Hi,
I have migrated from R24.1 to R24.2 by installing JetPack 2.3.
When I compare between
- R24.1: /usr/local/cuda-7.0/targets/armv7-linux-gnueabihf/include/
- R24.2: /usr/local/cuda-8.0/targets/aarch64-linux/include/
I could not find
- cuviddec.h
- nvcuvid.h
in CUDA V8.0(L4T R24.2).
What should I do ?
I can’t answer the question, but I do notice something odd…
The R24.1 listing is for the 32-bit armv7-linux-gnueabihf…although R24.1 had a 32-bit user space available, the kernel space in older L4T on JTX1 releases was still 64-bit, and CUDA 7 requires 64-bit kernels, so perhaps there was some sort of glue between 32-bit user space (armv7-linux-gnueabihf) and 64-bit kernel space (aarch64-linux).
R24.2 is purely 64-bit, and has no 32-bit user space (you’d have to manually add a foreign architecture, which is probably not what you want). CUDA 8 is also purely 64-bit, and aarch64-linux is where it should appear.
So the question is do you really want the armv7-linux-gnueabihf version? What is the use case? It seems you probably need to migrate from a 32-bit version to a 64-bit version, and do not really want to put legacy 32-bit support in. If those header files existed previously, and no longer show up, then it is likely a need to update a CUDA 7 app to use CUDA 8 instead.