Use maxint to bound integers.

We don't actually support arbitrary precision integers.

Signed-off-by: Edward Z. Yang <ezyangmeta.com>

ghstack-source-id: cefcf018e6c1fcebc48b1678142711ce09596465
Pull Request resolved: #96121

torch/fx/experimental/symbolic_shapes.py

@@ -1334,7 +1334,7 @@ def create_unbacked_symfloat(self):
     def create_unbacked_symint(self):
         symbol = sympy.Symbol(f"i{next(self.unbacked_symint_counter)}", integer=True)
         self.var_to_stack[symbol] = ''.join(traceback.format_list(traceback.extract_stack()[:-1]))
-        self.var_to_range[symbol] = ValueRanges.unknown()
+        self.var_to_range[symbol] = ValueRanges(-sys.maxsize - 1, sys.maxsize)
         return SymInt(SymNode(symbol, self, int, None))
 
     # This is guaranteed to return a symbol or its negation is a sympy.Symbol,
@@ -1361,7 +1361,10 @@ def create_symbol(self, val: int, source: Source, dyn=False) -> "sympy.Expr":
 
             # We also infer that it must be not 0/1
             lower = 2 if self.specialize_zero_one else 0
-            self.var_to_range[sympy_expr] = ValueRanges(lower, sympy.oo)
+            # NB: sys.maxsize is NOT allowed for sizes, because we use MAX_INT
+            # as a sentinel sometimes.  Your sizevar isn't going to be
+            # anywhere near the max 64-bit integer anyway.
+            self.var_to_range[sympy_expr] = ValueRanges(lower, sys.maxsize - 1)
 
         if not dyn and self.duck_shape:
             # This implements duck-shaping: input sizes that match are assigned