Distinguish ops from kernels in tf.profile

[tafsiri]

This makes it easier to process the list of kernels for a custom build since ops implemented with customGrad don't have true backend kernels.

To see the logs from the Cloud Build CI, please join either our discussion or announcement mailing list.

---

This change is 

[tafsiri]

cc @nsthorat in case there is any corner case or other use of the scope names as we had them before.

[annxingyuan]

Just to make sure I understand - in the example described in the PR description, the model calls the FusedConv2D kernel and calls fusedConv2D through custom grad?

Reviewable status: complete! 1 of 1 approvals obtained (waiting on @annxingyuan and @tafsiri)

---

tfjs/tools/custom_bundle/cli.ts, line 107 at r1 (raw file):

  // kernels so we need to filter them out.
  function isNotCustomOp(kernelName: string) {
    return !kernelName.endsWith('__op');

could we maybe pull out __op into a variable and reference it elsewhere?

---

tfjs-core/src/ops/operation.ts, line 41 at r1 (raw file):

  }

  // an an __op suffix to distinguish ops from kernels in tf.profile

Add an __op

[tafsiri]

@annxingyuan fusedConv2d the op is called which starts a scope with the op name (this name won't be replaced by a kernel name when you enter the first engine.runKernelFunc). Later on this op will call another engine.runKernelFunc with FusedConv2d which will also get added to tf.profile (it will start a new scope).

For reference above are the kernels captured for blazeface by tf.profile (sorry I should have put that in the original discussion).