tensor.extract operand #0 does not dominate this use

[AmosLewis]

What happened?

Found this when lowering onnx opt-125M-awq model nod-ai/SHARK-TestSuite#258

opt-125M-awq.default.onnx.linalg.elide.mlir:1078:22: error: operand #0 does not dominate this use
    %extracted_365 = tensor.extract %282[] : tensor<i64>
                     ^
opt-125M-awq.default.onnx.linalg.elide.mlir:23:3: note: called from
  func.func @main_graph(%arg0: tensor<?x?xi64>, %arg1: tensor<?x?xi64>, %arg2: tensor<?x12x?x64xf32>, %arg3: tensor<?x12x?x64xf32>, %arg4: tensor<?x12x?x64xf32>, %arg5: tensor<?x12x?x64xf32>, %arg6: tensor<?x12x?x64xf32>, %arg7: tensor<?x12x?x64xf32>, %arg8: tensor<?x12x?x64xf32>, %arg9: tensor<?x12x?x64xf32>, %arg10: tensor<?x12x?x64xf32>, %arg11: tensor<?x12x?x64xf32>, %arg12: tensor<?x12x?x64xf32>, %arg13: tensor<?x12x?x64xf32>, %arg14: tensor<?x12x?x64xf32>, %arg15: tensor<?x12x?x64xf32>, %arg16: tensor<?x12x?x64xf32>, %arg17: tensor<?x12x?x64xf32>, %arg18: tensor<?x12x?x64xf32>, %arg19: tensor<?x12x?x64xf32>, %arg20: tensor<?x12x?x64xf32>, %arg21: tensor<?x12x?x64xf32>, %arg22: tensor<?x12x?x64xf32>, %arg23: tensor<?x12x?x64xf32>, %arg24: tensor<?x12x?x64xf32>, %arg25: tensor<?x12x?x64xf32>) -> (tensor<?x?x50272xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>, tensor<?x12x?x64xf32>) {
  ^
opt-125M-awq.default.onnx.linalg.elide.mlir:1078:22: note: see current operation: %425 = "tensor.extract"(%426#1) : (tensor<i32>) -> i32
    %extracted_365 = tensor.extract %282[] : tensor<i64>
                     ^
opt-125M-awq.default.onnx.linalg.elide.mlir:1087:12: note: operand defined here (op in the same block)
    %289 = linalg.

Steps to reproduce your issue

iree-compile --iree-input-demote-i64-to-i32 --iree-hal-target-backends=llvm-cpu opt-125M-awq.default.onnx.linalg.elide.mlir > opt-125M-awq.default.vmfb

opt-125M-awq.default.onnx.linalg.elide.mlir

What component(s) does this issue relate to?

No response

Version information

candidate-20240630.940

Additional context

No response

[AmosLewis]

Redundant filed #17759