| // Copyright (c) 2019, the R8 project authors. Please see the AUTHORS file | |
| // for details. All rights reserved. Use of this source code is governed by a | |
| // BSD-style license that can be found in the LICENSE file. | |
| package backport; | |
| import java.math.BigInteger; | |
| public class StrictMathBackportJava9Main { | |
| public static void main(String[] args) { | |
| testMultiplyExactLongInt(); | |
| testMultiplyFull(); | |
| testFloorDivLongInt(); | |
| testFloorModLongInt(); | |
| } | |
| public static void testMultiplyExactLongInt() { | |
| assertEquals(8L, StrictMath.multiplyExact(2L, 4)); | |
| assertEquals(Long.MAX_VALUE, StrictMath.multiplyExact(Long.MAX_VALUE, 1)); | |
| assertEquals(Long.MIN_VALUE, StrictMath.multiplyExact(Long.MIN_VALUE / 2L, 2)); | |
| try { | |
| throw new AssertionError(StrictMath.multiplyExact(Long.MAX_VALUE, 2)); | |
| } catch (ArithmeticException expected) { | |
| } | |
| try { | |
| throw new AssertionError(StrictMath.multiplyExact(Long.MIN_VALUE, 2)); | |
| } catch (ArithmeticException expected) { | |
| } | |
| } | |
| public static void testMultiplyFull() { | |
| assertEquals(8L, StrictMath.multiplyFull(2, 4)); | |
| assertEquals(4611686014132420609L, | |
| StrictMath.multiplyFull(Integer.MAX_VALUE, Integer.MAX_VALUE)); | |
| assertEquals(-4611686016279904256L, | |
| StrictMath.multiplyFull(Integer.MAX_VALUE, Integer.MIN_VALUE)); | |
| assertEquals(4611686018427387904L, | |
| StrictMath.multiplyFull(Integer.MIN_VALUE, Integer.MIN_VALUE)); | |
| } | |
| public static void testMultiplyHigh() { | |
| long[] interestingValues = { | |
| Long.MIN_VALUE, Long.MAX_VALUE, | |
| Integer.MIN_VALUE, Integer.MAX_VALUE, | |
| Short.MIN_VALUE, Short.MAX_VALUE, | |
| Byte.MIN_VALUE, Byte.MAX_VALUE, | |
| 0L, | |
| -1L, 1L, | |
| -42L, 42L | |
| }; | |
| for (long x : interestingValues) { | |
| for (long y : interestingValues) { | |
| long expected = BigInteger.valueOf(x) | |
| .multiply(BigInteger.valueOf(y)) | |
| .shiftRight(64) | |
| .longValue(); | |
| assertEquals(expected, StrictMath.multiplyHigh(x, y)); | |
| } | |
| } | |
| } | |
| public static void testFloorDivLongInt() { | |
| assertEquals(1L, StrictMath.floorDiv(4L, 4)); | |
| assertEquals(1L, StrictMath.floorDiv(-4L, -4)); | |
| assertEquals(-1L, StrictMath.floorDiv(-4L, 4)); | |
| assertEquals(-1L, StrictMath.floorDiv(4L, -4)); | |
| assertEquals(1L, StrictMath.floorDiv(4L, 3)); | |
| assertEquals(1L, StrictMath.floorDiv(-4L, -3)); | |
| assertEquals(-2L, StrictMath.floorDiv(-4L, 3)); | |
| assertEquals(-2L, StrictMath.floorDiv(4L, -3)); | |
| // Spec edge case: result is actually MAX_VALUE+1 which becomes MIN_VALUE. | |
| assertEquals(Long.MIN_VALUE, StrictMath.floorDiv(Long.MIN_VALUE, -1)); | |
| } | |
| public static void testFloorModLongInt() { | |
| assertEquals(0, StrictMath.floorMod(4L, 4)); | |
| assertEquals(0, StrictMath.floorMod(-4L, -4)); | |
| assertEquals(0, StrictMath.floorMod(-4L, 4)); | |
| assertEquals(0, StrictMath.floorMod(4L, -4)); | |
| assertEquals(1, StrictMath.floorMod(4L, 3)); | |
| assertEquals(-1, StrictMath.floorMod(-4L, -3)); | |
| assertEquals(2, StrictMath.floorMod(-4L, 3)); | |
| assertEquals(-2, StrictMath.floorMod(4L, -3)); | |
| } | |
| private static void assertEquals(int expected, int actual) { | |
| if (expected != actual) { | |
| throw new AssertionError("Expected <" + expected + "> but was <" + actual + '>'); | |
| } | |
| } | |
| private static void assertEquals(long expected, long actual) { | |
| if (expected != actual) { | |
| throw new AssertionError("Expected <" + expected + "> but was <" + actual + '>'); | |
| } | |
| } | |
| } |