[llvm] [KernelInfo] Implement new LLVM IR pass for GPU code analysis (PR #102944)

[Johannes Doerfert via llvm-commits]

================
@@ -772,6 +773,15 @@ void AMDGPUTargetMachine::registerPassBuilderCallbacks(PassBuilder &PB) {
           return onlyAllocateVGPRs;
         return nullptr;
       });
+
+  PB.registerFullLinkTimeOptimizationLastEPCallback(
+      [](ModulePassManager &PM, OptimizationLevel Level) {
+        if (KernelInfoEndLTO) {
----------------
jdoerfert wrote:

Move the conditional outside.

https://github.com/llvm/llvm-project/pull/102944