CANN: Use smart pointers to manage ACL objects

[hipudding]

Previously, ACL objects were managed via manual destruction, which led to multiple memory-leak issues during runtime.
This patch replaces manual memory management with smart pointers so that ACL objects are properly released and ownership is clearly defined.

Note that the ownership of an ACL object belongs to the function that creates it.
Other internal functions should operate on these ACL objects using raw pointers to avoid unintended ownership transfers.

Additionally, since aclTensorList automatically frees its contained aclTensor objects, any aclTensor added to a tensor list must release ownership to avoid double free operations.

This PR also removes the asynchronous task submission mechanism.
Due to changes in recent CANN versions, tiling time has significantly decreased. Even with a dual-thread submission model, the dispatch overhead still falls on the critical path, making async submission less beneficial.
Moreover, aclGraph support provides a much better path to reducing operator dispatch latency.

Make sure to read the contributing guidelines before submitting a PR

[Copilot]

Pull Request Overview

This PR refactors the CANN backend to replace manual memory management of ACL objects with smart pointers (std::unique_ptr with custom deleters), eliminating memory leaks and clarifying ownership semantics. Additionally, it removes the asynchronous task submission mechanism which is no longer beneficial due to CANN version improvements.

Key changes:

• Introduces smart pointer wrappers (acl_tensor_ptr, acl_scalar_ptr, acl_int_array_ptr, acl_tensor_list_ptr) with custom deleters for automatic ACL resource cleanup
• Removes task queue infrastructure (cann_task_queue, cann_task, aclnn_task, etc.)
• Replaces all manual aclDestroy* calls with automatic cleanup via smart pointers
• Replaces async helper functions with direct ACL API calls
• Removes the Doxyfile configuration (documentation generation file)

Reviewed Changes

Copilot reviewed 5 out of 7 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
acl_tensor.h	Defines smart pointer types and custom deleters for ACL objects; adds factory functions returning smart pointers
acl_tensor.cpp	Implements factory functions returning smart pointers; removes unreachable return statement
aclnn_ops.h	Updates function signatures to use smart pointers; removes task queue classes and async helper functions
aclnn_ops.cpp	Converts all ACL object creation/usage to smart pointers; removes manual resource cleanup calls
ggml-cann.cpp	Updates tensor creation calls to use smart pointers; replaces async helpers with direct ACL calls; removes task queue usage
common.h	Removes task queue and async task infrastructure; updates constructor
Doxyfile	Removes Doxygen configuration file (2579 lines deleted)

[hipudding]

Test case passwd：

2701/2701 tests passed
Backend CANN0: OK
Backend 2/3: CANN1
Skipping
Backend 3/3: CPU
Skipping
3/3 backends

[noemotiovon]

Thanks for your contribution! This PR is really exciting — we finally don’t have to manually release host-side ACL tensors anymore!

[noemotiovon]

Additionally, I think we still need to verify that there are currently no memory leaks.

[hipudding]

Additionally, I think we still need to verify that there are currently no memory leaks.

I ran Valgrind on llama-server and didn’t find any leaks. There were a few false positives, though, coming from inside CANN.