Diptorup Deb

@mingjie-intel the information in the ticket are useful if we want to go one level lower than SPIR-V and look at the IGC compiled binaries.

@roxx30198 can you update the PR description a bit to state what are the changes you did and why they were needed?

I think it is a good idea and will be a good contribution to numba. We need to evaluate where and when we can fit it in our schedule.

@ZzEeKkAa take a look at https://github.com/IntelPython/numba-dpex/blob/8fc7fbddb2b73d80b98a0e629fddb6647aa456f8/numba_dpex/tests/experimental/codegen/test_intenum_literal_codegen.py#L15. It is possible to only compile a kernel provided we generate the right signature. Perhaps a solution can be to have a CLI tool...

@ZzEeKkAa thank you for the reproducer. The issue with the overload PR is now clear to me. The issue happens because overloads are not compiled to SPIR-V. We do SPIR-V...

@oleksandr-pavlyk there are some compile time warnings after these changes.

@oleksandr-pavlyk @ZzEeKkAa I believe the original issue was that the conda-recipe `meta.yaml` should not pin dependency versions rather the version requirements for a dependency should be specified as an environment...

@roxx30198 Can you please update the docs as suggested?

@fcharras In general our approach has been to only apply lightweight optimizations to the LLVM IR generated for kernels. The idea is to generate a relatively unmolested LLVM IR and...

> The thing that I failed to state is my GPU environment: > > ``` > (dppy_stack2) opavlyk@opavlyk-mobl:~/tmp$ sycl-ls > [opencl:acc:0] Intel(R) FPGA Emulation Platform for OpenCL(TM), Intel(R) FPGA Emulation...