[Mlir-commits] [mlir] [mlir][gpu] Introduce `gpu.dynamic_shared_memory` Op (PR #71546)

[Oleksandr Alex Zinenko]

================
@@ -433,6 +433,29 @@ def GPU_GPUFuncOp : GPU_Op<"func", [
   let hasVerifier = 1;
 }
 
+def GPU_DynamicSharedMemoryOp : GPU_Op<"dynamic_shared_memory", [] > {
+  let summary = "Get the memref for dynamic shared memory";
+  
+  let description = [{
+    This operation provides a memref pointer to the start of dynamic shared 
+    memory, often referred to as workgroup memory. It's important to note that
+     this dynamic shared memory needs to be allocated at kernel launch. One can 
+     conveniently utilize `the dynamic_shared_memory_size` parameter of 
+     `gpu.launch` for this purpose.
+   
+    Examples: 
+    ```mlir        
+    %0 = gpu.dynamic.shared.memory : memref<?xi8, 3>
+    %1 = memref.view %0[%c8192][] : memref<?xi8, 3> to memref<32x64xf32, #gpu.address_space<workgroup>>
+    %2 = memref.view %0[%c16384][] : memref<?xi8, 3> to memref<32x64xf32, #gpu.address_space<workgroup>>
----------------
ftynse wrote:

Nit: the magic change of address space from `3` to `workgroup` will raise questions from the reader of documentation. 

https://github.com/llvm/llvm-project/pull/71546