[Mlir-commits] [mlir] [mlir][amdgpu] Add lowering for make_dma_descriptor (PR #169955)

[Krzysztof Drewniak]

================
@@ -730,37 +737,77 @@ LogicalResult MakeDmaBaseOp::verify() {
 LogicalResult MakeDmaDescriptorOp::verify() {
   ArrayRef<int64_t> globalStaticStrides = getGlobalStaticStrides();
 
-  if (globalStaticStrides.empty()) {
+  if (globalStaticStrides.empty())
     return emitOpError("strides must not be empty.");
-  }
-  if (globalStaticStrides.back() != 1) {
+  if (globalStaticStrides.back() != 1)
     return emitOpError("strides for the innermost dimension must be 1.");
-  }
 
   ArrayRef<int64_t> globalStaticSizes = getGlobalStaticSizes();
   size_t rank = globalStaticSizes.size();
-  if (rank != globalStaticStrides.size()) {
+  if (rank < 2)
+    return emitOpError("tensor and tile must be at least of rank 2.");
----------------
krzysz00 wrote:

They actually don't, I think

You just left pad the dimensions with 0s until you get to an appropriate size

https://github.com/llvm/llvm-project/pull/169955