Support for AMD GPUs

[jaywonchung]

AMD GPUs are seeing increased adoption. ROCm has nice compatibility layers with PyTorch, too. Plus, ROCm-SMI (apparently) has all the energy-related management APIs we need -- measuring power and energy, and setting the power limit and GPU frequency.

First, we should evaluate whether the ROCm-SMI APIs behave like their NVML counterparts and show similar time/energy behavior with NVIDIA GPUs.

Then, implementatino-wise, ROCm-SMI lacks a Python package (something like nvidia-smi-py that NVIDIA officially provides). We should probably package that ourselves (much like the pynvml, which is a community-supported Python binding for NVML), until AMD provides an official binding.

[jaywonchung]

Progress

• Abstraction layer over GPUs (Abstracting away the gpu #46)
• AMDSMI + ROCm 6.0 implementations of management APIs (Added AMD GPU Support to Zeus #57)
• Track down power/energy API issue ([Issue]: Power info and energy counter APIs always return the same info ROCm/amdsmi#22)
• The cumulative energy counter works on MI200, MI210, MI250, and MI300x but not on MI100 ([Issue]: Incorrect Energy Consumption Reported by amdsmi_get_energy_count() Method ROCm/amdsmi#38) -- Issue warning after checking GPU board info?
• Test measurement on AMD GPUs
• Test optimization on AMD GPUs