwise, please respond with a comment indicating any updates. This PR will be labeled inactive-90d if there is no activity in the next 60 days.

[harrism]

@mzient can we close this?

[github-actions[bot]]

This PR has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this PR if it is no longer required. Otherwise, please respond with a comment indicating any updates. This PR will be labeled inactive-90d if there is no activity in the next 60 days.

[mzient]

@harrism This PR is here for visibility - it's the code used by DALI. If you think it should be closed, it can be, but that's probably where I'll keep adjusting RMM to the proposed libcu++ memory_resource interfaces.

[github-actions[bot]]

This PR has been labeled inactive-30d due to no recent activity in the past 30 days. Please close this PR if it is no longer required. Otherwise, please respond with a comment indicating any updates. This PR will be labeled inactive-90d if there is no activity in the next 60 days.

[github-actions[bot]]

This PR has been labeled inactive-90d due to no recent activity in the past 90 days. Please close this PR if it is no longer required. Otherwise, please respond with a comment indicating any updates.

[harrism]

Superseded by #1095. Closing.,