[Mlir-commits] [mlir] [AMDGPU] Added support for Sparse WMMA ops (PR #183360)

[Jakub Kuderski]

================
@@ -1339,6 +1342,162 @@ static std::optional<StringRef> wmmaOpToIntrinsic(WMMAOp wmma,
   return std::nullopt;
 }
 
+/// Returns the `rocdl` intrinsic corresponding to a SparseWMMA operation
+/// `swmmac` if one exists. This includes checking to ensure the intrinsic is
+/// supported on the architecture you are compiling for.
+struct SparseWMMAOpInfo {
+  StringRef name;
+  bool useSign;
+  bool useReuse;
+  bool useClamp;
+};
+
+static std::optional<SparseWMMAOpInfo>
+sparseWMMAOpToIntrinsic(SparseWMMAOp swmmac, Chipset chipset) {
+
+  Type sourceAElem = getElementTypeOrSelf(swmmac.getSourceA().getType());
----------------
kuhar wrote:

```suggestion
sparseWMMAOpToIntrinsic(SparseWMMAOp swmmac, Chipset chipset) {
  Type sourceAElem = getElementTypeOrSelf(swmmac.getSourceA().getType());
```

https://github.com/llvm/llvm-project/pull/183360