Use a single loop and let the compiler split off the first row

Co-Authored-By: Claude Fable 5.1 <noreply@anthropic.com>

Author: eendebakpt

Objects/longobject.c

@@ -3999,30 +3999,7 @@ x_mul(PyLongObject *a, PyLongObject *b)
         }
     }
     else {      /* a is not the same as b -- gradeschool int mult */
-        /* No digit of z needs to start out zeroed: the i == 0 pass stores
-         * z[0:size_b+1] outright, pass i only reads z[i:i+size_b], which
-         * earlier passes have already written, and each pass stores its final
-         * carry into z[i+size_b], which no earlier pass has touched.
-         */
-        assert(size_a >= 1);
-        {
-            twodigits carry = 0;
-            twodigits f = a->long_value.ob_digit[0];
-            digit *pz = z->long_value.ob_digit;
-            digit *pb = b->long_value.ob_digit;
-            digit *pbend = b->long_value.ob_digit + size_b;
-
-            while (pb < pbend) {
-                carry += *pb++ * f;
-                *pz++ = (digit)(carry & PyLong_MASK);
-                carry >>= PyLong_SHIFT;
-                assert(carry <= PyLong_MASK);
-            }
-            /* The final carry fits in a digit, so no masking is needed. */
-            *pz = (digit)carry;
-            assert((carry >> PyLong_SHIFT) == 0);
-        }
-        for (i = 1; i < size_a; ++i) {
+        for (i = 0; i < size_a; ++i) {
             twodigits carry = 0;
             twodigits f = a->long_value.ob_digit[i];
             digit *pz = z->long_value.ob_digit + i;
@@ -4034,15 +4011,17 @@ x_mul(PyLongObject *a, PyLongObject *b)
                     return NULL;
                 });
 
+            /* Row 0 has nothing to add to yet (z is not zeroed); every
+               later row only reads digits that the previous row wrote.
+               Compilers split the i == 0 row off into its own loop. */
             while (pb < pbend) {
-                carry += *pz + *pb++ * f;
+                carry += (i ? *pz : 0) + *pb++ * f;
                 *pz++ = (digit)(carry & PyLong_MASK);
                 carry >>= PyLong_SHIFT;
                 assert(carry <= PyLong_MASK);
             }
-            /* The final carry fits in a digit, so no masking is needed. */
-            *pz = (digit)carry;
             assert((carry >> PyLong_SHIFT) == 0);
+            *pz = (digit)carry;
         }
     }
     return long_normalize(z);