Patrick Horgan <phorgan1@xxxxxxxxx> writes: [...] > If you look on the boost wiki you find a discussion of strict-aliasing > and how gcc generates awesome code if you do things right. (If you > like it I wrote that section, if you don't it was probably random > cosmic rays). > > https://svn.boost.org/trac/boost/wiki/Guidelines/WarningsGuidelines > I have 2 remarks: >From the -fstrict-alisiang part, one can read: In particular you're promising by using this flag that an object of one type won't reside at the same address as an object of an incompatible type I don't think this is exact since the standard allows one special case of this: please see the C's standard 6.5.2.3p5 The other thing is about the swaphalves() example. The main point is that -fno-strict-aliasing (without the use of union) produces incorrect code, the code size is pretty much the same for both cases. Reading this section, I feel like '-fno-strict-aliasing' without the union produces incorrect _and_ slow code for that particular example which is not true: $ gcc -O2 -fno-strict-aliasing swaphalves-with-union.c && objdump -D a.out [...] 0000000000400480 <swaphalves>: 400480: 89 f8 mov %edi,%eax 400482: c1 c0 10 rol $0x10,%eax 400485: c3 retq [...] $ gcc -O2 -fno-strict-aliasing swaphalves.c && objdump -D a.out [...] 0000000000400480 <swaphalves>: 400480: 89 f8 mov %edi,%eax 400482: 66 89 7c 24 fe mov %di,-0x2(%rsp) 400487: c1 e8 10 shr $0x10,%eax 40048a: 66 89 44 24 fc mov %ax,-0x4(%rsp) 40048f: 8b 44 24 fc mov -0x4(%rsp),%eax 400493: c3 retq [...] -- Francis
