Static variables in inline function

Question

I have a hash function defined in a header file which I want to make sure is being inlined, but it uses a very large const static buffer of ints. I am wondering whether this buffer will affect the ability of the function to be inlined. Here is the code,(it's based on a crc32 implementation found here).

inline HashId hash(const void *value, const size_t length,
   const HashId previous = 0)
{
   // Here is the really large const static buffer.
   static const unsigned int crcTable[256] = { /* 256 Unique ints */ };

   // Short hash calculation
   unsigned char *cast = (unsigned char*) value;
   unsigned int crc32 = previous ^ 0xFFFFFFFF;
   for (size_t i = 0; i < length; ++i)
   {
      crc32 = (crc32 >> 8) ^ crcTable[ (crc32 ^ cast[i]) & 0xFF];
   }
   return (crc32 ^ 0xFFFFFFFF);
}

My hope is, that if the input is known at compile time, then this call will get compiled away into a single value. Is this expecting too much of the compiler? I'm compiling with gcc 4.6 with -O2, but I'm also interested to hear how other compilers might treat this.

No, it will not get compiled away into a single value. You are asking too much of the compiler. The compiler will generally not evaluate anything at compile-time that involves dereferencing pointers. — Dietrich Epp
– Dietrich Epp, Commented Oct 16, 2011 at 11:57
@OliCharlesworth I did, although my assembler knowledge is pretty basic,it appears to be inlining the loop, but not unrolling it. I don't want to turn on -funroll-loops, as it hurts performance in other parts of the program. Is their any way I can unroll a single loop? — Darcy Rayner
– Darcy Rayner, Commented Oct 16, 2011 at 12:00
template meta programming? I'm not an expert, but probably it will be hard to do — Ruggero Turra
– Ruggero Turra, Commented Oct 16, 2011 at 12:28
<quote>I want to make sure is being inlined</quote>. Why do you care. The compiler is making a decision on what to do based on how to make the application as fast as possible. If the compiler decides its faster not to inline it do you still want to force the inline? Basically humans are notoriously bad at making this decision. The compiler is very good at it. Let the compiler do its job and forget about what it is doing. — Loki Astari
– Loki Astari, Commented Oct 16, 2011 at 17:40

edA-qa mort-ora-y · Accepted Answer · 2011-10-16 11:55:19Z

As Oli indicates, your best bet is to simply look at the compiler output. However, here are some things to consider:

If the contents of the static are truly constant you will be better putting this in an anonymous namespace. This ensures its initialization has nothing to do with calling the function.
type of the static. Your example is quite trivial and thus will probably be fine, but other types will be different. Note that the initialization of such statics may or must be deferred until the block is entered (puzzle through the standard at 6.7 if you want)
Multi-threading, relating to the previous this initialization may be required to use a mutex to prevent multiple threads from doing the initialization.

Given the possible scenarios it'd be best to not have statics inside your function if you wish it to be inlined. But each compiler is different, and all types are not equal.

Whether the compiler can actually reduce this function to a compile-time constant is just an extended issue to whether it can inline the static at all. You may also wish to look at constexpr specifier if you can use C++11.

Collectives™ on Stack Overflow

Static variables in inline function

1 Answer 1

1 Comment

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

1 Comment

Related