mul_u64_u64_div_u64: increase precision by conditionally swapping a and b

As indicated in the added comment, the algorithm works better if b is big.
As multiplication is commutative, a and b can be swapped.  Do this if a
is bigger than b.

Link: https://lkml.kernel.org/r/20240303092408.662449-2-u.kleine-koenig@pengutronix.de
Signed-off-by: Uwe Kleine-König <u.kleine-koenig@pengutronix.de>
Tested-by: Biju Das <biju.das.jz@bp.renesas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

diff --git a/lib/math/div64.c b/lib/math/div64.c
index 55a81782e2717f7bca4f8d53834458db960952a1..191761b1b623cc52c977c96a8b9a14cee1f03707 100644 (file)
--- a/lib/math/div64.c
+++ b/lib/math/div64.c
@@ -22,6 +22,7 @@
 #include <linux/export.h>
 #include <linux/math.h>
 #include <linux/math64.h>
+#include <linux/minmax.h>
 #include <linux/log2.h>
 
 /* Not needed on 64bit architectures */
@@ -190,6 +191,20 @@ u64 mul_u64_u64_div_u64(u64 a, u64 b, u64 c)
 
        /* can a * b overflow ? */
        if (ilog2(a) + ilog2(b) > 62) {
+               /*
+                * Note that the algorithm after the if block below might lose
+                * some precision and the result is more exact for b > a. So
+                * exchange a and b if a is bigger than b.
+                *
+                * For example with a = 43980465100800, b = 100000000, c = 1000000000
+                * the below calculation doesn't modify b at all because div == 0
+                * and then shift becomes 45 + 26 - 62 = 9 and so the result
+                * becomes 4398035251080. However with a and b swapped the exact
+                * result is calculated (i.e. 4398046510080).
+                */
+               if (a > b)
+                       swap(a, b);
+
                /*
                 * (b * a) / c is equal to
                 *