}
      return HashCode.fromBytesNoCopy(bytes);
    }

    @Override
    public int bits() {
      int bitSum = 0;
      for (HashFunction function : functions) {
        bitSum += function.bits();
      }
      return bitSum;
    }

    @Override
    public boolean equals(@CheckForNull Object object) {
      if (object instanceof ConcatenatedHashFunction) {

  }

  void getDependentDialects(DialectRegistry& registry) const override {
    registry.insert<TF::TensorFlowDialect, arith::ArithDialect>();
  }

  void runOnOperation() override;
};

// Generate an einsum equation from the given DotDimensionNumber.
std::string CreateEinsumEquation(
    const xla::DotDimensionNumbers& dot_dimension_numbers, const int lhs_rank,
    const int rhs_rank) {
  // Prepare necessary indices.

`archiveBaseName`::
Defaults to `$archivesBaseName`.

[[sec:base_plugin_conventions]]
== Conventions (deprecated)

The Base Plugin also adds conventions related to the creation of archives, such as ZIPs, TARs and JARs.
These are *deprecated* and superseded by the extension described above.

            tensor_spec.TensorSpec(
                shape=input_shape, dtype=dtypes.float32, name='input_tensor'
            )
        ),
    )
    return model

  # Prepares sample einsum input data shapes.
  # This function returns:
  # 1. Shape for input 1
  # 2. Shape for input 2
  # 3. Shape for bias
  # 4. Signature for input 1 (Could contain None dimension)

std::unique_ptr<OperationPass<func::FuncOp>> CreateUnrollBatchMatMulPassPass();

// Optional pass which will map TF BatchMatMul to TF Einsum
std::unique_ptr<OperationPass<func::FuncOp>> CreateBatchMatMulToEinsumPass();

// Pass that transform Einsum to other TF Ops for the supported variants.
std::unique_ptr<OperationPass<func::FuncOp>> CreateTransformEinsumPass();

// Optimizes Tensorflow graph.

        testDir = temporaryFolder.testDirectory
        distributionDir = new TestFile(testDir, 'someDistPath')
        gradleHomeDir = new TestFile(distributionDir, 'gradle-0.9')
        zipStore = new File(testDir, 'zips')
        zipDestination = new TestFile(zipStore, zipFileName)
        install = new Install(new Logger(true), download, pathAssembler)
    }

    private void initConfiguration() {

    %16 = stablehlo.broadcast_in_dim %6, dims = [0, 1, 2] : (tensor<1x1x1xi32>) -> tensor<8x16x4xi32>
    %17 = stablehlo.subtract %15, %16 : tensor<8x16x4xi32>  // q2 - z2
    // Corresponds to einsum expression: b i j, b j d -> b i d
    %18 = stablehlo.dot_general %14, %17, batching_dims = [0] x [0], contracting_dims = [2] x [1] : (tensor<8x16x16xi32>, tensor<8x16x4xi32>) -> tensor<8x16x4xi32>

SmallVector<Value> AppendToVector(const ArrayRef<Value> arguments,
                                  Value append) {
  SmallVector<Value> ret(arguments);
  ret.push_back(append);
  return ret;
}

// Check if the given einsum equation is supported by XlaDotV2.
// Conditions:
// 1. Two inputs & one output.
// 2. No ... in the equation.
// 3. Batch dimensions should be the same, or only the left equation should have

          quantize_func_count++;
        } else if (func_name.starts_with("dequantize_i")) {
          dequantize_func_count++;
        }
      } else if (auto einsum = llvm::isa<TF::EinsumOp>(op)) {
        if (IsInCompsiteFunction(op)) return;
        // Leftover Einsum ops are always non-quantized.
        auto op_name = op->getName().stripDialect();
        func_count_map[op_name].num_float++;
      }
    });

    MLIR pipeline in the future.
  }];
  let constructor = "TF::CreateGpuOpFusionPass()";
}

def BatchMatMulToEinsumPass : Pass<"tf-batch-matmul-to-tf-einsum", "mlir::func::FuncOp"> {
  let summary = "Replace TF BatchMatMul op by TF Einsum op.";
  let constructor = "TF::CreateBatchMatMulToEinsumPass()";
}