// DISABLE-ALL-DOT-GENERAL: @main
func.func @main(%arg0: tensor<1x1x167xf32>) -> tensor<1x1x64xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<167x64xf32>
  %1 = stablehlo.dot_general %arg0, %0, contracting_dims = [2] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x1x167xf32>, tensor<167x64xf32>) -> tensor<1x1x64xf32>
  return %1 : tensor<1x1x64xf32>
}

// DISABLE-ALL-DOT-GENERAL: %[[CONST:.+]] = stablehlo.constant dense<2.000000e+00>

func.func @dot_general_with_relu_fn(%arg0: tensor<1x1x167xf32>, %arg1: tensor<167x64xf32>) -> tensor<1x1x64xf32> {
  %0 = stablehlo.constant dense<2.000000e+00> : tensor<167x64xf32>
  %1 = stablehlo.constant dense<0.000000e+00> : tensor<1x1x64xf32>
  %2 = stablehlo.dot_general %arg0, %0, contracting_dims = [2] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x1x167xf32>, tensor<167x64xf32>) -> tensor<1x1x64xf32>
  %3 = stablehlo.maximum %2, %1 : tensor<1x1x64xf32>

    precision_config = #vhlo.array_v1<[#vhlo<precision_v1 DEFAULT>, #vhlo<precision_v1 DEFAULT>]>}> : (tensor<1x1x167xf32>, tensor<167x64xf32>) -> tensor<1x1x64xf32>
  return %0 : tensor<1x1x64xf32>
}

//CHECK:func.func private @dot_general(%arg0: tensor<1x1x167xf32>, %arg1: tensor<167x64xf32>) -> tensor<1x1x64xf32> {
//CHECK-NEXT: %0 = "vhlo.dot_general_v1"(%arg0, %arg1) <{

func.func @reshape_vector_shape(tensor<4x4x4xf32>) -> tensor<16x4xf32> {
^bb0(%arg0: tensor<4x4x4xf32>) :
  %shape0 = arith.constant dense<[[16, 4]]> : tensor<1x2xi32>
  // expected-error @+1 {{'tfl.reshape' op requires 'shape' to be rank 1, but got 2}}
  %1 = "tfl.reshape"(%arg0, %shape0) : (tensor<4x4x4xf32>, tensor<1x2xi32>) -> tensor<16x4xf32>
  func.return %1 : tensor<16x4xf32>
}

// -----

    %0 = stablehlo.constant dense<0.000000e+00> : tensor<1x64xf32>
    %1 = stablehlo.dot_general %arg0, %arg1, contracting_dims = [1] x [0], precision = [DEFAULT, DEFAULT] : (tensor<1x3xf32>, tensor<3x64xf32>) -> tensor<1x64xf32>
    %2 = stablehlo.add %1, %arg2 : tensor<1x64xf32>
    %3 = stablehlo.maximum %2, %0 : tensor<1x64xf32>
    return %3 : tensor<1x64xf32>
  }
}

  %split_dim_2 = arith.constant dense<1> : tensor<1xi32>
  %4, %5 = "tfl.split"(%split_dim_2, %arg0) {num_splits = 2 : i32} : (tensor<1xi32>, tensor<16x4xf32>) -> (tensor<16x2xf32>, tensor<16x2xf32>)
  %6:2 = "tfl.split"(%split_dim_2, %arg0) {num_splits = 2 : i32} : (tensor<1xi32>, tensor<16x4xf32>) -> (tensor<16x2xf32>, tensor<16x?xf32>)

func.func @FuseReshapeAroundBMMLHSNegative(%arg0: tensor<1x64xf32>, %arg1: tensor<1x64x1024xf32> ) -> (tensor<1x1024xf32> )  {
  %cst = arith.constant dense<[1, 1024]> : tensor<2xi32>
  %cst_0 = arith.constant dense<[1, 1, 64]> : tensor<3xi32>
  %0 = "tfl.reshape"(%arg0, %cst_0) : (tensor<1x64xf32>, tensor<3xi32>) -> tensor<1x1x64xf32>
  %1 = "tfl.batch_matmul"(%0, %arg1) {adj_x = false, adj_y = false} : (tensor<1x1x64xf32>, tensor<1x64x1024xf32>) -> tensor<1x1x1024xf32>

// dimensions into a single dimension. For example,
//
//   %shape = arith.constant dense<[1, 128, 64]> : tensor<3xi32>
//   %reshape = tfl.reshape(%input, %shape) // %input: tensor<128x64xf32>
//   %fc = tfl.fully_connected(%reshape, %filter, %bias)
//           {keep_num_dims = false, weights_format = "DEFAULT"}
//
// can be canonicalized to
//