Implement cuGraphExecUpdate_v2

[zluda-violet]

No description provided.

[Copilot]

Pull Request Overview

This PR implements the cuGraphExecUpdate_v2 CUDA function by adding it to the CUDA function declarations and providing a HIP backend implementation. This function updates an executable graph with a new graph and provides detailed information about any update failures.

• Adds cuGraphExecUpdate_v2 to the CUDA function declarations
• Implements the HIP backend wrapper that converts HIP graph update results to CUDA equivalents
• Adds necessary type mappings for CUgraphExecUpdateResultInfo

Reviewed Changes

Copilot reviewed 3 out of 3 changed files in this pull request and generated 3 comments.

File	Description
zluda_common/src/lib.rs	Adds CUgraphExecUpdateResultInfo type mapping for CUDA to HIP conversion
zluda/src/lib.rs	Declares the cuGraphExecUpdate_v2 function in the CUDA function list
zluda/src/impl/graph.rs	Implements the HIP backend for cuGraphExecUpdate_v2 with result conversion logic

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[vosen]

Is there a way to use FromCuda instead of transmute here and below?
In case we change the definition of CUgraphExec to be more complicated than transmute of hipGraphExec_t

[zluda-violet]

How's the new code look? It still uses transmute but in a wrapper struct returned from FromCuda

[vosen]

This is very involved, how about something like this (with an appropriate comment why we do these acrobatics instead of mem::transmute):

    let error_node: *mut hipGraphNode_t = FromCuda::from_cuda(std::ptr::from_mut(&mut result_info.errorNode))?;
    let error_from_node: *mut hipGraphNode_t = FromCuda::from_cuda(std::ptr::from_mut(&mut result_info.errorFromNode))?;
    *error_node = h_error_node;
    *error_from_node = h_error_node;

[zluda-violet]

Sure, done.

[vosen]

This looks wrong to me, my reading is that

• We load bits from result_info.errorNode into h_error_node on stack
• hipGraphExecUpdate writes into h_error_node on stack
• We copy bits from result_info.errorNode (without writing into it) into result_info.errorFromNode

[zluda-violet]

Yeah, you're correct. The code you wrote works I think, I just misread it before.