[Mlir-commits] [mlir] [mlir][ROCDL] adds wmma scaled intrinsics for gfx1250 (PR #165915)

[Jakub Kuderski]

================
@@ -1028,6 +1028,224 @@ llvm.func @rocdl.ds.read.tr(%ptr : !llvm.ptr<3>) -> vector<4xf16> {
   llvm.return %r3 : vector<4xf16>
 }
 
+llvm.func @rocdl.wmma.scale.f32.16x16x128.f8f6f4(%arg0 : i32,
+                   %arg1 : vector<4 x f32>, %arg2 : vector<8xi32>,
+                   %arg3 : vector<12xi32>, %arg4 : vector<4xi32>, 
+                   %arg5 : vector<16xi32>, %arg6 : vector<12xi32>,
+                   %arg7 : vector<8xi64>, %arg8 : i64) -> vector<4 x f32> {
+  %cst0 = llvm.mlir.constant(0 : i32) : i32
+  %cst1 = llvm.mlir.constant(1 : i32) : i32
+  %cst2 = llvm.mlir.constant(2 : i32) : i32
+  %cst3 = llvm.mlir.constant(3 : i32) : i32
+  %cst4 = llvm.mlir.constant(4 : i32) : i32
+  %cst0_i16 = llvm.mlir.constant(0 : i16) : i16
+  %zero = llvm.mlir.constant(false) : i1
+  // CHECK-LABEL: rocdl.wmma.scale.f32.16x16x128.f8f6f4
+  
+  
+  // fp8 * fp8
----------------
kuhar wrote:

```suggestion
  // CHECK-LABEL: rocdl.wmma.scale.f32.16x16x128.f8f6f4
  
  // fp8 * fp8
```

https://github.com/llvm/llvm-project/pull/165915