// RUN: tf-quant-opt %s -split-input-file -quant-insert-quantized-functions -quant-quantize-composite-functions='quantization-method=ptq target-opset=XLA' | FileCheck %s
// RUN: tf-quant-opt %s -split-input-file -quant-insert-quantized-functions -quant-quantize-composite-functions='quantization-method=ptq target-opset=XLA enable-per-channel-quantization=true' | FileCheck --check-prefix=PerChannel %s

module {

// CHECK-DAG: %[[w_quant:.*]] = "tf.Const"() <{value = dense<{{\[\[\[\[}}40, 20]
// CHECK-DAG: {{\[\[\[}}-87, -42]

// CHECK: %[[quantize:.*]] = "tf.PartitionedCall"(%arg0, %[[in_scale]], %[[in_zp]])
// CHECK-SAME: f = @quantize_i8
// CHECK: %[[conv_quant:.*]] = "tf.PartitionedCall"(%[[quantize]], %[[w_quant]], %[[b_quant]],
// CHECK-SAME: %[[in_scale]], %[[in_zp]], %[[w_scale]], %[[w_zp]],

	for i := 0; i < samples; i++ {
		window := time.Duration(math.Exp(float64(i)/(samples-1)*(logMax-logMin) + logMin))
		if quantiles == nil {
			plot[i] = make([]float64, 2)
			plot[i][1] = mmuCurve.MMU(window)
		} else {
			plot[i] = make([]float64, 1+len(quantiles))
			copy(plot[i][1:], mmuCurve.MUD(window, quantiles))
		}
		plot[i][0] = float64(window)
	}

	// Create JSON response.

      Map<Integer, Double> quantiles = algorithm.multipleQuantiles(indexes, 100, dataset.clone());
      assertWithMessage("Wrong keys from " + algorithm).that(quantiles.keySet()).isEqualTo(indexes);
      for (int i : indexes) {
        assertWithMessage("Mismatch between %s and %s at %s", algorithm, REFERENCE_ALGORITHM, i)
            .that(quantiles.get(i))
            .isWithin(ALLOWED_ERROR)

// of "apply<Name of the Quantization Algorithm>Quantization".
// After applying the function, a quantize/dequantize functions are created
// where the body of each function contains a specific quantization algorithm.
// The input of the quantize function has one operand of
// IsValueWithQuantizablePrecision and the output is a tensor with supported

  func.return %rst : tensor<8xf32>

// CHECK: %[[CONSTANT:.*]] = "tf.Const"() <{value = dense<0.000000e+00> : tensor<8xf32>}>
// CHECK: %[[QUANTIZE:.*]] = "quantfork.qcast"(%[[CONSTANT]]) : (tensor<8xf32>) -> tensor<8x!quant.uniform<i4:f32, 1.000000e+00:-8>>
// CHECK: %[[DEQUANTIZE:.*]] = "quantfork.dcast"(%[[QUANTIZE]])
// CHECK: return %[[DEQUANTIZE]] : tensor<8xf32>
}

// CHECK-LABEL: fakeQuantNotFolded

                                       int32_t qmax) {
  auto quantize = [scale, zero_point](float f) {
    return zero_point + static_cast<int32_t>(std::round(f / scale));
  };

  if (rmin.has_value() && rmax.has_value()) {
    return {std::max(qmin, quantize(rmin.value())),
            std::min(qmax, quantize(rmax.value()))};
  } else if (rmin.has_value()) {
    return {std::max(qmin, quantize(rmin.value())), qmax};
  } else if (rmax.has_value()) {

// RUN: tf-quant-opt %s -split-input-file -quant-lift-quantizable-spots-as-functions -quant-prepare-quantize-drq -quant-quantize='weight-quantization=true' -verify-each=false | FileCheck %s

// -----

module {
  func.func @matmul(%arg0: tensor<1x2x2x3xf32>) -> (tensor<*xf32>) {
    %cst_0 = "tf.Const"() {value = dense<0.000000e+00> : tensor<2x1024xf32>} : () -> tensor<2x1024xf32>

message QuantizationInfo {
  // min/max of the per axis value range. To quantize the value, the metadata
  // of the target properties should be specified or read from the ops
  // quantization specification.
  message MinMax {
    float min = 1;
    float max = 2;
  }

  // Affine parameters to quantize the per axis value. The metadata of the
  // target properties should be specified as well.

    // This is the argument used to refer to the pass in
    // the textual format (on the commandline for example).
    return "quant-quantize-weights";
  }

  StringRef getDescription() const final {
    // This is a brief description of the pass.
    return "Quantize weights used by quantizable ops.";
  }

  void getDependentDialects(DialectRegistry& registry) const override {