[compiled_autograd] workaround windows compilation issue

zou3519 · zou3519 · commit 53b154d828c3 · 2025-03-04T09:26:26.000-08:00
torch.compile doesn't work on windows so we can ifdef-away the problem. I do not know what the root cause actually is. Most notably, the pytorch windows build is fine, but some third-party projects that use pytorch headers on windows (e.g. torchaudio) have issues. Test Plan: - wait for CI ghstack-source-id: be04459 Pull Request resolved: #148454
diff --git a/torch/csrc/dynamo/compiled_autograd.h b/torch/csrc/dynamo/compiled_autograd.h
@@ -1057,6 +1057,13 @@ struct IValuePacker {
   // That's what the TypePtr is for: it contains the information to do the
   // parsing. See torch::jit::toIValue for more information.
   static at::TypePtr packed_type() {
+#ifdef _WIN32
+    // NB: the if-constexpr usage triggers compilation errors on Windows
+    // with certain compiler settings
+    // (see https://github.com/pytorch/pytorch/pull/144707 for examples).
+    // It's not clear what the problem is, so we're going to ignore it for now.
+    TORCH_INTERNAL_ASSERT(false, "torch.compile not supported on Windows");
+#else
     if constexpr (::std::is_same_v<T, at::Tensor>) {
       return at::TensorType::get();
     } else if constexpr (::std::is_same_v<T, int64_t>) {
@@ -1095,6 +1102,7 @@ struct IValuePacker {
       TORCH_INTERNAL_ASSERT(false, "IValuePacker not implemented for type");
       return at::NoneType::get();
     }
+#endif
   }
 };
 
