将GCC内联程序集CMOV转换为Visual Studio汇编器 [英] Convert GCC Inline Assembly CMOV to Visual Studio Assembler

问题描述

在文章线性与二进制搜索，它是使用CMOV指令的二进制搜索的快速实现.我想在VC ++中实现此功能，因为我正在处理的应用程序依赖于Binary Search的性能.

In the article Linear vs. Binary Search, there is a fast implementation of binary search which uses the CMOV instruction. I would like to implement this in VC++ as the application I'm working on relies on the performance of Binary Search.

该实现具有一些GCC内联汇编器，其声明如下:

The implementation has some GCC inline assembler which is declared as follows:

static int binary_cmov (const int *arr, int n, int key) {
    int min = 0, max = n;
    while (min < max) {
            int middle = (min + max) >> 1;

            asm ("cmpl %3, %2\n\tcmovg %4, %0\n\tcmovle %5, %1"
                 : "+r" (min),
                   "+r" (max)
                 : "r" (key), "g" (arr [middle]),
                   "g" (middle + 1), "g" (middle));

            // Equivalent to 
            // if (key > arr [middle])
            //    min = middle + 1;
            // else
            //    max = middle;
    }
    return min;
}

我想将此GCC汇编程序转换为与Microsoft Visual Studio兼容的汇编程序，但作为GCC菜鸟不知道从哪里开始.

I'd like to convert this GCC Assembler to Microsoft Visual Studio compatible assembler, but as a GCC noob wouldn't know where to start.

任何人都可以帮忙，或者至少解释一下GCC内联汇编程序吗? TIA！

Can anyone help, or at least explain the GCC Inline Assembler? TIA!

推荐答案

重构代码以更好地向编译器表达意图:

refactoring the code in order to better express intent to the compiler:

int binary_cmov (const int *arr, int n, int key) 
{
    int min = 0, max = n;
    while (min < max) 
    {
      int middle = (min + max) >> 1;
      min = key > arr[middle] ? middle + 1 : min;
      max = key > arr[middle] ? max : middle;
    }
    return min;
}

使用gcc5.3 -O3，产量:

With gcc5.3 -O3, yields:

binary_cmov(int const*, int, int):
        xorl    %eax, %eax
        testl   %esi, %esi
        jle     .L4
.L3:
        leal    (%rax,%rsi), %ecx
        sarl    %ecx
        movslq  %ecx, %r8
        leal    1(%rcx), %r9d
        movl    (%rdi,%r8,4), %r8d
        cmpl    %edx, %r8d
        cmovl   %r9d, %eax
        cmovge  %ecx, %esi
        cmpl    %eax, %esi
        jg      .L3
        rep ret
.L4:
        rep ret

故事的寓意-不要嵌入汇编器.您要做的就是使代码不可移植.

moral of the story - don't embed assembler. All you do is make the code non-portable.

走得更远...

为什么不更明确地表达意图?

why not express intent even more explicitly?

#include <utility>

template<class...Ts>
  auto sum(Ts...ts)
{
  std::common_type_t<Ts...> result = 0;
  using expand = int[];
  void(expand{ 0, ((result += ts), 0)... });
  return result;
}

template<class...Ts>
  auto average(Ts...ts)
{
  return sum(ts...) / sizeof...(Ts);
}

int binary_cmov (const int *arr, int n, int key) 
{
    int min = 0, max = n;
    while (min < max) 
    {
      int middle = average(min, max);
      auto greater = key > arr[middle];
      min = greater ? middle + 1 : min;
      max = greater ? max : middle;
    }
    return min;
}

编译器输出:

binary_cmov(int const*, int, int):
        xorl    %eax, %eax
        testl   %esi, %esi
        jle     .L4
.L3:
        leal    (%rax,%rsi), %ecx
        movslq  %ecx, %rcx
        shrq    %rcx
        movslq  %ecx, %r8
        leal    1(%rcx), %r9d
        movl    (%rdi,%r8,4), %r8d
        cmpl    %edx, %r8d
        cmovl   %r9d, %eax
        cmovge  %ecx, %esi
        cmpl    %eax, %esi
        jg      .L3
        rep ret
.L4:
        rep ret

这篇关于将GCC内联程序集CMOV转换为Visual Studio汇编器的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！