func.func @QuantizeGatherWeightOnly(%arg0: tensor<3xi32>) -> tensor<3x3x3x3xf32> {
  %w = arith.constant dense<1.270000e+02> : tensor<64x3x3x3xf32>
  %emb = "tfl.gather"(%w, %arg0) {axis = 0 : i32, batch_dims = 0 : i32} : (tensor<64x3x3x3xf32>, tensor<3xi32>) -> tensor<3x3x3x3xf32>
  %emb_s = "quantfork.stats"(%emb) {layerStats = dense<[0.000000e+00, 1.000000e+01]> : tensor<2xf32>} : (tensor<3x3x3x3xf32>) -> tensor<3x3x3x3xf32>
  func.return %emb_s : tensor<3x3x3x3xf32>

func.func @QuantizeGatherWeightOnly(%arg0: tensor<3xi32>) -> tensor<3x3x3x3xf32> {
  %w = arith.constant dense<1.270000e+02> : tensor<64x3x3x3xf32>
  %emb = "tfl.gather"(%w, %arg0) {axis = 0 : i32, batch_dims = 0 : i32} : (tensor<64x3x3x3xf32>, tensor<3xi32>) -> tensor<3x3x3x3xf32>
  %emb_s = "quantfork.stats"(%emb) {layerStats = dense<[0.000000e+00, 1.000000e+01]> : tensor<2xf32>} : (tensor<3x3x3x3xf32>) -> tensor<3x3x3x3xf32>
  func.return %emb_s : tensor<3x3x3x3xf32>

  %w = arith.constant dense<127.0> : tensor<32x3x3x3xf32>
  %b = arith.constant dense<0.0> : tensor<32xf32>
  %dc = "tfl.depthwise_conv_2d"(%arg0, %w, %b) {depth_multiplier = 4 : i32, dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "VALID", stride_h = 4 : i32, stride_w = 5 : i32} : (tensor<1x224x224x3xf32>, tensor<32x3x3x3xf32>, tensor<32xf32>) -> tensor<1x112x112x32xf32>

  %0 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<16x3x3x3xf32>} : () -> tensor<16x3x3x3xf32>
  %1 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<16xf32>} : () -> tensor<16xf32>
  %2 = func.call @fold_all_test(%arg0, %0, %1) : (tensor<256x32x32x3xf32>, tensor<16x3x3x3xf32>, tensor<16xf32>) -> tensor<256x30x30x16xf32>
  func.return %2 : tensor<256x30x30x16xf32>
}

  %0 = stablehlo.constant dense<2.000000e+00> : tensor<4x2x3x3xf32>
  %1 = stablehlo.constant dense<3.000000e+00> : tensor<1x4x5x5xf32>
  %2 = stablehlo.convolution(%arg0, %0) dim_numbers = [b, f, 0, 1]x[o, i, 0, 1]->[b, f, 0, 1], window = {pad = [[1, 1], [1, 1]]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64} : (tensor<1x2x5x5xf32>, tensor<4x2x3x3xf32>) -> tensor<1x4x5x5xf32>
  %3 = stablehlo.add %2, %1 : tensor<1x4x5x5xf32>

  %w = arith.constant dense<-1.0> : tensor<32x3x3x3xf32>
  %3 = "tfl.quantize"(%w) {qtype = tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>} : (tensor<32x3x3x3xf32>) -> tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>
  %4 = "tfl.dequantize"(%3) : (tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>) -> tensor<32x3x3x3xf32>

  %w = arith.constant dense<-1.0> : tensor<32x3x3x3xf32>
  %3 = "tfl.quantize"(%w) {qtype = tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>} : (tensor<32x3x3x3xf32>) -> tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>
  %4 = "tfl.dequantize"(%3) : (tensor<32x3x3x3x!quant.uniform<u8<1:255>:f32, 0.1>>) -> tensor<32x3x3x3xf32>

  func.return %rst : tensor<256x8x7x4xf32>

// LOBIT-DAG: %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<4xf32>
// LOBIT-DAG: %[[CONSTANT0:.*]] = arith.constant dense<0.000000e+00> : tensor<4x3x3x3xf32>
// LOBIT: %[[QUANTIZE:.*]] = "tfl.quantize"(%[[CONSTANT0]]) <{qtype = tensor<4x3x3x3x!quant.uniform<u4<1:15>:f32:0, {1.000000e+00:1,1.000000e+00:2,1.000000e+00:7,1.000000e+00:15}>>}>

  %cst = "tf.Const"() {value = dense<1.000000e+00> : tensor<2x3x3x3xf32>} : () -> tensor<2x3x3x3xf32>
  %cst_0 = "tf.Const"() {value = dense<0.500000e+00> : tensor<1x3x2x3xf32>} : () -> tensor<1x3x2x3xf32>
  %0 = "tf.Conv2D"(%arg0, %cst) {data_format = "NHWC", dilations = [1, 1, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x3xf32>) -> tensor<1x3x2x3xf32>

  func.return %rst : tensor<256x8x7x16xf32>

// CHECK-DAG: %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<16xf32>
// CHECK-DAG: %[[CONSTANT0:.*]] = arith.constant dense<0.000000e+00> : tensor<16x3x3x3xf32>
// CHECK: %[[QUANTIZE:.*]] = "tfl.quantize"(%[[CONSTANT0]]) <{qtype = tensor<16x3x3x3x!quant.uniform<u8:f32, 1.000000e+00>>}>
// CHECK: %[[DEQUANTIZE:.*]] = "tfl.dequantize"(%[[QUANTIZE]])