Benchmark suite measuring entity performance at scale: - B1: Entity creation (measures allocation overhead) - B2: Full iteration (measures cache locality) - B3: Single range query (measures O(n) scan cost) - B4: N×N visibility (the "what can everyone see" problem) - B5: Movement churn (baseline for spatial index overhead) Key findings at 2,000 entities on 100×100 grid: - Creation: 75k entities/sec - Range query: 0.05ms (O(n) - checks all entities) - N×N visibility: 128ms total (O(n²)) - EntityCollection iteration 60× slower than direct iteration Addresses #115, addresses #117 🤖 Generated with [Claude Code](https://claude.ai/code) Co-Authored-By: Claude <noreply@anthropic.com>
This commit is contained in:
parent
3e07334aa5
commit
fcc0376f31
1 changed files with 387 additions and 0 deletions
387
tests/benchmarks/entity_scale_benchmark.py
Normal file
387
tests/benchmarks/entity_scale_benchmark.py
Normal file
|
|
@ -0,0 +1,387 @@
|
|||
#!/usr/bin/env python3
|
||||
"""Entity Performance Benchmark for #115 (SpatialHash) and #117 (Memory Pool)
|
||||
|
||||
Target: 10,000 entities on 1000×1000 grid
|
||||
Goal: Establish baseline metrics for spatial query and memory performance
|
||||
|
||||
Scenarios:
|
||||
B1: Entity creation stress (measures allocation overhead)
|
||||
B2: Full iteration (measures cache locality)
|
||||
B3: Single range query (measures current O(n) cost)
|
||||
B4: N-to-N visibility (the "what can everyone see" problem)
|
||||
B5: Movement churn (measures update cost for spatial index)
|
||||
|
||||
Usage:
|
||||
cd build && ./mcrogueface --headless --exec ../tests/benchmarks/entity_scale_benchmark.py
|
||||
|
||||
Expected output: Baseline timings and projected gains from #115/#117 implementations.
|
||||
"""
|
||||
|
||||
import mcrfpy
|
||||
import sys
|
||||
import time
|
||||
import random
|
||||
|
||||
# Configuration
|
||||
# Use smaller grid for denser entity distribution (more realistic visibility tests)
|
||||
GRID_SIZE = (100, 100) # 10,000 cells - entities will actually see each other
|
||||
|
||||
# Full suite - may timeout on large counts due to O(n²) visibility
|
||||
# ENTITY_COUNTS = [100, 500, 1000, 2500, 5000, 10000]
|
||||
|
||||
# Quick suite for initial baseline (on 100x100 grid, these give densities of 1-20%)
|
||||
ENTITY_COUNTS = [100, 500, 1000, 2000]
|
||||
QUERY_RADIUS = 15 # Smaller radius for smaller grid
|
||||
MOVEMENT_PERCENT = 0.10 # 10% of entities move each frame
|
||||
N2N_SAMPLE_SIZE = 50 # Sample size for N×N visibility test
|
||||
|
||||
results = {}
|
||||
texture = None
|
||||
|
||||
|
||||
def setup_grid_with_entities(n_entities):
|
||||
"""Create a grid and populate with n entities at random positions."""
|
||||
global texture
|
||||
|
||||
scene_name = f"bench_{n_entities}"
|
||||
mcrfpy.createScene(scene_name)
|
||||
ui = mcrfpy.sceneUI(scene_name)
|
||||
|
||||
# Create grid - minimal rendering size since we're testing entity operations
|
||||
grid = mcrfpy.Grid(grid_size=GRID_SIZE, pos=(0, 0), size=(100, 100))
|
||||
ui.append(grid)
|
||||
|
||||
# Load texture once
|
||||
if texture is None:
|
||||
try:
|
||||
texture = mcrfpy.Texture("assets/kenney_tinydungeon.png", 16, 16)
|
||||
except Exception as e:
|
||||
print(f"ERROR: Could not load texture: {e}")
|
||||
return None, None
|
||||
|
||||
# Create entities
|
||||
for i in range(n_entities):
|
||||
x = random.randint(0, GRID_SIZE[0] - 1)
|
||||
y = random.randint(0, GRID_SIZE[1] - 1)
|
||||
entity = mcrfpy.Entity((x, y), texture, 0, grid)
|
||||
grid.entities.append(entity)
|
||||
|
||||
mcrfpy.setScene(scene_name)
|
||||
return grid, scene_name
|
||||
|
||||
|
||||
def benchmark_creation(n_entities):
|
||||
"""B1: Measure time to create N entities from scratch."""
|
||||
global texture
|
||||
|
||||
scene_name = "bench_create_test"
|
||||
mcrfpy.createScene(scene_name)
|
||||
ui = mcrfpy.sceneUI(scene_name)
|
||||
|
||||
grid = mcrfpy.Grid(grid_size=GRID_SIZE, pos=(0, 0), size=(100, 100))
|
||||
ui.append(grid)
|
||||
|
||||
if texture is None:
|
||||
try:
|
||||
texture = mcrfpy.Texture("assets/kenney_tinydungeon.png", 16, 16)
|
||||
except Exception as e:
|
||||
print(f"ERROR: Could not load texture: {e}")
|
||||
return None
|
||||
|
||||
# Time just the entity creation loop
|
||||
start = time.perf_counter()
|
||||
|
||||
for i in range(n_entities):
|
||||
x = random.randint(0, GRID_SIZE[0] - 1)
|
||||
y = random.randint(0, GRID_SIZE[1] - 1)
|
||||
entity = mcrfpy.Entity((x, y), texture, 0, grid)
|
||||
grid.entities.append(entity)
|
||||
|
||||
elapsed = time.perf_counter() - start
|
||||
return elapsed * 1000 # ms
|
||||
|
||||
|
||||
def benchmark_iteration(grid):
|
||||
"""B2: Measure time to iterate all entities and read position."""
|
||||
entities_list = list(grid.entities) # Convert to list first
|
||||
n = len(entities_list)
|
||||
|
||||
start = time.perf_counter()
|
||||
|
||||
total_x = 0.0
|
||||
for entity in entities_list:
|
||||
total_x += entity.x # Force read of position
|
||||
|
||||
elapsed = time.perf_counter() - start
|
||||
return elapsed * 1000, n # ms, count
|
||||
|
||||
|
||||
def benchmark_iteration_via_collection(grid):
|
||||
"""B2b: Measure iteration directly through EntityCollection."""
|
||||
start = time.perf_counter()
|
||||
|
||||
total_x = 0.0
|
||||
for entity in grid.entities:
|
||||
total_x += entity.x
|
||||
|
||||
elapsed = time.perf_counter() - start
|
||||
return elapsed * 1000 # ms
|
||||
|
||||
|
||||
def benchmark_range_query(entity, radius):
|
||||
"""B3: Measure single visible_entities call (current O(n) implementation)."""
|
||||
start = time.perf_counter()
|
||||
|
||||
visible = entity.visible_entities(radius=radius)
|
||||
|
||||
elapsed = time.perf_counter() - start
|
||||
return elapsed * 1000, len(visible) # ms, count
|
||||
|
||||
|
||||
def benchmark_n_to_n_visibility(grid, radius, sample_size):
|
||||
"""B4: Measure visibility queries for a sample of entities.
|
||||
|
||||
This simulates "what can every entity see" which is O(N²) currently.
|
||||
We sample to avoid timeouts on large entity counts.
|
||||
"""
|
||||
entities_list = list(grid.entities)
|
||||
n = len(entities_list)
|
||||
actual_sample = min(sample_size, n)
|
||||
sample = random.sample(entities_list, actual_sample)
|
||||
|
||||
start = time.perf_counter()
|
||||
|
||||
total_visible = 0
|
||||
for entity in sample:
|
||||
visible = entity.visible_entities(radius=radius)
|
||||
total_visible += len(visible)
|
||||
|
||||
elapsed = time.perf_counter() - start
|
||||
avg_visible = total_visible / actual_sample if actual_sample > 0 else 0
|
||||
|
||||
return elapsed * 1000, actual_sample, avg_visible # ms, sample_size, avg_found
|
||||
|
||||
|
||||
def benchmark_movement(grid, move_percent):
|
||||
"""B5: Move a percentage of entities to random positions.
|
||||
|
||||
Currently this is just position assignment (O(1) per entity).
|
||||
With SpatialHash, this would include hash bucket updates.
|
||||
"""
|
||||
entities_list = list(grid.entities)
|
||||
n = len(entities_list)
|
||||
to_move_count = int(n * move_percent)
|
||||
to_move = random.sample(entities_list, to_move_count)
|
||||
|
||||
start = time.perf_counter()
|
||||
|
||||
for entity in to_move:
|
||||
entity.x = random.randint(0, GRID_SIZE[0] - 1)
|
||||
entity.y = random.randint(0, GRID_SIZE[1] - 1)
|
||||
|
||||
elapsed = time.perf_counter() - start
|
||||
return elapsed * 1000, to_move_count # ms, count
|
||||
|
||||
|
||||
def run_single_scale(n_entities):
|
||||
"""Run all benchmarks for a single entity count."""
|
||||
print(f"\n{'='*60}")
|
||||
print(f" {n_entities:,} ENTITIES on {GRID_SIZE[0]}x{GRID_SIZE[1]} grid")
|
||||
print(f"{'='*60}")
|
||||
|
||||
# B1: Creation benchmark
|
||||
print("\n[B1] Entity Creation...")
|
||||
create_ms = benchmark_creation(n_entities)
|
||||
if create_ms is None:
|
||||
print(" FAILED: Could not load texture")
|
||||
return None
|
||||
|
||||
entities_per_sec = n_entities / (create_ms / 1000) if create_ms > 0 else 0
|
||||
print(f" Time: {create_ms:.2f}ms")
|
||||
print(f" Rate: {entities_per_sec:,.0f} entities/sec")
|
||||
|
||||
# Setup grid for remaining tests
|
||||
grid, scene_name = setup_grid_with_entities(n_entities)
|
||||
if grid is None:
|
||||
return None
|
||||
|
||||
# B2: Iteration benchmark
|
||||
print("\n[B2] Full Iteration (read all positions)...")
|
||||
iter_ms, count = benchmark_iteration(grid)
|
||||
iter_per_sec = count / (iter_ms / 1000) if iter_ms > 0 else 0
|
||||
print(f" Time: {iter_ms:.3f}ms")
|
||||
print(f" Rate: {iter_per_sec:,.0f} reads/sec")
|
||||
|
||||
# B2b: Iteration via collection
|
||||
iter_coll_ms = benchmark_iteration_via_collection(grid)
|
||||
print(f" Via EntityCollection: {iter_coll_ms:.3f}ms")
|
||||
|
||||
# B3: Single range query
|
||||
print(f"\n[B3] Single Range Query (radius={QUERY_RADIUS})...")
|
||||
entities_list = list(grid.entities)
|
||||
# Pick entity near center of grid for representative query
|
||||
center_entities = [e for e in entities_list
|
||||
if 400 < e.x < 600 and 400 < e.y < 600]
|
||||
if center_entities:
|
||||
test_entity = center_entities[0]
|
||||
else:
|
||||
test_entity = entities_list[len(entities_list)//2]
|
||||
|
||||
query_ms, found = benchmark_range_query(test_entity, QUERY_RADIUS)
|
||||
print(f" Time: {query_ms:.3f}ms")
|
||||
print(f" Found: {found} entities in range")
|
||||
print(f" Checked: {n_entities} entities (O(n) scan)")
|
||||
|
||||
# B4: N-to-N visibility
|
||||
print(f"\n[B4] N×N Visibility (sample={N2N_SAMPLE_SIZE})...")
|
||||
n2n_ms, sample_size, avg_visible = benchmark_n_to_n_visibility(
|
||||
grid, QUERY_RADIUS, N2N_SAMPLE_SIZE
|
||||
)
|
||||
per_query_ms = n2n_ms / sample_size if sample_size > 0 else 0
|
||||
print(f" Sample time: {n2n_ms:.2f}ms ({sample_size} queries)")
|
||||
print(f" Per query: {per_query_ms:.3f}ms")
|
||||
print(f" Avg visible: {avg_visible:.1f} entities")
|
||||
|
||||
# Extrapolate to full N×N
|
||||
full_n2n_ms = per_query_ms * n_entities
|
||||
print(f" Estimated full N×N: {full_n2n_ms:,.0f}ms ({full_n2n_ms/1000:.1f}s)")
|
||||
|
||||
# B5: Movement
|
||||
print(f"\n[B5] Movement ({MOVEMENT_PERCENT*100:.0f}% of entities)...")
|
||||
move_ms, moved = benchmark_movement(grid, MOVEMENT_PERCENT)
|
||||
move_per_sec = moved / (move_ms / 1000) if move_ms > 0 else 0
|
||||
print(f" Time: {move_ms:.3f}ms ({moved} entities)")
|
||||
print(f" Rate: {move_per_sec:,.0f} moves/sec")
|
||||
|
||||
return {
|
||||
'n': n_entities,
|
||||
'create_ms': create_ms,
|
||||
'create_rate': entities_per_sec,
|
||||
'iter_ms': iter_ms,
|
||||
'iter_coll_ms': iter_coll_ms,
|
||||
'query_ms': query_ms,
|
||||
'query_found': found,
|
||||
'n2n_sample_ms': n2n_ms,
|
||||
'n2n_per_query_ms': per_query_ms,
|
||||
'n2n_avg_visible': avg_visible,
|
||||
'n2n_full_estimate_ms': full_n2n_ms,
|
||||
'move_ms': move_ms,
|
||||
'move_count': moved,
|
||||
}
|
||||
|
||||
|
||||
def print_summary_table():
|
||||
"""Print a summary table of all results."""
|
||||
print("\n" + "=" * 80)
|
||||
print("SUMMARY TABLE")
|
||||
print("=" * 80)
|
||||
|
||||
header = f"{'Entities':>10} {'Create':>10} {'Iterate':>10} {'Query':>10} {'N×N Est':>12} {'Move':>10}"
|
||||
print(header)
|
||||
print(f"{'':>10} {'(ms)':>10} {'(ms)':>10} {'(ms)':>10} {'(ms)':>12} {'(ms)':>10}")
|
||||
print("-" * 80)
|
||||
|
||||
for n in sorted(results.keys()):
|
||||
r = results[n]
|
||||
print(f"{r['n']:>10,} {r['create_ms']:>10.1f} {r['iter_ms']:>10.2f} "
|
||||
f"{r['query_ms']:>10.2f} {r['n2n_full_estimate_ms']:>12,.0f} {r['move_ms']:>10.2f}")
|
||||
|
||||
|
||||
def print_analysis():
|
||||
"""Print performance analysis and projected gains."""
|
||||
print("\n" + "=" * 80)
|
||||
print("ANALYSIS: Projected Gains from #115 (SpatialHash) and #117 (Memory Pool)")
|
||||
print("=" * 80)
|
||||
|
||||
# Find largest test run
|
||||
max_n = max(results.keys())
|
||||
r = results[max_n]
|
||||
|
||||
print(f"\nBaseline at {max_n:,} entities:")
|
||||
print("-" * 40)
|
||||
|
||||
# Creation analysis (#117)
|
||||
print(f"\n[#117 Memory Pool] Entity Creation:")
|
||||
print(f" Current: {r['create_ms']:.1f}ms ({r['create_rate']:,.0f}/sec)")
|
||||
projected_create = r['create_ms'] / 25
|
||||
print(f" Projected: ~{projected_create:.1f}ms (25× faster)")
|
||||
print(f" Rationale: Pool allocation eliminates malloc overhead per entity")
|
||||
|
||||
# Iteration analysis (#117)
|
||||
print(f"\n[#117 Memory Pool] Iteration:")
|
||||
print(f" Current: {r['iter_ms']:.2f}ms")
|
||||
projected_iter = r['iter_ms'] / 5
|
||||
print(f" Projected: ~{projected_iter:.2f}ms (5× faster)")
|
||||
print(f" Rationale: Contiguous memory improves CPU cache utilization")
|
||||
|
||||
# Range query analysis (#115)
|
||||
print(f"\n[#115 SpatialHash] Range Query:")
|
||||
print(f" Current: {r['query_ms']:.2f}ms (checks {max_n:,} entities)")
|
||||
print(f" Found: {r['query_found']} entities in radius {QUERY_RADIUS}")
|
||||
|
||||
# Calculate expected speedup
|
||||
# With spatial hash, we only check entities in nearby buckets
|
||||
# Bucket size typically 32-64 cells, so for radius 25 we check ~4-16 buckets
|
||||
# Each bucket has ~10 entities at 0.01 density
|
||||
expected_checks = max(r['query_found'] * 5, 100) # Conservative estimate
|
||||
speedup = max_n / expected_checks
|
||||
projected_query = r['query_ms'] / speedup
|
||||
print(f" Projected: ~{projected_query:.3f}ms ({speedup:.0f}× faster)")
|
||||
print(f" Rationale: Only check ~{expected_checks} entities in nearby hash buckets")
|
||||
|
||||
# N×N analysis (#115)
|
||||
print(f"\n[#115 SpatialHash] N×N Visibility:")
|
||||
print(f" Current: {r['n2n_full_estimate_ms']:,.0f}ms ({r['n2n_full_estimate_ms']/1000:.1f}s)")
|
||||
projected_n2n = r['n2n_full_estimate_ms'] / speedup
|
||||
print(f" Projected: ~{projected_n2n:,.0f}ms ({projected_n2n/1000:.2f}s)")
|
||||
print(f" Rationale: Each of N queries benefits from {speedup:.0f}× speedup")
|
||||
|
||||
# Combined benefit
|
||||
print(f"\n[Combined] Per-Frame Budget Analysis:")
|
||||
print(f" Target: 16.67ms (60 FPS)")
|
||||
current_frame = r['iter_ms'] + r['move_ms'] + (r['n2n_per_query_ms'] * 100) # 100 AI entities
|
||||
print(f" Current (100 AI queries): ~{current_frame:.1f}ms")
|
||||
projected_frame = projected_iter + r['move_ms'] + (projected_query * 100)
|
||||
print(f" Projected: ~{projected_frame:.1f}ms")
|
||||
if current_frame > 16.67 and projected_frame < 16.67:
|
||||
print(f" Result: ENABLES 60 FPS with 100 AI entities")
|
||||
|
||||
# Movement overhead warning
|
||||
print(f"\n[#115 SpatialHash] Movement Overhead:")
|
||||
print(f" Current: {r['move_ms']:.2f}ms (no index to update)")
|
||||
print(f" Projected: ~{r['move_ms'] * 1.5:.2f}ms (+50% for hash updates)")
|
||||
print(f" Note: This overhead is acceptable given query speedups")
|
||||
|
||||
|
||||
def run_benchmarks(runtime=None):
|
||||
"""Main benchmark runner."""
|
||||
global results
|
||||
|
||||
print("=" * 80)
|
||||
print("Entity Scale Benchmark Suite")
|
||||
print(f"Grid: {GRID_SIZE[0]}×{GRID_SIZE[1]}, Query Radius: {QUERY_RADIUS}")
|
||||
print(f"Testing entity counts: {ENTITY_COUNTS}")
|
||||
print("=" * 80)
|
||||
|
||||
for n in ENTITY_COUNTS:
|
||||
result = run_single_scale(n)
|
||||
if result:
|
||||
results[n] = result
|
||||
|
||||
if results:
|
||||
print_summary_table()
|
||||
print_analysis()
|
||||
|
||||
print("\nBenchmark complete.")
|
||||
sys.exit(0)
|
||||
|
||||
|
||||
# Entry point
|
||||
if __name__ == "__main__":
|
||||
# For headless mode, run immediately
|
||||
# For interactive mode, use timer to let render loop start
|
||||
import sys
|
||||
if "--headless" in sys.argv or True: # Always run immediately for benchmarks
|
||||
run_benchmarks()
|
||||
else:
|
||||
mcrfpy.setTimer("run_bench", run_benchmarks, 100)
|
||||
Loading…
Add table
Add a link
Reference in a new issue