Skip to content

Releases: KomputeProject/kompute

Release v0.3.1

20 Sep 16:37
5fc36e3
Compare
Choose a tag to compare
  • #61 Fixed bug in OpAlgoRhsLhsOut
  • #60 Add example of how vulkan kompute can be used for ML in Godot Game Engine
  • #59 Changed c++ to 14 from 17 for support with older frameworks

Release v0.3.0

13 Sep 11:12
Compare
Choose a tag to compare
  • #50 - Added documentation and testing for reusing recorded commands
  • #19 - Added Machine Learning example with Kompute
  • #40 - Provide further granularity for handling tensor data
  • #39 - Create OpTensorSyncLocal and OpTensorSyncDevice operations
  • #43 - Renamed OpCreateTensor to OpTensorCreate to align with tensor operations
  • #43 - Fixed OpTensorCreate not mapping data for host tensors
  • #47 - Add preSubmit function to OpBase to account for multiple eval commands
  • #58 - Add standalone examples that show using Kompute from scratch
  • #58 - Make Kompute installable locally
  • #56 - Remove OpAlgoBase copy tensor functionality to delegate to OpTensorSync

Release v0.2.0

05 Sep 10:31
Compare
Choose a tag to compare
  • #18 - Improve access to underlying data for speed and ease of access
  • #17 - Enabled for compute shaders to be passed as files or strings
  • #13 - Enable OpCreateTensor to receive more than 1 tensor
  • #11 - Add default specialisation data to algorithm holding all tensor sizes
  • #9 - Added documentation automated with doxygen and sphinx
  • #15 - Memory profiling and ensured no memory leaks explicitly
  • #30 - Removed spdlog as required dependency and into optional
  • #37 - Migrated to GTest

Release v0.1.0

29 Aug 13:03
Compare
Choose a tag to compare

Vulkan Kompute Release 0.1.0

The 0.1.0 release of Vulkan Kompute, the General Purpose Vulkan Compute Framework.

Features

  • Automatic detection and selection of phyiscal GPU device
  • Single header import with dynamic library available for easy integration
  • Uses vcpkg with optional manifest to download / fetch dependencies
  • Provides high level kompute interface via Manager and Sequence
  • GPU processing data abstracted with Tensors
  • Compute shaders and pipelines are abstracted via algorithms
  • Base options for operations including create tensor and multiplication as example
  • Ability to initialise components with external vulkan resources
  • Conversion of glsl shaders into SPIR-V and native HPP files for static building
  • DEBUG option for verbose builds that cover execution path in detail with SPDLOG
  • Docker image to build the library in linux environments
  • CMAKE support with tested functionality in linux and windows
  • Documentation containing autodoc with Doxygen using Sphinx
  • Unit and integration testing using Catch2