[llvm-branch-commits] [clang] [lld] [llvm] [mlir] [RFC][AMDGPU] Add BARRIER address space (PR #209746)

[Sameer Sahasrabuddhe via llvm-branch-commits]

================
@@ -165,6 +157,26 @@ unsigned AMDGPUMachineFunctionInfo::allocateLDSGlobal(const DataLayout &DL,
   return Offset;
 }
 
+unsigned
+AMDGPUMachineFunctionInfo::allocateBarrierGlobal(const DataLayout &DL,
+                                                 const GlobalVariable &GV) {
+  assert(AMDGPU::isNamedBarrier(GV));
+  std::optional<unsigned> BarAddr =
+      get32BitAbsoluteAddress(GV, AMDGPUAS::BARRIER);
+  if (!BarAddr)
+    llvm_unreachable("named barrier should have an assigned address");
----------------
ssahasra wrote:

I don't think we are supposed to worry about "bad IR" reaching this point, even in a release build. The verifier is always run on external input, right? Otherwise we would have to upgrade a whole bunch of asserts all across the LLVM codebase.

https://github.com/llvm/llvm-project/pull/209746