util.func public @rank_reduced_extract_transposed(%arg0: tensor<80x80x1203xf16>) -> tensor<1203x5x80xf16> {
  %0 = tensor.empty() : tensor<1203x80x80xf16>
  %1 = tensor.empty() : tensor<1203x80x80xf16>
  %2 = tensor.empty() : tensor<1203x80x80xf16>
  %transposed = linalg.transpose ins(%arg0 : tensor<80x80x1203xf16>) outs(%0 : tensor<1203x80x80xf16>) permutation = [2, 0, 1]
  %expanded = tensor.expand_shape %transposed [[0, 1], [2], [3, 4]] output_shape [1, 1203, 80, 80, 1] : tensor<1203x80x80xf16> into tensor<1x1203x80x80x1xf16>
  %extracted_slice = tensor.extract_slice %expanded[0, 0, 0, 0, 0] [1, 1203, 5, 80, 1] [1, 1, 1, 1, 1] : tensor<1x1203x80x80x1xf16> to tensor<1203x5x80xf16>
  util.return %extracted_slice : tensor<1203x5x80xf16>
}

../iree-build/tools/iree-opt --pass-pipeline="builtin.module(util.func(iree-global-opt-propagate-linalg-transpose))" propagate_extract_slice.mlir

--- After sinking transpose ops down ---
expected type to be 'tensor<5x80x1x1x1203xf16>' or a rank-reduced version. (size mismatch) 
[mlir-asm-printer AsmPrinter.cpp:2073 1] util.func' failed to verify and will be printed in generic form
"util.func"() <{function_type = (tensor<80x80x1203xf16>) -> tensor<1203x5x80xf16>, sym_name = "rank_reduced_extract_transposed_unit_dim", sym_visibility = "public", tied_operands = [-1 : index]}> ({
^bb0(%arg0: tensor<80x80x1203xf16>):
  %0 = "tensor.expand_shape"(%arg0) <{reassociation = [[0], [1, 2], [3, 4]], static_output_shape = array<i64: 80, 80, 1, 1, 1203>}> : (tensor<80x80x1203xf16>) -> tensor<80x80x1x1x1203xf16>
  %1 = "tensor.extract_slice"(%0) <{operandSegmentSizes = array<i32: 1, 0, 0, 0>, static_offsets = array<i64: 0, 0, 0, 0, 0>, static_sizes = array<i64: 5, 80, 1, 1, 1203>, static_strides = array<i64: 1, 1, 1, 1, 1>}> : (tensor<80x80x1x1x1203xf16>) -> tensor<80x1203x1203xf16>
  %2 = "tensor.empty"() : () -> tensor<1203x80x80xf16>
  %3 = "linalg.transpose"(%1, %2) <{permutation = array<i64: 1, 0, 0>}> ({
  ^bb0(%arg1: f16, %arg2: f16):
    "linalg.yield"(%arg1) : (f16) -> ()
  }) : (tensor<80x1203x1203xf16>, tensor<1203x80x80xf16>) -> tensor<1203x80x80xf16>
  "util.return"(%3) : (tensor<1203x80x80xf16>) -> ()
}) : () -> ()

propagate_extract_slice.mlir:30:22: error: expected type to be 'tensor<5x80x1x1x1203xf16>' or a rank-reduced version. (size mismatch) 
  %extracted_slice = tensor.extract_slice %expanded[0, 0, 0, 0, 0] [1, 1203, 5, 80, 1] [1, 1, 1, 1, 1] : tensor<1x1203x80x80x1xf16> to tensor<1203x5x80xf16>
                     ^
propagate_extract_slice.mlir:30:22: note: see current operation: %1 = "tensor.extract_slice"(%0) <{operandSegmentSizes = array<i32: 1, 0, 0, 0>, static_offsets = array<i64: 0, 0, 0, 0, 0>, static_sizes = array<i64: 5, 80, 1, 1, 1203>, static_strides = array<i64: 1, 1, 1, 1, 1>}> : (tensor<80x80x1x1x1203xf16>) -> tensor<80x1203x1203xf16>
[pass-manager Pass.cpp:683 1] Pipeline failed for pass 'PropagateLinalgTransposePass' on operation 'expected type to be 'tensor<5x80x1x1x1203xf16>' or a rank-reduced version. (size mismatch) 
[mlir-asm-printer AsmPrinter.cpp:2073 1] util.func' failed to verify and will be printed in generic form
"util.func"() <{function_type = (tensor<80x80x1203xf16>) -> tensor<1203x5x80xf16>, sym_name = "rank_reduced_extract_transposed_unit_dim", sym_visibility = "public", tied_operands = [-1 : index]}> ({...}) : () -> ()'
[pass-manager Pass.cpp:683 1] Pipeline failed for pass 'mlir::detail::OpToOpPassAdaptor' on operation 'expected type to be 'tensor<5x80x1x1x1203xf16>' or a rank-reduced version. (size mismatch) 
[mlir-asm-printer AsmPrinter.cpp:2073 1] builtin.module' failed to verify and will be printed in generic form
"builtin.module"() ({...}) : () -> ()'
[pass-manager Pass.cpp:1102 2] PassManager run completed with result: failure