[llvm] AMDGPU: Perform libcall recognition to replace fast OpenCL pow (PR #182135)

[Matt Arsenault via llvm-commits]

================
@@ -680,12 +709,21 @@ bool AMDGPULibCalls::fold(CallInst *CI) {
           {CI->getType(), CI->getArgOperand(1)->getType()}));
       return true;
     }
-    case AMDGPULibFunc::EI_POW: {
+    case AMDGPULibFunc::EI_POW:
+    case AMDGPULibFunc::EI_POW_FAST: {
----------------
arsenm wrote:

I don't think it's likely to grow in the future. The todo is a reordering question. The ultimate goal is to delete all of the custom expansion stuff, but that's probably needs to wait for a year 

https://github.com/llvm/llvm-project/pull/182135