Introduce FB_CUDACHECKTHROW_EX_NOCOMM macro for enhanced error reporting #157
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
added 2 commits
January 9, 2026 08:06
Summary: **TL;DR:** Adds a new `FB_CUDACHECKTHROW_EX` macro that throws `ctran::utils::Exception` instead of `std::runtime_error` on CUDA failures, enabling richer error context (rank, comm hash, description) for improved fault tolerance debugging. --- # Detailed Overview ## Context & Motivation As part of MCCL's fault tolerance efforts (T247416168), we are migrating error handling in CTRAN from `std::runtime_error` to `ctran::utils::Exception`. This migration enables structured error reporting with additional metadata (rank, comm hash, operation description) that is critical for diagnosing failures in distributed communication workloads. The existing `FB_CUDACHECKTHROW` macro throws a generic `std::runtime_error` when CUDA calls fail, which lacks the context needed for effective fault tolerance debugging. This diff introduces an extended version that preserves backward compatibility while enabling enhanced error reporting for callers that have access to communicator context. Differential Revision: D90191648
Summary: **TL;DR:** Adds a convenience macro `FB_CUDACHECKTHROW_EX_NOCOMM` that wraps `FB_CUDACHECKTHROW_EX` with `std::nullopt` values for use in contexts where communicator information (rank, commHash, commDesc) is not available. --- # Detailed Overview ## Context & Motivation As part of MCCL's fault tolerance efforts (T247416168), we are migrating CUDA error handling in CTRAN from `std::runtime_error` to `ctran::utils::Exception`. The previous diff (D90191648) introduced `FB_CUDACHECKTHROW_EX`, which requires rank, comm hash, and description parameters for enhanced error reporting. However, there are contexts within the codebase where these communicator-specific values are not available (e.g., utility functions, initialization code, or standalone CUDA operations). Requiring callers to manually pass `std::nullopt` for all three parameters in these cases is verbose and error-prone. Reviewed By: arttianezhu Differential Revision: D90192768
meta-cla
bot
added
the
CLA Signed
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
TL;DR: Adds a convenience macro
FB_CUDACHECKTHROW_EX_NOCOMMthat wrapsFB_CUDACHECKTHROW_EXwithstd::nulloptvalues for use in contexts where communicator information (rank, commHash, commDesc) is not available.Detailed Overview
Context & Motivation
As part of MCCL's fault tolerance efforts (T247416168), we are migrating CUDA error handling in CTRAN from
std::runtime_errortoctran::utils::Exception. The previous diff (D90191648) introducedFB_CUDACHECKTHROW_EX, which requires rank, comm hash, and description parameters for enhanced error reporting.However, there are contexts within the codebase where these communicator-specific values are not available (e.g., utility functions, initialization code, or standalone CUDA operations). Requiring callers to manually pass
std::nulloptfor all three parameters in these cases is verbose and error-prone.Reviewed By: arttianezhu
Differential Revision: D90192768