b9803

ggml-org/llama.cppb9803Jun 26, 2026by github-actions[bot]same day as b9802

What changed since b9802

low risk

Single OpenCL profiling fix flushes incomplete batches at shutdown so interrupted runs still emit profiling data.

Improvements

  • OpenCL backend now flushes any partially accumulated profiling batch during shutdown, improving completeness of profiling output.

Bug fixes

  • Profiling data from interrupted or incomplete OpenCL batches is no longer lost when the process exits.

Upgrade notes

This is a build-to-build increment with no API, config, or model-format changes; existing binaries and scripts work unchanged. Only users relying on OpenCL profiling output will notice a difference, and only in the form of more complete traces. No migration steps required.

Analyzed Nov 1, 2026

Upgrading?

Release notes summary

Improves OpenCL profiling by flushing incomplete batches at shutdown, ensuring profiling data is captured even when processing is interrupted.

Key Highlights

  • Flush profiling batch at shutdown for incomplete batches
  • Ensures profiling data capture for interrupted batches

New Features

  • OpenCL profiling batch flush mechanism

Release Notes (excerpt)

<details open> opencl: flush profiling batch at shutdown for incomplete batches (#25016) </details> **macOS/iOS:** - macOS Apple Silicon (arm64) - macOS Apple Silicon (arm64, KleidiAI enabled) DISABLED - macOS Intel (x64) - iOS XCFramework **Linux:** - Ubuntu x64 (CPU) - Ubuntu arm64 (CPU) - Ubuntu s390x (CPU) - Ubuntu x64 (Vulkan) - Ubuntu arm64 (Vulkan) - Ubuntu x64 (ROCm 7.2) - Ubuntu x64 (OpenVINO) - Ubuntu x64 (SYCL FP32) - Ubuntu x64 (SYCL FP16) **Android:** - Android arm64 (CPU)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories