(TF_ConstOp (GetI64ScalarElementsAttr<-1>)))),
  (TF_SoftmaxCrossEntropyWithLogitsOp $features, $adjusted_labels)]>;

//===----------------------------------------------------------------------===//
// Dequantize op patterns.
//===----------------------------------------------------------------------===//

def DequantizeHalfRange : NativeCodeCall<
  "DequantizeHalfRange(&$_builder, $0)">;

  func.return %0 : tensor<8x8x8x8xf32>

  // CHECK-LABEL: fakeQuantArgsFalse
  // CHECK: "tfl.quantize"(%arg0) <{qtype = tensor<8x8x8x8x!quant.uniform<u8:f32, 0.0011764706057660721:85>>}>
  // CHECK: %1 = "tfl.dequantize"(%0) : (tensor<8x8x8x8x!quant.uniform<u8:f32, 0.0011764706057660721:85>>) -> tensor<8x8x8x8xf32>
}

    const mlir::TFL::PassConfig& pass_config,
    mlir::OpPassManager* pass_manager) {
  // This pass wraps all the tf.FakeQuant ops in a custom op so they are not
  // folded before being converted to tfl.quantize and tfl.dequantize ops.
  auto wrapped_ops = mlir::TFL::AllTfFakeQuantOps();
  pass_manager->addNestedPass<mlir::func::FuncOp>(
      mlir::TFL::CreateRaiseCustomOpsPass(wrapped_ops));

								}
							}
						}
					}

					if d.progressive {
						// Save the coefficients.
						d.progCoeffs[compIndex][by*mxx*hi+bx] = b
						// At this point, we could call reconstructBlock to dequantize and perform the
						// inverse DCT, to save early stages of a progressive image to the *image.YCbCr
						// buffers (the whole point of progressive encoding), but in Go, the jpeg.Decode

  %cst_0 = arith.constant dense<[1.0, 2.0, 3.0]> : tensor<3xf32>
  %w = arith.constant dense<2.0> : tensor<3x3x3x3xf32>
  %q = "tfl.quantize"(%w) {qtype = tensor<3x3x3x3x!quant.uniform<i8<-127:127>:f32:0,{1.0,2.0,3.0}>>} : (tensor<3x3x3x3xf32>) -> tensor<3x3x3x3x!quant.uniform<i8<-127:127>:f32:0,{1.0,2.0,3.0}>>
  %dq = "tfl.dequantize"(%q) : (tensor<3x3x3x3x!quant.uniform<i8<-127:127>:f32:0,{1.0,2.0,3.0}>>) -> tensor<3x3x3x3xf32>

// set of acceptable options.
// LINT.IfChange
enum BuiltinOperator : int32 {
  ADD = 0,
  AVERAGE_POOL_2D = 1,
  CONCATENATION = 2,
  CONV_2D = 3,
  DEPTHWISE_CONV_2D = 4,
  DEPTH_TO_SPACE = 5,
  DEQUANTIZE = 6,
  EMBEDDING_LOOKUP = 7,
  FLOOR = 8,
  FULLY_CONNECTED = 9,
  HASHTABLE_LOOKUP = 10,
  L2_NORMALIZATION = 11,
  L2_POOL_2D = 12,
  LOCAL_RESPONSE_NORMALIZATION = 13,
  LOGISTIC = 14,

      if (fc_op.getFilter() != filter) {
        // This filter goes through quantize and dequantize ops. Then we just
        // need to update the weight to the quantize op.
        filter.replaceAllUsesWith(new_filter_op);
      } else {
        // This filter doesn't go through quantize and dequantize ops, Then
        // we update the weight of the affine op directly.

  %2 = "tfl.dequantize"(%1) : (tensor<1x384x384x!quant.uniform<i8:f32, 0.003:-128>>) -> tensor<1x384x384xf32>
  %3 = "tfl.pseudo_const"() {value = dense<1.000000e+00> : tensor<1x1x384xf32>} : () -> tensor<1x384x384xf32>

// set of acceptable options.
// LINT.IfChange
enum BuiltinOperator : int32 {
  ADD = 0,
  AVERAGE_POOL_2D = 1,
  CONCATENATION = 2,
  CONV_2D = 3,
  DEPTHWISE_CONV_2D = 4,
  DEPTH_TO_SPACE = 5,
  DEQUANTIZE = 6,
  EMBEDDING_LOOKUP = 7,
  FLOOR = 8,
  FULLY_CONNECTED = 9,
  HASHTABLE_LOOKUP = 10,
  L2_NORMALIZATION = 11,
  L2_POOL_2D = 12,
  LOCAL_RESPONSE_NORMALIZATION = 13,
  LOGISTIC = 14,

    ([CVE-2022-21728](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2022-21728))
*   Fixes a heap OOB access in `Dequantize`
    ([CVE-2022-21726](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2022-21726))
*   Fixes an integer overflow in shape inference for `Dequantize`
    ([CVE-2022-21727](https://cve.mitre.org/cgi-bin/cvename.cgi?name=CVE-2022-21727))
*   Fixes a heap OOB access in `FractionalAvgPoolGrad`