[Mlir-commits] [mlir] [mlir][amdgpu] Define an amdgpu.scaling_mfma wrapper (PR #137498)

[Jakub Kuderski]

================
@@ -506,6 +506,49 @@ LogicalResult GatherToLDSOp::verify() {
   return success();
 }
 
+LogicalResult ScaledMFMAOp::verify() {
+  unsigned opselA = getOpselA();
+  unsigned opselB = getOpselB();
+
+  opselA >>= 8;
+  opselB >>= 8;
+
+  if (opselA != 0)
+    return emitOpError("Opsel A must be a zero extended 8 bit value.");
+
+  if (opselB != 0)
+    return emitOpError("Opsel B must be a zero extended 8 bit value.");
+
+  auto validType = [&](Type mlirElemType) {
+    return llvm::TypeSwitch<Type, bool>(mlirElemType)
+        .Case([](Float8E4M3FNType) { return true; })
+        .Case([](Float8E5M2Type) { return true; })
+        .Case([](Float6E2M3FNType) { return true; })
+        .Case([](Float6E3M2FNType) { return true; })
+        .Case([](Float4E2M1FNType) { return true; })
+        .Default([](Type) { return false; });
+  };
----------------
kuhar wrote:

I think you can simplify this to:
```suggestion
  auto isValidType = llvm::IsaPred<Float8E4M3FNType, Float8E5M2Type, Float6E2M3FNType, Float6E3M2FNType, Float4E2M1FNType>;
```

https://github.com/llvm/llvm-project/pull/137498