Re: [RFC][PATCH] new byteorder primitives - ..._{replace,get}_bits()

From: Jakub Kicinski
Date: Tue Dec 12 2017 - 21:44:39 EST


On Wed, 13 Dec 2017 01:51:25 +0000, Al Viro wrote:
> On Tue, Dec 12, 2017 at 05:35:28PM -0800, Jakub Kicinski wrote:
>
> > It used to be __always_inline, but apparently LLVM/clang doesn't
> > propagate constants :(
> >
> > 4e59532541c8 ("nfp: don't depend on compiler constant propagation")
>
> Doesn't propagate constants or doesn't have exact same set of
> rules for __builtin_constant_p()? IOW, if you dropped that
> BUILD_BUG_ON(), what would be left after optimizations?

Hm. You're right. It just doesn't recognize the parameter as constant
in __builtin_constant_p(). I haven't compiled the actual code, but
here is a trivial test:

18:36 ~$ clang --version
clang version 4.0.1 (tags/RELEASE_401/final)
Target: x86_64-unknown-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

18:36 ~$ cat /tmp/test.c
#include <stdio.h>

static inline int test_thing(unsigned int a)
{
printf("const: %d\n", __builtin_constant_p(a));
if (a > 10)
return a - 1;
return a;
}

int main()
{
printf("const: %d\n", __builtin_constant_p(0));
return test_thing(0);
}
18:36 ~$ clang -g /tmp/test.c -O2 -Wall -W -o /tmp/test
18:36 ~$ /tmp/test
const: 1
const: 0
18:36 ~$ objdump -S /tmp/test
...
00000000004004e0 <main>:
return a;
}

int main()
{
printf("const: %d\n", __builtin_constant_p(0));
4004e0: 50 push %rax
4004e1: bf a0 05 40 00 mov $0x4005a0,%edi
4004e6: be 01 00 00 00 mov $0x1,%esi
4004eb: 31 c0 xor %eax,%eax
4004ed: e8 fe fe ff ff callq 4003f0 <printf@plt>
printf("const: %d\n", __builtin_constant_p(a));
4004f2: bf a0 05 40 00 mov $0x4005a0,%edi
4004f7: 31 f6 xor %esi,%esi
4004f9: 31 c0 xor %eax,%eax
4004fb: e8 f0 fe ff ff callq 4003f0 <printf@plt>
return test_thing(0);
400500: 31 c0 xor %eax,%eax
400502: 59 pop %rcx
400503: c3 retq
400504: 66 2e 0f 1f 84 00 00 nopw %cs:0x0(%rax,%rax,1)
40050b: 00 00 00
40050e: 66 90 xchg %ax,%ax

...