b9941

ggml-org/llama.cppb9941Jul 9, 2026by github-actions[bot]same day as b9940

What changed since b9940

low risk

Internal compile-time and arithmetic optimization in the indexing path; no user-facing API changes between b9940 and b9941.

Improvements

  • Indexing logic no longer depends on the compiler to optimize away local memory, making behavior more predictable across toolchains.
  • Replaced conditional control flow with multiply/add arithmetic in the affected code path, which is cheaper to execute.
  • Should yield marginally faster and more consistent builds/runs on toolchains with weaker optimization passes.

Upgrade notes

This is a drop-in binary/source update with no config or API changes; the offline flag added in b9940 for llama-bench remains available. Rebuild from source if you rely on specific compiler optimization behavior, since the change alters how the indexing code is expressed. No migration steps required.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Optimized compilation by avoiding reliance on compile-time optimizations and using cheaper arithmetic operations.

Key Highlights

  • Avoids relying on compile-time optimization for local memory
  • Uses multiply/add instead of control flow statements
  • Cheaper than issuing control flow statements

New Features

  • Compilation optimizations
  • Reduced reliance on compile-time optimization

Release Notes (excerpt)

<details open> Only index by compile times + always multiply/add (#25445) The first one avoids relying on compile to optimize local memory away, and the second is cheaper than issuing control flow statements </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)…

Published by ggml-org on GitHub.

Read the full release notes

Related Repositories