[Mlir-commits] [mlir] [mlir][AMDGPU] Add scaled wmma ops for gfx1250 (PR #169854)

[Jakub Kuderski]

================
@@ -447,101 +447,100 @@ LogicalResult WMMAOp::verify() {
 //===----------------------------------------------------------------------===//
 
 LogicalResult ScaledWMMAOp::verify() {
+  // Helper functions for type classification.
+  auto isF8 = [](Type t) {
+    return isa<Float8E4M3FNType, Float8E5M2Type, Float8E8M0FNUType,
+               Float8E4M3FNUZType, Float8E5M2FNUZType>(t);
+  };
----------------
kuhar wrote:

```suggestion
  auto isF8 = llvm::IsaPred<Float8E4M3FNType, Float8E5M2Type, Float8E8M0FNUType,
               Float8E4M3FNUZType, Float8E5M2FNUZType>;
```

also, do we actually need to consider all these float types? I thought that scaled wmma/mfma is only over ocp fp8 types?

https://github.com/llvm/llvm-project/pull/169854