Clean up

[dan-zheng]

A big list of clean up opportunities, vaguely ordered by importance.
Feel free to add/edit.

• Revamp tensor initializers, especially ones that take both "data" and "shape" arguments.
  • "Data" arguments should be clearly distinguished from "shape" arguments.
  • Tensor.ones(2, 3) is fine. Tensor.fill(1, 2, 3) is not.
  • Use multiple parameter lists for disambiguation: Tensor.fill(2, 3) { is => idx += 1; idx }
• Done. Use either Manifest or Typ consistently. Typ is not native for our version of LMS.
• Improve testing.
  • Currently, all tests roughly look like this:

      test("foobar") {
        val test1 = new DslDriverC[String, Unit] with TensorExp {
          def snippet(a: Rep[String]): Rep[Unit] = {
            // `a` is a dummy that's not actually used.
            Tensor.assertEqual(...)
          }
        }
        test1.eval("a")
      }

    This is not ideal because the Rep[String] argument is never used. Can we make this nicer (does DslDriverC[Unit, Unit] work)?
  • Done. Create a helper test utility class (e.g. LanternTestSuite) that extends org.scalatest.FunSuite. This class can define helpers like def runTest(snippet: DslDriverC[String, Unit]).
• Done. Rename files.
  • The filenames ad_lms.scala and ad_lms_vector.scala aren't very clear to me. Perhaps ScalarDifferentiation.scala and TensorDifferentiation.scala are more clear.
  • Pascal case all filenames.

For the most part, these items are low priority. We can pursue them lazily.

[GSAir]

I am not sure how to add an item to the list:

• be consistent with usage Manifest or Typ. Typ is not native for the version of LMS we are using.