just_test.mlir

func.func @vector_expansion(
    %47: vector<1xf32>
) -> f32 {
    %cst_0 = arith.constant dense<0.000000e+00> : vector<1xf32>
    %cst_3 = arith.constant  0.000000e+00 : f32
    %48 = vector.extract %47[0] : f32 from vector<1xf32>
    %49 = gpu.subgroup_reduce  add %48 cluster(size = 8) : (f32) -> f32
    %50 = vector.insert %49, %cst_0 [0] : f32 into vector<1xf32>
    %51 = vector.shape_cast %50 : vector<1xf32> to vector<1xf32>
    %52 = vector.broadcast %cst_3 : f32 to vector<1xf32>
    %53 = arith.addf %51, %52 : vector<1xf32>
    %54 = vector.extract %53[0] : f32 from vector<1xf32>
    %55 = vector.broadcast %54 : f32 to vector<f32>
    %56 = vector.transpose %55, [] : vector<f32> to vector<f32>
    %60 = vector.extract %56[] : f32 from vector<f32>
  return %60 : f32
}

builtin.module attributes { transform.with_named_sequence } {
  transform.named_sequence @__transform_main(%variant_op: !transform.any_op {transform.readonly}) {
    %top_level_func = transform.structured.match ops{["func.func"]} in %variant_op : (!transform.any_op) -> !transform.any_op
    transform.iree.test_gpu_vector_distribution %top_level_func : !transform.any_op
    transform.yield
  }
}