NOTES TO SELF atomic struct mthread_status { unsigned int area_covered bool undevidable bool searching bool division_syn bool division_ack bool division_syn_ack } Thread finishes its job. 1: set "searching" local atomic struct to true 2: for each thread: if done=true; exit if searing, division_syn, division_ack, undevidable: go to next if all seem to be searching, set global done=true if not: continue below 3: record area_covered in own array, if more threads proceed to next, else continue below 4: pick thread with largest area_covered: 5: if division_syn or division_ack or searching or undevidable, go to next 5: set division_syn to true, wait for division_ack 6: divide(): copy over half of x/y, copy divisions, check and set undevidable, start to do work thread is working. 1: for y: update area_covered check division_syn if true: 2 inline here for x: do some math stuff here searching = true, "if thread finishes its job" inline here. 2: set division_ack wait for division_syn_ack check and set undevidable // there should be no conflict with updating variable y is dependent on, // as it's always more then current, and test will be accurate next loop. thread is starting. 1: for y: update area_covered check division_syn if true: 2 inline here for x: do some math stuff here searching = true, "if thread finishes its job" inline here. PREFORMANCE: without tasks helping eachother (commit c30fc7596810f7033dfd9a7452c808153bd2e14a): const uint32_t WIDTH = 1920; const uint32_t HEIGHT = 1080; const int JOBS = 6; //test uneven stuff const std::complex b_min (-0.7463-0.005, 0.1102-0.005); const std::complex b_max (-0.7463+0.005, 0.1102+0.005); #define MAX_ITER 1000 #define INF_CUTOFF 256 #define COLOR_RAMP 100 /usr/bin/time -f '%p' -p ./mandelbrot real 36.41 user 105.00 sys 0.02 real 36.38 user 104.98 sys 0.02 real 36.32 user 105.00 sys 0.05 With tasks helping eachother: (commit 3663966b88681f44ec8939e39e33ef922227b7a7): real 19.21 user 111.70 sys 0.03 real 19.29 user 111.61 sys 0.02 real 19.30 user 111.84 sys 0.02 //CONTENDOR_HI REPLACEMENT - 50000 iterations //const complex DEFAULT_B_MIN (-0.74364386269 - 0.00000003000, 0.13182590271 - 0.00000003000); //const complex DEFAULT_B_MAX (-0.74364386269 + 0.00000003000, 0.13182590271 + 0.00000003000); //CONTENDOR_ZOOM- 50000 iterations //const complex DEFAULT_B_MIN (-0.74364386269 - 0.00000001000, 0.13182590271 - 0.00000001000); //const complex DEFAULT_B_MAX (-0.74364386269 + 0.00000001000, 0.13182590271 + 0.00000001000);