b9810

ggml-org/llama.cppb9810Jun 26, 2026by github-actions[bot]same day as b9803

What changed since b9803

low risk

Adds a cublasSgemmBatched compatibility mapping so HIP/MUSA vendor headers build against CUDA-named APIs.

New features

  • cublasSgemmBatched is now mapped in the HIP/MUSA vendor header shims, letting code written against the CUDA name compile on AMD and Moore Threads backends.

Improvements

  • Improved cross-platform parity between the CUDA and HIP/MUSA build paths for batched GEMM calls.

Upgrade notes

This is a build-time header addition only; no runtime behavior, CLI flags, or model formats change. Rebuild from source if you compile against HIP or MUSA and previously hit missing cublasSgemmBatched symbols. Prebuilt binaries for other backends are unaffected.

Analyzed Oct 31, 2026

Upgrading?

Release notes summary

Adds cublasSgemmBatched mapping for HIP/MUSA vendor headers, improving cross-platform compatibility between CUDA and HIP/MUSA.

Key Highlights

  • Added cublasSgemmBatched mapping for HIP/MUSA
  • Improved cross-platform CUDA/HIP compatibility

Release Notes (excerpt)

<details open> CUDA: add cublasSgemmBatched mapping for HIP/MUSA vendor headers (#25033) </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