Matthew Woehlke wrote:
> Andrew Haley wrote:
>> Matthew Woehlke wrote:
>>> Using gcc-4.3.2-7.x86_64 (Fedora 10).
>>>
>>> I have these functions:
>>>
>>> void my_ntoh_float(float *f)
>>> {
>>> int p = (int*)f;
>>> *p = ntohl(*p);
>>> }
>>> [snip snip]
>>>
>>> Is the bug in the code, or in gcc? In either case, is there a way
>>> (besides dropping optimizations) to fix this that doesn't involve
>>> non-portable code?
>>
>> It's in your code. You're accessing an object (p) as an incompatible
>> type.
>> See C99, Section 6.3.2.3 for all the details.
>
> That's what I suspected.
>
>> Do this:
>>
>> float my_ntoh_float(float f)
>> {
>
> That should be 'float * f'.
>
>> union
>> {
>> int i;
>> float f;
>> } u;
>
> D'oh, should have thought of that. Brain must not be working :-).
>
> Anyway, I saw Brian's reply first, and already got it working with a
> variant of this (i.e. not changing the signature of my_ntoh_float, which
> I can't do for compatibility reasons!!). Thanks both for the quick replies!
>
>> Which generates this:
>>
>> my_get_float:
>> pushl %ebp
>> movl %esp, %ebp
>> subl $4, %esp
>> movl 8(%ebp), %eax
>> movl (%eax), %eax
>> bswap %eax
>> movl %eax, -4(%ebp)
>> flds -4(%ebp)
>> leave
>> ret
>
> Hmm, I still get this:
>
> ror $0x8,%ax
> ror $0x10,%eax
> ror $0x8,%ax
>
> ...which is interesting. Are you building 32-bit code? Looks like I
> maybe need to see if there is a -march that we aren't using that might
> help our performance (not just here, but in general).
>
> (I do get bswap with 64-bit code.)
It's this macro in byteswap.h:
# if __WORDSIZE == 64 || (defined __i486__ || defined __pentium__ \
|| defined __pentiumpro__ || defined __pentium4__ \
|| defined __k8__ || defined __athlon__ \
|| defined __k6__ || defined __nocona__ \
|| defined __core2__ || defined __geode__ \
|| defined __amdfam10__)
/* To swap the bytes in a word the i486 processors and up provide the
`bswap' opcode. On i386 we have to use three instructions. */
# define __bswap_32(x) \
(__extension__ \
({ register unsigned int __v, __x = (x); \
if (__builtin_constant_p (__x)) \
__v = __bswap_constant_32 (__x); \
else \
__asm__ ("bswap %0" : "=r" (__v) : "0" (__x)); \
__v; }))
# else
# define __bswap_32(x) \
(__extension__ \
({ register unsigned int __v, __x = (x); \
if (__builtin_constant_p (__x)) \
__v = __bswap_constant_32 (__x); \
else \
__asm__ ("rorw $8, %w0;" \
"rorl $16, %0;" \
"rorw $8, %w0" \
: "=r" (__v) \
: "0" (__x) \
: "cc"); \
__v; }))
# endif
Andrew.
