diff options
| author | Marshall Lochbaum <mwlochbaum@gmail.com> | 2022-03-27 22:13:40 -0400 |
|---|---|---|
| committer | Marshall Lochbaum <mwlochbaum@gmail.com> | 2022-03-27 22:14:37 -0400 |
| commit | ab0c110db931360231647e512d3323e82c06c345 (patch) | |
| tree | 020d89491bdbc82d917005f60916fbbc34c00a50 /docs | |
| parent | a51466b3e13299cd89b3b84305837607514b6817 (diff) | |
2021 AoC comparison shows BQN has very few icache misses on AoC-style code as well
Diffstat (limited to 'docs')
| -rw-r--r-- | docs/implementation/kclaims.html | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/docs/implementation/kclaims.html b/docs/implementation/kclaims.html index 837ac1c2..79731f87 100644 --- a/docs/implementation/kclaims.html +++ b/docs/implementation/kclaims.html @@ -106,4 +106,4 @@ <span class='Number'>1.245378356</span> <span class='Value'>seconds</span> <span class='Value'>time</span> <span class='Value'>elapsed</span> </pre> -<p>The stalls are less than 1% here, so maybe the smaller executable is paying off in some way. I can't be sure, because the programs being run are very different: <code><span class='Number'>19</span><span class='Value'>.k</span></code> is 10 lines while the others are hundreds of lines long. But I don't have a longer K program handy to test with (and you could always argue the result doesn't apply to Whitney's K anyway). Again, it doesn't matter much: the point is that the absolute most the other interpreters could gain from being more L1-friendly is about 10% on those fairly representative programs.</p> +<p>The stalls are less than 1% here, but it seems this is largely due to the different nature of the program: <code><span class='Number'>19</span><span class='Value'>.k</span></code> is 10 lines while the others are hundreds of lines long. Now that <a href="../community/aoc.html">Advent of Code</a> 2021 has run, dzaima points out that his solutions are comparable in intent to ngn's, and I measure very close to 0.5% icache stalls in both (27 of 5,404 million cycles in BQN and 34 of 6,600 in ngn/k, problems 23 and 24 omitted). But I don't have a longer K program handy to test with, and you could always argue the result doesn't apply to Whitney's K. Again, it doesn't matter much: the point is that the absolute most the other interpreters could gain from being more L1-friendly is about 10% on those fairly representative programs.</p> |