ludic build on every core: the IR split with llvm-split and compiled by a clang per part at once

Most of a build was one clang -O2 on one .ll (the game: 32 s of a 41 s headless build, one core).
Both paths that assemble - the CLI's (build.ludic: ludicc --emit-llvm, then clang) and ludicc's own
(-o, which ludic bundle and the examples use) - now cut the program's IR into N parts with llvm-split
(externalizing what the parts share), compile them with one clang each in parallel (-x ir -O<opt>
-mmacosx-version-min=11.0, the link's own clang taking the objects where it took the .ll), and remove
the parts and objects after. N is $LUDIC_JOBS, else min(cores, free GB / 1.5).

It needs an llvm-split and a clang of the same LLVM (Homebrew's LLVM 22 writes attributes Apple's
clang 17 cannot read): $LUDIC_LLVM, else /opt/homebrew/opt/llvm/bin. With either missing, on Windows
(its shell cannot run the parts at once yet), with LUDIC_SPLIT=0, or when a part fails, it compiles the
.ll whole as before.

$LUDIC_OPT=1 is a developer's faster build; ludic bundle sets LUDIC_OPT=2 for its compile whatever the
shell says.

Measured before the compile-only rule (this Mac, 12 cores, one build at a time):
- the game headless: 37-41 s -> 13-15.5 s (8 parts / by free memory), peak 2.1 GB -> 1.0-1.1 GB;
- the lab headless: 43.1 s -> 12.7 s, peak 2.4 GB -> 1.0 GB;
- the game at LUDIC_OPT=1, split: 11.8 s (fps cost not measured).
Both built and linked clean; the goldens and a headless shot of the result are not run here.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

selfhost/main.ludic

@@ -280,7 +280,11 @@ entry {
   # -Wno-override-module: the emitted IR carries no target triple, so clang
   # substitutes the host's and warns about it on every single build. There is
   # nothing to act on — the host default is exactly what we want — so it is off.
-  var cmd = `{cc} -O2 -Wno-override-module {ll}`
+  # $LUDIC_OPT=1 is a developer's faster build (never a bundle's: ludic bundle sets 2)
+  let opt = getenv_or("LUDIC_OPT", "2")
+  let objs = split_compile(ll, out, opt)           # on every core when it can (split_build.ludic)
+  var cmd = `{cc} -O{opt} -Wno-override-module {ll}`
+  if objs != null { cmd = `{cc} -O{opt} -Wno-override-module{objs}` }
   if g_target_win {
     # The Windows window: win32.ll (user32, gdi32 for the software present, xinput
     # for pads), and audio_win.ll when the program uses Audio.*, as on macOS.
@@ -389,6 +393,7 @@ entry {
   var line = cmd
   if host_is_windows() { line = `"{cmd}"` }
   let rc = run(line)
+  if objs != null { split_clean(out) }
   if not save { remove_file(ll) }
   if rc != 0 { die("ludicc: link failed\n") }
   if natives_out != null {

selfhost/split_build.ludic

@@ -0,0 +1,87 @@
+# split_build.ludic - the program's IR compiled on every core: llvm-split cuts it into N parts, one
+# clang each runs at once, and the link takes their objects where it took the one .ll. Most of a
+# build was one clang -O2 on one file (the game: 32 s of 41). Needs an llvm-split and a clang of
+# the same LLVM (a newer LLVM's parts do not read in an older clang): $LUDIC_LLVM (a bin directory),
+# else Homebrew's. Anything missing, or a part that does not compile, falls back to the one clang.
+
+function split_there(p: pointer) -> bool {
+  let f = file_open(p, "rb")
+  if f == null { return false }
+  file_close(f)
+  return true
+}
+
+# the LLVM bin directory to split with, or null (Windows' shell cannot run the parts at once yet)
+function split_dir() -> pointer {
+  if host_is_windows() { return null }
+  if getenv_or("LUDIC_SPLIT", "1") == "0" { return null }
+  let d = getenv_or("LUDIC_LLVM", "/opt/homebrew/opt/llvm/bin")
+  if not split_there(`{d}/llvm-split`) or not split_there(`{d}/clang`) { return null }
+  return d
+}
+
+# how many parts: $LUDIC_JOBS, else the cores, no more than the free memory allows at 1.5 GB a clang
+function split_jobs(tmp: pointer) -> int {
+  let j = getenv("LUDIC_JOBS")
+  if j != null { return split_int(j) }
+  let probe = `{tmp}.cores`
+  let aw = "awk '/Pages free/{f=$3} /Pages inactive/{i=$3} END{print int((f+i)*p/1073741824)}'"
+  run(`echo $(sysctl -n hw.ncpu) $(vm_stat | {aw} p=$(sysctl -n hw.pagesize)) > "{probe}" 2>/dev/null`)
+  let f = file_open(probe, "rb")
+  if f == null { return 1 }
+  let buf = bytes(64)
+  let n = file_read(f, buf, 63)
+  file_close(f)
+  remove_file(`"{probe}"`)
+  if n <= 0 { return 1 }
+  buf[n] = 0
+  var cores = 0
+  var i = 0
+  while i < n and buf[i] >= '0' and buf[i] <= '9' { cores = cores * 10 + (buf[i] - '0');  i += 1 }
+  while i < n and buf[i] == ' ' { i += 1 }
+  var gb = 0
+  while i < n and buf[i] >= '0' and buf[i] <= '9' { gb = gb * 10 + (buf[i] - '0');  i += 1 }
+  let by_mem = (gb * 2) / 3
+  if by_mem < cores { return by_mem }
+  return cores
+}
+function split_int(s: pointer) -> int {
+  var v = 0
+  var i = 0
+  while i < len(s) and s[i] >= '0' and s[i] <= '9' { v = v * 10 + (s[i] - '0');  i += 1 }
+  return v
+}
+
+# the IR at `ll` compiled in parts at -O`opt`: their objects, quoted and space-separated, for the
+# link to take in place of the .ll; null when it was not split (the caller compiles the .ll whole)
+function split_compile(ll: pointer, out: pointer, opt: pointer) -> pointer {
+  let d = split_dir()
+  if d == null { return null }
+  let n = split_jobs(out)
+  if n < 2 { return null }
+  let pre = `{out}.part`
+  if run(`"{d}/llvm-split" -j {n} -o "{pre}" "{ll}"`) != 0 { return null }
+  var sh = ""
+  var i = 0
+  while i < n {
+    sh = sh + `"{d}/clang" -x ir -O{opt} -mmacosx-version-min=11.0 -Wno-override-module -Wno-missing-sysroot -c "{pre}{itoa(i)}" -o "{pre}{itoa(i)}.o" & `
+    i += 1
+  }
+  run(`/bin/sh -c '{sh}wait'`)
+  var objs = ""
+  var ok = true
+  i = 0
+  while i < n {
+    remove_file(`"{pre}{itoa(i)}"`)
+    if not split_there(`{pre}{itoa(i)}.o`) { ok = false }
+    objs = objs + ` "{pre}{itoa(i)}.o"`
+    i += 1
+  }
+  if not ok {
+    split_clean(out)
+    return null
+  }
+  return objs
+}
+# the parts' objects, once linked (or when a part failed)
+function split_clean(out: pointer) -> void { run(`rm -f "{out}".part*.o`) }