// PerTensor-LABEL: QuantizeMatmulWithActConst
func.func @QuantizeMatmulWithActConst(%arg0: tensor<1x3x3x512xf32>) -> tensor<1x3x3x12xf32> {
  %w = arith.constant dense<127.0> : tensor<512x12xf32>
  %mm = "tfl.batch_matmul"(%arg0, %w) {adj_x = false, adj_y = false} : (tensor<1x3x3x512xf32>, tensor<512x12xf32>) -> tensor<1x3x3x12xf32>
  func.return %mm : tensor<1x3x3x12xf32>

}
// CHECK: func.func @main(%[[ARG:.+]]: tensor<1x3x4x3xf32>) -> tensor<1x3x2x2xf32> {
// CHECK-DAG: %[[CONST_0:.*]] = stablehlo.constant dense<[{{.*}}]> : tensor<2xf32>
// CHECK-DAG: %[[CONST_1:.*]] = stablehlo.constant dense<[{{.*}}]> : tensor<2x3x3x2xf32>
// CHECK-DAG: %[[CONV:.*]] = stablehlo.convolution(%[[ARG]], %[[CONST_1]]) {{.*}} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32>

  %1 = "tf.Maximum"(%0, %cst_0) : (tensor<1x3x4x2xf32>, tensor<f32>) -> tensor<1x3x4x2xf32>
  %2 = "tf.Minimum"(%1, %cst_1) : (tensor<1x3x4x2xf32>, tensor<f32>) -> tensor<1x3x4x2xf32>
  func.return %2 : tensor<1x3x4x2xf32>
// CHECK-DAG: %[[CONST_0:.*]] = "tf.Const"() <{value = dense<{{.*}}> : tensor<1x1x3x2xf32>}> : () -> tensor<1x1x3x2xf32>

func.func @NotQuantizeBatchMatmulWithActAct(%arg0: tensor<1x3x3x512xf32>) -> tensor<1x3x3x3xf32> {
  %0 = "quantfork.stats"(%arg0) {layerStats = dense<[0.000000e+00, 1.000000e+01]> : tensor<2xf32>} : (tensor<1x3x3x512xf32>) -> tensor<1x3x3x512xf32>
  %mm = "tfl.batch_matmul"(%0, %0) {adj_x = false, adj_y = true} : (tensor<1x3x3x512xf32>, tensor<1x3x3x512xf32>) -> tensor<1x3x3x3xf32>

      } : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x4x2xf32>
    return %0 : tensor<1x3x4x2xf32>
  }

  // CHECK: func.func private @qdq_for_conv_weight_per_channel(%[[ARG0:.+]]: tensor<1x3x4x3xf32>)
  // CHECK: %[[CST:.+]] = "tf.Const"() <{value = dense<3.000000e-01> : tensor<2x3x3x2xf32>}> : () -> tensor<2x3x3x2xf32>

    %3 = "tf.Identity"(%2) {device = ""} : (tensor<1x3x1x1xf32>) -> tensor<1x3x1x1xf32>
    return %3 : tensor<1x3x1x1xf32>
  }

// CHECK-LABEL: func @multiple_quantizable_ops_in_graph

    %cst_1 = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x3x1xf32>} : () -> tensor<2x3x3x1xf32>
    %cst_2 = "tf.Const"() {value = dense<3.000000e+00> : tensor<2x3x3x2xf32>} : () -> tensor<2x3x3x2xf32>
    %0 = "tf.PartitionedCall"(%arg0, %cst_1) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_depthwise_conv2d_fn} : (tensor<1x3x4x3xf32>, tensor<2x3x3x1xf32>) -> tensor<*xf32>

  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 5, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<16xf32>, tensor<16xf32>) -> tensor<3x3x3x16xf32>
  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>
  func.return %rst : tensor<256x8x7x16xf32>

  %fq = "tf.FakeQuantWithMinMaxVarsPerChannel"(%in, %mini, %maxi) {num_bits = 3, narrow_range = false} : (tensor<3x3x3x16xf32>, tensor<16xf32>, tensor<16xf32>) -> tensor<3x3x3x16xf32>
  %rst = "tf.Conv2D"(%arg, %fq) {T = "tfdtype$DT_FLOAT", data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x7x16xf32>
  func.return %rst : tensor<256x8x7x16xf32>

  // CHECK: %[[VAL_1:.*]] = "tf.Identity"(%[[VAL_0]]) : (tensor<*xf32>) -> tensor<*xf32>
  // CHECK: return %[[VAL_1]]
  %0 = "tf.Conv2D"(%arg2, %arg1) {data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true} : (tensor<8x32x32x3xf32>, tensor<1x1x3x128xf32>) -> tensor<*xf32>