%1 = "tf.Conv2D"(%0, %cst) {data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32>
    %2 = "tf.Relu"(%1) {device = ""} : (tensor<1x3x2x2xf32>) -> tensor<1x3x2x2xf32>

      %outputs_4, %control_5 = tf_executor.island wraps "tf.Conv2D"(%outputs, %outputs_2) {data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 1, 1], use_cudnn_on_gpu = true} : (tensor<*xf32>, tensor<5x5x32x16xf32>) -> tensor<*xf32>

  %conv = "tf.Conv2D"(%dq_input, %dq_weight) {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "VALID", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>

  // CHECK-DAG: %[[RESULT:.+]] = 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<1x256x256x3xf32>, tensor<1x1x3x2xf32>) -> tensor<1x256x256x2xf32>

type blas64_Triangular struct {
	Stride int
	Uplo   int
}

// TriDense represents an upper or lower triangular matrix in dense storage
// format.
type TriDense struct {
	mat blas64_Triangular
}

func NewTriDense() *TriDense {
	return &TriDense{
		mat: blas64_Triangular{
			Stride: 3,
			Uplo:   blas_Upper,
		},
	}
}

func (t *TriDense) isUpper() bool {

    %0 = "tf.Cast"(%arg0) {Truncate = false, device = ""} : (tensor<1x3x4x3xf32>) -> tensor<1x3x4x3xf32>
    %1 = "tf.Conv2D"(%0, %cst) {data_format = "NHWC", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true}
        : (tensor<1x3x4x3xf32>, tensor<2x3x3x2xf32>) -> tensor<1x3x2x2xf32> loc(fused["Conv2D:", "Model/conv2d"])

  %5 = "tf.AddV2"(%strides, %strides) {device = ""} : (tensor<3xi32>, tensor<3xi32>) -> tensor<3xi32>

  return %0 : tensor<8x4x14x14x16xf32>
}

// Check that the constants that are input to XlaConvV2's 3rd, 4th, 5th, 6th
// and 7th arguments are cloned with same value.
// CHECK-DAG: %[[STRIDES:.*]] = "tf.Const"()
// CHECK-SAME: value = dense<[3, 1, 1]> : tensor<3xi32>

  func.return %0 : tensor<256x30x30x16xf32>
}

// -----

func.func @testConv2D(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<3x3x3x16xf32>) -> tensor<256x30x30x16xf32> {
  // expected-error @+1 {{requires positive strides}}

      CompileTimeConstantOperand<TF::MaxOp, 1>,  // $reduction_indices
      // $ksize, $strides
      CompileTimeConstantOperand<TF::MaxPoolGradGradV2Op, 3, 4>,
      // $ksize, $strides
      CompileTimeConstantOperand<TF::MaxPoolGradV2Op, 2, 3>,
      CompileTimeConstantOperand<TF::MaxPoolV2Op, 1, 2>,   // $ksize, $strides
      CompileTimeConstantOperand<TF::MeanOp, 1>,           // $reduction_indices

			} else if nrgba != nil {
				stride, pix = nrgba.Stride, nrgba.Pix
			}
			if stride != 0 {
				j0 := (y - b.Min.Y) * stride
				j1 := j0 + b.Dx()*4
				for j := j0; j < j1; j += 4 {
					cr0[i+0] = pix[j+0]
					cr0[i+1] = pix[j+1]
					cr0[i+2] = pix[j+2]
					i += 3
				}
			} else {
				for x := b.Min.X; x < b.Max.X; x++ {
					r, g, b, _ := m.At(x, y).RGBA()