diff options
| -rw-r--r-- | Objects/listsort.txt | 14 |
1 files changed, 10 insertions, 4 deletions
diff --git a/Objects/listsort.txt b/Objects/listsort.txt index c561288..139c17c 100644 --- a/Objects/listsort.txt +++ b/Objects/listsort.txt @@ -113,8 +113,8 @@ Comparison with Python's Samplesort Hybrid 4377402 262437 262459 416347 1457945 524286 1.99% 1577.82% -0.06% 967.83% -24.01% 774.44% - 524288 9205096 9453356 9408463 524468 9441930 2218577 9692015 - 9278734 524580 524633 837947 2916107 1048574 + 524288 9205096 9453356 9408463 524468 9441930 2218577 9692015 + 9278734 524580 524633 837947 2916107 1048574 1.88% 1693.52% -0.03% 1026.79% -23.92% 824.30% 1048576 19458756 19950272 19838588 1048766 19912134 4430649 20434212 @@ -431,7 +431,7 @@ at-a-time mode. A refinement: The MergeState struct contains the value of min_gallop that controls when we enter galloping mode, initialized to MIN_GALLOP. -merge_lo() and merge_hi() adjust this higher when gallooping isn't paying +merge_lo() and merge_hi() adjust this higher when galloping isn't paying off, and lower when it is. @@ -549,7 +549,7 @@ that merge_lo and merge_hi adjust: the longer we stay in galloping mode, the smaller min_gallop gets, making it easier to transition back to galloping mode (if we ever leave it in the current merge, and at the start of the next merge). But whenever the gallop loop doesn't pay, -min_gallop is increased by one, making it harder to transition to back +min_gallop is increased by one, making it harder to transition back to galloping mode (and again both within a merge and across merges). For random data, this all but eliminates the gallop penalty: min_gallop grows large enough that we almost never get into galloping mode. And for cases @@ -576,6 +576,12 @@ probably a better guess at the final result than either 0 or 9999. But it's unclear how to generalize that intuition usefully, and merging of wildly unbalanced runs already enjoys excellent performance. +~sort is a good example of when balanced runs could benefit from a better +hint value: to the extent possible, this would like to use a starting +offset equal to the previous value of acount/bcount. Doing so saves about +10% of the compares in ~sort. However, doing so is also a mixed bag, +hurting other cases. + Comparing Average # of Compares on Random Arrays ------------------------------------------------ |