Optimisation of classic FizzBuzz test.
The least efficient implementation, with 3 ifs and two printfs per number. It is so inefficient that I don't even use it for comparison.
A little bit optimised version, with 2 ifs (not counting the loop condition) and 1 printf per number.
Single loop iteration for 15 numbers, it means for 15 numbers just one branch (vs 45 branches for naive) and 1 printf (vs 15 printfs for naive).
Generic printf replaced with custom print routine, tailored for this particular task.
Like customprint, but buffer is filled in reverse, and adjacent Fizz and Buzz words are merged into single memcpy, as a result number of memcpy calls per iteration goes down from 15 to just 4. Courtesy of kariya-mitsuru.
Reuse buffer from previous iteration, update only the changed characters. Use x86_64 vector instructions for comparing buffers.
Reuse buffer from previous iteration, update only the changed characters. Use x86_64 vector instructions for comparing buffers. Fill buffer in reverse to reduce number of memcpy calls. Courtesy of kariya-mitsuru.
Use worker threads to process the sets of numbers in parallel. Buffer is filled in reverse, like in reusebuf2.
All tests are performed on Dell Latitude 7480 with Core i7-7600U (2 cores with hyperthreading) and 16 Gb RAM, OpenSUSE Leap 15.1, kernel 4.12.14, gcc 7.5.0.
Output redirected to /dev/null. Multithreaded implementation uses 4 worker threads, with load of 3M number per thread.
|Implementation||Time (sec.millisec)||Relative to naive||Relative to previous|