S_uv_mul_overflow in inline.h: Add casts to make multiplications done in UV

If C compiler doesn't know __builtin_mul_overflow, S_uv_mul_overflow
will be implemented with fallback "long multiplication" algorithm,
but it had a bug that elemental multiplications were done in unsigned
long precision instead of UV precision.  It will lead wrong result
when unsigned long is narrower than UV (for example -Duse64bitint
on 32-bit platform).

Author: t-a-k

inline.h

@@ -3546,6 +3546,8 @@ S_uv_mul_overflow (UV auv, UV buv, UV *const result)
     const UV botmask = ~topmask;
 
 #    if UVSIZE > LONGSIZE && UVSIZE <= 2 * LONGSIZE
+    /* If UV is double-word integer, declare these variables as single-word
+       integers to help compiler to avoid double-word multiplication.  */
     unsigned long alow, ahigh, blow, bhigh;
 #    else
     UV alow, ahigh, blow, bhigh;
@@ -3567,7 +3569,7 @@ S_uv_mul_overflow (UV auv, UV buv, UV *const result)
         /* One operand is large, 1 small */
         /* Either ahigh or bhigh is zero here, so the addition below
            can't overflow.  */
-        product_middle = ahigh * blow + alow * bhigh;
+        product_middle = (UV)ahigh * blow + (UV)alow * bhigh;
         if (product_middle & topmask)
             return true;
         /* OK, product_middle won't lose bits when we shift it.  */
@@ -3576,7 +3578,7 @@ S_uv_mul_overflow (UV auv, UV buv, UV *const result)
     /* else: eg 32 bit is at most 0xFFFF * 0xFFFF == 0xFFFE0001
        so the unsigned multiply cannot overflow.  */
 
-    UV product_low = alow * blow;
+    UV product_low = (UV)alow * blow;
     return S_uv_add_overflow(product_middle, product_low, result);
 }
 #  endif