cp-library
This documentation is automatically generated by online-judge-tools/verification-helper
perf/rank.py
- View this file on GitHub
- Last update: 2025-07-11 23:11:42+09:00
Depends on
cp_library/alg/dp/max2_fn.py
- cp_library/alg/iter/rank/irank_fn.py
- cp_library/alg/iter/rank/irank_multi_fn.py
- cp_library/bit/pack/packer3_cls.py
- cp_library/bit/pack/packer_cls.py
- cp_library/perf/benchmark.py
Code
#!/usr/bin/env python3
"""
Simple ranking benchmark using the new declarative framework.
Compares irank vs irank_multi performance across different data patterns.
"""
import random
import sys
import os
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
from cp_library.perf.benchmark import Benchmark, BenchmarkConfig
from cp_library.alg.iter.rank.irank_fn import irank
from cp_library.alg.iter.rank.irank_multi_fn import irank as irank_multi
# Configure benchmark
config = BenchmarkConfig(
name="rank",
sizes=[1000000, 100000, 10000, 1000, 100, 10, 1], # Reverse order to warm up JIT
operations=['construction', 'random', 'sorted', 'duplicates', 'reverse'],
iterations=5,
warmup=3,
output_dir="./output/benchmark_results/rank"
)
# Create benchmark instance
benchmark = Benchmark(config)
# Data generator
@benchmark.data_generator("default")
def generate_rank_data(size: int, operation: str):
"""Generate ranking data in different patterns"""
if operation == 'random':
data = [random.randint(1, size) for _ in range(size)]
elif operation == 'sorted':
data = list(range(size))
elif operation == 'duplicates':
# Many duplicates (10% unique values)
unique_count = max(1, size // 10)
data = [random.randint(1, unique_count) for _ in range(size)]
elif operation == 'reverse':
data = list(range(size, 0, -1))
else:
raise ValueError(f"Unknown operation: {operation}")
# Pre-initialize data for fair timing (exclude copy overhead)
preinitialized = {
'data_copy1': list(data),
'data_copy2': list(data),
'distinct': False
}
return {
'data': data,
'distinct': False,
'size': size,
'operation': operation,
'preinitialized': preinitialized
}
# Construction operation
@benchmark.implementation("irank", "construction")
def construction_irank(data):
"""Construct data copy for irank"""
data_copy = list(data['data'])
checksum = 0
for x in data_copy:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "construction")
def construction_irank_multi(data):
"""Construct data copy for irank_multi"""
data_copy = list(data['data'])
checksum = 0
for x in data_copy:
checksum ^= x
return checksum
# Random operation
@benchmark.implementation("irank", "random")
def random_irank(data):
"""Standard irank implementation for random data"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "random")
def random_irank_multi(data):
"""Multi-pass irank implementation for random data"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Sorted operation
@benchmark.implementation("irank", "sorted")
def sorted_irank(data):
"""Standard irank implementation for sorted data"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "sorted")
def sorted_irank_multi(data):
"""Multi-pass irank implementation for sorted data"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Duplicates operation
@benchmark.implementation("irank", "duplicates")
def duplicates_irank(data):
"""Standard irank implementation for data with duplicates"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "duplicates")
def duplicates_irank_multi(data):
"""Multi-pass irank implementation for data with duplicates"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Reverse operation
@benchmark.implementation("irank", "reverse")
def reverse_irank(data):
"""Standard irank implementation for reverse data"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "reverse")
def reverse_irank_multi(data):
"""Multi-pass irank implementation for reverse data"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Additional benchmark with distinct=True
@benchmark.data_generator("distinct")
def generate_rank_data_distinct(size: int, operation: str):
"""Generate ranking data with distinct=True"""
base_data = generate_rank_data(size, operation)
base_data['distinct'] = True
base_data['preinitialized']['distinct'] = True
return base_data
def irank_distinct_implementation(data):
"""irank with distinct=True"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=True)
checksum = 0
for x in result:
checksum ^= x
return checksum
def irank_multi_distinct_implementation(data):
"""irank_multi with distinct=True"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=True)
checksum = 0
for x in result:
checksum ^= x
return checksum
# Custom validator for rank results (now using XOR checksums)
@benchmark.validator("default")
def validate_rank_result(expected, actual):
"""Validate ranking results using XOR checksums"""
try:
# Compare XOR checksums directly
return int(expected) == int(actual)
except Exception:
return False
if __name__ == "__main__":
# Parse command line args and run appropriate mode
runner = benchmark.parse_args()
runner.run()#!/usr/bin/env python3
"""
Simple ranking benchmark using the new declarative framework.
Compares irank vs irank_multi performance across different data patterns.
"""
import random
import sys
import os
sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
"""
Declarative benchmark framework with minimal boilerplate.
Features:
- Decorator-based benchmark registration
- Automatic data generation and validation
- Built-in timing with warmup
- Configurable operations and sizes
- JSON results and matplotlib plotting
"""
import time
import json
import statistics
import argparse
from typing import Dict, List, Any, Callable, Union
from dataclasses import dataclass
from pathlib import Path
from collections import defaultdict
@dataclass
class BenchmarkConfig:
"""Configuration for benchmark runs"""
name: str
sizes: List[int] = None
operations: List[str] = None
iterations: int = 10
warmup: int = 2
output_dir: str = "./output/benchmark_results"
save_results: bool = True
plot_results: bool = True
plot_scale: str = "loglog" # Options: "loglog", "linear", "semilogx", "semilogy"
progressive: bool = True # Show results operation by operation across sizes
# Profiling mode
profile_mode: bool = False
profile_size: int = None
profile_operation: str = None
profile_implementation: str = None
def __post_init__(self):
if self.sizes is None:
self.sizes = [100, 1000, 10000, 100000]
if self.operations is None:
self.operations = ['default']
class Benchmark:
"""Declarative benchmark framework using decorators"""
def __init__(self, config: BenchmarkConfig):
self.config = config
self.data_generators = {}
self.implementations = {}
self.validators = {}
self.setups = {}
self.results = []
def profile(self, operation: str = None, size: int = None, implementation: str = None):
"""Create a profiling version of this benchmark"""
profile_config = BenchmarkConfig(
name=f"{self.config.name}_profile",
sizes=self.config.sizes,
operations=self.config.operations,
profile_mode=True,
profile_operation=operation,
profile_size=size,
profile_implementation=implementation,
save_results=False,
plot_results=False
)
profile_benchmark = Benchmark(profile_config)
profile_benchmark.data_generators = self.data_generators
profile_benchmark.implementations = self.implementations
profile_benchmark.validators = self.validators
profile_benchmark.setups = self.setups
return profile_benchmark
def parse_args(self):
"""Parse command line arguments for profiling mode"""
parser = argparse.ArgumentParser(
description=f"Benchmark {self.config.name} with optional profiling mode",
formatter_class=argparse.RawDescriptionHelpFormatter,
epilog="""
Examples:
# Normal benchmark mode
python benchmark.py
# Profile specific operation and implementation
python benchmark.py --profile --operation random_access --implementation grid
# Profile with specific size
python benchmark.py --profile --size 1000000
# Profile all implementations of an operation
python benchmark.py --profile --operation construction
"""
)
parser.add_argument('--profile', action='store_true',
help='Run in profiling mode (minimal overhead for profilers)')
parser.add_argument('--operation', type=str,
help=f'Operation to profile. Options: {", ".join(self.config.operations)}')
parser.add_argument('--size', type=int,
help=f'Size to profile. Options: {", ".join(map(str, self.config.sizes))}')
parser.add_argument('--implementation', type=str,
help='Specific implementation to profile (default: all)')
args = parser.parse_args()
# If profile mode requested, return a profiling benchmark
if args.profile:
return self.profile(
operation=args.operation,
size=args.size,
implementation=args.implementation
)
# Otherwise return self for normal mode
return self
def data_generator(self, name: str = "default"):
"""Decorator to register data generator"""
def decorator(func):
self.data_generators[name] = func
return func
return decorator
def implementation(self, name: str, operations: Union[str, List[str]] = None):
"""Decorator to register implementation"""
if operations is None:
operations = ['default']
elif isinstance(operations, str):
operations = [operations]
def decorator(func):
for op in operations:
if op not in self.implementations:
self.implementations[op] = {}
self.implementations[op][name] = func
return func
return decorator
def validator(self, operation: str = "default"):
"""Decorator to register custom validator"""
def decorator(func):
self.validators[operation] = func
return func
return decorator
def setup(self, name: str, operations: Union[str, List[str]] = None):
"""Decorator to register setup function that runs before timing"""
if operations is None:
operations = ['default']
elif isinstance(operations, str):
operations = [operations]
def decorator(func):
for op in operations:
if op not in self.setups:
self.setups[op] = {}
self.setups[op][name] = func
return func
return decorator
def measure_time(self, func: Callable, data: Any, setup_func: Callable = None) -> tuple[Any, float]:
"""Measure execution time with warmup and optional setup"""
# Warmup runs
for _ in range(self.config.warmup):
try:
if setup_func:
setup_data = setup_func(data)
func(setup_data)
else:
func(data)
except Exception:
# If warmup fails, let the main measurement handle the error
break
# Actual measurement
start = time.perf_counter()
for _ in range(self.config.iterations):
if setup_func:
setup_data = setup_func(data)
result = func(setup_data)
else:
result = func(data)
elapsed_ms = (time.perf_counter() - start) * 1000 / self.config.iterations
return result, elapsed_ms
def validate_result(self, expected: Any, actual: Any, operation: str) -> bool:
"""Validate result using custom validator or default comparison"""
if operation in self.validators:
return self.validators[operation](expected, actual)
return expected == actual
def run(self):
"""Run all benchmarks"""
if self.config.profile_mode:
self._run_profile_mode()
else:
self._run_normal_mode()
def _run_normal_mode(self):
"""Run normal benchmark mode"""
print(f"Running {self.config.name}")
print(f"Sizes: {self.config.sizes}")
print(f"Operations: {self.config.operations}")
print("="*80)
# Always show progressive results: operation by operation across all sizes
for operation in self.config.operations:
for size in self.config.sizes:
self._run_single(operation, size)
# Save and plot results
if self.config.save_results:
self._save_results()
if self.config.plot_results:
self._plot_results()
# Print summary
self._print_summary()
def _run_profile_mode(self):
"""Run profiling mode with minimal overhead for use with vmprof"""
operation = self.config.profile_operation or self.config.operations[0]
size = self.config.profile_size or max(self.config.sizes)
impl_name = self.config.profile_implementation
print(f"PROFILING MODE: {self.config.name}")
print(f"Operation: {operation}, Size: {size}")
if impl_name:
print(f"Implementation: {impl_name}")
print("="*80)
print("Run with vmprof: vmprof --web " + ' '.join(sys.argv))
print("="*80)
# Generate test data
generator = self.data_generators.get(operation, self.data_generators.get('default'))
if not generator:
raise ValueError(f"No data generator for operation: {operation}")
test_data = generator(size, operation)
# Get implementations
impls = self.implementations.get(operation, {})
if not impls:
raise ValueError(f"No implementations for operation: {operation}")
# Filter to specific implementation if requested
if impl_name:
if impl_name not in impls:
raise ValueError(f"Implementation '{impl_name}' not found for operation '{operation}'")
impls = {impl_name: impls[impl_name]}
# Run with minimal overhead - no timing, no validation
for name, func in impls.items():
print(f"\nRunning {name}...")
sys.stdout.flush()
# Setup if needed
setup_func = self.setups.get(operation, {}).get(name)
if setup_func:
data = setup_func(test_data)
else:
data = test_data
# Run the actual function (this is what vmprof will profile)
result = func(data)
print(f"Completed {name}, result checksum: {result}")
sys.stdout.flush()
def _run_single(self, operation: str, size: int):
"""Run a single operation/size combination"""
print(f"\nOperation: {operation}, Size: {size}")
print("-" * 50)
sys.stdout.flush()
# Generate test data
generator = self.data_generators.get(operation,
self.data_generators.get('default'))
if not generator:
raise ValueError(f"No data generator for operation: {operation}")
test_data = generator(size, operation)
# Get implementations for this operation
impls = self.implementations.get(operation, {})
if not impls:
print(f"No implementations for operation: {operation}")
return
# Get setup functions for this operation
setups = self.setups.get(operation, {})
# Run reference implementation first
ref_name, ref_impl = next(iter(impls.items()))
ref_setup = setups.get(ref_name)
expected_result, _ = self.measure_time(ref_impl, test_data, ref_setup)
# Run all implementations
for impl_name, impl_func in impls.items():
try:
setup_func = setups.get(impl_name)
result, time_ms = self.measure_time(impl_func, test_data, setup_func)
correct = self.validate_result(expected_result, result, operation)
# Store result
self.results.append({
'operation': operation,
'size': size,
'implementation': impl_name,
'time_ms': time_ms,
'correct': correct,
'error': None
})
status = "OK" if correct else "FAIL"
print(f" {impl_name:<20} {time_ms:>8.3f} ms {status}")
sys.stdout.flush()
except Exception as e:
self.results.append({
'operation': operation,
'size': size,
'implementation': impl_name,
'time_ms': float('inf'),
'correct': False,
'error': str(e)
})
print(f" {impl_name:<20} ERROR: {str(e)[:40]}")
sys.stdout.flush()
def _save_results(self):
"""Save results to JSON"""
output_dir = Path(self.config.output_dir)
output_dir.mkdir(parents=True, exist_ok=True)
filename = output_dir / f"{self.config.name}_{int(time.time())}.json"
with open(filename, 'w') as f:
json.dump(self.results, f, indent=2)
print(f"\nResults saved to {filename}")
def _plot_results(self):
"""Generate plots using matplotlib if available"""
try:
import matplotlib.pyplot as plt
output_dir = Path(self.config.output_dir)
output_dir.mkdir(parents=True, exist_ok=True)
# Group and prepare data for plotting
data_by_op = self._group_results_by_operation()
# Create plots for each operation
for operation, operation_data in data_by_op.items():
self._create_performance_plot(plt, operation, operation_data, output_dir)
except ImportError:
print("Matplotlib not available - skipping plots")
except Exception as e:
print(f"Plotting failed: {e}")
def _group_results_by_operation(self) -> Dict[str, Dict[int, List[Dict[str, Any]]]]:
"""Group results by operation and size for plotting"""
data_by_op = defaultdict(lambda: defaultdict(list))
for r in self.results:
if r['time_ms'] != float('inf') and r['correct']:
data_by_op[r['operation']][r['size']].append({
'implementation': r['implementation'],
'time_ms': r['time_ms']
})
return data_by_op
def _create_performance_plot(self, plt, operation: str, operation_data: Dict[int, List[Dict[str, Any]]], output_dir: Path):
"""Create a performance plot for a single operation"""
sizes = sorted(operation_data.keys())
implementations = set()
for size_data in operation_data.values():
for entry in size_data:
implementations.add(entry['implementation'])
implementations = sorted(implementations)
plt.figure(figsize=(10, 6))
for impl in implementations:
impl_times = []
impl_sizes = []
for size in sizes:
times = [entry['time_ms'] for entry in operation_data[size]
if entry['implementation'] == impl]
if times:
impl_times.append(statistics.mean(times))
impl_sizes.append(size)
if impl_times:
plt.plot(impl_sizes, impl_times, 'o-', label=impl)
plt.xlabel('Input Size')
plt.ylabel('Time (ms)')
plt.title(f'{self.config.name} - {operation} Operation')
plt.legend()
plt.grid(True, alpha=0.3)
# Apply the configured scaling
if self.config.plot_scale == "loglog":
plt.loglog()
elif self.config.plot_scale == "linear":
pass # Default linear scale
elif self.config.plot_scale == "semilogx":
plt.semilogx()
elif self.config.plot_scale == "semilogy":
plt.semilogy()
else:
# Default to loglog if invalid option
plt.loglog()
plot_file = output_dir / f"{self.config.name}_{operation}_performance.png"
plt.savefig(plot_file, dpi=300, bbox_inches='tight')
plt.close()
print(f"Plot saved: {plot_file}")
def _print_summary(self):
"""Print performance summary"""
print("\n" + "="*80)
print("PERFORMANCE SUMMARY")
print("="*80)
# Group by operation
by_operation = defaultdict(lambda: defaultdict(list))
for r in self.results:
if r['error'] is None and r['time_ms'] != float('inf'):
by_operation[r['operation']][r['implementation']].append(r['time_ms'])
print(f"{'Operation':<15} {'Best Implementation':<20} {'Avg Time (ms)':<15} {'Speedup':<10}")
print("-" * 70)
for op, impl_times in sorted(by_operation.items()):
# Calculate averages
avg_times = [(impl, statistics.mean(times))
for impl, times in impl_times.items()]
avg_times.sort(key=lambda x: x[1])
if avg_times:
best_impl, best_time = avg_times[0]
worst_time = avg_times[-1][1]
speedup = worst_time / best_time if best_time > 0 else 0
print(f"{op:<15} {best_impl:<20} {best_time:<15.3f} {speedup:<10.1f}x")
'''
╺━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╸
https://kobejean.github.io/cp-library
'''
def irank(A: list[int], distinct = False):
P = Packer(len(A)-1); V = P.enumerate(A); V.sort()
if distinct:
for r, ai in enumerate(V): a, i = P.dec(ai); A[i], V[r] = r, a
elif V:
r, p = -1, V[-1]+1 # set p to unique value to trigger `if a != p` on first elm
for ai in V:
a, i = P.dec(ai)
if a!=p: V[r:=r+1] = p = a
A[i] = r
del V[r+1:]
return V
class Packer:
__slots__ = 's', 'm'
def __init__(P, mx: int): P.s = mx.bit_length(); P.m = (1 << P.s) - 1
def enc(P, a: int, b: int): return a << P.s | b
def dec(P, x: int) -> tuple[int, int]: return x >> P.s, x & P.m
def enumerate(P, A, reverse=False): P.ienumerate(A:=list(A), reverse); return A
def ienumerate(P, A, reverse=False):
if reverse:
for i,a in enumerate(A): A[i] = P.enc(-a, i)
else:
for i,a in enumerate(A): A[i] = P.enc(a, i)
def indices(P, A: list[int]): P.iindices(A:=list(A)); return A
def iindices(P, A):
for i,a in enumerate(A): A[i] = P.m&a
def max2(a, b):
return a if a > b else b
def irank(*A: list[int], distinct = False):
N = mxj = 0
for Ai in A: N += len(Ai); mxj = max2(mxj, len(Ai))
P = Packer3(len(A)-1, mxj); V = P.enumerate(A, N); V.sort()
if distinct:
for r,aij in enumerate(V):a,i,j=P.dec(aij);A[i][j],V[r]=r,a
elif V:
r, p = -1, V[-1]+1 # set p to unique value to trigger `if a != p` on first elm
for aij in V:
a,i,j=P.dec(aij)
if a!=p:V[r:=r+1]=p=a
A[i][j]=r
del V[r+1:]
return V
class Packer3:
def __init__(P, mxb: int, mxc: int):
bb, bc = mxb.bit_length(), mxc.bit_length()
P.mc, P.mb, P.sb, P.sa = (1<<bc)-1, (1<<bb)-1, bc, bc+bb
def enc(P, a: int, b: int, c: int): return a << P.sa | b << P.sb | c
def dec(P, x: int) -> tuple[int, int, int]: return x >> P.sa, (x >> P.sb) & P.mb, x & P.mc
def enumerate(P, A, N, reverse=False):
V, k = [0]*N, 0
if reverse:
for i,Ai in enumerate(A):
for j, a in enumerate(Ai):V[k]=P.enc(-a, i, j);k+=1
else:
for i,Ai in enumerate(A):
for j, a in enumerate(Ai):V[k]=P.enc(a, i, j);k+=1
return V
# Configure benchmark
config = BenchmarkConfig(
name="rank",
sizes=[1000000, 100000, 10000, 1000, 100, 10, 1], # Reverse order to warm up JIT
operations=['construction', 'random', 'sorted', 'duplicates', 'reverse'],
iterations=5,
warmup=3,
output_dir="./output/benchmark_results/rank"
)
# Create benchmark instance
benchmark = Benchmark(config)
# Data generator
@benchmark.data_generator("default")
def generate_rank_data(size: int, operation: str):
"""Generate ranking data in different patterns"""
if operation == 'random':
data = [random.randint(1, size) for _ in range(size)]
elif operation == 'sorted':
data = list(range(size))
elif operation == 'duplicates':
# Many duplicates (10% unique values)
unique_count = max(1, size // 10)
data = [random.randint(1, unique_count) for _ in range(size)]
elif operation == 'reverse':
data = list(range(size, 0, -1))
else:
raise ValueError(f"Unknown operation: {operation}")
# Pre-initialize data for fair timing (exclude copy overhead)
preinitialized = {
'data_copy1': list(data),
'data_copy2': list(data),
'distinct': False
}
return {
'data': data,
'distinct': False,
'size': size,
'operation': operation,
'preinitialized': preinitialized
}
# Construction operation
@benchmark.implementation("irank", "construction")
def construction_irank(data):
"""Construct data copy for irank"""
data_copy = list(data['data'])
checksum = 0
for x in data_copy:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "construction")
def construction_irank_multi(data):
"""Construct data copy for irank_multi"""
data_copy = list(data['data'])
checksum = 0
for x in data_copy:
checksum ^= x
return checksum
# Random operation
@benchmark.implementation("irank", "random")
def random_irank(data):
"""Standard irank implementation for random data"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "random")
def random_irank_multi(data):
"""Multi-pass irank implementation for random data"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Sorted operation
@benchmark.implementation("irank", "sorted")
def sorted_irank(data):
"""Standard irank implementation for sorted data"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "sorted")
def sorted_irank_multi(data):
"""Multi-pass irank implementation for sorted data"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Duplicates operation
@benchmark.implementation("irank", "duplicates")
def duplicates_irank(data):
"""Standard irank implementation for data with duplicates"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "duplicates")
def duplicates_irank_multi(data):
"""Multi-pass irank implementation for data with duplicates"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Reverse operation
@benchmark.implementation("irank", "reverse")
def reverse_irank(data):
"""Standard irank implementation for reverse data"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
@benchmark.implementation("irank_multi", "reverse")
def reverse_irank_multi(data):
"""Multi-pass irank implementation for reverse data"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=pre['distinct'])
checksum = 0
for x in result:
checksum ^= x
return checksum
# Additional benchmark with distinct=True
@benchmark.data_generator("distinct")
def generate_rank_data_distinct(size: int, operation: str):
"""Generate ranking data with distinct=True"""
base_data = generate_rank_data(size, operation)
base_data['distinct'] = True
base_data['preinitialized']['distinct'] = True
return base_data
def irank_distinct_implementation(data):
"""irank with distinct=True"""
pre = data['preinitialized']
result = irank(pre['data_copy1'], distinct=True)
checksum = 0
for x in result:
checksum ^= x
return checksum
def irank_multi_distinct_implementation(data):
"""irank_multi with distinct=True"""
pre = data['preinitialized']
result = irank_multi(pre['data_copy2'], distinct=True)
checksum = 0
for x in result:
checksum ^= x
return checksum
# Custom validator for rank results (now using XOR checksums)
@benchmark.validator("default")
def validate_rank_result(expected, actual):
"""Validate ranking results using XOR checksums"""
try:
# Compare XOR checksums directly
return int(expected) == int(actual)
except Exception:
return False
if __name__ == "__main__":
# Parse command line args and run appropriate mode
runner = benchmark.parse_args()
runner.run()