Skip to content

feat: add Ascend950 direct SDMA backend - #85

Open
Kur0x wants to merge 7 commits into
mainfrom
codex/a5-direct-sdma-main
Open

feat: add Ascend950 direct SDMA backend#85
Kur0x wants to merge 7 commits into
mainfrom
codex/a5-direct-sdma-main

Conversation

@Kur0x

@Kur0x Kur0x commented Jul 29, 2026

Copy link
Copy Markdown
Member

Summary

  • add a runtime-selected Ascend950/A5 SDMA backend using CANN's built-in ShmemSdmaStarsQuery, 48 STARS channels, Host RTSQ mappings, and direct two-SQE AIV submission
  • preserve the existing A2/A3 PTO path and best-effort communicator behavior, including retryable resource cleanup and per-communicator failure isolation
  • migrate the supported driver floor to 25.1.rc1, extend the SDMA demo/tests and CI with an Ascend950 compile-only gate, and update both the draw.io source and exported SVG

Validation

  • python3 -m unittest discover -s tests/ci -p 'test_*.py': 141 passed, 1 skipped
  • ruby tests/ci/test_workflows.rb: passed
  • bash tests/ci/test_common_env_override.sh: passed
  • C++14 compile and execution: SDMA metadata, A5 query validation, communicator wiring, and source guard tests passed
  • bash syntax checks, git diff --check, and draw.io/SVG XML validation passed
  • exported architecture SVG visually inspected; no clipping or layout regressions
  • independent code review completed after fixes; final focused review reported no findings

Remaining Validation

  • a full CANN 9.1 Host build and dav-3510 compile must run in CI; the local macOS workspace has no CANN installation
  • the documented Ascend950 data-plane matrix was not rerun in this task because the validation host was unavailable
  • the approved plan's fully injectable Host runtime failure matrix is not yet implemented; this PR remains draft until that gap and CI results are resolved

Base

  • targets main directly; the branch contains one feature commit on top of origin/main

@Kur0x
Kur0x marked this pull request as ready for review July 29, 2026 06:22
@Kur0x
Kur0x requested a review from a team as a code owner July 29, 2026 06:22

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1bd5c0bd5e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread src/include/tilexr_sdma_a5.h Outdated
Comment thread src/comm/tilexr_comm.cpp

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 42ec9c76e0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread tests/sdma/demo/run_tilexr_sdma_demo.sh
Comment thread tests/sdma/demo/run_tilexr_sdma_demo.sh
Comment thread tests/sdma/demo/tilexr_sdma_demo.cpp
Comment thread src/comm/tilexr_comm.cpp
Comment thread src/include/tilexr_sdma_a5.h
Comment thread tests/sdma/demo/tilexr_sdma_demo.cpp
@Kur0x
Kur0x requested a review from chaowick July 30, 2026 06:32
Comment thread src/comm/tilexr_comm.cpp
Comment thread src/include/tilexr_sdma_a5.h

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: ce12aa912e

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread src/comm/tilexr_comm.cpp
Comment thread tests/sdma/CMakeLists.txt
Comment thread tests/sdma/demo/tilexr_sdma_benchmark.cpp
Comment thread tests/sdma/demo/tilexr_sdma_benchmark.cpp

@chatgpt-codex-connector chatgpt-codex-connector Bot left a comment

Copy link
Copy Markdown

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 6f7662bc01

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

  • Open a pull request for review
  • Mark a draft as ready
  • Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

Comment thread scripts/ci/control/build_blue.sh
@Kur0x
Kur0x requested a review from SilentEmotion July 30, 2026 11:59
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants