[THEME] CUDA Runtime Modernization

[jrhemstad]

Is this a duplicate?

• I confirmed there appear to be no duplicate issues for this request and that I agree to the Code of Conduct

Area

General CCCL

Is your feature request related to a problem? Please describe.

As a CUDA C++ developer, I would like more modern, idiomatic C++ abstractions for interacting with the various bits of the CUDA Runtime. The existing C APIs are overly verbose, error-prone, and overall not very delightful.

Describe the solution you'd like

I'd like idiomatic, modernized C++ abstractions for the most common functionality in the CUDA Runtime. We can group this functionality into three main areas:

Tasks

• cudax MVP Example #2159
• Additional cudax samples #2365

Memory Allocation

• [EPIC] Roadmap for cuda/memory_resource #1502

Kernel Launch & Programming Model

• cuda::launch kernel-launch API #2038
• [EPIC] CUDA Next asynchronous programming model #2041
• [EPIC] Modern CUDA Object Model #2837
• Stream ordered host execution #2872
• Expose runtime compilation of kernels #2875

Resources

• Add stream abstraction #2036
• Add event abstractions #2034
• CUDA Device/Context management #2039
• Modern versions of cudaMemset and cudaMemcpy #2000

Describe alternatives you've considered

No response

Additional context

No response

General infrastructure

• Add internal wrapper for CUDA driver APIs #2042
• Add exception type/hierarchy #2044
• Codify compute capabilities as statically and dynamically queryable traits #2097