varTimeDoubleBaseMultMatchesBaseMult := func(x, y Scalar) bool {
		var p, q1, q2, check Point

		p.VarTimeDoubleScalarBaseMult(&x, B, &y)

		q1.ScalarBaseMult(&x)
		q2.ScalarBaseMult(&y)
		check.Add(&q1, &q2)

		checkOnCurve(t, &p, &check, &q1, &q2)
		return p.Equal(&check) == 1
	}

	if err := quick.Check(varTimeDoubleBaseMultMatchesBaseMult, quickCheckConfig(32)); err != nil {
		t.Error(err)
	}

  func.return %1 : tensor<1x28x23x2xf32>
  // CHECK: %cst_0 = arith.constant dense<3.000000e+00> : tensor<23x2xf32>
  // CHECK: %0 = "tf.Conv2D"(%arg0, %cst) <{data_format = "NHWC", dilations = [1, 2, 3, 1], padding = "SAME", strides = [1, 4, 5, 1]}> {T = "tfdtype$DT_FLOAT"}
  // CHECK: %1 = "tf.Mul"(%0, %cst_0) : (tensor<1x28x23x2xf32>, tensor<23x2xf32>) -> tensor<1x28x23x2xf32>
  // CHECK: return %1 : tensor<1x28x23x2xf32>
}

// CHECK-LABEL: depthwiseConv2D
// CHECK-DAG:  %[[CONSTANT:.*]] = arith.constant dense<0.000000e+00> : tensor<12xf32>
// CHECK-DAG:  %[[CONSTANT0:.*]] = arith.constant dense<[1, 3, 3, 12]> : tensor<4xi32>
// CHECK:  %0 = "tf.Reshape"(%arg1, %[[CONSTANT0]]) : (tensor<3x3x3x4xf32>, tensor<4xi32>) -> tensor<1x3x3x12xf32>

  func.return %4 : i32
// CHECK-LABEL: squeezeAndReshape
// CHECK:  "tfl.squeeze"(%arg0) <{squeeze_dims = [0]}> : (tensor<1x1x10xf32>) -> tensor<1x10xf32>
// CHECK:  %1 = "tfl.squeeze"(%arg1) <{squeeze_dims = []}> : (tensor<?x10xf32>) -> tensor<*xf32>
// CHECK:  %cst = arith.constant dense<[2, 5]> : tensor<2xi32>
// CHECK:  %2 = "tfl.reshape"(%0, %cst) : (tensor<1x10xf32>, tensor<2xi32>) -> tensor<2x5xf32>

// CHECK:       %0 = "tfl.while"(%arg0) ({
// CHECK:       ^bb0(%arg1: tensor<i32>):
// CHECK:         %1 = tfl.add %arg0, %arg0 {fused_activation_function = "NONE", tac.device = "CPU", tac.inference_type = "FLOAT"} : tensor<i32>
// CHECK:         %2 = func.call @func_1_GPU_FLOAT(%1) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_1"} : (tensor<i32>) -> tensor<i32>

  return %2 : tensor<1x1024xf32>
  // CHECK: %cst = arith.constant dense<[1, 1024]> : tensor<2xi32>
  // CHECK: %cst_0 = arith.constant dense<[1, 1, 64]> : tensor<3xi32>
  // CHECK: %0 = "tfl.reshape"(%arg0, %cst_0) : (tensor<1x64xf32>, tensor<3xi32>) -> tensor<1x1x64xf32>

  // CHECK: %1 = "tfl.transpose"(%0, %[[cst_0]]) : (tensor<7x6x5x4x3x2xf32>, tensor<6xi32>) -> tensor<2x3x4x5x6x7xf32>
  // CHECK: %2 = "tfl.reshape"(%1, %[[cst_1]]) : (tensor<2x3x4x5x6x7xf32>, tensor<7xi32>) -> tensor<1x2x3x4x5x6x7xf32>
  // CHECK: return %2
}

// CHECK-LABEL: OptimizeTranposeWithRank7orMoreEffectiveRank4
func.func @OptimizeTranposeWithRank7orMoreEffectiveRank4(%arg0: tensor<56x8x56x1x1x1x7xf32> ) -> (tensor<1x1x8x56x56x7x1xf32>)  {

  func.return %4#0, %4#1 : tensor<i16>, tensor<i32>

  // CHECK:  %0 = mhlo.constant dense<false> : tensor<i1>
  // CHECK:  %1 = "mhlo.iota"() <{iota_dimension = 0 : i64}> : () -> tensor<2xi32>
  // CHECK-DAG:  %2 = mhlo.constant dense<32767> : tensor<i16>
  // CHECK:  %3 = mhlo.constant dense<0> : tensor<i32>
  // CHECK:  %cst = arith.constant dense<0> : tensor<1xi32>

  // CHECK: %0 = "tf.Bitcast"(%arg0) : (tensor<8x16x32x64xf32>) -> tensor<8x16x32x64x2xi16>
  // CHECK: %1 = "tf.Bitcast"(%arg0) {device = "/job:localhost/replica:0/task:0/device:GPU:0"} : (tensor<8x16x32x64xf32>) -> tensor<8x16x32x64xi32>
  // CHECK: %2 = "tf.Identity"(%0) : (tensor<8x16x32x64x2xi16>) -> tensor<8x16x32x64x2xi16>
  // CHECK: return %1, %2
}

// CHECK-LABEL: testSameCastType

// CHECK:         %0 = "tf.Sub"(%arg0, %arg1) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>
// CHECK:         %1 = "tf.Mul"(%0, %0) : (tensor<4xf32>, tensor<4xf32>) -> tensor<4xf32>
// CHECK:         return %1 : tensor<4xf32>
// CHECK:       }

// -----

func.func @unrollSplit(%arg0: tensor<i32>, %arg1: tensor<1x8x8x1024xf32>) -> (tensor<1x8x8x256xf32>, tensor<1x8x8x256xf32>, tensor<1x8x8x256xf32>) {