[Mlir-commits] [mlir] [mlir][tosa] Enhance folder for Tosa binary operators (PR #128059)

[Tai Ly]

Tai78641 wrote:

> I wonder if it's worth limiting the size of the binary operations that are folded - would the runtime of folding a larger input, say `tensor<1024x1024x1024x1024xf32>`, be reasonable?

I looked in the code base for examples of limiting fold by number of elements, with no luck.
And I found that TFLite op's ConstFoldBinaryOp imposes no limit on size of number of elements

https://github.com/llvm/llvm-project/pull/128059