v3.8.4

Change logs

• Restore cublas argument. (injector)
• Implement some Dark API functions (#41), Thanks @SEt-t
• Implement mul24 instruction.
• Add environment variable ZLUDA_COMGR_LOG_LEVEL.
  If you set ZLUDA_COMGR_LOG_LEVEL=1, it will give you minimal logs that demonstrate the kernel code compilation is under progress.
• Bring back the minimal implementations of runtime API. (#45)
  It is enough to run stable-diffusion-webui-forge with CUDA Stream enabled.

v3.8.3

Change logs

• Add --version argument for zluda.exe. (zluda_injector)
• Revive HIP SDK 5.7 support for Windows.

※ Please move to HIP SDK 6.1 if possible. 5.7 support can be discontinued at any time.

v3.8.2

Change logs

• Add support of Rust 1.80.
• Fix bug of hiprt-sys. Thanks @mobile-bungalow.

v3.8.1

This release is automatically generated by GitHub Actions.

v3.8

Change logs

• Add ROCm 6.1 support.
• Discontinue ROCm <6 support. v3.8 for ROCm 5.7

※ The future release will target ROCm 6 or later.

v3.7.2

Change logs

• Merge upstream commits.
  • vosen#153
  • vosen#211
  • vosen#225
  • vosen#229

v3.8-rocm5

Implement cuda runtime api. (cudart) (#17)

* [WIP] Implement cudart.

* wip

* wip

* Implement cudart.

* wip

* Ready to merge.

v3.7.1

Change logs

• Implement cublasDotEx.
• Implement nvmlDeviceGetCudaComputeCapability.
• Rewrite nvrtcVersion.
• Merge upstream commits.
  • vosen#203
  • vosen#206
  • vosen#216

v3.7.1-pre2-dnn

v3.8-pre2-dnn

Implement cudnnGetProperty.

v3.7.1-pre2

Change logs

• Implement nvmlDeviceGetCudaComputeCapability.
• Rewrite nvrtcVersion.
• Merge upstream commits.
  • vosen#203