// CHECK: return %[[RESULT]] : tensor<2xi32>
  %starts = "tf.Const"() {value = dense<[1]> : tensor<1xi64>} : () -> (tensor<1xi64>)
  %sizes = "tf.Const"() {value = dense<[2]> : tensor<1xi64>} : () -> (tensor<1xi64>)
  %0 = "tf.Slice"(%arg0, %starts, %sizes) : (tensor<4xi32>, tensor<1xi64>, tensor<1xi64>) -> tensor<2xi32>
  func.return %0 : tensor<2xi32>
}

// -----

// CHECK-LABEL: slice_i32_consts

tensorflow::TensorShape ToTensorShape(llvm::ArrayRef<T> sizes) {
  return tensorflow::TensorShape(
      llvm::SmallVector<int64_t, num_dims>(sizes.begin(), sizes.end()));
}

template <typename T, int num_dims>
tensorflow::TensorShape ToTensorShape(
    llvm::iterator_range<DenseElementsAttr::ElementIterator<T>> sizes) {
  return tensorflow::TensorShape(
      llvm::SmallVector<int64_t, num_dims>(sizes.begin(), sizes.end()));
}

		n = cl + 1
	}
	if n < 1 {
		return 1
	}
	return int(n) // doesn't truncate; max is 512K
}

// Seven bufPools manage different frame sizes. This helps to avoid scenarios where long-running
// streaming requests using small frame sizes occupy large buffers initially allocated for prior
// requests needing big buffers. The size ranges are as follows:

<a href="https://en.wikipedia.org/wiki/Two's_complement">two's complement arithmetic</a>.
</p>

<p>
There is also a set of predeclared integer types with implementation-specific sizes:
</p>

<pre class="grammar">
uint     either 32 or 64 bits
int      same size as uint
uintptr  an unsigned integer large enough to store the uninterpreted bits of a pointer value
</pre>

<p>