加入我们,或者,欢迎回来。
您需要 登录 才可以下载或查看,没有帐号?注册会员
x
你们知道我在搞RGSS的物理引擎
最近在研究JBox2D的源码
看到里面有这一段package org.jbox2d.common; /** * Contains methods from MathUtils that rely on JVM features. These are separated out from * MathUtils so that they can be overridden when compiling for GWT. */ class PlatformMathUtils { private static final float SHIFT23 = 1 << 23; private static final float INV_SHIFT23 = 1.0f / SHIFT23; public static final float fastPow(float a, float b) { float x = Float.floatToRawIntBits(a); x *= INV_SHIFT23; x -= 127; float y = x - (x >= 0 ? (int) x : (int) x - 1); b *= x + (y - y * y) * 0.346607f; y = b - (b >= 0 ? (int) b : (int) b - 1); y = (y - y * y) * 0.33971f; return Float.intBitsToFloat((int) ((b + 127 - y) * SHIFT23)); } }
package org.jbox2d.common;
/**
* Contains methods from MathUtils that rely on JVM features. These are separated out from
* MathUtils so that they can be overridden when compiling for GWT.
*/
class PlatformMathUtils {
private static final float SHIFT23 = 1 << 23;
private static final float INV_SHIFT23 = 1.0f / SHIFT23;
public static final float fastPow(float a, float b) {
float x = Float.floatToRawIntBits(a);
x *= INV_SHIFT23;
x -= 127;
float y = x - (x >= 0 ? (int) x : (int) x - 1);
b *= x + (y - y * y) * 0.346607f;
y = b - (b >= 0 ? (int) b : (int) b - 1);
y = (y - y * y) * 0.33971f;
return Float.intBitsToFloat((int) ((b + 127 - y) * SHIFT23));
}
}
这是个用于快速计算乘方的算法,精度不高,根号2算出来大概是1.415
为了测试这个算法到底有多快,我写了一段代码进行测试public class Test { private static final float SHIFT23 = 1 << 23; private static final float INV_SHIFT23 = 1.0f / SHIFT23; public static final float fastPow(float a, float b) { float x = Float.floatToRawIntBits(a); x *= INV_SHIFT23; x -= 127; float y = x - (x >= 0 ? (int) x : (int) x - 1); b *= x + (y - y * y) * 0.346607f; y = b - (b >= 0 ? (int) b : (int) b - 1); y = (y - y * y) * 0.33971f; return Float.intBitsToFloat((int) ((b + 127 - y) * SHIFT23)); } public static long times(int timesCount, Runnable runnable) { long time = System.currentTimeMillis(); for (int i = 0; i < timesCount; i++) { runnable.run(); } return System.currentTimeMillis() - time; } public static void main(String[] args) { System.out.println(Test.times(100000000, () -> Test.fastPow(1.5f, 3.0f))); System.out.println(Test.times(100000000, () -> StrictMath.pow(1.5f, 3.0f))); } }
public class Test {
private static final float SHIFT23 = 1 << 23;
private static final float INV_SHIFT23 = 1.0f / SHIFT23;
public static final float fastPow(float a, float b) {
float x = Float.floatToRawIntBits(a);
x *= INV_SHIFT23;
x -= 127;
float y = x - (x >= 0 ? (int) x : (int) x - 1);
b *= x + (y - y * y) * 0.346607f;
y = b - (b >= 0 ? (int) b : (int) b - 1);
y = (y - y * y) * 0.33971f;
return Float.intBitsToFloat((int) ((b + 127 - y) * SHIFT23));
}
public static long times(int timesCount, Runnable runnable) {
long time = System.currentTimeMillis();
for (int i = 0; i < timesCount; i++) {
runnable.run();
}
return System.currentTimeMillis() - time;
}
public static void main(String[] args) {
System.out.println(Test.times(100000000, () -> Test.fastPow(1.5f, 3.0f)));
System.out.println(Test.times(100000000, () -> StrictMath.pow(1.5f, 3.0f)));
}
}
原理是,执行一亿次新算法,输出消耗的时间,再执行一亿次StrictMath自带的算法,输出消耗时间
结果吓我一大跳:新算法可以在7毫秒内执行一亿次乘方运算,比标准库自带算法快了2400多倍?
这么好的算法一定要移植到Ruby中用啊!SHIFT23 = (1 << 23).to_f INV_SHIFT23 = 1.0 / SHIFT23 def fast_pow(a, b) x = a.int_bits x *= INV_SHIFT23 x -= 0x7f y = x - (x >= 0 ? x.to_i : x.to_i - 1) b *= x + (y - y * y) * 0.346607 y = b - (b >= 0 ? b.to_i : b.to_i - 1) y = (y - y * y) * 0.33971 Float.get_from_int_bits(((b + 0x7f - y) * SHIFT23).to_i) end class Float def int_bits [self].pack('g').unpack('B*').first.to_i(2) end class << self def get_from_int_bits(i) [sprintf("%032b", i)].pack('B*').unpack('g').first end end end def times(n, &block) time = Time.now n.times(&block) Time.now - time end puts(times(100000000) { fast_pow(1.5, 3.0) }) puts(times(100000000) { 1.5 ** 3.0 })
SHIFT23 = (1 << 23).to_f
INV_SHIFT23 = 1.0 / SHIFT23
def fast_pow(a, b)
x = a.int_bits
x *= INV_SHIFT23
x -= 0x7f
y = x - (x >= 0 ? x.to_i : x.to_i - 1)
b *= x + (y - y * y) * 0.346607
y = b - (b >= 0 ? b.to_i : b.to_i - 1)
y = (y - y * y) * 0.33971
Float.get_from_int_bits(((b + 0x7f - y) * SHIFT23).to_i)
end
class Float
def int_bits
[self].pack('g').unpack('B*').first.to_i(2)
end
class << self
def get_from_int_bits(i)
[sprintf("%032b", i)].pack('B*').unpack('g').first
end
end
end
def times(n, &block)
time = Time.now
n.times(&block)
Time.now - time
end
puts(times(100000000) { fast_pow(1.5, 3.0) })
puts(times(100000000) { 1.5 ** 3.0 })
代码内还附带了根据IEEE 754浮点“单一格式”位布局返回指定浮点值的表示形式的方法,和返回对应于给定位表示形式的浮点数值的方法
运行结果再次让我大吃一惊: |