Re: [PATCH 2/8] compat: make gcc bswap an inline function

From: Junio C Hamano <gitster@pobox.com>
Date: Mar 16, 2011, 19:44 UTC
Message-ID: <7v8vwe97f9.fsf@alter.siamese.dyndns.org>
In-Reply-To: <20110316093108.GB7886@elie>

Jonathan Nieder <jrnieder@gmail.com> writes:

Show 13 quoted lines

> Johannes Sixt wrote:
>> Am 3/16/2011 8:00, schrieb Jonathan Nieder:
>
>>> +static inline uint32_t git_bswap32(uint32_t x)
>>> +{
>>> +	uint32_t result;
>>> +	if (__builtin_constant_p(x))
>>
>> Can this predicate ever be true? Isn't it false even if the function is
>> inlined?
>
> It's true if x is a constant.
> ... demonstration ...

And this trivial function:

    int f(void) {
            return git_bswap32(0x01000000);
    }

compiles down to:

    .globl f
            .type   f, @function
    f:
    .LFB2:
            .cfi_startproc
            movl    $1, %eax
            ret
            .cfi_endproc

on my amd64, with -Os, -O1, or -O2 (gcc 4.4.5). Of course -O0 gives a true call to git_bswap32() and leaves the definitions of two helper functions in the object, but that is to be expected.