Performance Optimizations for Tape Operations

[rghouzra]

Title

Optimize memory usage in Tape operations

Description

While profiling our tape operations, I noticed some opportunities to reduce memory allocations and copying. This PR makes targeted optimizations to callback handling and adjoint resets without changing the core behavior.

What Changed

• Changed insertCallback() to use emplace_back() to construct elements in-place
• Updated getAndResetOutputAdjoint() to use move semantics for derivative transfers

Why

The previous implementation was doing unnecessary memory allocations during callback insertions and making copies of derivative values where moves would suffice. These changes should help reduce memory

[coveralls]

Pull Request Test Coverage Report for Build 12105754388

Details

• 8 of 8 (100.0%) changed or added relevant lines in 1 file are covered.
• No unchanged relevant lines lost coverage.
• Overall coverage remained the same at 98.961%

---

Totals
Change from base Build 12099899696:	0.0%
Covered Lines:	2763
Relevant Lines:	2792

---

💛 - Coveralls

[xcelerit-team]

Could you also please update the description of the pull request please? Thanks.

[rghouzra]

Could you also please update the description of the pull request please? Thanks.

done

[github-actions]

Test Results

    29 files  ±0      29 suites  ±0   9m 29s ⏱️ +3s
 1 359 tests ±0   1 359 ✅ ±0  0 💤 ±0  0 ❌ ±0 
39 279 runs  ±0  39 279 ✅ ±0  0 💤 ±0  0 ❌ ±0 

Results for commit b3a07e6. ± Comparison against base commit 5fedca2.

[auto-differentiation-dev]

Hi @rghouzra ,

Thank you for this PR and for your effort in contributing to the project - it's great to see your interest in improving the codebase!

After reviewing the changes, I noticed:

• insertCallback(): Switching to emplace doesn’t provide a performance gain in this case, as copying or moving is the same for a pointer.
• getAndResetOutputAdjoint(): The move semantics won’t significantly improve performance here, particularly because the function is rarely called (once per checkpoint) and there is no impact for first-order derivatives as copy or move is the same for double values.

Given these points, I don’t think merging this particular PR would benefit the repository right now. However, we truly appreciate your enthusiasm, and there are several open issues in the repo where your skills and ideas could make a significant impact. I’d love for you to take a look and see if anything catches your interest: https://github.com/auto-differentiation/xad/issues.

Thanks again for your contribution, and I hope to see more of your work in the future! Let me know if you have any questions or would like guidance on specific issues.

[rghouzra]

Hi @rghouzra ,

Thank you for this PR and for your effort in contributing to the project - it's great to see your interest in improving the codebase!

After reviewing the changes, I noticed:

* **`insertCallback()`**: Switching to `emplace` doesn’t provide a performance gain in this case, as copying or moving is the same for a pointer.

* **`getAndResetOutputAdjoint()`**: The move semantics won’t significantly improve performance here, particularly because the function is rarely called (once per checkpoint) and there is no impact for first-order derivatives as copy or move is the same for `double` values.

Given these points, I don’t think merging this particular PR would benefit the repository right now. However, we truly appreciate your enthusiasm, and there are several open issues in the repo where your skills and ideas could make a significant impact. I’d love for you to take a look and see if anything catches your interest: https://github.com/auto-differentiation/xad/issues.

Thanks again for your contribution, and I hope to see more of your work in the future! Let me know if you have any questions or would like guidance on specific issues.

Hi @rghouzra ,

Thank you for this PR and for your effort in contributing to the project - it's great to see your interest in improving the codebase!

After reviewing the changes, I noticed:

* **`insertCallback()`**: Switching to `emplace` doesn’t provide a performance gain in this case, as copying or moving is the same for a pointer.

* **`getAndResetOutputAdjoint()`**: The move semantics won’t significantly improve performance here, particularly because the function is rarely called (once per checkpoint) and there is no impact for first-order derivatives as copy or move is the same for `double` values.

Given these points, I don’t think merging this particular PR would benefit the repository right now. However, we truly appreciate your enthusiasm, and there are several open issues in the repo where your skills and ideas could make a significant impact. I’d love for you to take a look and see if anything catches your interest: https://github.com/auto-differentiation/xad/issues.

Thanks again for your contribution, and I hope to see more of your work in the future! Let me know if you have any questions or would like guidance on specific issues.

Hi @auto-differentiation-dev
Thanks for the feedback! I’ll check out the open issues.