Kövesdi György wrote: > Hi, > > I would like to use C++ for 8-bit AVR CPUs, so saving space would be > important. > I found that the constructors and destructors always have two instances, which > are exactly the same. I tried it for X86, X86/64, M68k, and AVR platforms, > with the same result, so it is not platform-dependent. > > Here is a disassembled example: > > 000000f8 <valami::valami()>: > f8: fc 01 movw r30, r24 > fa: 84 e0 ldi r24, 0x04 ; 4 > fc: 91 e0 ldi r25, 0x01 ; 1 > fe: 91 83 std Z+1, r25 ; 0x01 > 100: 80 83 st Z, r24 > 102: 13 82 std Z+3, r1 ; 0x03 > 104: 12 82 std Z+2, r1 ; 0x02 > 106: 08 95 ret > > 00000108 <valami::valami()>: > 108: fc 01 movw r30, r24 > 10a: 84 e0 ldi r24, 0x04 ; 4 > 10c: 91 e0 ldi r25, 0x01 ; 1 > 10e: 91 83 std Z+1, r25 ; 0x01 > 110: 80 83 st Z, r24 > 112: 13 82 std Z+3, r1 ; 0x03 > 114: 12 82 std Z+2, r1 ; 0x02 > 116: 08 95 ret > > Is it possible to switch off this behavior? AFAIR gcc produces up to three versions of a constructor (internally named somewhat with C1, C2, C3) which are invoked for different situations (e.g. being a derived or superclass in a class hierachy that uses virtual functions or even virtual derivation.). I don't know the exact details, but assume that if one of these situations is not the case, gcc still produces all versions of the constructor even though the resulting machine code is identical. To proof this theory you might want to take a look at the generated symbol names without demangling them. They should be different. If that is the case you might try compiling with -ffunction-sections and linking with --gc-sections. I suppose you are not using virtual functions, so either version is not actually used and the linker removes it during garbage collection. Daniel
