[Mlir-commits] [mlir] [MLIR][NVVM] Add support for narrow-fp to bf16x2 conversions (PR #200157)

[Durgadoss R]

================
@@ -2185,32 +2185,54 @@ class NVVM_ConvertToFP16x2Op_Base <string srcType, Type srcArgType, string dstTy
     !tolower(dstType) # [{.
 
     The result `dst` is represented as a vector of }] # !tolower(dstType) # [{ elements.
-    }] #
-    !if(!eq(dstType, "F16"),
-    [{The `relu` attribute, when set, lowers to the '.relu' variant of 
-    the cvt instruction."}], "") # [{
-    
+
+    The `relu` attribute, when set, lowers to the '.relu' variant of
+    the cvt instruction.}] #
+
+    !if(!eq(dstType, "BF16"),
+    [{
+
+    The `sat` attribute specifies the saturation mode.
+
+    The optional scaling-factors for each of the inputs are provided through
+    the operand `scaleFactor` as a packed i16 type. Only `ue8m0` is supported
+    as the type of the scale-factor currently.}], "") # [{
+
     [For more information, see PTX ISA](https://docs.nvidia.com/cuda/parallel-thread-execution/index.html#data-movement-and-conversion-instructions-cvt)
   }];
   let results = (outs VectorOfLengthAndType<[2], [!cast<Type>(dstType)]>:$dst);
   let arguments = !if(!eq(dstType, "F16"),
     (ins srcArgType:$src,
-         DefaultValuedAttr<BoolAttr, "false">:$relu,
-         TypeAttr:$srcType),
+         TypeAttrOf<AnyTypeOf<supportedTypes>>:$srcType,
----------------
durga4github wrote:

I spent a few minutes trying to remember what is the difference between `srcType` and `srcArgType`..;-)

May be, let us rename srcArgType to srcStorageType/srcContainerType...that is more representative of what it is?

https://github.com/llvm/llvm-project/pull/200157