Compile CUDA using the LLVM backend

We can compile CUDA to PTX using the LLVM backend

llc produces PTX files without complaining.

changed title from WIP: LLVM experiments to WIP: Compile CUDA using the LLVM backend

changed the description

OK, apparently one has to mark pointers to be in address space (1) and functions as kernels https://llvm.org/docs/NVPTXUsage.html#marking-functions-as-kernels

added 1 commit

• ca9c30ea - llvm: Use addressspace 1 (global memory) for nvvm_target

Compare with previous version

added 1 commit

• df7f4a7b - llvm: Mark CUDA kernels and load/call resulting ptx with pycuda

Compare with previous version

Update: address space 1 is now used for device pointers and kernel functions are now correctly marked. Loading the generated PTX with pycuda now yields the same results as numpy.

TODO: change hard-coded lld-10 to generic name of binary.

Test is failing because lcc is not installed. Numba uses CUDA's nvvm library instead of llvm's lcc. They need to perform some hacks to convert the IR to an older IR version nvcc uses.

added 2 commits

• 99347f9f - Add lcc to default config
• 52650ffb - Avoid hard-coded lcc command

Compare with previous version

added 1 commit

• 31501f85 - llvm: Add llc to pystencils' config

Compare with previous version

added 1 commit

• 21db981f - llvm: Add llc to pystencils' config

Compare with previous version

Tests should be working after installing llvm.

added 5 commits

• a526fe47 - Make compilation to PTX work (loading of function is WIP)
• 2e6f3efe - llvm: Use addressspace 1 (global memory) for nvvm_target
• a57a164a - llvm: Mark CUDA kernels and load/call resulting ptx with pycuda
• 8d22fddf - llvm: Add llc to pystencils' config
• 23e8e04b - Skip llvm cuda compilation if llc not in $PATH

Compare with previous version

unmarked as a Work In Progress

Tests works on two machines of mine with llc-10 and llc-9. Test is skipped now if no llc is found

merged

mentioned in commit 6d875789