^bb0(%arg0: tensor<8xf32>, %arg3: tensor<f32>, %arg4: tensor<f32>):
  %1 = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg3, %arg4) {num_bits = 3, narrow_range = false} : (tensor<8xf32>, tensor<f32>, tensor<f32>) -> tensor<8xf32>
  func.return %1 : tensor<8xf32>

// CHECK: %0 = "tf.FakeQuantWithMinMaxVars"(%arg0, %arg1, %arg2)
// CHECK: return %0 : tensor<8xf32>
}

		{{end}}

		// round 1
		{{range $i, $s := dup 4 .Shift1 -}}
			{{printf "arg0 = arg1 + bits.RotateLeft32((((arg2^arg3)&arg1)^arg3)+arg0+x%x+%#08x, %d)" (idx 1 $i) (index $.Table1 $i) $s | relabel}}
			{{rotate -}}
		{{end}}

		// round 2
		{{range $i, $s := dup 4 .Shift2 -}}
			{{printf "arg0 = arg1 + bits.RotateLeft32((((arg1^arg2)&arg3)^arg2)+arg0+x%x+%#08x, %d)" (idx 2 $i) (index $.Table2 $i) $s | relabel}}
			{{rotate -}}

		{name: "FNMSUBS", argLength: 3, reg: fp31, asm: "FNMSUBS"}, // -arg0 + (arg1 * arg2)
		{name: "FNMSUBD", argLength: 3, reg: fp31, asm: "FNMSUBD"}, // -arg0 + (arg1 * arg2)
		{name: "MADD", argLength: 3, reg: gp31, asm: "MADD"},       // +arg0 + (arg1 * arg2)
		{name: "MADDW", argLength: 3, reg: gp31, asm: "MADDW"},     // +arg0 + (arg1 * arg2), 32-bit
		{name: "MSUB", argLength: 3, reg: gp31, asm: "MSUB"},       // +arg0 - (arg1 * arg2)

func.func @num_replicas_replicated(%arg0: tensor<i32>, %arg1: tensor<i32>, %arg2: tensor<i32>, %arg3: tensor<i32>) -> (tensor<i32>, tensor<i32>) {
  %0:2 = tf_executor.graph {
    %control = tf_executor.island() wraps "tf.TPUReplicateMetadata"() {_tpu_replicate = "cluster", device = "/device:TPU:0", num_replicas = 2, topology = "topology"} : () -> ()

// CHECK:           return %[[VAL_9]] : tensor<384x512xf32>
// CHECK:         }

// -----

func.func @padConcatTo4D(%arg0: tensor<384x384xf32>, %arg1: tensor<384x384xf32>, %arg2: tensor<384x384xf32>, %arg3: tensor<384x384xf32>) -> tensor<1536x384xf32> {

module {
func.func @main(%arg0: tensor<1xf32>, %arg1: tensor<1xf32>, %arg2: tensor<1xf32>, %arg3: tensor<1xf32>) -> tensor<2x1xf32> attributes {tf.entry_function = {inputs = "input0,input1,input2,input3", outputs = "output"}} {
  %0 = "tfl.add"(%arg0, %arg1) {fused_activation_function = "RELU6"} : (tensor<1xf32>, tensor<1xf32>) -> tensor<1xf32>

module {
  func.func @main(%arg0: tensor<100xf32>, %arg1: tensor<100xf32>, %arg2: tensor<100xf32>, %arg3: tensor<100xf32>) -> tensor<2x100xf32> {
    %0 = func.call @func_0_GPU_FLOAT(%arg0, %arg1, %arg2) {tac.device = "GPU", tac.inference_type = "FLOAT", tac.interface_name = "func_0"} : (tensor<100xf32>, tensor<100xf32>, tensor<100xf32>) -> tensor<100xf32>

// CHECK: %[[dw:.*]] = "tfl.depthwise_conv_2d"(%arg0, %arg1, %arg2)
// CHECK: %[[q:.*]] = "tfl.quantize"(%[[dw]])
// CHECK: %[[dq:.*]] = "tfl.dequantize"(%[[q]])
// CHECK: return %[[dq]], %[[dq]]
}

// Series of values needing requantization -- first the args then the results
// of concatenation operations. concat(concat(arg2, arg0), concat(arg1, arg0)),
// concat(concat(arg2, arg0), arg3)). arg0 should be requantized twice --

      ```mlir
        func.func @remove_first_result(%arg0, %arg1) {
          %0:2 = "tf.WhileRegion"(%arg0, %arg1) ({
          ^bb0(%arg2, %arg3):
            %1 = "tf.OpA"() {is_stateless = true}
            "tf.Yield"(%1)
          }, {
          ^bb0(%arg2, %arg3):
            %1 = "tf.OpB"(%arg2) {is_stateless = true}
            %2 = "tf.OpC"(%arg3) {is_stateless = true}
            "tf.Yield"(%1, %2)

  func.func @static_and_dynamic_shapes(%arg0: tensor<*xi32>, %arg1: tensor<8xi32>, %arg2: tensor<*xi32>, %arg3: tensor<8xi32>) -> tensor<8xi32> {
    // CHECK-NOT:  "tf.Shape"(%[[ARG_1]])
    // CHECK-NOT:  "tf.Shape"(%[[ARG_3]])
    // CHECK:      %[[ARG_0_SHAPE:[0-9]*]] = "tf.Shape"(%[[ARG_0]])
    // CHECK:      %[[ARG_2_SHAPE:[0-9]*]] = "tf.Shape"(%[[ARG_2]])