Weekly GitHub Report for Xla

Thank you for subscribing to our weekly newsletter! Each week, we deliver a comprehensive summary of your GitHub project's latest activity right to your inbox, including an overview of your project's issues, pull requests, contributors, and commit activity.

Table of Contents

• I. News
  • 1.1. Recent Version Releases
  • 1.2. Other Noteworthy Updates
• II. Issues
  • 2.1. Top 5 Active Issues
  • 2.2. Top 5 Stale Issues
  • 2.3. Open Issues
  • 2.4. Closed Issues
  • 2.5. Issue Discussion Insights

• III. Pull Requests
  • 3.1. Open Pull Requests
  • 3.2. Closed Pull Requests
  • 3.3. Pull Request Discussion Insights

• IV. Contributors
  • 4.1. Contributors

I. News

1.1 Recent Version Releases:

No recent version releases were found.

1.2 Version Information:

Please provide the version release information you would like me to analyze and summarize.

II. Issues

2.1 Top 5 Active Issues:

We consider active issues to be issues that that have been commented on most frequently within the last week. Bot comments are omitted.

1. [ERR: RUNTIME] Segfaults due to FFI type handling: This issue reports segmentation faults occurring in multiple JAX tests, traced back to a recent pull request affecting FFI type handling, specifically involving the GpuTransposePlanCache. The problem appears to stem from incorrect type registrations causing the runtime to delete objects with the wrong deleter, leading to crashes during test execution.

   • The comments discuss attempts to diagnose the issue using AddressSanitizer and logging changes, confirm that different types are mistakenly assigned the same type ID, and note ongoing efforts to fix undefined behavior in the code; one user reports the test passing with the latest JAX and XLA versions but with a different build method.
   • Number of comments this week: 7

2. Graph viewer on Xprof missing: This issue reports that when running JAX near the latest commit, the Xprof tool no longer displays the graph viewer, which previously worked with jaxlib version 0.9.0, indicating a recent regression. The user provides a reproducible example and notes that clicking on graph viewer links redirects back to the overview page, and also shares observations about differences in the trace directory contents between the working and non-working versions.

   • The comment provides additional debugging information by comparing the trace directory contents between the older jaxlib version 0.9.0 and the newer version, highlighting that the newer version lacks individual module hlo_proto files and instead contains a single NO_MODULE.hlo_proto.pb file, which may be related to the missing graph viewer functionality.
   • Number of comments this week: 1

Since there were fewer than 5 open issues, all of the open issues have been listed above.

2.2 Top 5 Stale Issues:

We consider stale issues to be issues that has had no activity within the last 30 days. The team should work together to get these issues resolved and closed as soon as possible.

As of our latest update, there are no stale issues for the project this week.

2.3 Open Issues

This section lists, groups, and then summarizes issues that were created within the last week in the repository.

Issues Opened This Week: 2

Summarized Issues:

• Segmentation faults and FFI type handling: Multiple JAX tests are experiencing segmentation faults caused by issues with Foreign Function Interface (FFI) type handling and conflicts in type registration. These problems are particularly linked to the deletion of GpuTransposePlanCache objects using incorrect deleters, leading to instability in the system.
• issues/37752

• Profiling tool regression in Xprof: The Xprof tool no longer displays the graph viewer in profiling traces when running JAX near the latest commit, indicating a regression. This issue contrasts with earlier versions of jaxlib where the graph viewer was available, affecting profiling usability.
• issues/37774

2.4 Closed Issues

This section lists, groups, and then summarizes issues that were closed within the last week in the repository. This section also links the associated pull requests if applicable.

Issues Closed This Week: 3

Summarized Issues:

• Conversion to StableHLO: The user is seeking guidance on converting an XLA-optimized HloModule to StableHLO, noting familiarity with xla_translate for MHLO conversion but wanting to adopt StableHLO as it replaces MHLO/HLO. This reflects a need for clearer documentation or tooling support for StableHLO conversion workflows.
• issues/36492

• Placeholder or Minimal Issues: There is an issue titled "Test" that appears to be a placeholder or minimal entry without any additional description or comments. This indicates some issues may be used for testing or tracking without substantive content.
• issues/37590

• CUDA Error Message Suppression on ROCm: The issue discusses suppressing CUDA error messages during training when using ROCm to improve user experience and reduce confusion. This highlights efforts to handle platform-specific error reporting more gracefully.
• issues/37741

2.5 Issue Discussion Insights

This section will analyze the tone and sentiment of discussions within this project's open and closed issues that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.

Based on our analysis, there are no instances of toxic discussions in the project's open or closed issues from the past week.

III. Pull Requests

3.1 Open Pull Requests

This section provides a summary of pull requests that were opened in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.

Pull Requests Opened This Week: 15

Key Open Pull Requests

1. [xla:gpu] Add a HangWatchdog to detect deadlocks in XLA:GPU execution: This pull request introduces a HangWatchdog safety mechanism to detect and abort deadlocks during XLA:GPU execution, including checks for debug options, handling of empty duration strings, and stress tests to fix race conditions.

• URL: pull/37789

• Associated Commits: 75c9a, 6da6d, 96f17, 711d9

2. [ROCm] Explicitly skip failing test cases : This pull request explicitly marks all unit tests that fail on ROCm with GTEST_SKIP annotations to make their exclusion clear and systematic, rather than relying on excluding test targets during full test suite runs, while noting that some of these skipped tests are expected to be fixed in upcoming related pull requests.

• URL: pull/37511

• Associated Commits: 6ba3a, 402e0, daab4

3. [ROCm] Add option to disable automatic solib and rocm rpaths adding to the final library: This pull request introduces a new toolchain feature called no_solib that allows disabling the automatic addition of solib and ROCm runtime library rpaths to the final library, which is necessary for building ROCm JAX plugins with custom rpath settings.

• URL: pull/37513

• Associated Commits: d87f5, a36a5, 2d35d

Other Open Pull Requests

• ROCm GPU Backend Improvements: Multiple pull requests enhance ROCm GPU support by enabling native FP8 Triton GEMM operations, improving register spill information propagation for autotuning, fixing compilation pipeline pass nesting, and expanding unit test coverage with platform detection and device configurations. These changes collectively improve performance, correctness, and test reliability on AMD GPUs within the XLA project.
  • pull/37573, pull/37629, pull/37732, pull/37854

• GPU Backend and Codegen Enhancements: Several pull requests improve GPU backend functionality by adding SYCL all_runtime dependency to fix client creation errors, supporting sub-byte types in DynamicMemcpyFusion with byte-aligned strides, and refining GPU emitters' loop unrolling heuristics to reduce register pressure and boost fusion performance. These updates enhance GPU codegen stability, compatibility, and execution efficiency.
  • pull/37769, pull/37773, pull/37806

• Build Process and Tooling: One pull request introduces building and using the xxd tool from source to remove the only non-hermetic, non-POSIX dependency in the PJRT plugins build process, enabling fully remote and hermetic builds on a vanilla Debian Docker image. This change simplifies the build environment and improves reproducibility.
  • pull/37816

• Parser and Execution Stability: A pull request adds a configurable recursion depth limit to the HLO parser to prevent stack overflow crashes from deeply nested inputs, replacing crashes with graceful error messages and including unit tests to verify this behavior. Another pull request adds state management support in all execution stages of the FFI handler, introducing per-instance and per-execution state types to improve execution control.
  • pull/37766, pull/37783

• CPU Operation Support: One pull request enables the in-place oneDNN softmax operation on CPU within the XLA project, expanding CPU operation capabilities and performance.
  • pull/37537

• New Container Implementation: A pull request introduces the tsl::UniqueAny container, an std::any-like container designed to support move-only types, accompanied by benchmark results demonstrating its performance. This addition provides a new utility for handling move-only types efficiently.
  • pull/37821

3.2 Closed Pull Requests

This section provides a summary of pull requests that were closed in the repository over the past week. The top three pull requests with the highest number of commits are highlighted as 'key' pull requests. Other pull requests are grouped based on similar characteristics for easier analysis. Up to 25 pull requests are displayed in this section, while any remaining pull requests beyond this limit are omitted for brevity.

Pull Requests Closed This Week: 15

Key Closed Pull Requests

1. [GPU][NFC] Rename H100 and B200 test data for clarity.: This pull request aims to rename the H100 and B200 GPU test data to improve clarity by accurately reflecting that these are not RTX models, as part of a cleanup effort.

• URL: pull/37309

• Associated Commits: 93a4f, b0b5c

• Associated Commits: 93a4f, b0b5c

2. [Rocm] Add flag to control hipblaslt swish activation fusion.: This pull request introduces a new flag that allows users to enable or disable the hipblaslt swish activation fusion in order to address performance degradation observed in certain large language models when this fusion is enabled.

• URL: pull/36826

• Associated Commits: 933bc

• Associated Commits: 933bc

3. [ROCm] add missing clang flag for hipcc: This pull request adds the missing --offload-compress clang flag to the hipcc compiler to enable compression of offloaded code objects, thereby reducing binary size and resolving a linking error in hlo_runner_main_build_test.

• URL: pull/37370

• Associated Commits: b5186

• Associated Commits: b5186

Other Closed Pull Requests

• PjRt and XLA Identifier Consistency: This pull request proposes switching PjRt to use the same strongly-typed identifiers—ProcessId, DeviceId, and ChipId—as those used in xla/runtime to ensure consistency across PjRt and XLA. This change aims to unify identifier usage and reduce potential mismatches between the two components.
  pull/37395

• CollectiveMemory API Enhancements: Multiple pull requests focus on improving the use of the CollectiveMemory API in the XLA GPU backend. These include testing the CollectiveKernel by verifying memory space assignments and adding APIs to acquire peer addresses to maintain consistency with symmetric and multimem collective memory implementations.
  pull/37414, pull/37490

• TSL Concurrency Module API Addition: A new API called Future::Flatten() is proposed for the tsl:concurrency module. This addition aims to enhance concurrency utilities by providing a method to flatten futures, as detailed in the associated commit.
  pull/37473

• XLA FFI API Refactoring and Improvements: Several pull requests address improvements and refactoring of the XLA FFI APIs. These include splitting the FFI APIs into two focused libraries to improve manageability and renaming functions for consistency, extracting a separate library for handler registration to avoid global static issues, and updating the internal FFI library to expose static handler and type registries for safer custom call library construction.
  pull/37541, pull/37601, pull/37670

• CUDA and Related Library Upgrades: This pull request upgrades CCCL to version 3.2.0 and RAFT/RMM to version 26.02.00 to improve compatibility with CUDA versions 12.9, 13.0, and 13.1+. It addresses issues arising from XLA's implicit use of CCCL from the CUDA Toolkit and enables XLA to build out of the box for CUDA 13.1+ environments.
  pull/37560

• ROC Automation Testing: A pull request proposes testing the ROC automation process within the project. The goal is to validate changes through unit and execution tests to ensure correctness and performance improvements.
  pull/37569

• XLA GPU Backend Testing and API Additions: One pull request adds a unified Walk API for traversing sequences of commands and thunks in the XLA GPU backend, incorporating templates for early termination via absl::Status, though it was not merged. Another enables Triton MLIR lit tests by uncommenting test suite targets following the availability of necessary binaries for open-source builds.
  pull/37677, pull/37740

• Error Handling Improvements in Type Registration: This pull request improves error handling by ensuring that type registration errors are not ignored. This prevents hard-to-debug runtime crashes and promotes early failure with logged errors.
  pull/37764

3.3 Pull Request Discussion Insights

This section will analyze the tone and sentiment of discussions within this project's open and closed pull requests that occurred within the past week. It aims to identify potentially heated exchanges and to maintain a constructive project environment.

Based on our analysis, there are no instances of toxic discussions in the project's open or closed pull requests from the past week.

IV. Contributors

4.1 Contributors

Active Contributors:

We consider an active contributor in this project to be any contributor who has made at least 1 commit, opened at least 1 issue, created at least 1 pull request, or made more than 2 comments in the last month.

If there are more than 10 active contributors, the list is truncated to the top 10 based on contribution metrics for better clarity.

Access Last Week's Newsletter:

• Link