mirror of
https://github.com/boykopovar/AnyPS5.git
synced 2026-10-29 12:15:39 +00:00
Celegans12
2801497e0c
An APS5_AB run interleaves phases A and B, and every periodic report covers the time since the previous report of its kind, so its numbers need sorting before A and B can be compared. tools/ab_split.py reads the [ab] lines and counts a [present], [gputime], [draw] pipelined or [syncwait] report for a phase only when its whole span lies inside the phase after its settled line. It prints presents per 10 s and frame time per phase, GPU busy and idle time per frame, GPU time per present (also per --program key and --class), committer, drain and hook-wait times, the A and B means, and each B phase against the mean of its neighbouring A phases, which cancels a linear drift. AI-assisted: yes
137 lines
8.0 KiB
Python
137 lines
8.0 KiB
Python
import json
|
|
import os
|
|
import subprocess
|
|
import sys
|
|
import tempfile
|
|
import unittest
|
|
from pathlib import Path
|
|
|
|
sys.path.insert(0, str(Path(__file__).resolve().parents[1]))
|
|
import ab_split
|
|
|
|
|
|
def present(busy, gaps, presents=40, dropped=0):
|
|
return (f"[present] {presents} presents over 10 s ({dropped} dropped: no free swapchain image; 1 may trail on the GPU); "
|
|
f"inflight_wait p50 0.00 ms, p90 0.18 ms, max 0.37 ms over {presents}; per frame, presented or dropped: "
|
|
f"GPU busy {busy:.1f} ms, idle gaps {gaps:.1f} ms (0 batches without a completion record); per flip: 3.0 batches "
|
|
"recorded (0.0 unsignaled at the flip); per present: 1.0 batches ahead of the blit, 0.0 after it; after the flip "
|
|
"packet: last submit 0.1 ms, blit submit 0.2 ms; in-place reads overwritten before execution: 0 of 0 checked")
|
|
|
|
|
|
def gputime(total, program, staging, presents=40):
|
|
return (f"[gputime] {total:.0f} ms of GPU time in 90 batches over 10 s (batch 900 ms first to last command; classes 300 ms, "
|
|
f"all ranges 800 ms, untimed 100 ms outside every range; {presents} presents; 0 ranges dropped at the 512 cap); "
|
|
f"by program: 0x1000 x10 {program:.0f}ms 0x2000 x4 5ms; by class: staging-in x80 {staging:.1f}ms 2.0MiB "
|
|
"(per present x2.0 9.99ms 0.05MiB) present-blit x40 4.0ms 0.0MiB (per present x1.0 0.10ms 0.00MiB)")
|
|
|
|
|
|
def draw(per_commit, room, drain, commits=1000, enqueued=1000):
|
|
return (f"[draw] pipelined (10 s, depth 64): {commits} commits at {per_commit:.1f} us each on the committer, 0 failed; "
|
|
f"{enqueued} enqueued behind 2.00 on average, waiting {room:.1f} us per draw for room; 0 read ranges, 0 stores held; "
|
|
f"drains: packet 10 ({drain:.1f} ms) submission 2 (1.0 ms); packet drains by opcode: 0x49 x3 0 read ranges, 0 stores held")
|
|
|
|
|
|
def syncwait(first, second):
|
|
return ("[syncwait] unlocked hook waits by read site (10 s; count/GPU wait, avg batches up to the target, avg of them "
|
|
"unsignaled at the start, target was the newest submission, avg target submit-to-signal ms):"
|
|
f" copy 3/{first:.0f}ms 1.0 0.0 100% 2.0 fill 1/{second:.0f}ms 1.0 0.0 0% 1.0; batches ahead by queue: 0x0 4")
|
|
|
|
|
|
def reports(busy, total, program, staging, per_commit, wait):
|
|
return [present(busy, 50.0), gputime(total, program, staging), draw(per_commit, 2.0, 5.0), syncwait(wait, 10.0)]
|
|
|
|
|
|
def phase_lines(kind, index, begin, seconds, busy, total, program, staging, per_commit, wait, presents=40):
|
|
lines = [f"[ab] phase {kind} {index} begins at present {begin} (t {seconds:.3f} s): APS5_NO_BARRIER_BATCH {'set' if kind == 'A' else 'unset'}"]
|
|
lines += reports(999.0, 999.0, 999.0, 999.0, 999.0, 999.0)
|
|
lines.append(f"[ab] phase {kind} {index} settled at present {begin + 2} (t {seconds + 2:.3f} s)")
|
|
lines += reports(888.0, 888.0, 888.0, 888.0, 888.0, 888.0)
|
|
lines += reports(busy, total, program, staging, per_commit, wait)
|
|
lines += reports(busy, total, program, staging, per_commit, wait)
|
|
return lines
|
|
|
|
|
|
def log():
|
|
lines = ["[gpu] startup", "[ab] in-run A/B: phases of 30.0 s from present 100, settle 2.0 s; A: APS5_NO_BARRIER_BATCH set; B: APS5_NO_BARRIER_BATCH unset"]
|
|
lines += reports(777.0, 777.0, 777.0, 777.0, 777.0, 777.0)
|
|
lines += phase_lines("A", 0, 100, 10.0, 100.0, 400.0, 40.0, 8.0, 200.0, 30.0)
|
|
lines += phase_lines("B", 1, 140, 40.0, 130.0, 480.0, 60.0, 9.0, 180.0, 50.0)
|
|
lines += phase_lines("A", 2, 180, 70.0, 110.0, 420.0, 44.0, 8.0, 200.0, 34.0)
|
|
lines.append("[ab] phase B 3 begins at present 220 (t 100.000 s): APS5_NO_BARRIER_BATCH unset")
|
|
return [line + "\n" for line in lines]
|
|
|
|
|
|
class AbSplitTests(unittest.TestCase):
|
|
def test_phases_count_only_reports_inside_a_settled_phase(self):
|
|
result = ab_split.split(log(), ["0x1000"], ["staging-in"])
|
|
self.assertTrue(result["config"].startswith("[ab] in-run A/B: phases of 30.0 s"))
|
|
self.assertEqual([(phase["kind"], phase["index"], phase["begin_present"], phase["settled_present"], phase["end_present"])
|
|
for phase in result["phases"]],
|
|
[("A", 0, 100, 102, 140), ("B", 1, 140, 142, 180), ("A", 2, 180, 182, 220), ("B", 3, 220, None, None)])
|
|
self.assertEqual(result["phases"][0]["clean_reports"], {"present": 2, "gputime": 2, "draw": 2, "syncwait": 2})
|
|
self.assertEqual(result["phases"][3]["clean_reports"], {"present": 0, "gputime": 0, "draw": 0, "syncwait": 0})
|
|
first = result["phases"][0]["metrics"]
|
|
self.assertAlmostEqual(first["presents/10s"], 38 / 28 * 10)
|
|
self.assertAlmostEqual(first["frame ms"], 28 / 38 * 1000)
|
|
self.assertAlmostEqual(first["GPU busy ms/frame"], 100.0)
|
|
self.assertAlmostEqual(first["GPU idle gaps ms/frame"], 50.0)
|
|
self.assertAlmostEqual(first["gputime ms/present"], 10.0)
|
|
self.assertAlmostEqual(first["0x1000 ms/present"], 1.0)
|
|
self.assertAlmostEqual(first["staging-in ms/present"], 0.2)
|
|
self.assertAlmostEqual(first["committer us/commit"], 200.0)
|
|
self.assertAlmostEqual(first["room wait us/draw"], 2.0)
|
|
self.assertAlmostEqual(first["drain ms/report"], 6.0)
|
|
self.assertAlmostEqual(first["syncwait GPU ms/report"], 40.0)
|
|
|
|
def test_b_is_compared_with_its_neighbouring_a_phases(self):
|
|
rows = {row["metric"]: row for row in ab_split.split(log())["metrics"]}
|
|
busy = rows["GPU busy ms/frame"]
|
|
self.assertAlmostEqual(busy["a"], 105.0)
|
|
self.assertAlmostEqual(busy["b"], 130.0)
|
|
self.assertAlmostEqual(busy["change"], 25.0)
|
|
self.assertAlmostEqual(busy["change_pct"], 25 / 105 * 100)
|
|
self.assertEqual(busy["noise"], {"mean": 25.0, "se": None, "sd": None, "n": 1})
|
|
self.assertAlmostEqual(rows["syncwait GPU ms/report"]["noise"]["mean"], 60.0 - 42.0)
|
|
|
|
def test_classes_are_read_without_their_per_present_figures(self):
|
|
row = ab_split.parse_report("gputime", gputime(400.0, 40.0, 8.0))
|
|
self.assertEqual(row["classes"], {"staging-in": 8.0, "present-blit": 4.0})
|
|
self.assertEqual(row["programs"], {"0x1000": 40.0, "0x2000": 5.0})
|
|
|
|
def test_dropped_presents_count_as_frames(self):
|
|
row = ab_split.parse_report("present", present(90.0, 10.0, presents=30, dropped=10))
|
|
self.assertEqual(row, {"frames": 40, "busy": 90.0, "gaps": 10.0})
|
|
|
|
def test_unrecognised_reports_are_skipped(self):
|
|
self.assertIsNone(ab_split.parse_report("gputime", "[gputime] disabled"))
|
|
self.assertIsNone(ab_split.parse_report("draw", "[draw] pipelined depth 64"))
|
|
self.assertIsNone(ab_split.parse_report("present", "[present] flip 3"))
|
|
|
|
def test_a_log_without_phases_is_refused(self):
|
|
with self.assertRaisesRegex(ValueError, "no \\[ab\\] phase lines"):
|
|
ab_split.split([present(1.0, 1.0) + "\n"])
|
|
|
|
def test_command_line(self):
|
|
with tempfile.TemporaryDirectory() as directory:
|
|
path = Path(directory) / "run.err.log"
|
|
path.write_text("".join(log()), encoding="utf-8")
|
|
run = lambda *args: subprocess.run([sys.executable, str(Path(ab_split.__file__)), *args], capture_output=True, text=True,
|
|
encoding="utf-8", timeout=10, env=dict(os.environ, PYTHONIOENCODING="ascii"))
|
|
text = run(str(path), "--program", "0x01000")
|
|
self.assertEqual(text.returncode, 0, text.stderr)
|
|
self.assertIn("0x1000 ms/present", text.stdout)
|
|
self.assertRegex(text.stdout, r"GPU busy ms/frame +105\.00 +130\.00 +\+25\.00 +\+23\.8% +\+25\.00 \(n 1\)")
|
|
self.assertIn("unfinished", text.stdout)
|
|
data = json.loads(run(str(path), "--json").stdout)
|
|
self.assertEqual(len(data["phases"]), 4)
|
|
self.assertEqual(run(str(path), "--program", "1000").returncode, 2)
|
|
empty = Path(directory) / "empty.log"
|
|
empty.write_text("[gpu] nothing\n", encoding="utf-8")
|
|
failed = run(str(empty))
|
|
self.assertEqual(failed.returncode, 2)
|
|
self.assertIn("FAIL: no [ab] phase lines", failed.stderr)
|
|
|
|
|
|
if __name__ == "__main__":
|
|
unittest.main()
|