[inductor] support unbacked symint divisors in vars_and_sizes (pytorch#130595)

Scenario:
```
>>> nodes
IterationRangesEntry(
    x2,
    divisor=192*u0 + 192576,
    length=s1,
    (xindex//(192*u0 + 192576)),
    {x0: 192, x1: u0 + 1003, x2: s1, x3: 192*s1*u0 + 192576*s1, x4: 192*u0 + 192576})
IterationRangesEntry(
    x1,
    divisor=192,
    length=u0 + 1003,
    ModularIndexing(xindex, 192, u0 + 1003),
    {x0: 192, x1: u0 + 1003, x2: s1, x3: 192*s1*u0 + 192576*s1, x4: 192*u0 + 192576})
IterationRangesEntry(
    x0,
    divisor=1,
    length=192,
    ModularIndexing(xindex, 1, 192),
    {x0: 192, x1: u0 + 1003, x2: s1, x3: 192*s1*u0 + 192576*s1, x4: 192*u0 + 192576})
```

Think about whether using fallback is safe here. I think it's safe because the divisor of one IterationRangesEntry should be the product of the lengths of the preceding IterationRangesEntry? Unless, one of the lengths divides by an unbacked symint?

Pull Request resolved: pytorch#130595
Approved by: https://github.com/aakhundov, https://github.com/ezyang

Author: ColinPeppler

test/inductor/test_unbacked_symints.py

@@ -252,6 +252,23 @@ def fn(x, w, repeats, is_bmm):
         expected = fn(*example_inputs)
         torch.testing.assert_close(actual, expected)
 
+    @torch._dynamo.config.patch(capture_scalar_outputs=True)
+    def test_unbacked_range_tree_divisor(self, device):
+        def fn(x, num):
+            u0 = num.item()
+            torch._check_is_size(u0)
+            zeros = torch.zeros(u0, device=device, dtype=torch.int)
+            return (torch.ops.aten.index(x, [None, zeros]),)
+
+        example_inputs = (
+            torch.randn(16, 16, device=device),
+            torch.tensor(3, device=device),
+        )
+
+        actual = torch.compile(fn, fullgraph=True)(*example_inputs)
+        expected = fn(*example_inputs)
+        torch.testing.assert_close(actual, expected)
+
 
 instantiate_device_type_tests(
     TestUnbackedSymints, globals(), only_for=(GPU_TYPE, "cpu")

torch/_inductor/codegen/simd.py

@@ -200,7 +200,11 @@ def vars_and_sizes(self, index: sympy.Expr):
         """Figure out vars from this tree used in index"""
         nodes = [V.kernel.range_tree_nodes.get(s) for s in index.free_symbols]
         nodes = [n for n in nodes if n and n.prefix == self.prefix]
-        nodes.sort(key=lambda x: V.graph.sizevars.size_hint(x.divisor))
+        nodes.sort(
+            key=lambda x: V.graph.sizevars.size_hint(
+                x.divisor, fallback=config.unbacked_symint_fallback
+            )
+        )
         divisor = sympy.Integer(1)
         index_vars = []
         sizes = []