[llvm-branch-commits] [clang] [llvm] [mlir] [RFC][AMDGPU] Add BARRIER address space (PR #195613)

Pierre van Houtryve via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Wed May 6 07:18:58 PDT 2026


================
@@ -47,6 +46,14 @@ enum : unsigned {
   BUFFER_STRIDED_POINTER = 9, ///< Address space for 192-bit fat buffer
                               ///< pointers with an additional index.
 
+  RESERVED_0 = 10,
+  RESERVED_1 = 11,
+  RESERVED_2 = 12,
+  RESERVED_3 = 13,
+  RESERVED_4 = 14,
+
+  BARRIER = 15, ///< Address space for modeling barrier IDs as addresses.
----------------
Pierre-vh wrote:

Thinking about this, should this be much higher so that if we add more similar AS, they can be contiguous ?
e.g. make them start at addrspace 50 ? That makes some logic a bit more annoying (like the stuff that uses MAX_AMDGPU_ADDRESS) but may be worth it. I'm not sure.

https://github.com/llvm/llvm-project/pull/195613


More information about the llvm-branch-commits mailing list