同时除法和求余数?

发布于 2024-09-26 17:40:39 字数 1706 浏览 4 评论 0 原文

如果你对这篇内容有疑问,欢迎到本站社区发帖提问 参与讨论,获取更多帮助,或者扫码二维码加入 Web 技术交流群。

扫码二维码加入Web技术交流群

发布评论

需要 登录 才能够评论, 你可以免费 注册 一个本站的账号。

评论(9

奈何桥上唱咆哮 2024-10-03 17:40:39

C 有 divldiv 。它们是否为商和余数生成单独的指令将取决于您特定的标准库实现以及编译器和优化设置。从 C99 开始,您还可以使用 lldiv 来表示更大的数字。

C has div and ldiv. Whether these generate separate instructions for the quotient and remainder will depend on your particular standard library implementation and compiler and optimization settings. Starting with C99, you also have lldiv for larger numbers.

轮廓§ 2024-10-03 17:40:39

Python 确实如此。

>>> divmod(9, 4)
(2, 1)

这很奇怪,因为 Python 是一种高级语言。

Ruby 也是如此:

11.divmod(3) #=> [3, 2]

*** 编辑***

应该注意的是,这些运算符的目的可能不是尽可能高效地完成工作,更可能的是这些函数的存在是出于正确性/可移植性的原因。

对于那些感兴趣的人,我相信这是代码整数 divmod 的 Python 实现:

static enum divmod_result
i_divmod(register long x, register long y,
     long *p_xdivy, long *p_xmody)
{
long xdivy, xmody;

if (y == 0) {
    PyErr_SetString(PyExc_ZeroDivisionError,
                    "integer division or modulo by zero");
    return DIVMOD_ERROR;
}
/* (-sys.maxint-1)/-1 is the only overflow case. */
if (y == -1 && UNARY_NEG_WOULD_OVERFLOW(x))
    return DIVMOD_OVERFLOW;
xdivy = x / y;
/* xdiv*y can overflow on platforms where x/y gives floor(x/y)
 * for x and y with differing signs. (This is unusual
 * behaviour, and C99 prohibits it, but it's allowed by C89;
 * for an example of overflow, take x = LONG_MIN, y = 5 or x =
 * LONG_MAX, y = -5.)  However, x - xdivy*y is always
 * representable as a long, since it lies strictly between
 * -abs(y) and abs(y).  We add casts to avoid intermediate
 * overflow.
 */
xmody = (long)(x - (unsigned long)xdivy * y);
/* If the signs of x and y differ, and the remainder is non-0,
 * C89 doesn't define whether xdivy is now the floor or the
 * ceiling of the infinitely precise quotient.  We want the floor,
 * and we have it iff the remainder's sign matches y's.
 */
if (xmody && ((y ^ xmody) < 0) /* i.e. and signs differ */) {
    xmody += y;
    --xdivy;
    assert(xmody && ((y ^ xmody) >= 0));
}
*p_xdivy = xdivy;
*p_xmody = xmody;
return DIVMOD_OK;
}

Python does.

>>> divmod(9, 4)
(2, 1)

Which is odd, because Python is such a high level language.

So does Ruby:

11.divmod(3) #=> [3, 2]

*** EDIT ***

It should be noted that the purpose of these operators is probably not to do the work as efficiently as possible, it is more likely the functions exist for correctness/portability reasons.

For those interested, I believe this is the code of the Python implementation for integer divmod:

static enum divmod_result
i_divmod(register long x, register long y,
     long *p_xdivy, long *p_xmody)
{
long xdivy, xmody;

if (y == 0) {
    PyErr_SetString(PyExc_ZeroDivisionError,
                    "integer division or modulo by zero");
    return DIVMOD_ERROR;
}
/* (-sys.maxint-1)/-1 is the only overflow case. */
if (y == -1 && UNARY_NEG_WOULD_OVERFLOW(x))
    return DIVMOD_OVERFLOW;
xdivy = x / y;
/* xdiv*y can overflow on platforms where x/y gives floor(x/y)
 * for x and y with differing signs. (This is unusual
 * behaviour, and C99 prohibits it, but it's allowed by C89;
 * for an example of overflow, take x = LONG_MIN, y = 5 or x =
 * LONG_MAX, y = -5.)  However, x - xdivy*y is always
 * representable as a long, since it lies strictly between
 * -abs(y) and abs(y).  We add casts to avoid intermediate
 * overflow.
 */
xmody = (long)(x - (unsigned long)xdivy * y);
/* If the signs of x and y differ, and the remainder is non-0,
 * C89 doesn't define whether xdivy is now the floor or the
 * ceiling of the infinitely precise quotient.  We want the floor,
 * and we have it iff the remainder's sign matches y's.
 */
if (xmody && ((y ^ xmody) < 0) /* i.e. and signs differ */) {
    xmody += y;
    --xdivy;
    assert(xmody && ((y ^ xmody) >= 0));
}
*p_xdivy = xdivy;
*p_xmody = xmody;
return DIVMOD_OK;
}
究竟谁懂我的在乎 2024-10-03 17:40:39

In C#/.NET you've got Math.DivRem:
http://msdn.microsoft.com/en-us/library/system.math.divrem.aspx

But according to this thread this isn't that much an optimization.

帝王念 2024-10-03 17:40:39

在 Java 中(自 1.5 起),类 BigDecimal 具有操作 divideAndRemainder,返回一个包含 2 个元素的数组,其中包含除法的结果和余数。

BigDecimal bDecimal = ...
BigDecimal[] result = bDecimal.divideAndRemainder(new BigDecimal(60));

Java 17 Javadoc: https://docs.oracle.com/en/java/javase/17/docs/api/java.base/java/math/BigDecimal.html#除法余数(java.math.BigDecimal)

In Java (since 1.5) the class BigDecimal has the operation divideAndRemainder returning an array of 2 elements with the result and de remainder of the division.

BigDecimal bDecimal = ...
BigDecimal[] result = bDecimal.divideAndRemainder(new BigDecimal(60));

Java 17 Javadoc: https://docs.oracle.com/en/java/javase/17/docs/api/java.base/java/math/BigDecimal.html#divideAndRemainder(java.math.BigDecimal)

暮色兮凉城 2024-10-03 17:40:39

.NET 框架具有 Math.DivRem

int mod, div = Math.DivRem(11, 3, out mod);
// mod = 2, div = 3

虽然,DivRem 只是这样的包装:(

int div = x / y;
int mod = x % y;

我不知道抖动是否可以/确实将此类事情优化为单个指令。)

The .NET framework has Math.DivRem:

int mod, div = Math.DivRem(11, 3, out mod);
// mod = 2, div = 3

Although, DivRem is just a wrapper around something like this:

int div = x / y;
int mod = x % y;

(I have no idea whether or not the jitter can/does optimise that sort of thing into a single instruction.)

当爱已成负担 2024-10-03 17:40:39

正如 Stringer Bell 提到的,DivRem 未优化 最高 .NET 3.5。

在 .NET 4.0 它使用 NGen

我用 Math.DivRem 得到的结果 (debug;release = ~11000ms)

11863
11820
11881
11859
11854

我用 MyDivRem 得到的结果 (debug;release = ~11000ms)

29177
29214
29472
29277
29196

针对 x86 的项目。


Math.DivRem 使用示例

int mod1;
int div1 = Math.DivRem(4, 2, out mod1);

方法签名

DivRem(Int32, Int32, Int32&) : Int32
DivRem(Int64, Int64, Int64&) : Int64

.NET 4.0 代码

[TargetedPatchingOptOut("Performance critical to inline across NGen image boundaries")]
public static int DivRem(int a, int b, out int result)
{
    result = a % b;
    return (a / b);
}

.NET 4.0 IL

.custom instance void System.Runtime.TargetedPatchingOptOutAttribute::.ctor(string) = { string('Performance critical to inline across NGen image boundaries') }
.maxstack 8
L_0000: ldarg.2 
L_0001: ldarg.0 
L_0002: ldarg.1 
L_0003: rem 
L_0004: stind.i4 
L_0005: ldarg.0 
L_0006: ldarg.1 
L_0007: div 
L_0008: ret 

MSDN 参考

As Stringer Bell mentioned there is DivRem which is not optimized up to .NET 3.5.

On .NET 4.0 it uses NGen.

The results I got with Math.DivRem (debug; release = ~11000ms)

11863
11820
11881
11859
11854

Results I got with MyDivRem (debug; release = ~11000ms)

29177
29214
29472
29277
29196

Project targeted for x86.


Math.DivRem Usage example

int mod1;
int div1 = Math.DivRem(4, 2, out mod1);

Method signatures

DivRem(Int32, Int32, Int32&) : Int32
DivRem(Int64, Int64, Int64&) : Int64

.NET 4.0 Code

[TargetedPatchingOptOut("Performance critical to inline across NGen image boundaries")]
public static int DivRem(int a, int b, out int result)
{
    result = a % b;
    return (a / b);
}

.NET 4.0 IL

.custom instance void System.Runtime.TargetedPatchingOptOutAttribute::.ctor(string) = { string('Performance critical to inline across NGen image boundaries') }
.maxstack 8
L_0000: ldarg.2 
L_0001: ldarg.0 
L_0002: ldarg.1 
L_0003: rem 
L_0004: stind.i4 
L_0005: ldarg.0 
L_0006: ldarg.1 
L_0007: div 
L_0008: ret 

MSDN Reference

软的没边 2024-10-03 17:40:39

Haskell具有 后者直接对应于机器指令(根据 Integral运营商引号与Div Divmod 可能不会。

Haskell has both divMod and quotRem that latter of which corresponds directly to the machine instruction (according to Integral operators quot vs. div) while divMod may not.

糖果控 2024-10-03 17:40:39
    int result,rest;
    _asm
    {
        xor edx, edx // pone edx a cero; edx = 0
        mov eax, result// eax = 2AF0
        mov ecx, radix // ecx = 4
        div ecx
        mov val, eax
        mov rest, edx
    }
    int result,rest;
    _asm
    {
        xor edx, edx // pone edx a cero; edx = 0
        mov eax, result// eax = 2AF0
        mov ecx, radix // ecx = 4
        div ecx
        mov val, eax
        mov rest, edx
    }
~没有更多了~
我们使用 Cookies 和其他技术来定制您的体验包括您的登录状态等。通过阅读我们的 隐私政策 了解更多相关信息。 单击 接受 或继续使用网站,即表示您同意使用 Cookies 和您的相关数据。
原文