from C++ to C# // Structure Memory Optimization

Question

@Zbyl I have seen your answer in this thread (Bit fields in C#) and i really like the Bitvector32 method , but for purpose of optimization : what if i have plenty of structures of a size of 8 bits / 12 bits (less than 32 bits) , is there anyway to do it in a manner less than Bitvector32 because that would be a lot of exagerated allocation of memory that would never be used : i will only use the first 8 bits of Bitvector32 . Here is an example of structure that i want to make in C# :

   struct iec_qualif
{
    unsigned char var :2;
    unsigned char res :2;
    unsigned char bl :1; // blocked/not blocked
    unsigned char sb :1; // substituted/not substituted
    unsigned char nt :1; // not topical/topical
    unsigned char iv :1; // valid/invalid
};

@AhmedAekBenJemia he means you should have used `unsigned char blocked :1;` instead of `unsigned char bl :1; // blocked/not blocked`. — Marian Spanik, Apr 22 '16 at 07:30
How many of these 8 bit / 12 bit structures does your program hold in memory? — peterchen, Apr 22 '16 at 07:37
Have you considered placing up to four of your structs into a single BitVector32? — Aconcagua, Apr 22 '16 at 07:38
like 14 structs of 8 bits /8 structs of 12 bits/ 10 structs of 16 bits / 10 structs of 24 bits / — Ahmed Aek Ben Jemia, Apr 22 '16 at 07:41
@Aconcagua is it possible to do that with separate structs ?? because the structs are declared out of the class , so a declared Bitvector outside of the struct will never be recognized .. — Ahmed Aek Ben Jemia, Apr 22 '16 at 07:43
@AhmedAekBenJemia I see what you mean - you want to wrap a class around the BitVector. Will get difficult then, if not impossible... — Aconcagua, Apr 22 '16 at 07:58
C# has totally distinct memory model. Similar (in memory model) to C is only struct of basic types, all other allocates objects, uses references etc. Maybe suggest data in simple types and accessors as properties? — Jacek Cz, Apr 22 '16 at 08:45

score 1 · Accepted Answer · edited May 23 '17 at 11:59

At first, I would consider the number of structs you are actually dealing with. Today's GB-memory-equipped hardware should easily be capable to cope with several thousands of your structs, even if you waste three bytes for each (1 000 000 of your structs would then occupy 4 MB of potentially > 1000 MB available). With this in mind, you could even live without the bit field entirely, having just normal bytes for each field (your setter should check range then, however), resulting in 6 bytes instead of four (and possibly two more for alignment issues), but giving you faster access to the values (getters), as no bit fiddling is necessary.

On the other hand: Bit fields in the end are nothing more than a convenient way to let the compiler write the code that you otherwise would have to write yourself. But with a little practise, it is not a too difficult task to handle the bits yourself, see this answer in the topic you referred to yourself (the 'handmade accessors' part), storing internally all data in a variable of type byte and accessing it via bitshifting and masking.

from C++ to C# // Structure Memory Optimization

1 Answers1