• davidl (48/48) Sep 10 2007 I read this :

davidl <davidl 126.com> writes:

I read this :
http://dlang.group.javaeye.com/group/topic/2806
C++ code:
#include <stdio.h>
#include <exception>

volatile int turn=0;
volatile int count;

void fooFunc(void)
{
      ++ count;
};

typedef void (*foo_t)(void);

void test(int c1, int c2, foo_t foo)
{
      for (int j=0;j<c1;j++)
      {
          ++ turn;
          count = 0;
          for (int i=0;i<c2;i++)
          {
              try
              {
                  foo();
              }
              catch (std::exception &e)
              {

              }
          }
      }

      printf("\nturn %d, count %d", turn, count);

}

int main()
{
      test(8*1024, 1048576, fooFunc);
      return 0;
}

C++ code runs way faster than D, why??

Finally, the culprit comes out that D code doesn't optimize the args of i,
j to registers. also the function pointer func is not inlined, MSVC seems
totally make the args 'instanciated', so the func looks like a template
gened code.

And why we need to access i, j from memory?

Because the exception system is not context safe.
So we lose a big chance of optimzing those vars to registers(Even if
there's not exception get thrown, compiler can not assume the context
consistent).






-- 
使用 Opera 革命性的电子邮件客户程序: http://www.opera.com/mail/