[Mlir-commits] [mlir] [mlir][tosa] Lower AVG_POOL2D_ADAPTIVE to Linalg (PR #225637)

[Luke Hutton]

================
@@ -1141,5 +1141,8 @@ void mlir::tosa::populateTosaToLinalgNamedConversionPatterns(
   patterns->add<
       MaxPool2dConverter
     >(converter, patterns->getContext());
+
+  tosa::AvgPool2dAdaptiveOp::getCanonicalizationPatterns(
----------------
lhutton1 wrote:

I think it would be reasonable to assume canonicalization is run before tosa-to-linalg, rather than include the pattern here directly. I presume there are no plans yet to support the "adaptive" nature of this operation in the lowering?

https://github.com/llvm/llvm-project/pull/225637