c79ca8f · terrain-lookdev

Changed files (3) +119 -27

bench_diffusion.py +62 -20

…	…	@@ -27,7 +27,8 @@ PORT = 8819
27	27	MODEL = "xandergos/terrain-diffusion-30m"
28	28	VENV = "/storage/projects/code/godot/terrain-diffusion/.venv/bin/python"
29	29	CWD = "/storage/projects/code/godot/terrain-diffusion"
30		ORIGIN = 21_000_000 # far from anything this session has touched
	30	ORIGIN = 51_000_000
	31	CHECK_AT = 55_555_000 # one fixed region every config is validated on # far from anything this session has touched
31	32
32	33
33	34	def wait_up(proc, timeout=600):
…	…	@@ -43,6 +44,38 @@ def wait_up(proc, timeout=600):
43	44	return False
44	45
45	46
	47	REF = {} # reference tile from a config known to be correct
	48
	49
	50	def tile_data(i1, j1, n):
	51	"""Elevation + climate, for CORRECTNESS checking."""
	52	import numpy as np
	53	u = (f"http://127.0.0.1:{PORT}/terrain?i1={i1}&j1={j1}"
	54	f"&i2={i1+n}&j2={j1+n}&scale=1")
	55	with urllib.request.urlopen(u, timeout=3600) as r:
	56	h = int(r.headers["X-Height"]); w = int(r.headers["X-Width"]); b = r.read()
	57	e = np.frombuffer(b[:hw2], dtype="<i2").astype(float).reshape(h, w)
	58	c = np.frombuffer(b[hw2:], dtype="<f4").reshape(h, w, 4) if len(b) > hw2 else None
	59	return e, c
	60
	61
	62	def correctness(i1, j1, n=256):
	63	"""A config that returns ZEROS is infinitely fast. This is the gate that was
	64	missing from the first three runs, and its absence shipped a 14x result that
	65	was 70% blank tile."""
	66	import numpy as np
	67	e, c = tile_data(i1, j1, n)
	68	zeros = float((e == 0).mean())
	69	nans = int(np.isnan(c).sum()) if c is not None else 0
	70	corr = float("nan")
	71	if "e" in REF and REF["e"].shape == e.shape:
	72	a, b_ = REF["e"].ravel(), e.ravel()
	73	if a.std() > 0 and b_.std() > 0:
	74	corr = float(np.corrcoef(a, b_)[0, 1])
	75	ok = zeros < 0.02 and nans == 0 and (corr != corr or corr > 0.98)
	76	return dict(zeros=zeros, nans=nans, corr=corr, ok=ok)
	77
	78
46	79	def tile(i1, j1, n):
47	80	u = (f"http://127.0.0.1:{PORT}/terrain?i1={i1}&j1={j1}"
48	81	f"&i2={i1+n}&j2={j1+n}&scale=1")
…	…	@@ -103,11 +136,17 @@ def run(label, extra, n=512, reps=3, offset=0):
103	136	warm = tile(base, base, 256) # pay compile / lazy init here
104	137	ts = [tile(base + 20_000(k+1), base + 20_000(k+1), n) for k in range(reps)]
105	138	mem = gpu_mem_of(p.pid)
	139	# ALWAYS the same coordinates, so every config is checked on one region.
	140	ck = correctness(CHECK_AT, CHECK_AT)
	141	if not REF and ck["zeros"] < 0.02 and ck["nans"] == 0:
	142	REF["e"], _ = tile_data(CHECK_AT, CHECK_AT, 256)
106	143	best, avg = min(ts), sum(ts)/len(ts)
107	144	km2 = (n30/1000)*2
108		print(f" {label:<28} warm {warm:6.1f}s tile {avg:6.2f}s (best {best:5.2f})"
109		f" {km2/avg:6.1f} km2/s {mem:5d} MiB")
110		return dict(label=label, warm=warm, avg=avg, best=best, mem=mem)
	145	flag = "ok " if ck["ok"] else "BAD "
	146	print(f" {label:<26} {flag} tile {avg:6.2f}s {km2/avg:6.1f} km2/s "
	147	f"{mem:5d} MiB zeros {100*ck['zeros']:5.1f}% nan {ck['nans']:6d} "
	148	f"corr {ck['corr']:.4f}")
	149	return dict(label=label, avg=avg, best=best, mem=mem, **ck)
111	150	finally:
112	151	try: os.killpg(os.getpgid(p.pid), signal.SIGTERM)
113	152	except Exception: pass
…	…	@@ -119,28 +158,31 @@ BASELINE_PIDS = set(gpu_apps().keys()) # the always-on service, not ours to wa
119	158
120	159	if __name__ == "__main__":
121	160	print(f"512 tile = 15.36 km across = 236 km2. Fresh coordinates every time.\n")
122		print(f" {'config':<28} {'model load':>11} {'per tile':>22} {'throughput':>12} {'VRAM':>9}")
	161	print(f" {'config':<26} {'ok?':<4} {'per tile':>12} {'throughput':>11} "
	162	f"{'VRAM':>9} correctness")
123	163	rows = []
	164	os.environ.pop("TD_COMPILE_MODE", None)
124	165	F16 = ["--no-compile", "--dtype", "fp16", "--kwarg", "onestep_latent=True"]
125		D = lambda n: F16 + ["--kwarg", f"decoder_tile_size={n}"]
126		C = lambda n: ["--compile", "--dtype", "fp16", "--kwarg", "onestep_latent=True",
127		"--kwarg", f"decoder_tile_size={n}"]
128	166	for i, (label, extra) in enumerate([
129		("decoder_tile 256 (winner)", D(256)),
130		("decoder_tile 192", D(192)),
131		("decoder_tile 128", D(128)),
132		("decoder_tile 64", D(64)),
133		("256 + compile default", C(256)),
134		("256 + compile reduce-ovh", C(256)),
	167	("fp16+onestep (reference)", F16), # known good
	168	(" + decoder_tile 256", F16 + ["--kwarg", "decoder_tile_size=256"]),
	169	(" + decoder_tile 128", F16 + ["--kwarg", "decoder_tile_size=128"]),
	170	(" + compile only", ["--compile", "--dtype", "fp16",
	171	"--kwarg", "onestep_latent=True"]),
	172	(" + compile + dtile 256", ["--compile", "--dtype", "fp16",
	173	"--kwarg", "onestep_latent=True",
	174	"--kwarg", "decoder_tile_size=256"]),
	175	("fp32 plain (control)", ["--no-compile"]),
135	176	]):
136		if "reduce-ovh" in label:
137		os.environ["TD_COMPILE_MODE"] = "reduce-overhead"
138		else:
139		os.environ.pop("TD_COMPILE_MODE", None)
140	177	r = run(label, extra, offset=i)
141	178	if r: rows.append(r)
142	179	if rows:
143	180	b = rows[0]["avg"]
144		print("\n speedup vs baseline:")
	181	print("\n speedup vs the reference config — ONLY meaningful where ok:")
145	182	for r in rows:
146		print(f" {r['label']:<28} {b/r['avg']:5.2f}x (VRAM {r['mem']} MiB)")
	183	print(f" {r['label']:<26} {'ok ' if r['ok'] else 'BAD'} "
	184	f"{b/r['avg']:5.2f}x VRAM {r['mem']:5d} MiB")
	185	bad = [r["label"] for r in rows if not r["ok"]]
	186	if bad:
	187	print("\n DISCARD (wrong output, so the timing is meaningless): "
	188	+ ", ".join(x.strip() for x in bad))

index.html +41 -4

…	…	@@ -100,8 +100,11 @@ kbd{font-family:"IBM Plex Mono",monospace;background:#0e1216;border:1px solid va
100	100	<label>Diffusion i1 (px)<input type="number" id="i1" value="0" step="256"></label>
101	101	<label>Diffusion j1 (px)<input type="number" id="j1" value="2048" step="256"></label>
102	102	<label>Diffusion seed<input type="number" id="seed" value="1234" step="1">
103		<span class="d">a new seed is a whole new world. it rebuilds the pipeline and
104		clears every cached tile, so it is much slower than moving i1/j1.</span></label>
	103	<span class="d">a whole new world, and it flushes every cached tile — so it
	104	is applied only on Enter or the button below, never by scrolling or
	105	arrow keys. Active seed is shown next to it.</span></label>
	106	<div class="row"><button id="applyseed">Apply seed</button>
	107	<span class="v" id="activeseed" style="align-self:center">1234</span></div>
105	108	<div class="row">
106	109	<button id="newloc">New location</button>
107	110	<button id="newseed">New seed</button>
…	…	@@ -153,6 +156,10 @@ kbd{font-family:"IBM Plex Mono",monospace;background:#0e1216;border:1px solid va
153	156
154	157	<main>
155	158	<div class="bar">
	159	<label style="display:flex;gap:5px;align-items:center;font-size:11.5px">
	160	<input type="checkbox" id="lockramp" style="width:auto"> lock ramp
	161	<span id="ramptxt" style="font-family:'IBM Plex Mono',monospace;font-size:10px;color:var(--dim)"></span>
	162	</label>
156	163	<div class="views" id="modes">
157	164	<button data-m="2d" aria-pressed="true">2D</button>
158	165	<button data-m="3d" aria-pressed="false">3D</button>
…	…	@@ -335,6 +342,8 @@ function updateDerived() {
335	342	}
336	343	let view = "shaded-height";
337	344	let mode = "2d";
	345	let RAMP = null; // pinned colour ramp [lo, hi], null = fit per tile
	346	let ACTIVE_SEED = "1234"; // what the server is actually on; the field is a draft
338	347
339	348	for (const [grp, items] of Object.entries(GROUPS)) {
340	349	const host = $("grp_" + grp);
…	…	@@ -379,6 +388,12 @@ async function load(which, url, imgId, stId, busyId) {
379	388	if (old.startsWith("blob:")) URL.revokeObjectURL(old);
380	389	const h = r.headers;
381	390	const ms = (performance.now() - t0).toFixed(0);
	391	// Capture the ramp the first time it is locked, so both panels and every
	392	// subsequent pan use the SAME scale.
	393	if ($("lockramp").checked && !RAMP && h.get("X-Ramp")) {
	394	RAMP = h.get("X-Ramp").split(",").map(Number);
	395	$("ramptxt").textContent = `${RAMP[0].toFixed(0)} .. ${RAMP[1].toFixed(0)} m`;
	396	}
382	397	let s = `min ${h.get("X-Min")} max ${h.get("X-Max")} ` +
383	398	`mean ${h.get("X-Mean")} sd ${h.get("X-Sd")} m ${ms} ms`;
384	399	if (h.get("X-Climate")) s += `\n${h.get("X-Climate")}`;
…	…	@@ -546,12 +561,15 @@ function render() {
546	561	syncLabels();
547	562	const p = erosionQuery();
548	563	const q = new URLSearchParams({n: $("n").value, i1: $("i1").value,
549		j1: $("j1").value, seed: $("seed").value});
	564	j1: $("j1").value, seed: ACTIVE_SEED});
550	565	if ($("erode").checked) {
551	566	q.set("erode", "1");
552	567	for (const [key] of GROUPS.erode) q.set(key, $(key).value);
553	568	}
554	569
	570	if (RAMP && $("lockramp").checked) {
	571	for (const u of [p, q]) { u.set("vmin", RAMP[0]); u.set("vmax", RAMP[1]); }
	572	}
555	573	if (mode === "3d") {
556	574	load3d("/api/erosion.bin?" + p, "cv_e", "st_e", "busy_e");
557	575	load3d("/api/diffusion.bin?" + q, "cv_d", "st_d", "busy_d");
…	…	@@ -571,9 +589,28 @@ function schedule() {
571	589	}
572	590
573	591	$("erode").addEventListener("change", render);
	592	$("lockramp").addEventListener("change", () => {
	593	if (!$("lockramp").checked) { RAMP = null; $("ramptxt").textContent = ""; }
	594	render();
	595	});
574	596	document.addEventListener("input", e => {
	597	if (e.target.id === "seed") return; // deliberate only — see below
575	598	if (e.target.type === "range" \|\| e.target.type === "number") schedule();
576	599	});
	600	// Wheel over a number input changes its value in most browsers. For the seed
	601	// that means silently regenerating the world, so refuse the gesture outright.
	602	$("seed").addEventListener("wheel", e => { e.preventDefault(); $("seed").blur(); },
	603	{passive: false});
	604	$("seed").addEventListener("keydown", e => {
	605	if (e.key === "ArrowUp" \|\| e.key === "ArrowDown") e.preventDefault();
	606	if (e.key === "Enter") { e.preventDefault(); applySeed(); }
	607	});
	608	function applySeed() {
	609	ACTIVE_SEED = $("seed").value;
	610	$("activeseed").textContent = ACTIVE_SEED;
	611	render();
	612	}
	613	$("applyseed").onclick = applySeed;
577	614	$("render").onclick = render;
578	615	// Two different ways to get different terrain, and they cost very differently.
579	616	$("newloc").onclick = () => {
…	…	@@ -583,7 +620,7 @@ $("newloc").onclick = () => {
583	620	};
584	621	$("newseed").onclick = () => {
585	622	$("seed").value = Math.floor(Math.random() * 1e9);
586		render();
	623	applySeed();
587	624	};
588	625	$("reset").onclick = () => {
589	626	for (const items of Object.values(GROUPS))

server.py +16 -3

…	…	@@ -257,7 +257,18 @@ def to_png(rgb):
257	257
258	258
259	259	def colourise(h, mpp, view, vmin=None, vmax=None):
260		"""view: shade \| height \| slope \| shaded-height"""
	260	"""view: shade \| height \| slope \| shaded-height
	261
	262	vmin/vmax PIN the ramp. Without them the ramp is fitted to this tile, so
	263	panning one step re-normalises and the colours jump even though the terrain
	264	is identical -- which is exactly what terrain-diffusion issue #19 reported
	265	as "tile inconsistencies" before the author identified it as the colour map.
	266	An instrument that changes its scale between readings is not an instrument.
	267	"""
	268	if vmin in ("", None): vmin = None
	269	if vmax in ("", None): vmax = None
	270	vmin = float(vmin) if vmin is not None else None
	271	vmax = float(vmax) if vmax is not None else None
261	272	finite = h[np.isfinite(h)]
262	273	lo = float(vmin) if vmin is not None else (float(finite.min()) if finite.size else 0.0)
263	274	hi = float(vmax) if vmax is not None else (float(finite.max()) if finite.size else 1.0)
…	…	@@ -314,7 +325,8 @@ def api_erosion():
314	325	steps = int(_clamp(float(a.get("steps", 256)), 32, 1024))
315	326	knobs = parse_knobs(a)
316	327	h = render_erosion(ox, oz, n, mpp, knobs, steps, 40)
317		png, lo, hi = colourise(h, mpp, a.get("view", "shaded-height"))
	328	png, lo, hi = colourise(h, mpp, a.get("view", "shaded-height"),
	329	a.get("vmin"), a.get("vmax"))
318	330	return stats_headers(Response(png, mimetype="image/png"), h, lo, hi,
319	331	{"X-Source": "erosion_render"})
320	332	except Exception as e:
…	…	@@ -336,7 +348,8 @@ def api_diffusion():
336	348	h, _ridge = erode_field(h, mpp_d, a)
337	349	extra["X-Source"] = "terrain-diffusion + erosion_filter"
338	350	extra["X-Mpp"] = repr(mpp_d)
339		png, lo, hi = colourise(h, mpp_d, a.get("view", "shaded-height"))
	351	png, lo, hi = colourise(h, mpp_d, a.get("view", "shaded-height"),
	352	a.get("vmin"), a.get("vmax"))
340	353	if climate is not None:
341	354	extra["X-Climate"] = (f"{climate[...,0].mean():.1f}C,"
342	355	f"{climate[...,1].mean():.0f},"

terrain-lookdev commits / c79ca8f

Changed files (3) +119 -27