keep up-to-date with recent pytorch

.github/workflows/test_pytorch.yml

@@ -17,7 +17,7 @@ jobs:
     strategy:
       fail-fast: false
       matrix:
-        python-version: ["3.8", "3.9", "3.10", "3.11"] # Python 3.11 bug waits for fix https://github.com/thuml/depyf/actions/runs/7004325219/job/19051829613 .
+        python-version: ["3.9", "3.10", "3.11"] # Python 3.11 bug waits for fix https://github.com/thuml/depyf/actions/runs/7004325219/job/19051829613 .
 
     steps:
     - uses: actions/checkout@v3

depyf/explain/utils.py

@@ -90,16 +90,27 @@
 
         cpp_guard = False
 
+        # starting from https://github.com/pytorch/pytorch/pull/138896 ,
+        # pytorch uses `guard_manager` instead of `check_fn` to store the
+        # guards
+        attr_name = "guard_manager" if hasattr(cache, "guard_manager") else "check_fn"
+
+        guard_manager = getattr(cache, attr_name)
+
         try:
-            from torch._dynamo.guards import GuardManager
-            cpp_guard = isinstance(cache.check_fn, GuardManager)
+            klass = getattr(torch._dynamo.guards, "GuardManagerWrapper", None) or \
+                getattr(torch._C._dynamo.guards, "GuardManager", None)
+            assert klass is not None
+            cpp_guard = isinstance(guard_manager, klass)
         except Exception:
             pass
 
         if not cpp_guard:
-            guard = cache.check_fn.code_parts
-            freevar_names = cache.check_fn.__code__.co_freevars
-            freevar_values = [x.cell_contents for x in cache.check_fn.__closure__]
+            # for old version of pytorch,
+            # `guard_manager` is a plain python function
+            guard = guard_manager.code_parts
+            freevar_names = guard_manager.__code__.co_freevars
+            freevar_values = [x.cell_contents for x in guard_manager.__closure__]
         else:
             # keep the logic synced with
             # https://github.com/pytorch/pytorch/blob/7b6b10417d8616ebd7a42b06528c5c2b2fded55a/torch/_dynamo/guards.py#L262
@@ -118,14 +129,14 @@
                 for child in root.get_child_managers():
                     visit(child, ans)
             guard = []
-            root = cache.check_fn.root
+            root = guard_manager.root
             visit(root, guard)
-            if cache.check_fn.closure_vars is None:
+            if guard_manager.closure_vars is None:
                 freevar_names = tuple()
                 freevar_values = []
             else:
-                freevar_names = tuple(cache.check_fn.closure_vars.keys())
-                freevar_values = list(cache.check_fn.closure_vars.values())
+                freevar_names = tuple(guard_manager.closure_vars.keys())
+                freevar_values = list(guard_manager.closure_vars.values())
 
         self.guard = guard
         self.freevars = {name: value for name, value in zip(freevar_names, freevar_values)}

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_1 Backward graph 0.py

@@ -2,18 +2,19 @@
 
 
 
-def forward(self, primals_1: "f32[10]", div: "f32[10]", tangents_1: "f32[10]"):
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+def forward(self, primals_1: "f32[10]", tangents_1: "f32[10]"):
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
     neg: "f32[10]" = torch.ops.aten.neg.default(tangents_1)
     abs_1: "f32[10]" = torch.ops.aten.abs.default(primals_1)
     add: "f32[10]" = torch.ops.aten.add.Tensor(abs_1, 1);  abs_1 = None
-    div_2: "f32[10]" = torch.ops.aten.div.Tensor(div, add);  div = None
+    div_1: "f32[10]" = torch.ops.aten.div.Tensor(primals_1, add)
+    div_2: "f32[10]" = torch.ops.aten.div.Tensor(div_1, add);  div_1 = None
     mul: "f32[10]" = torch.ops.aten.mul.Tensor(neg, div_2);  neg = div_2 = None
     div_3: "f32[10]" = torch.ops.aten.div.Tensor(tangents_1, add);  tangents_1 = add = None
     sgn: "f32[10]" = torch.ops.aten.sgn.default(primals_1);  primals_1 = None
     mul_1: "f32[10]" = torch.ops.aten.mul.Tensor(mul, sgn);  mul = sgn = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
     add_1: "f32[10]" = torch.ops.aten.add.Tensor(div_3, mul_1);  div_3 = mul_1 = None
-    return [add_1, None]
+    return (add_1, None)
 

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_1 Captured Graph 0.py

@@ -6,12 +6,12 @@ def forward(self, L_a_: "f32[10]", L_b_: "f32[10]"):
     l_a_ = L_a_
     l_b_ = L_b_
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
     abs_1: "f32[10]" = torch.abs(l_a_)
     add: "f32[10]" = abs_1 + 1;  abs_1 = None
     x: "f32[10]" = l_a_ / add;  l_a_ = add = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
     sum_1: "f32[]" = l_b_.sum();  l_b_ = None
     lt: "b8[]" = sum_1 < 0;  sum_1 = None
     return (x, lt)

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_1 Forward graph 0.py

@@ -3,13 +3,13 @@
 
 
 def forward(self, primals_1: "f32[10]", primals_2: "f32[10]"):
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
     abs_1: "f32[10]" = torch.ops.aten.abs.default(primals_1)
     add: "f32[10]" = torch.ops.aten.add.Tensor(abs_1, 1);  abs_1 = None
     div: "f32[10]" = torch.ops.aten.div.Tensor(primals_1, add);  add = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
     sum_1: "f32[]" = torch.ops.aten.sum.default(primals_2);  primals_2 = None
     lt: "b8[]" = torch.ops.aten.lt.Scalar(sum_1, 0);  sum_1 = None
-    return [div, lt, primals_1, div]
+    return (div, lt, primals_1)
 

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_1 Joint graph 0.py

@@ -6,16 +6,16 @@ def forward(self, primals, tangents):
     primals_1: "f32[10]"; primals_2: "f32[10]"; tangents_1: "f32[10]"; 
 
     primals_1, primals_2, tangents_1, = fx_pytree.tree_flatten_spec([primals, tangents], self._in_spec)
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
     abs_1: "f32[10]" = torch.ops.aten.abs.default(primals_1)
     add: "f32[10]" = torch.ops.aten.add.Tensor(abs_1, 1);  abs_1 = None
     div: "f32[10]" = torch.ops.aten.div.Tensor(primals_1, add)
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
     sum_1: "f32[]" = torch.ops.aten.sum.default(primals_2);  primals_2 = None
     lt: "b8[]" = torch.ops.aten.lt.Scalar(sum_1, 0);  sum_1 = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
     neg: "f32[10]" = torch.ops.aten.neg.default(tangents_1)
     div_1: "f32[10]" = torch.ops.aten.div.Tensor(primals_1, add)
     div_2: "f32[10]" = torch.ops.aten.div.Tensor(div_1, add);  div_1 = None
@@ -24,7 +24,7 @@ def forward(self, primals, tangents):
     sgn: "f32[10]" = torch.ops.aten.sgn.default(primals_1);  primals_1 = None
     mul_1: "f32[10]" = torch.ops.aten.mul.Tensor(mul, sgn);  mul = sgn = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
     add_1: "f32[10]" = torch.ops.aten.add.Tensor(div_3, mul_1);  div_3 = mul_1 = None
     return pytree.tree_unflatten([div, lt, add_1, None], self._out_spec)
 

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_1 pre insert_deferred_runtime_asserts __compiled_fn_1 0.py

@@ -0,0 +1,18 @@
+from __future__ import annotations
+
+
+
+def forward(self, L_a_: "f32[10]", L_b_: "f32[10]"):
+    l_a_ = L_a_
+    l_b_ = L_b_
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+    abs_1: "f32[10]" = torch.abs(l_a_)
+    add: "f32[10]" = abs_1 + 1;  abs_1 = None
+    x: "f32[10]" = l_a_ / add;  l_a_ = add = None
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
+    sum_1: "f32[]" = l_b_.sum();  l_b_ = None
+    lt: "b8[]" = sum_1 < 0;  sum_1 = None
+    return (x, lt)
+

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_1 tensorify_python_scalars 0.py

@@ -0,0 +1,30 @@
+from __future__ import annotations
+
+
+
+def forward(self, primals, tangents):
+    primals_1: "f32[10]"; primals_2: "f32[10]"; tangents_1: "f32[10]"; 
+
+    primals_1, primals_2, tangents_1, = fx_pytree.tree_flatten_spec([primals, tangents], self._in_spec)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+    abs_1: "f32[10]" = torch.ops.aten.abs.default(primals_1)
+    add: "f32[10]" = torch.ops.aten.add.Tensor(abs_1, 1);  abs_1 = None
+    div: "f32[10]" = torch.ops.aten.div.Tensor(primals_1, add)
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:5 in toy_function, code: if b.sum() < 0:
+    sum_1: "f32[]" = torch.ops.aten.sum.default(primals_2);  primals_2 = None
+    lt: "b8[]" = torch.ops.aten.lt.Scalar(sum_1, 0);  sum_1 = None
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+    neg: "f32[10]" = torch.ops.aten.neg.default(tangents_1)
+    div_1: "f32[10]" = torch.ops.aten.div.Tensor(primals_1, add)
+    div_2: "f32[10]" = torch.ops.aten.div.Tensor(div_1, add);  div_1 = None
+    mul: "f32[10]" = torch.ops.aten.mul.Tensor(neg, div_2);  neg = div_2 = None
+    div_3: "f32[10]" = torch.ops.aten.div.Tensor(tangents_1, add);  tangents_1 = add = None
+    sgn: "f32[10]" = torch.ops.aten.sgn.default(primals_1);  primals_1 = None
+    mul_1: "f32[10]" = torch.ops.aten.mul.Tensor(mul, sgn);  mul = sgn = None
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:4 in toy_function, code: x = a / (torch.abs(a) + 1)
+    add_1: "f32[10]" = torch.ops.aten.add.Tensor(div_3, mul_1);  div_3 = mul_1 = None
+    return pytree.tree_unflatten([div, lt, add_1, None], self._out_spec)
+

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_11 Backward graph 0.py

@@ -2,17 +2,17 @@
 
 
 
-def forward(self, primals_1: "f32[8]", primals_2: "f32[8]", tangents_1: "f32[8]"):
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
-    mul_2: "f32[8]" = torch.ops.aten.mul.Tensor(tangents_1, primals_2);  primals_2 = None
+def forward(self, primals_1: "Sym(s0)", primals_2: "f32[s0]", primals_3: "f32[s0]", tangents_1: "f32[s0]"):
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
+    mul_4: "f32[s0]" = torch.ops.aten.mul.Tensor(tangents_1, primals_3);  primals_3 = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
-    mul: "f32[8]" = torch.ops.aten.mul.Tensor(primals_1, -1);  primals_1 = None
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    mul: "f32[s0]" = torch.ops.aten.mul.Tensor(primals_2, -1);  primals_2 = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
-    mul_3: "f32[8]" = torch.ops.aten.mul.Tensor(tangents_1, mul);  tangents_1 = mul = None
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
+    mul_5: "f32[s0]" = torch.ops.aten.mul.Tensor(tangents_1, mul);  tangents_1 = mul = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
-    mul_4: "f32[8]" = torch.ops.aten.mul.Tensor(mul_2, -1);  mul_2 = None
-    return [mul_4, mul_3]
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    mul_6: "f32[s0]" = torch.ops.aten.mul.Tensor(mul_4, -1);  mul_4 = None
+    return (None, mul_6, mul_5)
 

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_11 Captured Graph 0.py

@@ -2,14 +2,14 @@
 
 
 
-def forward(self, L_b_: "f32[8]", L_x_: "f32[8]"):
+def forward(self, s0: "Sym(s0)", L_b_: "f32[s0]", L_x_: "f32[s0]"):
     l_b_ = L_b_
     l_x_ = L_x_
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
-    b: "f32[8]" = l_b_ * -1;  l_b_ = None
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    b: "f32[s0]" = l_b_ * -1;  l_b_ = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
-    mul_1: "f32[8]" = l_x_ * b;  l_x_ = b = None
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
+    mul_1: "f32[s0]" = l_x_ * b;  l_x_ = b = None
     return (mul_1,)
 

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_11 Forward graph 0.py

@@ -2,11 +2,11 @@
 
 
 
-def forward(self, primals_1: "f32[8]", primals_2: "f32[8]"):
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
-    mul: "f32[8]" = torch.ops.aten.mul.Tensor(primals_1, -1)
+def forward(self, primals_1: "Sym(s0)", primals_2: "f32[s0]", primals_3: "f32[s0]"):
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    mul: "f32[s0]" = torch.ops.aten.mul.Tensor(primals_2, -1)
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
-    mul_1: "f32[8]" = torch.ops.aten.mul.Tensor(primals_2, mul);  mul = None
-    return [mul_1, primals_1, primals_2]
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
+    mul_2: "f32[s0]" = torch.ops.aten.mul.Tensor(primals_3, mul);  mul = None
+    return (mul_2, primals_2, primals_3, primals_1)
 

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_11 Joint graph 0.py

@@ -3,18 +3,18 @@
 
 
 def forward(self, primals, tangents):
-    primals_1: "f32[8]"; primals_2: "f32[8]"; tangents_1: "f32[8]"; 
+    primals_1: "Sym(s0)"; primals_2: "f32[s0]"; primals_3: "f32[s0]"; tangents_1: "f32[s0]"; 
 
-    primals_1, primals_2, tangents_1, = fx_pytree.tree_flatten_spec([primals, tangents], self._in_spec)
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
-    mul: "f32[8]" = torch.ops.aten.mul.Tensor(primals_1, -1);  primals_1 = None
+    primals_1, primals_2, primals_3, tangents_1, = fx_pytree.tree_flatten_spec([primals, tangents], self._in_spec)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    mul: "f32[s0]" = torch.ops.aten.mul.Tensor(primals_2, -1);  primals_2 = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
-    mul_1: "f32[8]" = torch.ops.aten.mul.Tensor(primals_2, mul)
-    mul_2: "f32[8]" = torch.ops.aten.mul.Tensor(tangents_1, primals_2);  primals_2 = None
-    mul_3: "f32[8]" = torch.ops.aten.mul.Tensor(tangents_1, mul);  tangents_1 = mul = None
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
+    mul_2: "f32[s0]" = torch.ops.aten.mul.Tensor(primals_3, mul)
+    mul_4: "f32[s0]" = torch.ops.aten.mul.Tensor(tangents_1, primals_3);  primals_3 = None
+    mul_5: "f32[s0]" = torch.ops.aten.mul.Tensor(tangents_1, mul);  tangents_1 = mul = None
 
-    # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
-    mul_4: "f32[8]" = torch.ops.aten.mul.Tensor(mul_2, -1);  mul_2 = None
-    return pytree.tree_unflatten([mul_1, mul_4, mul_3], self._out_spec)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    mul_6: "f32[s0]" = torch.ops.aten.mul.Tensor(mul_4, -1);  mul_4 = None
+    return pytree.tree_unflatten([mul_2, None, mul_6, mul_5], self._out_spec)
 

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_11 pre insert_deferred_runtime_asserts __compiled_fn_11 0.py

@@ -0,0 +1,15 @@
+from __future__ import annotations
+
+
+
+def forward(self, s0: "Sym(s0)", L_b_: "f32[s0]", s1: "Sym(s0)", L_x_: "f32[s0]"):
+    l_b_ = L_b_
+    l_x_ = L_x_
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    b: "f32[s0]" = l_b_ * -1;  l_b_ = None
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
+    mul_1: "f32[s0]" = l_x_ * b;  l_x_ = b = None
+    return (mul_1,)
+

tests/depyf_output/debug_function_aot_eager_with_dynamic_shape_with_grad/__compiled_fn_11 tensorify_python_scalars 0.py

@@ -0,0 +1,20 @@
+from __future__ import annotations
+
+
+
+def forward(self, primals, tangents):
+    primals_1: "Sym(s0)"; primals_2: "f32[s0]"; primals_3: "f32[s0]"; tangents_1: "f32[s0]"; 
+
+    primals_1, primals_2, primals_3, tangents_1, = fx_pytree.tree_flatten_spec([primals, tangents], self._in_spec)
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    mul: "f32[s0]" = torch.ops.aten.mul.Tensor(primals_2, -1);  primals_2 = None
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:7 in torch_dynamo_resume_in_toy_function_at_5, code: return x * b
+    mul_2: "f32[s0]" = torch.ops.aten.mul.Tensor(primals_3, mul)
+    mul_4: "f32[s0]" = torch.ops.aten.mul.Tensor(tangents_1, primals_3);  primals_3 = None
+    mul_5: "f32[s0]" = torch.ops.aten.mul.Tensor(tangents_1, mul);  tangents_1 = mul = None
+
+     # File: /Users/youkaichao/data/DeepLearning/depyf/tests/test_pytorch/test_pytorch.py:6 in torch_dynamo_resume_in_toy_function_at_5, code: b = b * -1
+    mul_6: "f32[s0]" = torch.ops.aten.mul.Tensor(mul_4, -1);  mul_4 = None
+    return pytree.tree_unflatten([mul_2, None, mul_6, mul_5], self._out_spec)
+