CUDA Python 11.7.1 Release notes#

Released on June 29, 2022

Hightlights#

Fix error propagation in CUDA Runtime bindings
Resolves issue #22

Limitations#

Source builds#

CUDA Python no longer re-declares CUDA types, instead it uses the types from CUDA C headers. As such source builds now need to access to latest CTK headers. In particular:

“$CUDA_HOME/include” has latest CTK headers
CTK headers have all types defined

(2) Certain CUDA types are not declared on mobile platforms and may face a “has not been declared” error during source builds. A temporary workaround is to use the headers found in https://gitlab.com/nvidia/headers/cuda. In particular CUDA Python needs the following headers and their dependencies:

cuda.h
cudaProfiler.h
driver_types.h
cuda_runtime.h
nvrtc.h

This a short-term limitation and will be relaxed in a future release.

CUDA Functions Not Supported in this Release#

Symbol APIs
- cudaGraphExecMemcpyNodeSetParamsFromSymbol
- cudaGraphExecMemcpyNodeSetParamsToSymbol
- cudaGraphAddMemcpyNodeToSymbol
- cudaGraphAddMemcpyNodeFromSymbol
- cudaGraphMemcpyNodeSetParamsToSymbol
- cudaGraphMemcpyNodeSetParamsFromSymbol
- cudaMemcpyToSymbol
- cudaMemcpyFromSymbol
- cudaMemcpyToSymbolAsync
- cudaMemcpyFromSymbolAsync
- cudaGetSymbolAddress
- cudaGetSymbolSize
- cudaGetFuncBySymbol
Launch Options
- cudaLaunchKernel
- cudaLaunchCooperativeKernel
- cudaLaunchCooperativeKernelMultiDevice
cudaSetValidDevices
cudaVDPAUSetVDPAUDevice