func.func @main_graph$async_dispatch_33_softmax_Dx28xDxDxf32_generic() {
          %c32_i64 = arith.constant 32 : i64
          %cst = arith.constant 0.000000e+00 : f32
          %cst_0 = arith.constant 0xFF800000 : f32
          %c0 = arith.constant 0 : index
          %0 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(0) : i32
          %1 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(1) : i32
          %2 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(2) : i32
          %3 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(3) : i32
          %4 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(4) : i32
          %5 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(5) : i32
          %6 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(6) : i32
          %7 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(7) : i32
          %8 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(8) : i32
          %9 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(9) : i32
          %10 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(10) : i32
          %11 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(11) : i32
          %12 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(12) : i32
          %13 = hal.interface.constant.load layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) ordinal(13) : i32
          %14 = arith.extui %0 : i32 to i64
          %15 = arith.extui %1 : i32 to i64
          %16 = arith.shli %15, %c32_i64 : i64
          %17 = arith.ori %14, %16 : i64
          %18 = arith.index_castui %17 : i64 to index
          %19 = arith.extui %2 : i32 to i64
          %20 = arith.extui %3 : i32 to i64
          %21 = arith.shli %20, %c32_i64 : i64
          %22 = arith.ori %19, %21 : i64
          %23 = arith.index_castui %22 : i64 to index
          %24 = arith.extui %4 : i32 to i64
          %25 = arith.extui %5 : i32 to i64
          %26 = arith.shli %25, %c32_i64 : i64
          %27 = arith.ori %24, %26 : i64
          %28 = arith.index_castui %27 : i64 to index
          %29 = arith.extui %6 : i32 to i64
          %30 = arith.extui %7 : i32 to i64
          %31 = arith.shli %30, %c32_i64 : i64
          %32 = arith.ori %29, %31 : i64
          %33 = arith.index_castui %32 : i64 to index
          %34 = arith.extui %8 : i32 to i64
          %35 = arith.extui %9 : i32 to i64
          %36 = arith.shli %35, %c32_i64 : i64
          %37 = arith.ori %34, %36 : i64
          %38 = arith.index_castui %37 : i64 to index
          %39 = arith.extui %10 : i32 to i64
          %40 = arith.extui %11 : i32 to i64
          %41 = arith.shli %40, %c32_i64 : i64
          %42 = arith.ori %39, %41 : i64
          %43 = arith.index_castui %42 : i64 to index
          %44 = arith.extui %12 : i32 to i64
          %45 = arith.extui %13 : i32 to i64
          %46 = arith.shli %45, %c32_i64 : i64
          %47 = arith.ori %44, %46 : i64
          %48 = arith.index_castui %47 : i64 to index
          %49:7 = util.assume.int 
              %18<udiv = 7168>, 
              %23<umin = 0, umax = 9007199254740991>, 
              %28<umin = 0, umax = 18014398509481982>, 
              %33<umin = 0, umax = 9007199254740991>, 
              %38<umin = 0, umax = 18014398509481982>, 
              %43<umin = 0, umax = 9007199254740991>, 
              %48<umin = 0, umax = 18014398509481982>
            : index, index, index, index, index, index, index
          %50 = iree_tensor_ext.dispatch.workload.ordinal %49#1, 0 : index
          %51 = iree_tensor_ext.dispatch.workload.ordinal %49#3, 2 : index
          %52 = iree_tensor_ext.dispatch.workload.ordinal %49#4, 3 : index
          %53 = iree_tensor_ext.dispatch.workload.ordinal %49#5, 4 : index
          %54 = iree_tensor_ext.dispatch.workload.ordinal %49#6, 5 : index
          %55 = hal.interface.binding.subspan layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) binding(0) alignment(64) offset(%c0) flags("ReadOnly|Indirect") : !iree_tensor_ext.dispatch.tensor<readonly:tensor<?x1x?x?xi8>>{%50, %51, %52}
          %56 = hal.interface.binding.subspan layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) binding(1) alignment(64) offset(%49#0) flags("ReadOnly|Indirect") : !iree_tensor_ext.dispatch.tensor<readonly:tensor<?x28x?x?xf16>>{%53, %51, %54}
          %57 = hal.interface.binding.subspan layout(<constants = 14, bindings = [#hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, "ReadOnly|Indirect">, #hal.pipeline.binding<storage_buffer, Indirect>], flags = Indirect>) binding(2) alignment(64) offset(%c0) flags(Indirect) : !iree_tensor_ext.dispatch.tensor<writeonly:tensor<?x28x?x?xf32>>{%53, %51, %54}
          %58 = iree_tensor_ext.dispatch.workload.ordinal %49#2, 1 : index
          %59 = iree_tensor_ext.dispatch.tensor.load %56, offsets = [0, 0, 0, 0], sizes = [%53, 28, %51, %54], strides = [1, 1, 1, 1] : !iree_tensor_ext.dispatch.tensor<readonly:tensor<?x28x?x?xf16>>{%53, %51, %54} -> tensor<?x28x?x?xf16>
          %60 = tensor.empty(%53, %51, %54) : tensor<?x28x?x?xf32>
          %61 = iree_tensor_ext.dispatch.tensor.load %55, offsets = [0, 0, 0, 0], sizes = [%50, 1, %51, %58], strides = [1, 1, 1, 1] : !iree_tensor_ext.dispatch.tensor<readonly:tensor<?x1x?x?xi8>>{%50, %51, %52} -> tensor<?x?x?xi8>
          %62 = arith.trunci %61 : tensor<?x?x?xi8> to tensor<?x?x?xi1>
          %63 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%59, %62 : tensor<?x28x?x?xf16>, tensor<?x?x?xi1>) outs(%60 : tensor<?x28x?x?xf32>) {
          ^bb0(%in: f16, %in_1: i1, %out: f32):
            %66 = arith.select %in_1, %cst, %cst_0 : f32
            %67 = arith.extf %in : f16 to f32
            %68 = arith.addf %67, %66 : f32
            linalg.yield %68 : f32
          } -> tensor<?x28x?x?xf32>
          %64 = linalg.softmax dimension(3) ins(%63 : tensor<?x28x?x?xf32>) outs(%60 : tensor<?x28x?x?xf32>) -> tensor<?x28x?x?xf32>
          %65 = linalg.generic {indexing_maps = [affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>, affine_map<(d0, d1, d2, d3) -> (d0, d1, d2, d3)>], iterator_types = ["parallel", "parallel", "parallel", "parallel"]} ins(%64 : tensor<?x28x?x?xf32>) outs(%60 : tensor<?x28x?x?xf32>) {
          ^bb0(%in: f32, %out: f32):
            %66 = arith.cmpf une, %in, %in : f32
            %67 = arith.select %66, %cst, %in : f32
            linalg.yield %67 : f32
          } -> tensor<?x28x?x?xf32>
          iree_tensor_ext.dispatch.tensor.store %65, %57, offsets = [0, 0, 0, 0], sizes = [%53, 28, %51, %54], strides = [1, 1, 1, 1] : tensor<?x28x?x?xf32> -> !iree_tensor_ext.dispatch.tensor<writeonly:tensor<?x28x?x?xf32>>{%53, %51, %54}
          return
        }