managedCUDA Changelog
What's new in managedCUDA 4.1
Apr 7, 2012- fixed bugs
- Added a reset() method to directly update a texture reference without disposing the old and recreating a new CudaTextureLinearPitched2D / CudaTextureLinear1D object.
- dim3 has new constructors taking int arguments and not just uint
- All Debug outputs changed to String.Format for .net pre 4.0 compatibility
- Removed "using System.Linq" for .net pre 4.0 compatibility
- Added indexer properties to CudaDeviceVariable and CudaPitchedDeviceVariable: Access elements directly from host
- Implicit conversion operators for CudaDeviceVariable (and some for CudaPitchedDeviceVariable): Alloc and copy in only one assignment operation.
- Memset now also for CudaDeviceVariable and CudaPitchedDeviceVariable and not only inderectly using CudaContext.
- Offsets for copy methods in device variables / pinned host variables
- Added CudaRand
- Added CudaBlas
- Added NPP
- New Samples
New in managedCUDA 4.0 (Jun 28, 2011)
- Improved stability
- Handy access to page-locked host memory accompanied by all async cuda capabilities
- Wrapper classes for DirectX and OpenGL interop.
- The complete CUDA driver API is now accessible through convenient wrappers. Not one direct API call is needed for anything (if I haven't missed something ;-)
- ATTENTION: big change! By mistake the order of width and height arguments were sometimes swapped in the previous release. This is now adjusted: It is always first width then height, just as in the original CUDA driver API.
- Versioning will from now on be associated to CUDA: this release is therefor managedCUDA 4.0.x.y