Change8

b9534

📦 llama-cppView on GitHub →
2 features🐛 2 fixes🔧 1 symbols

Summary

This release introduces FWHT support for Vulkan on Intel with shmem reduction and includes several platform-specific adjustments, such as disabling certain shaders or optimizations due to driver issues or bugs.

✨ New Features

  • Added FWHT (Fast Walsh-Hadamard Transform) support for Vulkan on Intel systems utilizing shared memory reduction.
  • Disabled subgroup shuffle functionality on MoltenVK builds targeting AMD GPUs.

🐛 Bug Fixes

  • Prevented the use of N as workgroup size in Vulkan operations.
  • Disabled FWHT shader execution on Intel Windows due to an identified driver bug.

Affected Symbols