[llvm] [OpenMP][OMPIRBuilder] Add support to omp target parallel (PR #67000)

[via llvm-commits]

shraiysh wrote:

To be sure I understand this correctly, in the testcase, `ParallelSimpleGPU`, the IR emitted is supposed to be the IR for GPU for the outlined function, but host IR for the main function `func`. In this IR, we are emitting and checking for the host code i.e. the `@__kmpc_parallel_51` call in the `func` function, and for the outlined code we are checking that it is what is expected on the device end. We are not testing for proper handling the generation of outlined code on host or generation of caller code in the target LLVM IR (the call to the outlined function in target IR), right?

https://github.com/llvm/llvm-project/pull/67000