_{Comment @coderabbitai help to get the list of available commands and usage tips.}

Greptile Summary

This PR adds a new C++ and CUDA Workflows user guide page (cpp_cuda_workflows.rst) that consolidates public extension entry points — native snippets via wp.func_native, AOT compilation, APIC replay, and native library headers. The differentiability.rst page is trimmed to autodiff-specific guidance and links out to the new hub, and cross-links are added in basics.rst, codegen.rst, debugging.rst, and allocators.rst. The func_native docstring in context.py is also significantly expanded.

• New cpp_cuda_workflows.rst covers four extension workflows with runnable examples, inline PTX with assertion checks, and a limitations list; the page is wired into index.rst between the configuration and debugging entries.
• differentiability.rst's Custom Native Functions section is refactored: the long CUDA shared-memory and atomic-counter demos are removed and replaced with a focused CPU-capable axpy example that retains the skipif: wp.get_cuda_device_count() == 0 guard inherited from the original section.
• context.py's func_native docstring is rewritten with three code examples, a Note block, and a See Also cross-reference.

Confidence Score: 5/5

This is a documentation-only change with no runtime code modifications; it is safe to merge.

All changes are documentation and docstrings. The new page's testcode blocks are either guarded by skipif or verified with np.testing.assert_array_equal. All cross-references resolve to existing targets. The only finding is a skipif guard on a CPU-capable doctest that is more restrictive than necessary — it doesn't cause any incorrect behavior, just reduces test coverage on CPU-only CI.

No files require special attention.

Important Files Changed

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[User needs non-Python integration] --> B{What kind?}
    B -->|Embed CUDA/C++ in kernel| C[cpp_cuda_workflows.rst\nNative Snippets section]
    B -->|Compile kernels ahead of time| D[cpp_cuda_workflows.rst\nAOT Workflows section]
    B -->|Replay from C++ without Python| E[cpp_cuda_workflows.rst\nAPI Capture Replay section]
    C --> F[wp.func_native in context.py\nExpanded docstring + examples]
    C --> G[differentiability.rst\nAdjoint / replay_snippet guidance]
    D --> H[codegen.rst\nahead_of_time_compilation_workflows label]
    E --> I[runtime.rst\napic_save_load label]
    F --> J[basics.rst / debugging.rst\nAllocators.rst cross-links]

_{Reviews (7): Last reviewed commit: "Document C++ and CUDA workflows" | Re-trigger Greptile}