(UpdateShapeWithAxis<-1> $qtype, $old_value))),
          [(CanUpdateShapeWithAxis<-1> $qtype, $old_value)]>;

// The axis is set to 0 because the transpose is from the legalization of
// tf.conv2d and the new channel axis is the first dimension.
def ReorderTransposeDequantQuantUsedByConv :
      Pat<(TF_TransposeOp:$old_value
              (TFL_DequantizeOp (TFL_QuantizeOp $input, $qtype)), $perm),

    %conv = "tf.Conv2D"(%arg0, %arg1) {attr_map = "0:strides,1:use_cudnn_on_gpu,2:padding,3:explicit_paddings,4:dilations", data_format = "NHWC", device = "", dilations = [1, 2, 2, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<1x2x2x3xf32>, tensor<2x3x3x2xf32>) -> tensor<*xf32>
    return %conv : tensor<*xf32>
  }

        # One pure conv
        out = nn_ops.conv2d(
            out,
            self.conv_filters,
            strides=(1, 1, 2, 1),
            dilations=(1, 1, 1, 1),
            padding='SAME',
            data_format='NHWC',
        )

        # One fakequant attached conv
        if is_qat_model:
          out = array_ops.fake_quant_with_min_max_args(

      if (!input_type || !input_type.hasStaticShape()) {
        return false;
      }
      total_count += input_type.getNumElements();
    }
    *count = total_count;
    return true;
  }

  // For conv2d/depthwise_conv/fully_connected ops.
  // This algorithm actually comes from TOCO tooling_util.cc
  static bool GetArithmeticCountForConvAndFullyconnectedOp(mlir::Operation* op,

                         %input : tensor<*xf32>, %weight : tensor<*x!tf_type.qint8>,
                         %weight_scale : tensor<*xf32>, %weight_zp : tensor<*xi32>) -> tensor<*xf32>
      attributes {tf_quant.quantized_ops = ["Conv2D"]} {

    %out = "tf.UniformQuantizedConvolutionHybrid"(%input, %weight,
                           %weight_scale, %weight_zp) {
        Tlhs = "tfdtype$DT_FLOAT",
        Trhs = "tfdtype$DT_QINT8",

  %conv1 = "tfl.conv_2d"(%1, %2, %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} : (tensor<1x224x224x3xf32>, tensor<32x3x3x3xf32>, tensor<32xf32>) -> tensor<1x112x112x32xf32>

handle git

env GIT_AUTHOR_NAME='Jay Conrod'
env GIT_AUTHOR_EMAIL='******@****.***'
env GIT_COMMITTER_NAME=$GIT_AUTHOR_NAME
env GIT_COMMITTER_EMAIL=$GIT_AUTHOR_EMAIL

at 2020-10-09T13:37:47-04:00

git init

git add go.mod p.go
git commit -m 'create module retract-pseudo'
git branch -m main
git tag v1.0.0

git mv p.go q.go
git commit -m 'trivial change'

git show-ref --tags --heads
cmp stdout .git-refs

func.func @testConv(%arg0: tensor<256x32x32x3xf32>, %arg1: tensor<16x3x3x3xf32>, %arg2: tensor<16xf32>) -> tensor<256x30x30x16xf32> {
  // CHECK: tac.device = "GPU", tac.inference_type = "FLOAT"

// DEBUG: %[[act:.*]] = "tfl.dequantize"(%arg0) : (tensor<1x224x224x3x!quant.uniform<u8:f32, 7.812500e-03:128>>) -> tensor<1x224x224x3xf32>
// DEBUG: %[[f_conv:.*]] = "tfl.conv_2d"(%[[act]], %[[wt]], %[[bias]])
// DEBUG: %[[q_conv:.*]] = "tfl.conv_2d"
// DEBUG: "tfl.NumericVerify"(%[[q_conv]], %[[f_conv]]) <{log_if_failed = true, tolerance = 5.000000e+00 : f32}>

  %1 = "tf.FakeQuantWithMinMaxArgs"(%arg0) {device = "", max = 2.000000e-01 : f32, min = -1.000000e-01 : f32, narrow_range = false, num_bits = 8 : i64} : (tensor<1x3x4x3xf32>) -> tensor<*xf32>
  %2 = "tf.Conv2D"(%1, %0) {data_format = "NHWC", device = "", dilations = [1, 1, 1, 1], explicit_paddings = [], padding = "SAME", strides = [1, 1, 2, 1], use_cudnn_on_gpu = true} : (tensor<*xf32>, tensor<*xf32>) -> tensor<*xf32>