cuda.core 1.4.0 Release Notes#

Fixes and enhancements#

  • StridedMemoryView now verifies the actual Tensor type before using the PyTorch native bridge. Protocol adapters with a torch-prefixed module name and a data_ptr attribute now use their declared data exchange protocol instead of failing or being passed to the native Tensor interface.

  • Kernels, Buffer.fill() and asynchronous copies now work on PinnedMemoryResource buffers on every device that supports the pool’s kind of host memory pool, as on cudaMallocHost memory. In cuda.core 1.3.0, only the device that was current when the resource was created could access its buffers. A process that imported the pool, through PinnedMemoryResource.from_allocation_handle() or by unpickling, granted no device access at all, because pool access is per process. There, kernels failed with CUDA_ERROR_ILLEGAL_ADDRESS, fill failed with CUDA_ERROR_INVALID_VALUE, and copies silently blocked the host until earlier work on the stream completed. The creating and the importing process now both grant read and write access to every supporting device. (#3415, #3501)