This is not a rigorous experiment, just trying to see if I am improving or regressing the software. I will list the mean and stddev of 5 iterations of "the actual sort" run-time after 1 warm-up iteration.
mergesort:
- commit: a49f1cf
description: initial naive implementation
mean: 13.335
std: 0.497
- commit: 20b7263
description: saved pointer to chunk
mean: 7.078
std: 0.265
- commit: 4bc0628
description: saved pointer to column data
mean: 7.310
std: 0.265
- commit: 017955f
description: >
corrected logical error in mergesort.
Does more copying. Note that the actual sorting time did not change much (even
got smaller), but the time to make the out array got big. This time is not
reported in "actual sort".
mean: 6.889
std: 0.214
? |-
merge and
countingsort
:
- commit: 3a40125
description: initial trial
mean: 6.345
std: 0.206
? |-
counting and
mergesort
:
- commit: 2c978a0
description: initial trial
mean: 2.375
std: 0.092
countingsort:
- commit: 3d26044
description: initial trial
mean: 1.531
std: 0.066
- commit: 4bc0628
description: saved pointer to column data
mean: 1.521
std: 0.085