[Mlir-commits] [mlir] [ROCDL] Add the global.atomic.fadd intrinsic in ROCDL (PR #94486)

[Kunwar Grover]

================
@@ -516,7 +516,7 @@ def ROCDL_RawBufferAtomicCmpSwap :
 }
 
 //===---------------------------------------------------------------------===//
-// MI-100 and MI-200 buffer atomic floating point add intrinsic
+// MI-100, MI-200 and MI-300 global/buffer atomic floating point add intrinsic
----------------
Groverkss wrote:

nit: at some point, we should start using architecture names (gfx9, etc.) to match llvm architecture models better

https://github.com/llvm/llvm-project/pull/94486