[llvm] [AMDGPU] Store resource analysis result into SIMachineFunctionInfo (PR #124040)

[Gang Chen via llvm-commits]

cmc-rep wrote:

> MachineFunctionInfo should not be the all-purpose cache, for something that occurs internally in one pass. Any caching should be done local to the analysis or the AsmPrinter.

There is reason we need to do this. I will send you the explanation.

https://github.com/llvm/llvm-project/pull/124040