Update on "Avoid calling allclose in the backward if there are tensor subclasses"

lezcano · lezcano · commit cbb629d72f18 · 2022-12-28T15:47:42.000Z
`allclose` it's data-dependent (returns a bool) so it does not play well with functorch. We are skipping that check in the context of subclasses to avoid hard errors. Partially fixes #90499 cc zou3519 Chillee samdow soumith [ghstack-poisoned]
diff --git a/torch/csrc/autograd/FunctionsManual.cpp b/torch/csrc/autograd/FunctionsManual.cpp
@@ -3550,8 +3550,8 @@ Tensor linalg_eig_backward(
   const auto diag_VhgV = VhgV.diagonal(0, -2, -1);
 
   if (V.is_complex() && !at::isTensorSubclassLike(diag_VhgV)) {
-    // Check invariance of the loss function wrt the transformation 
-    // V -> V * e^{i\phi} for an arbitrary phi \in \RR^n
+    // Check invariance of the loss function wrt the transformation
+    // V -> V * e^{i\phi} for an arbitrary phi in RR^n
     const auto imdiag_VhgV = at::imag(diag_VhgV);
     TORCH_CHECK(
         at::allclose(
