hi, If you want to align data during an optimization of a (C++) program, But should I align to (2^n) or to ((2^n)-1) ?? My structure has at the moment a size of 24, so I want to align it to 32, or should I start counting from zero and align it to 31?? cheers

The first byte of the structure should be on an address that is a multiple of 32.

cache is right. 0 through 31 is the first block of 32, so 32-63 is the next block, and so on.

If your struct only has "normal" (that is, no SSE) members, then aligning to 24 bytes will be enough, too. The largest "normal" datatype is 8 bytes, which works fine with 24 (and all others do anyway).
This will likely be good in respect of cache lines too. I am saying "likely" because some CPUs have a problem doing a single load/store across a cache line boundary, and I don't know what your struct looks like. If your struct is layed out the way it should be for normal aligned access, then no problem should occur in any case, it will fit evenly with the typical 32 byte cache lines. But if your layout is bad, you might get misaligned cache access in addition to misaligned access. Therefore, I'm careful saying something without having seen the struct.
On the other hand, having your data in less cache lines is obviously better, especially since the total number of cache lines is limited, and since the first 8 bytes of the struct are guaranteed to be available "instantly", and the CPU is likely to prefetch the next line while you're walking across the current one.

Preferrably, you should use _mm_malloc/_mm_free (for intel or gcc) or _aligned_malloc/_aligned_free (for Microsoft) with an alignment parameter of 24 (which will take care of the low level pointer whacking stuff) and allocate an array of structs for the sake of efficiency, rather than one by one.
If you allocate your structs on the stack, you'd use the respective __attribute__ or __declspec, of course.

ok, thx for the replies :)