Re: [RFC PATCH] fix a reshape checking logic inside make_stripe_request()

[Yu Kuai <yukuai1@xxxxxxxxxxxxxxx>]

Hi,

在 2025/05/06 10:34, Coly Li 写道:
> For the above code, I don’t see how only unlikely(previous) help on branch prediction and prefetch,
> IMHO the following form may help to achieve expected unlikely() result,
>
>            if (unlikely(previous && !stripe_ahead_of_reshape(mddev, conf, sh))) {
>
> Thanks.

What you mean you don't see *only unlikely*, for example:

int test0(int a, int b)
{
        if (a && b)
                return 1;
        return 0;
}

int test1(int a, int b)
{
        if (unlikely(a) && b)
                return 1;

        return 0;
}

You can see unlikely will generate setne/movzbl/test assemble code:

test0:
   0x0000000000401130 <+10>:    cmpl   $0x0,-0x4(%rbp)
   0x0000000000401134 <+14>:    je     0x401143 <test0+29>
   0x0000000000401136 <+16>:    cmpl   $0x0,-0x8(%rbp)
   0x000000000040113a <+20>:    je     0x401143 <test0+29>

test1:
   0x0000000000401154 <+10>:    cmpl   $0x0,-0x4(%rbp)
   0x0000000000401158 <+14>:    setne  %al
   0x000000000040115b <+17>:    movzbl %al,%eax
   0x000000000040115e <+20>:    test   %rax,%rax
   0x0000000000401161 <+23>:    je     0x401170 <test1+38>
   0x0000000000401163 <+25>:    cmpl   $0x0,-0x8(%rbp)

BTW, what you suggested is the same as:

if (unlikely(previous) && unlikely(!stripe_ahead_of_reshape(mddev, conf, 
sh)))

Thanks,
Kuai