<td align="left" width="10" style="padding:0;margin:0;line-height:1px;font-size:1px;"></td>
<td align="left" style="padding:0;margin:0;line-height:1px;font-size:1px;">
<table align="left" cellpadding="0" cellspacing="0" border="0" style="padding:0px; margin:0px;padding:0;margin:0;line-height:1px;font-size:1px;">
<tbody>
<tr>
<td align="left" style="padding:0;margin:0;line-height:1px;font-size:1px;">

    clear:left;
    float:left;
    padding:6px;
    text-transform:uppercase;
}
.sub-nav .sub-nav-list {
    padding-top:4px;
}
ul.nav-list {
    display:block;
    margin:0 25px 0 0;
    padding:0;
}
ul.sub-nav-list {
    float:left;
    margin:0 25px 0 0;
    padding:0;
}
ul.nav-list li {
    list-style:none;
    float:left;
    padding: 5px 6px;
    text-transform:uppercase;

	body.toc2 {
		padding-left: 20em;
		padding-right: 0;
	}

	#toc.toc2 {
		width: 20em;
	}

	#toc.toc2 #toctitle {
		font-size: 1.375em;
	}

	#toc.toc2>ul {
		font-size: 0.95em;
	}

	#toc.toc2 ul ul {
		padding-left: 1.25em;
	}

	body.toc2.toc-right {
		padding-left: 0;
		padding-right: 20em;
	}
}

   %2 = "tf.Conv2D"(%arg0, %arg1) {T = "tfdtype$DT_FLOAT", data_format = "NHWC",                           padding = "VALID", strides = [1, 4, 5, 1]} : (tensor<256x32x32x3xf32>, tensor<3x3x3x16xf32>) -> tensor<256x8x6x16xf32>
   // Unsupported padding

  // CHECK-DAG: [[PADDED:%.+]] = "tf.PadV2"(%arg0, [[FULL_PADDINGS]], [[PAD_DEFAULT]])
  // CHECK-DAG: [[PADDINGS:%.+]]:2 = "tf.Unpack"([[FULL_PADDINGS]]) <{axis = 1 : i64}>
  // CHECK-DAG: [[PADDINGS_SUM:%.+]] = "tf.AddV2"([[PADDINGS]]#0, [[PADDINGS]]#1)
  // CHECK-DAG: [[INPUT_SHAPE:%.+]] = "tf.Const"() <{value = dense<[3, 5, 7, 10]> : tensor<4xi64>}>

                 $w_factor, TFL_AF_None, $padding, $stride_h, $stride_w)),
    (TFL_Conv2DOp $input, $filter, $bias, $h_factor, $w_factor, ActFnAttr,
        $padding, $stride_h, $stride_w),
    [(HasOneUse $conv_out)]>;
  def FuseActivationFuncWithDepthwiseConv#ActFnOp#ActFnAttr : Pat<
    (ActFnOp (TFL_DepthwiseConv2DOp:$conv_out $input, $filter, $bias, $h_factor,
                $w_factor, TFL_AF_None, $padding, $stride_h, $stride_w,

//
// Before rewrite:
//   output = SpaceToBatchND(input, block_shape, paddings)
// Let:
//   [batch] + spatial_shape + remaining_shape = input.shape
//   M = spatial_shape.rank
// After rewrite:
//   padded = zero-pad input with paddings
//     The spatial_shape component of input.shape pads with paddings[*, 0]
//     before each dimension, and paddings[*, 1] after each dimension.
//   reshaped = reshape padded to:
//     [batch]

      name = "avgpool", filter_height = 3 : i32, filter_width = 6 : i32, fused_activation_function = "NONE", padding = "VALID", stride_h = 3 : i32, stride_w = 1 : i32
    } : (tensor<1x224x224x3xf32>) -> tensor<1x224x224x3xf32>
  %6 = "tfl.conv_2d"(%5, %4, %cst) {
      dilation_h_factor = 1 : i32, dilation_w_factor = 1 : i32, fused_activation_function = "NONE", padding = "SAME", stride_h = 2 : i32, stride_w = 2 : i32

	good &= good << 1
	good = uint8(int8(good) >> 7)

	// Zero the padding length on error. This ensures any unchecked bytes
	// are included in the MAC. Otherwise, an attacker that could
	// distinguish MAC failures from padding failures could mount an attack
	// similar to POODLE in SSL 3.0: given a good ciphertext that uses a
	// full block's worth of padding, replace the final block with another

        mean, variance = scale, offset
        out = nn_ops.conv2d(
            input_tensor,
            self.filters,
            strides=strides,
            dilations=dilations,
            padding=padding,
            data_format='NHWC',
        )
        if has_bias:
          out = nn_ops.bias_add(out, self.bias, data_format='NHWC')
        if has_batch_norm: