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

[Pierre van Houtryve via llvm-branch-commits]

================
@@ -2653,6 +2664,14 @@ bool AMDGPULegalizerInfo::legalizeAddrSpaceCast(
       if (!ApertureReg.isValid())
         return false;
 
+      if (SrcAS == AMDGPUAS::BARRIER) {
+        // barrier -> flat: add the barrier AS offset
+        SrcAsInt =
+            B.buildAdd(S32, SrcAsInt,
----------------
Pierre-vh wrote:

I'd rather not try to optimize this too much, as this would land then immediately get replaced by the synthetic apertures in the next PR of the stack.

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