On Tuesday 28 February 2012 21:55:57 Dimitry Andric wrote: > On 2012-02-28 20:39, Tijl Coosemans wrote: >> Author: tijl >> Date: Tue Feb 28 19:39:54 2012 >> New Revision: 232266 >> URL: http://svn.freebsd.org/changeset/base/232266 >> >> Log: >> Copy amd64 endian.h to x86 and merge with i386 endian.h. Replace >> amd64/i386/pc98 endian.h with stubs. >> >> In __bswap64_const(x) the conflict between 0xffUL and 0xffULL has been >> resolved by reimplementing the macro in terms of __bswap32(x). As a side >> effect __bswap64_var(x) is now implemented using two bswap instructions on >> i386 and should be much faster. __bswap32_const(x) has been reimplemented >> in terms of __bswap16(x) for consistency. > ... >> +#define __bswap32_const(_x) \ >> + (((__uint32_t)__bswap16(_x) << 16) | __bswap16((_x) >> 16)) >> + >> +#define __bswap32(_x) \ >> + (__builtin_constant_p(_x) ? \ >> + __bswap32_const((__uint32_t)(_x)) : __bswap32_var(_x)) >> + >> +#define __bswap64_const(_x) \ >> + (((__uint64_t)__bswap32(_x) << 32) | __bswap32((_x) >> 32)) > > Hmm, shouldn't __bswap32_const() be implemented in terms of > __bswap16_const(), and similarly for __bswap64_const(), which should be > implemented in terms of __bswap32_const()? > > The whole reason for the difficult dance with __builtin_constant_p is to > allow for compile-time resolving of bswap'd constants. Invoking the > regular __bswap() macros doesn't gain you anything here.
__bswap64_const is also used in __bswap64_var, so its argument isn't always a compile time constant and then __bswap32 becomes __bswap32_var. If it is a constant __bswap32 becomes __bswap32_const (even at -O0).
signature.asc
Description: This is a digitally signed message part.
