[Chapel Merge] support ref intent on gpuized forall w/ AMD

Branch: refs/heads/main
Revision: aec6d566693fe634f679ed306e294b6ce25c03bb
Author: stonea
Link: support ref intent on gpuized forall w/ AMD by stonea · Pull Request #25647 · chapel-lang/chapel · GitHub
Log Message:
support ref intent on gpuized forall w/ AMD (#25647)

This PR extends the work from #25594 so that ref intent scalars will
work with forall loops on AMD GPUs (the prior PR got it to work for
NVIDIA).

Since AMD GPUs (or at least the ones we're using today) don't have
universal addressing we need to convert the host pointer to the ref
intent'd parameter into a "device pointer" before it can be used by the
device. This PR performs this step.

[Reviewed by @e-kayrakli]

Compare: Comparing bee622825d221f9545e9c5af629cd3d123dbbd0a...0a08f355dd07ef23ef5f23c15d3975c245c22555 · chapel-lang/chapel · GitHub

Diff:
M runtime/include/chpl-gpu-impl.h
M runtime/src/chpl-gpu.c
M runtime/src/gpu/amd/gpu-amd.c
M runtime/src/gpu/nvidia/gpu-nvidia.c
D test/gpu/native/intents/forall_refIntentScalar.skipif
https://github.com/chapel-lang/chapel/pull/25647.diff