Skip to content

- This CL introduces the detect_layout_mismatch_copies tool, a new automated graph analysis tool for identifying copy operations causing physical layout mismatch, HBM materialization, and TPU lane boundary alignment overhead.#2813

Open
copybara-service[bot] wants to merge 1 commit into
masterfrom
test_918455327

Conversation

@copybara-service

@copybara-service copybara-service Bot commented Jun 15, 2026

Copy link
Copy Markdown
  • This CL introduces the detect_layout_mismatch_copies tool, a new automated graph analysis tool for identifying copy operations causing physical layout mismatch, HBM materialization, and TPU lane boundary alignment overhead.
  • The tool recursively traverses the HLO instruction graph context to catch sandwiched copies between upstream and downstream compute-intensive operations, while evaluating TPU lane size alignment constraints based on raw primitive data types.
  • This tool is exposed via the xprof_cli and the MCP server to support autonomous performance profiling discovery and layout optimization workflows.
  • It also updates the XProf smart suggestions reference (smart_suggestions.md) with the matching detection specifications and JAX-level refactoring recommendations.

@copybara-service copybara-service Bot force-pushed the test_918455327 branch 2 times, most recently from da1c14d to 1055261 Compare June 15, 2026 16:38
@copybara-service copybara-service Bot changed the title Project import generated by Copybara - This CL introduces the detect_layout_mismatch_copies tool, a new automated graph analysis tool for identifying copy operations causing physical layout mismatch, HBM materialization, and TPU lane boundary alignment overhead. Jun 15, 2026
…automated graph analysis tool for identifying copy operations causing physical layout mismatch, HBM materialization, and TPU lane boundary alignment overhead.

- The tool recursively traverses the HLO instruction graph context to catch sandwiched copies between upstream and downstream compute-intensive operations, while evaluating TPU lane size alignment constraints based on raw primitive data types.
- This tool is exposed via the `xprof_cli` and the MCP server to support autonomous performance profiling discovery and layout optimization workflows.
- It also updates the XProf smart suggestions reference (`smart_suggestions.md`) with the matching detection specifications and JAX-level refactoring recommendations.

PiperOrigin-RevId: 918455327
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.

0 participants