gh-156443: Keep PyLong loop carries as twodigits in shifts

Lib/test/test_long.py

@@ -204,7 +204,33 @@ def test_division(self):
         self.check_division(710031681576388032, 26769404391308)
         self.check_division(1933622614268221, 30212853348836)
 
-
+    @support.requires_IEEE_754
+    def test_intradigit_shift(self):
+        # Unit tests for v_lshift and v_rshift in longobject.c.
+        # These two functions are not used by Python << and >>,
+        # exercise them via float conversion and true division.
+
+        # Full limb values.
+        two_limbs = (1 << (2 * SHIFT)) - 1
+        three_limbs = (1 << (3 * SHIFT)) - 1
+        four_limbs = (1 << (4 * SHIFT)) - 1
+        # Multi-limb (not compact) but bit_length() still <= 55.
+        bits50 = (1 << 50) - 1
+
+        # _PyLong_Frexp (float(n)):
+        # v_lshift if n.bit_length() <= 55, else v_rshift.
+        self.check_float_conversion(bits50)
+        self.check_float_conversion(two_limbs)
+        self.check_float_conversion(four_limbs)
+        self.check_float_conversion(10**40)
+
+        # long_true_divide (a / b):
+        # - (a.bit_length() - b.bit_length()) <= 55  => v_lshift,
+        # - (a.bit_length() - b.bit_length()) > 55   => v_rshift.
+        self.check_truediv(three_limbs, two_limbs)
+        self.check_truediv(four_limbs, two_limbs)
+        self.check_truediv(four_limbs, three_limbs)
+        self.check_truediv(10**40, 10**20)
 
     def test_karatsuba(self):
         digits = list(range(1, 5)) + list(range(KARATSUBA_CUTOFF,

Misc/NEWS.d/next/Core_and_Builtins/2026-09-07-09-52-21.gh-issue-156443.tT40sm.rst

@@ -0,0 +1 @@
+Optimize shifts of large :class:`int` values.

Objects/longobject.c

@@ -1905,15 +1905,15 @@ static digit
 v_lshift(digit *z, digit *a, Py_ssize_t m, int d)
 {
     Py_ssize_t i;
-    digit carry = 0;
+    twodigits carry = 0;
 
     assert(0 <= d && d < PyLong_SHIFT);
     for (i=0; i < m; i++) {
         twodigits acc = (twodigits)a[i] << d | carry;
         z[i] = (digit)acc & PyLong_MASK;
-        carry = (digit)(acc >> PyLong_SHIFT);
+        carry = acc >> PyLong_SHIFT;
     }
-    return carry;
+    return (digit)carry;
 }
 
 /* Shift digit vector a[0:m] d bits right, with 0 <= d < PyLong_SHIFT.  Put
@@ -1923,16 +1923,16 @@ static digit
 v_rshift(digit *z, digit *a, Py_ssize_t m, int d)
 {
     Py_ssize_t i;
-    digit carry = 0;
-    digit mask = ((digit)1 << d) - 1U;
+    twodigits carry = 0;
+    twodigits mask = ((twodigits)1 << d) - 1U;
 
     assert(0 <= d && d < PyLong_SHIFT);
     for (i=m; i-- > 0;) {
-        twodigits acc = (twodigits)carry << PyLong_SHIFT | a[i];
-        carry = (digit)acc & mask;
+        twodigits acc = carry << PyLong_SHIFT | a[i];
+        carry = acc & mask;
         z[i] = (digit)(acc >> d);
     }
-    return carry;
+    return (digit)carry;
 }
 
 /* Divide long pin, w/ size digits, by non-zero digit n, storing quotient