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

[Krzysztof Drewniak]

================
@@ -954,6 +964,53 @@ struct MFMAOpLowering : public ConvertOpToLLVMPattern<MFMAOp> {
   }
 };
 
+struct ScaledMFMAOpLowering : public ConvertOpToLLVMPattern<ScaledMFMAOp> {
+  ScaledMFMAOpLowering(const LLVMTypeConverter &converter, Chipset chipset)
+      : ConvertOpToLLVMPattern(converter), chipset(chipset) {}
+
+  Chipset chipset;
+
+  LogicalResult
+  matchAndRewrite(ScaledMFMAOp op, ScaledMFMAOpAdaptor adaptor,
+                  ConversionPatternRewriter &rewriter) const override {
+    Location loc = op.getLoc();
+    Type outType = typeConverter->convertType(op.getDestD().getType());
+    Type intrinsicOutType = outType;
+    if (auto outVecType = dyn_cast<VectorType>(outType))
+      if (outVecType.getElementType().isBF16())
+        intrinsicOutType = outVecType.clone(rewriter.getI16Type());
+
+    if (chipset.majorVersion != 9 || chipset < kGfx908)
----------------
krzysz00 wrote:

`kGfx950` - this is a newer instruction

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