// CHECK: %[[CONV:.+]] = mhlo.convolution(%[[INPUT]], %[[NEW_FILTER]]) dim_numbers = [b, 0, 1, f]x[0, 1, i, o]->[b, 0, 1, f], window = {stride = [1, 1], pad = {{\[\[}}0, 0], [0, 0]], rhs_dilate = [1, 1]} {batch_group_count = 1 : i64, feature_group_count = 1 : i64} : (tensor<?x256x256x3xf32>, tensor<1x1x3x2xf32>) -> tensor<?x256x256x2xf32>
  // CHECK: %[[SHAPE:.+]] = shape.shape_of %[[CONV]] : tensor<?x256x256x2xf32> -> tensor<4xindex>

  Pat<(TF_MulOp:$mul (TF_Conv2DOp:$conv $input,
                          (Arith_ConstantOp:$filter F32ElementsAttr:$filter_value),
                          $strides, $use_cudnn, $padding, $explicit_padding,
                          IsDataFormatNHWC:$data_format, $dilations),
                     (Arith_ConstantOp:$multiplier F32ElementsAttr:$mul_value)),
// TODO(karimnosseir): Add check for $conv is of rank 4.
      (TF_Conv2DOp $input,

// CHECK: %[[CONV:.+]] = stablehlo.convolution(%[[ARG1]], %[[ARG2]])
// CHECK-SAME: (tensor<1x3x4x3xf32>, tensor<2x3x3x2x!quant.uniform<i8<-127:127>:f32, 0.0023622048182750312>>) -> tensor<1x3x4x2xf32>
// CHECK: return %[[CONV]]

// -----

// Test that per-channel weight-only quantized dot_general op is produced when

        String expectedValue = 'somevalue'

        when:
        project.convention.plugins.test = new TestConvention()
        project.conv = expectedValue

        then:
        project.conv == expectedValue
        project.convention.plugins.test.conv == expectedValue
        child1.conv == expectedValue
    }

    def setPropertyAndPropertyMissingWithProjectAndConventionProperty() {
        given:

// CHECK: %[[DEQUANTIZE:.*]] = "tfl.dequantize"(%[[QUANTIZE]])
// CHECK: %[[CONV:.*]] = "tfl.conv_2d"(%arg0, %[[DEQUANTIZE]], %[[CONSTANT]])
// CHECK: return %[[CONV]]
}

// CHECK-LABEL: perChannelFakeQuantWithConv2D
func.func @perChannelFakeQuantWithConv2D(tensor<256x32x32x3xf32>) -> (tensor<256x8x7x16xf32>) {

    %1 = "tf.PartitionedCall"(%arg0, %cst, %cst_0) {_tfl_quant_trait = "fully_quantizable", config = "", config_proto = "", executor_type = "", f = @composite_conv2d_with_bias_and_relu6_fn_1} : (tensor<1x2x2x3xf32>, tensor<2x2x3x2xf32>, tensor<2xf32>) -> tensor<*xf32> loc(callsite("test@conv"("Conv2D_1") at "QuantizationUnit(\12\08Conv2D_1\1a\04conv)"))

// CHECK: %[[DEQUANTIZE:.*]] = "tfl.dequantize"(%[[QUANTIZE]])
// CHECK: %[[CONV:.*]] = "tfl.conv_2d"(%arg0, %[[DEQUANTIZE]], %[[CONSTANT]])
// CHECK: return %[[CONV]]
}

// CHECK-LABEL: perChannelFakeQuantWithConv2D
func.func @perChannelFakeQuantWithConv2D(tensor<256x32x32x3xf32>) -> (tensor<256x8x7x16xf32>) {

  // dilation rate.
  // TF python library will rewrite dilated conv to
  // "SpaceToBatch->Conv->BatchToSpace" pattern, and the Conv in the middle
  // always has 'VALID' padding. The padding tensor in `SpaceToBatch` has two
  // parts of contributions, one is to reduce padding of CONV from 'SAME' to
  // 'VALID', and another is to make input shape multiples of dilation rate. The

    return %0 : tensor<1x3x4x2xf32>
  }
  // CHECK: func private @composite_conv_fn
  // CHECK: %[[CONV:.+]] = stablehlo.convolution
  // CHECK: return %[[CONV]]
}

// -----

// Test that q/dq pair with per-channel quantization parameter is inserted
// between constant and XlaCallModule op with `weight_only_ptq` method of

    func.return %7 : tensor<1x112x112x32x!quant.uniform<u8:f32, 1.0>>

// CHECK: %[[conv:.*]] = "tfl.conv_2d"(%arg0, %arg1, %arg2)
// CHECK-SAME: -> tensor<1x112x112x32x!quant.uniform<u8:f32, 0.0078431372549019607:128>>
// CHECK: %[[cst:.*]] = "tfl.pseudo_qconst"()
// CHECK: %[[add:.*]] = tfl.add(%[[conv]], %[[cst]])
// CHECK-SAME: -> tensor<1x112x112x32x!quant.uniform<u8:f32, 1.000000e+00>>
// CHECK: return %[[add]]
}