[Mlir-commits] [mlir] [mlir] [linalg] fix failure on specializing matmul with permuted loops (PR #184294)

[llvmlistbot at llvm.org]

ziereis wrote:

Whether the inputs are transposed does not matter for the loop ordering. It always needs to be:

(m,n,k) or (b,m,n,k)

We only need to care about transpositions for the tensor access dims and there we use the information from the  `IndexMatchResult` to construct them in the transposed order. 

https://github.com/llvm/llvm-project/pull/184294