Changes since v0.6.9:
Graph support
The wrappers library now supports the creation and manipulation of CUDA graphs - meriting a version number bump.
One can now:
- Construct graphs (= graph templates) directly
- Capture graphs (= graph templates) on streams
- Instantitate and launch graph templates
... all using a more convenient interface, similar to non-graph CUDA-API calls. Two examples of this kind of code have been added, both adaptations of NVIDIA CUDA samples:
The main class templates are: template_t, node_t, typed_node_t, instance_t - all in namespace cuda::graph.
Most, but not all, graph capabilities are supported.
Other minor changes
- #649: Respect deprecation of shared memory bank size setting as of CUDA 12.3
Build-related
- Avoiding more MSVC compilation warnings