[CodeGen][NewPM] Use proper NPM AtomicExpandPass in AMDGPU (#122086)

`PassRegistry.def` already has this entry, but the dummy definition was
being pulled instead.

I couldn't reproduce the build failures that FIXME referenced, maybe the
Dummy pass getting in the way was part of the cause.

Author: optimisan

llvm/include/llvm/Passes/MachinePassRegistry.def

@@ -194,7 +194,6 @@ MACHINE_FUNCTION_PASS_WITH_PARAMS(
 #ifndef DUMMY_FUNCTION_PASS
 #define DUMMY_FUNCTION_PASS(NAME, PASS_NAME)
 #endif
-DUMMY_FUNCTION_PASS("atomic-expand", AtomicExpandPass)
 #undef DUMMY_FUNCTION_PASS
 
 #ifndef DUMMY_MACHINE_MODULE_PASS

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -49,6 +49,7 @@
 #include "llvm/Analysis/CGSCCPassManager.h"
 #include "llvm/Analysis/CallGraphSCCPass.h"
 #include "llvm/Analysis/UniformityAnalysis.h"
+#include "llvm/CodeGen/AtomicExpand.h"
 #include "llvm/CodeGen/DeadMachineInstructionElim.h"
 #include "llvm/CodeGen/GlobalISel/CSEInfo.h"
 #include "llvm/CodeGen/GlobalISel/IRTranslator.h"
@@ -1957,8 +1958,7 @@ void AMDGPUCodeGenPassBuilder::addIRPasses(AddIRPass &addPass) const {
       (AMDGPUAtomicOptimizerStrategy != ScanOptions::None))
     addPass(AMDGPUAtomicOptimizerPass(TM, AMDGPUAtomicOptimizerStrategy));
 
-  // FIXME: Adding atomic-expand manages to break -passes=atomic-expand
-  // addPass(AtomicExpandPass(TM));
+  addPass(AtomicExpandPass(&TM));
 
   if (TM.getOptLevel() > CodeGenOptLevel::None) {
     addPass(AMDGPUPromoteAllocaPass(TM));