+1, Hard to argue with +10x. :)
Roger On 12/18/2018 01:03 PM, Brian Burkhalter wrote:
https://bugs.openjdk.java.net/browse/JDK-8183912 <https://bugs.openjdk.java.net/browse/JDK-8183912> For the patch included below, the JMH benchmark MovePointLeftZero attached to the above issue, when run with options "-f 1 -wi 5 -w 5 -i 10 -r 10" (one fork, five warmup iterations of five seconds each, ten measurement iterations of ten seconds each), shows an order of magnitude (10X) throughput improvement for a zero left shift. The effect on the positive left shift case appears to be at the level of measurement noise. Thanks, Brian --- a/src/java.base/share/classes/java/math/BigDecimal.java +++ b/src/java.base/share/classes/java/math/BigDecimal.java @@ -2871,6 +2871,8 @@ * @throws ArithmeticException if scale overflows. */ public BigDecimal movePointLeft(int n) { + if (n == 0) return this; + // Cannot use movePointRight(-n) in case of n==Integer.MIN_VALUE int newScale = checkScale((long)scale + n); BigDecimal num = new BigDecimal(intVal, intCompact, newScale, 0); @@ -2893,6 +2895,8 @@ * @throws ArithmeticException if scale overflows. */ public BigDecimal movePointRight(int n) { + if (n == 0) return this; + // Cannot use movePointLeft(-n) in case of n==Integer.MIN_VALUE int newScale = checkScale((long)scale - n); BigDecimal num = new BigDecimal(intVal, intCompact, newScale, 0);
