[Mlir-commits] [mlir] Introduce `arith.scaling_extf` and `arith.scaling_truncf` (PR #141965)

[Krzysztof Drewniak]

================
@@ -1215,6 +1215,44 @@ def Arith_ExtFOp : Arith_FToFCastOp<"extf", [DeclareOpInterfaceMethods<ArithFast
                           attr-dict `:` type($in) `to` type($out) }];
 }
 
+//===----------------------------------------------------------------------===//
+// Scaling ExtFOp
+//===----------------------------------------------------------------------===//
+def Arith_ScalingExtFOp
+    : Arith_Op<
+          "scaling_extf", [Pure, SameInputOutputTensorDims,
+                           DeclareOpInterfaceMethods<ArithFastMathInterface>,
+                           DeclareOpInterfaceMethods<CastOpInterface>]>,
+      Arguments<(ins FloatLike:$in, FloatLike:$scale,
+          OptionalAttr<Arith_FastMathAttr>:$fastmath)>,
+      Results<(outs FloatLike:$out)> {
+  let summary =
+      "cast from floating-point to larger floating-point using provided scales";
+  let description = [{
+    Implements micro-scaling floating point ExtF op. It expects both scales and input operand to be of same shape. 
+    Scale operand is expected to be of type f8E8M0. But that can be relaxed in future.  
+    Scale is usually calculated per block.  
+    Let's say originally input is shape <dim1 x dim2 x dim3 .. x dimN> then, given blockSize it can be reshaped to <dim1 x dim2 x ... (dimN/blockSize) x blockSize>. 
+    Scales will be calculated on the block axis. Therefore scale will be of shape <dim1 x dim2 x dim3 ... (dimN/blockSize) x 1>. 
+    Before calling into `arith.scaling_extf`, scales must be broadcasted appropariately to make it as same shape as input making `arith.scaling_extf` an elemenwise op.  
+    In above example. scales should be broadcasted to shape of <dim1 x dim2 x dim3 x ... (dimN/blockSize) x blockSize>.
----------------
krzysz00 wrote:

I think the description needs to be updated - this arith op is set up to do things elementwise because arith ops in general are elementwise and the broadcast scale thing is a special case that gets pattern-matched in a future ArithToAMDGPU

https://github.com/llvm/llvm-project/pull/141965