use gdb for a more accurate rss value

.gitignore

@@ -18,3 +18,4 @@ CMD
 VERSION
 STATS
 SIZE
+rss.txt

count.asm

@@ -47,8 +47,7 @@ program_header:                   ; Elf64_Phdr
     output db 10 dup(0), 10 ; max 32bit number is 10 digits (decimal) long, followed by newline
 
 _start:                 ; argc is at `rsp`, argv is at `rsp + 8`
-    mov rdi, [rsp + 8]  ; put argv into `rdi`
-    add rdi, 8          ; skip the first argument (program name)
+    mov rdi, [rsp + 16] ; put argv[1] into `rdi` (+ 8 to skip argv[0])
 
     xor rcx, rcx        ; zero
     xor rax, rax        ; zero

justfile

@@ -90,8 +90,9 @@ measure what:
   if [[ -f SIZE ]]; then
     jq '. += {"size":"'"$(cat SIZE)"'"}' "$out" | sponge "$out"
   fi
-  timers $(cat CMD) >/dev/null 2> STATS
-  jq '. += {"max_rss":'$(rg -oP '(?:max_rss:\s*)(\d+)' -r '$1' ./STATS)'}' "$out" | sponge "$out"
+
+  gdb -return-child-result --command rss.gdb --args $(cat CMD) >/dev/null
+  jq '. += {"rss":'$(( $(rg -oPr '$1' 'Rss:\s*(\d+)' ./rss.txt) * 1024 ))'}' "$out" | sponge "$out"
 
 measure-all:
   #!/usr/bin/env bash

rss.gdb

@@ -0,0 +1,19 @@
+# run this with:
+
+# catch program exit
+catch syscall exit
+run
+
+# capture program pid
+python
+gdb.execute("set $pid = " + str(gdb.selected_inferior().pid))
+end
+
+# while it's still running, extract rss from /proc
+eval "shell cat /proc/%d/smaps_rollup > rss.txt", $pid
+
+# allow program to exit, and exit gdb
+# when gdb is run with the `-return-child-result` flag it will cause gdb to exit
+# with the same code the child did
+continue
+exit

scripts/summary.js

@@ -39,7 +39,7 @@ await writeFile(
 <tr>
   <th>Execution time</th>
   <th>Binary size<sup>1</sup></th>
-  <th>Max Memory Usage<sup>2</sup></th>
+  <th>Max Memory Usage</th>
 </tr>
 <tr>
 <td>
@@ -90,11 +90,11 @@ ${[...sizeTypes.values()]
 
 ${markdownTable(
   [
-    ['#', 'name', 'max_rss'],
+    ['#', 'name', 'rss'],
     ...results
       .slice()
-      .sort((a, b) => a.max_rss - b.max_rss)
-      .map(({ name, max_rss }, i) => [i + 1, wrap(name), formatSize(max_rss, { minimumFractionDigits: 7 })]),
+      .sort((a, b) => a.rss - b.rss)
+      .map(({ name, rss }, i) => [i + 1, wrap(name), formatSize(rss, { minimumFractionDigits: 7 })]),
   ],
   {
     align: ['l', 'l', 'r'],
@@ -106,7 +106,6 @@ ${markdownTable(
 </table>
 
 > - <sup>1</sup>: only includes compiled files (i.e., does not include runtimes or libraries required for execution)
-> - <sup>2</sup>: Getting the \`max_rss\` isn't 100% reliable for very small binary sizes. This appears to be [a limitation of the linux kernel](https://github.com/acheronfail/timeRS/blob/master/LIMITATIONS.md).
 
 ${markdownTable(
   [