use callgrind for estimating cycles

...I expect this will be extremely slow for the slower langs, since even
just using this with assembly can take quite some time!

Author: acheronfail

.gitignore

@@ -9,6 +9,7 @@ results/*.json
 *.o
 *.exe
 *.hi
+*.txt
 count
 java/
 kotlin/
@@ -18,6 +19,7 @@ CMD
 VERSION
 STATS
 SIZE
+CYCLES
 rss.txt
 GPATH
 GRTAGS

docker/Dockerfile

@@ -4,7 +4,7 @@ ENV DEBIAN_FRONTEND="noninteractive" TZ="Europe/London"
 
 RUN apt-get update
 RUN apt-get install -y bc build-essential curl gdb lldb git jq moreutils tar \
-  unzip wget xz-utils                          && apt-get clean
+  unzip wget xz-utils valgrind                 && apt-get clean
 RUN apt-get install -y clang                   && apt-get clean
 RUN apt-get install -y default-jdk default-jre && apt-get clean
 RUN apt-get install -y erlang                  && apt-get clean

justfile

@@ -100,6 +100,10 @@ measure what: (_check "bc hyperfine max_rss jq sponge")
   max_rss -ro STATS -- $(cat CMD)
   jq '. += {"max_rss": '$(cat STATS)'}' "$out" | sponge "$out"
 
+  valgrind --tool=callgrind --callgrind-out-file=callgrind.txt $(cat CMD)
+  grep "summary:" callgrind.txt | awk '{print $2}' > CYCLES
+  jq '. += {"cycles": '$(cat CYCLES)'}' "$out" | sponge "$out"
+
 measure-all:
   #!/usr/bin/env bash
   set -exuo pipefail

scripts/summary.js

@@ -107,6 +107,22 @@ ${markdownTable(
 
 > - <sup>1</sup>: only includes compiled files (i.e., does not include runtimes or libraries required for execution)
 
+${markdownTable(
+  [
+    ['name', 'estimated cycles'],
+    ...results
+      .slice()
+      .sort((a, b) => a.cycles - b.cycles)
+      .map(({ name, cycles }) => [name, cycles]),
+  ],
+  {
+    align: ['l', 'r', 'l'],
+  }
+)}
+
+Note that cycles are counted with valgrind (using callgrind) and are only estimates of the actual CPU cycles used.
+
+
 ${markdownTable(
   [
     ['name', 'command', 'version'],