Isuru Fernando
Isuru Fernando
ah okay. since pocl changes the arguments to support local memory in CUDA, this isn't going to work. (NVIDIA's OpenCL seems to be using a non-public API to support local...
For Jetson Nano, you need https://github.com/pocl/pocl/pull/890
I rebased this PR on top of master branch. There are 2 tests failing though.
Two tests in pocl's internal test-suite run by `ctest`. ``` The following tests FAILED: 113 - runtime/clCreateSubDevices (Failed) 157 - EinsteinToolkit_SubDev (Subprocess aborted) Errors while running CTest ``` I guess...
Any suggestions on how to fix those tests? @pjaaskel, I looked through the suggestions, but I don't understand what to do in > In check_cmd_queue_for_device() - the for loop (DL_FOREACH)...
Is pocl not using standard LLVM passes that clang is using? I see pocl producing kernels with code similar to what `clang -O1 -x cl` produces and much different to...
CUDA backend doesn't support all the features that the pthread (CPU) backend does. Can you share `demo_float32.py` ?
`| ERROR | /home/nvidia/.cache/pocl/kcache//program.bc does not exist!` That's an error I haven't seen before. Not sure what is going on here
The path is wrong. `/home/nvidia/.cache/pocl/kcache//program.bc` should have been something like `/home/nvidia/.cache/pocl/kcache/AB/KMNAJOCCCKLCIDHODGOFINCNGMCALPPONOGCO/program.bc`
I could if I had access to a Jetson, but I don't have access.