Hi Brian,
On 10/17/2013 12:26 PM, Brian Burkhalter wrote:
This post concerns this issue:
https://bugs.openjdk.java.net/browse/JDK-4891331
I performed some tests using JMH [1] on Mac OS X [2] and Windows 7 [3]. The
tests were equivalent to calling multiply() with argument == this for bit
lengths from 32 through 448 without and with this patch applied:
--- a/src/share/classes/java/math/BigInteger.java Thu Oct 17 11:34:01
2013 -0400
+++ b/src/share/classes/java/math/BigInteger.java Thu Oct 17 11:45:42
2013 -0700
@@ -1374,6 +1374,10 @@
if (val.signum == 0 || signum == 0)
return ZERO;
+ if (val == this) {
+ return square();
+ }
+
int xlen = mag.length;
int ylen = val.mag.length;
A table of the ratios of throughput with the patch applied to throughput
without the patch applied is here:
Can you test the performance of a patch with:
if (val == this and mag.length > 8)
Do you have any measurements of the multiply performance of values that
are not ==? (Making sure there is no general regression.)
Thanks,
-Joe
