Re: Speed up _cpp_clean_line and _cpp_skip_block_comment

["Zack Weinberg" <zack at codesourcery dot com>]

Paolo Carlini <pcarlini@unitus.it> writes:

> ... another curiosity is that, comparing the code generated
> by 3.3.1 and current 3.4 for the original testcase:
>
>   void f(int* k, int j)
>   {
>     int i;
>
>     for (i = 0; i < j; ++i)
>       *k = *k + 2 * ((*k) - 1);
>   }

Er what?  That doesn't look anything like my test case.

> at -O2, the inner loops are:
>
> 3.3.1
> -----
>   10:   48                      dec    %eax
>   11:   8d 54 52 fe             lea    0xfffffffe(%edx,%edx,2),%edx
>   15:   75 f9                   jne    10 <f+0x10>
>
> 3.4
> ---
>   10:   8d 4c 52 fe             lea    0xfffffffe(%edx,%edx,2),%ecx
>   14:   48                      dec    %eax
>   15:   89 ca                   mov    %ecx,%edx
>   17:   75 f7                   jne    10 <f+0x10>
>
> isn't 3.3.1 better?

I would say so.  3.4 with -fnew-ra produces

.L5:
        decl    %edx
        leal    -2(%eax,%eax,2), %eax
        jne     .L5

zw