Pekka Jääskeläinen

Thanks and glad to see someone working on the RISC-V port again. Unfortunately we miss a RISC-V maintainer also, would you be interested to become one and submit a PR?

Great to hear, thanks.

This sounds suspicious. Does the problem occur with the single core device? (POCL_DEVICES=basic)? This might be explained if the kernel compiler somehow erroneously considers that private array as a global...

I have some (bitrotted) work towards improving the parallel region analysis in a private branch which I need to clean up and upstream. It should speed up the analysis of...

Note to myself, the pocl-standalone script still works. This can be reproduced with `pocl-standalone -h stencil.h -o stencil.bc stencil.cl`

Did you try with master? There has been bunch of updates to kernel compilation which might affect also this.

I have a hunch what causes it, but unfortunately haven't had enough concentration time to fix it, sorry. I'd guess that adding extra barriers between the ifs workarounds it too.

> Could BarrierTailReplication perhaps limit the depth `BarrierTailReplication` is called and fail on performing the transformation if the limit is reached? Yes that's something that might be useful: Graceful failures...

Can you also flatten out/rebase the commit history a bit?

Fails to build the CQ test: ```[ 89%] Building C object tests/runtime/CMakeFiles/test_command_buffer.dir/test_command_buffer.c.o /home/visit0r/src/pocl/tests/runtime/test_command_buffer.c: In function ‘main’: /home/visit0r/src/pocl/tests/runtime/test_command_buffer.c:36:5: error: unknown type name ‘clCreateCommandBufferKHR_fn’ 36 | clCreateCommandBufferKHR_fn clCreateCommandBufferKHR; | ^~~~~~~~~~~~~~~~~~~~~~~~~~~ /home/visit0r/src/pocl/tests/runtime/test_command_buffer.c:37:5: error:...