iree: error: pattern listener tracker fail; transform dialect interpreter failed

What happened?

The transform-dialect-based lowering seems to be hitting an issue with this test case.

Run with no_attention2.mlir from here

iree-compile --iree-hal-target-backends=cuda --iree-input-type=mhlo --iree-hal-cuda-llvm-target-arch=sm_80 no_attention2.mlir

Sorry for the large test case. The failing dispatch looks like this but for some reason isolating it into a small repro doesn’t reproduce the failure for me locally.

    builtin.module {
      func.func @_main_dispatch_342_generic_2048x2048_f32(%arg0: !flow.dispatch.tensor<readonly:tensor<2048x2048xf32>>, %arg1: !flow.dispatch.tensor<readonly:tensor<2048xf32>>, %arg2: !flow.dispatch.tensor<writeonly:tensor<2048xf32>>) {
        %cst = arith.constant 0.000000e+00 : f32
        %cst_0 = arith.constant 2.048000e+03 : f32
        %0 = flow.dispatch.tensor.load %arg0, offsets = [0, 0], sizes = [2048, 2048], strides = [1, 1] : !flow.dispatch.tensor<readonly:tensor<2048x2048xf32>> -> tensor<2048x2048xf32>
        %1 = flow.dispatch.tensor.load %arg1, offsets = [0], sizes = [2048], strides = [1] : !flow.dispatch.tensor<readonly:tensor<2048xf32>> -> tensor<2048xf32>
        %2 = tensor.empty() : tensor<2048xf32>
        %3 = linalg.fill ins(%cst : f32) outs(%2 : tensor<2048xf32>) -> tensor<2048xf32>
        %4 = linalg.generic {indexing_maps = [#map4, #map8], iterator_types = ["parallel", "reduction"]} ins(%0 : tensor<2048x2048xf32>) outs(%3 : tensor<2048xf32>) {
        ^bb0(%in: f32, %out: f32):
          %6 = arith.negf %in : f32
          %7 = arith.addf %out, %6 : f32
          linalg.yield %7 : f32
        } -> tensor<2048xf32>
        %5 = linalg.generic {indexing_maps = [#map, #map, #map, #map], iterator_types = ["parallel"]} ins(%4, %1, %3 : tensor<2048xf32>, tensor<2048xf32>, tensor<2048xf32>) outs(%2 : tensor<2048xf32>) {
        ^bb0(%in: f32, %in_1: f32, %in_2: f32, %out: f32):
          %6 = arith.addf %in, %in_1 : f32
          %7 = arith.divf %6, %cst_0 : f32
          %8 = arith.addf %in_2, %7 : f32
          linalg.yield %8 : f32
        } -> tensor<2048xf32>
        flow.dispatch.tensor.store %5, %arg2, offsets = [0], sizes = [2048], strides = [1] : tensor<2048xf32> -> !flow.dispatch.tensor<writeonly:tensor<2048xf32>>
        return
      }
    }

Full error log link

Steps to reproduce your issue

See above

What component(s) does this issue relate to?

Compiler

Version information

iree.git @ ab37989652aed11f7f46498c09b9ac515c83eaa3

Additional context

No response