[llvm] [AMDGPU] Add dynamic threshold for DPP atomic optimizer on integer LDS atomics (PR #186762)

[Shilei Tian via llvm-commits]

================
@@ -2205,6 +2205,15 @@ The AMDGPU backend supports the following LLVM IR attributes.
      "amdgpu-no-fwd-progress"                         Disable forward progress mode for wave priority
                                                       (enabled by default).
 
+     "amdgpu-atomic-optimizer-dpp-lds-threshold"="n"  Threshold that controls when the AMDGPUAtomicOptimizer uses a DPP-based
----------------
shiltian wrote:

This doesn't seem like something that we want to control via an attribute. Is it really necessary to control it on a per-function basis?

https://github.com/llvm/llvm-project/pull/186762