[Mlir-commits] [mlir] [mlir][tosa] Implement dynamic shape support for tosa.max_pool2d lowering (PR #87538)

[Spenser Bauman]

sabauma wrote:

For context, this was attempted once here: https://reviews.llvm.org/D133389

And then subsequently reverted here: https://reviews.llvm.org/D134370 due to incorrect codegen in PyTorch. If there are thoughts on how to mitigate the downstream risk, please let me know.

https://github.com/llvm/llvm-project/pull/87538