Koffi performance

[Koromix]

Hi,

I saw your package some time ago and it sent me into a rabbit hole when I saw how slow Koffi was on your memset benchmark. Thank you for noticing this problem :)

Part of the reason was that Koffi 2 uses external objects to represent pointers, instead of BigInt, and memset returns a pointer. So for each memset call, a V8 external object is created for the return value, and it is slow. But there are many other inefficiencies in Koffi 2 !

I will soon release Koffi 3.0, which uses BigInt pointers. But more than that, I've significantly reduced the overhead for normal synchronous calls. You can test the new version with npm install koffi@3.0.0-alpha.1 if you want.

With all the changes I've made, I get the following benchmark results on Windows with koffi@3.0.0-alpha.1:

╔════════════════════════════════════════════════════════════════╗
║                          SUMMARY                               ║
╠════════════════════════════════════════════════════════════════╣
║ ✗ abs(int32)                           2.99x slower vs koffi   ║
║ ✗ strlen(string)                       2.57x slower vs koffi   ║
║ ✗ sqrt(double)                         3.62x slower vs koffi   ║
║ ✗ GetTickCount()                       3.40x slower vs koffi   ║
║ ✗ memset(ptr,int,size)                 3.80x slower vs koffi   ║
║ ✗ sprintf(variadic)                    1.22x slower vs koffi   ║
║ ✗ struct read (2 fields)               1.20x slower vs koffi   ║
╠════════════════════════════════════════════════════════════════╣
║ Average ratio: 2.68x (koffi faster)                            ║
║ Wins: node-ctypes 0, koffi 7                                   ║
╚════════════════════════════════════════════════════════════════╝

[dmazzella]

Great results, well done!