13

In pre-.NET world I always assumed that int is faster than byte since this is how processor works.

Now it's matter habit of using int even when bytes could work, for example when byte is what is stored in database

Question: How .NET handles byte type versus int from point view of performance/memory.

Update: Thanks for the input. Unfortunately, nobody really answered the question. How .NET handles byte vs. int.

And if there is no difference in performance, then I like how chills42 put it: int for arithmetics bytes for binary Which I will continue to do.

Maxim Alexeyev
  • 1,021
  • 11
  • 24

5 Answers5

12

Your pre-.NET assumption were faulty -- there have always been plenty of computer systems around that, while nominally "byte-addressable", would have to set a single byte by reading a full word, masking it out to alter the one byte of it, write it all down -- slower than just setting a full word. It depends on the internals of the way the processor and memory are connected, not on the programmer-visible architecture.

Whether in .NET or native code, focus first on using the data as semantically correct for your application, not on trying to double guess the computer system's architect -- "Premature optimization is the root of all evil in programming", to quote Knuth quoting Hoare.

Alex Martelli
  • 854,459
  • 170
  • 1,222
  • 1,395
  • The quote is great, the trick is define "premature". While I agree that byte does not necessarily literally translates into reading a byte or using processor register in restricted way. Isnn't this "masking" comes with a price? – Maxim Alexeyev Jun 19 '09 at 21:30
  • 2
    @Maxim, of course it "comes with a price" -- one you _don't know whether you're paying_, ever since IBM introduced the concept of architecture (what the machine language programmer sees) as totally distinguished from actual hardware, with the IBM 360 series almost 50 years ago. If you're coding for any halfway-popular architecture (x86 is the worst for this, as it has by far the most, and most-divergent, implementations) you _just can't know_ how costly reading 1 byte vs 1 fullword is. So, don't even worry about it, and instead use straightforward, simple code. – Alex Martelli Jun 20 '09 at 02:44
  • So if I use int (32 bit), then it will run slower than __int64 on a 64 bit machine (both Android & PC)? – 123iamking Feb 02 '18 at 03:41
  • If int (32) run slower on 64bit machine, then byte & int will be exactly the same performance on a 64 machine? (but byte will be better for array store in that case?) – 123iamking Feb 02 '18 at 03:59
7

Are you talking about storage space or operations on a byte? If it's storage space, yes it takes up less space than an int (1 byte vs 4 bytes).

In terms of arithmetic operations on a byte, I don't have raw numbers and really only a profiler can give them to you. However you should consider that arithmetic operations are not done on raw byte instances. Instead they are promoted to int's and then the operation is done on an int. This is why you have to explicitly cast operations like the following

byte b1 = 4;
byte b2 = 6;
byte b3 = b1 + b2;  // Does not compile because the type is int

So in the general case I think it's safe to say that arithmetic operations on an int are faster than that of a byte. Simply because in the byte case you pay the (probably very small) cost of type promotion.

JaredPar
  • 733,204
  • 149
  • 1,241
  • 1,454
3

OK, I just opened disassembly window. There is nothing there but regular "mov byte"

So, .NET/CLR does not add anything to this. And all arithmetic operations done against int values, so no difference between bytes and int there.

Maxim Alexeyev
  • 1,021
  • 11
  • 24
2

Unless you are finished with your design and need to find clever ways to optimize, then just use what you need.

If you need a counter or are doing basic math, it is probably what you want, if you're working with binary data, go with a byte.

In the end, each type should be optimized for it's intended purposes, so you're better off to spend your time on design instead of optimization.

chills42
  • 14,201
  • 3
  • 42
  • 77
-1

Same as any other platform. Why would .NET change this? The code still has to run on the same CPU, which has the same performance characteristics as always.

And that means you should still use int by default.

jalf
  • 243,077
  • 51
  • 345
  • 550