Instruction timings

Here are some results of timings I've done of several instructions in C/C++, to compare several ways to do one thing.

Proceeding

FIXME

Summary of results

Depending on machines, floating point operations can be a lot slower than integer operations, or as fast, or even faster.

Multiprocessor configs and multicore processors doesn't bring any improvement on these tests, and that's normal. Indeed in order to exploit it you have to have several threads, which is not the case here, and not the purpose.

All results

All timings are given in nanoseconds (ns).

Title Code Cyril_F Cyril_P cJ_F MacGyver Infomob Almighty PC104
RGB → Y Conversion
Float imgg[k2++] = (0.299*img[k++] + 0.587*img[k++] + 0.114*img[k++]); 175 18.8 22.1 6.09 ??! 20.8 12.6 84.3
Integer imgg[k2++] = (19595*img[k++] + 38470*img[k++] + 7471*img[k++]) » 16; 91.1 8.92 8.30 25.1 4.03 153
Shift1 imgg[k2++] = (img[k++]»2 + img[k]»1 + img[k++]»3 + img[k++]»3); 50.4 9.75 11.2 10.6 5.97 29.3
Shift2 imgg[k2++] = (img[k]»2 + img[k++]»4 + img[k]»1 + img[k++]»4 + img[k++]»3); 56.7 12.5 13.3 11.7 6.65 43.4
Array indexes

Configs used

Summary

CPU Frequence Arith. Coproc. Bogomips FPU Cache GCC Version
Cyril_F Celeron Mendocino 500 MHz 1003 yes 128 KB3.
Cyril_P AMD64 3000+ 1.8 GHz 3593 yes 1024 KB 4.1.1
cJ_F AthlonXP 2400+ 2.0 GHz 4013 yes 256 KB 4.1.1
MacGyver P-m Dothan 1.7 GHz 3400 yes 2MB 4.1.1
Infomob Tri-Xeon 2.80 GHz 5570 yes 512 KB 3.2.2
Almighty PentiumD (DualCore64) 3.4 GHz 6800 yes 2048 KB 4.1.1
PC104 Transmeta Crusoe TM5400 500 MHz 997 yes 256 KB 3.4.6

/proc/cpuinfo

programming/instruction-timings.txt · Last modified: 2013/09/19 16:41 (external edit)
 
Except where otherwise noted, content on this wiki is licensed under the following license: CC Attribution-Share Alike 3.0 Unported
Recent changes RSS feed Donate Powered by PHP Valid XHTML 1.0 Valid CSS Driven by DokuWiki