Re: x86: A fast way to check capabilities of the current cpu
From: Miguel Ojeda
Date: Thu Dec 16 2010 - 05:17:35 EST
On Thu, Dec 16, 2010 at 7:25 AM, Miles Bader <miles@xxxxxxx> wrote:
> "H. Peter Anvin" <hpa@xxxxxxxxx> writes:
>>> In this case it this_cpu_*_test_bit() return an int, but they act as a
>>> bool and are used in if()s; where is the catch?
>>
>> If they aren't, and are stored in a variable for whatever reason, then
>> the || form will generate additional instructions to booleanize the
>> value for no good reason.
>
> It doesn't actually have to "booleanize" the value if it's used in a
> boolean context though (and, AFAICT, usually won't).
>
> My vague impression is that when used in a boolean context, gcc will
> often generate the same or "equivalent" code for both variants -- but
> sometimes a||b seems to generate better code; e.g.:
>
> static inline int test1a (int a, int b) { return a ? 1 : b; }
> int test1b (int a, int b) { if (test1a (a,b)) return a+b; else return 37; }
>
> static inline int test2a (int a, int b) { return a || b; }
> int test2b (int a, int b) { if (test2a (a,b)) return a+b; else return 37; }
>
I think hpa was talking about some code where gcc can not optimize out
the assignment (e.g. volatile, complex code, using the int outside
conditional expressions, etc.).
>=>
>
> test1b:
> testl %edi, %edi
> jne .L2
> movl $37, %eax
> testl %esi, %esi
> jne .L2
> rep
> ret
> .L2:
> leal (%rsi,%rdi), %eax
> ret
>
> test2b:
> leal (%rsi,%rdi), %edx
> movl $37, %eax
> orl %edi, %esi
> cmovne %edx, %eax
> ret
>
> .ident "GCC: (Debian 4.5.1-8) 4.5.1"
>
>
> -Miles
>
> --
> Is it true that nothing can be known? If so how do we know this? -Woody Allen
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/