fix(mge): correct trace outputs when grad does copy

GitOrigin-RevId: 65c8956a7d

imperative/python/megengine/jit/tracing.py

@@ -163,9 +163,9 @@ class trace:
         self._graph = None
         self._need_reset_nodes = None
         self._lazy_eval_graph = None
-        self._lazy_eval_tensors = set()
+        self._lazy_eval_tensors = {}
         self._lazy_eval_links = None
-        self._active_tensors = set()
+        self._active_tensors = {}
         self._tensor_remaps = None
         self._inputs_to_restore = None
         self._arg_bindings = None
@@ -249,8 +249,8 @@ class trace:
             y._compiled_info = CompiledTensorProxy(h)
             y.mixin_handle = h
             outputs += [y]
+            self._active_tensors[h] = TensorWeakRef(y)
         self._output_handles.update(ohandles)
-        self._active_tensors.update([TensorWeakRef(o) for o in outputs])
         return outputs
 
     def _apply_const(self, value, dtype, device):
@@ -303,9 +303,11 @@ class trace:
             x.mixin_handle = h
             x.recording = True
             x._trace_mixin_info = info
+            self._active_tensors[h] = TensorWeakRef(x)
+            if self._symbolic:
+                self._lazy_eval_tensors[h] = TensorWeakRef(x)
 
         self._seq.append((op, tuple(ihandles), tuple(ohandles)))
-        self._active_tensors.update([TensorWeakRef(o) for o in outputs])
 
     def _record_const(self, outputs):
         if skip_tracing:
@@ -327,6 +329,8 @@ class trace:
         x.mixin_handle = h
         x.recording = True
         x._trace_mixin_info = info
+        if self._symbolic:
+            self._lazy_eval_tensors[h] = TensorWeakRef(x)
         self._seq.append(("Const", tuple(), tuple(ohandles)))
 
     def _set_active(self, active: bool):
@@ -346,12 +350,12 @@ class trace:
             self._lazy_eval_links = ()
 
     def _take_escaped_tensors(self):
-        escaped_tensors = tuple(filter(lambda x: x() is not None, self._active_tensors))
+        escaped_tensors = tuple(filter(lambda x: x() is not None, self._active_tensors.values()))
         self._active_tensors.clear()
         return escaped_tensors
 
     def _lazy_eval(self, lazy_eval_graph, lazy_eval_tensors, lazy_eval_links):
-        lazy_eval_tensors = list(filter(lambda x: x() is not None, lazy_eval_tensors))
+        lazy_eval_tensors = list(filter(lambda x: x() is not None, lazy_eval_tensors.values()))
         readers = [G.OutputNode(x()._varnode).outputs[0] for x in lazy_eval_tensors]
         self._apply_graph_options(lazy_eval_graph)
         # FIXME
@@ -401,7 +405,7 @@ class trace:
                     # eval lazy eval tensors
                     self._lazy_eval(
                         self._lazy_eval_graph,
-                        tuple(self._lazy_eval_tensors),
+                        self._lazy_eval_tensors,
                         self._lazy_eval_links,
                     )
                     self._lazy_eval_graph = None
@@ -433,9 +437,10 @@ class trace:
             if not self._untraced and self._pc != len(self._seq):
                 raise TraceMismatchError("premature end")
             if not self._symbolic or not self._untraced:
-                for x in self._active_tensors:
+                for x in self._active_tensors.values():
                     if x() is not None:
                         x()._dev_tensor()
+                        x()._reset_varnode()
                         x().mixin_handle = -1
                         x().recording = False
 
@@ -459,7 +464,7 @@ class trace:
         if self._untraced:
             # conditionally reading a compiled tensor in excluded region
             # is permitted, so we have to assume every tensor might be read
-            for x in self._active_tensors:
+            for x in self._active_tensors.values():
                 info = self._tinfo[x().mixin_handle]
                 info.exported = True
                 info.data_read = True
@@ -626,8 +631,20 @@ class trace:
             if self._capture_as_const:
                 self._process_inputs(*args, **kwargs)
             outputs = self.__wrapped__(*args, **kwargs)
+            transform = False
+            if outputs is not None:
+                if not isinstance(outputs, collections.abc.Sequence):
+                    transform = True
+                    outputs = (outputs,)
+                for o in outputs:
+                    if o._copied:
+                        self._active_tensors[o.mixin_handle] = TensorWeakRef(o)
+                        if self._untraced and self._symbolic:
+                            self._lazy_eval_tensors[o.mixin_handle] = TensorWeakRef(o)
             if self._capture_as_const:
                 self._process_outputs(outputs)
+            if transform:
+                outputs = outputs[0]
             return outputs
 
     def dump(
@@ -1031,7 +1048,6 @@ def apply_symbolic_mode(op: OpDef, *args: RawTensor):
     if require_links:
         active_trace._lazy_eval_links = (G.VarNode(outputs[0]._varnode),)
 
-    active_trace._lazy_eval_tensors.update([TensorWeakRef(o) for o in outputs])
     return outputs
 
 
@@ -1042,7 +1058,6 @@ def apply_const_symbolic_mode(value, dtype, device):
     ret = RawTensor(graph.make_const(value, dtype=dtype, device=device))
     if np.array(value).ndim == 0:
         setscalar(ret)
-    active_trace._lazy_eval_tensors.add(TensorWeakRef(ret))
     return (ret,)
 
 

imperative/python/src/tensor.cpp

@@ -284,6 +284,11 @@ REGISTE_TENSORWRAPPER_FUNC(bool, recording)
 #undef REGISTE_TENSORWRAPPER_FUNC
 
 
+PyObject* TensorWrapper::copied() {
+    return py::cast(m_tensor->m_trace_info.copied).release().ptr();
+}
+
+
 #define REGISTE_TENSORWRAPPER_PYOBJECT_FUNC(member)                                 \
         PyObject* TensorWrapper::member() {                                         \
             return m_tensor->m_trace_info.member;                                   \
@@ -740,6 +745,7 @@ void init_tensor(py::module m) {
         .def<&TensorWrapper::_drop>("_drop")
         .def<&TensorWrapper::reset_varnode>("_reset_varnode")
         .def_getset<&TensorWrapper::varnode>("_varnode")
+        .def_getset<&TensorWrapper::copied>("_copied")
         .def_getset<&TensorWrapper::mixin_handle, &TensorWrapper::set_mixin_handle>("mixin_handle")
         .def_getset<&TensorWrapper::recording, &TensorWrapper::set_recording>("recording")
         .def_getset<&TensorWrapper::handle, &TensorWrapper::set_handle>("_handle")

imperative/python/src/tensor.h

@@ -161,6 +161,7 @@ struct TensorWrapper {
 
     PyObject* mixin_handle();
     PyObject* recording();
+    PyObject* copied();
 
     void set_mixin_handle(PyObject*);
     void set_recording(PyObject*);

imperative/python/src/trace_info.h

@@ -17,6 +17,7 @@ namespace mgb::imperative::python {
 struct TraceInfo {
     int64_t mixin_handle = -1;
     bool recording = false;
+    bool copied = false;
 
     PyObject* compiled_info = nullptr;
     PyObject* trace_mixin_info = nullptr;
@@ -32,6 +33,7 @@ struct TraceInfo {
         trace_mixin_info = that.trace_mixin_info;
         Py_XINCREF(trace_mixin_info);
 
+        copied = true;
         return *this;
     }