aprender-compute 0.29.0

High-performance SIMD compute library with GPU support, LLM inference engine, and GGUF model loading (was: trueno)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132

# Renacer Configuration for Trueno
# Performance assertions based on RTX 4090 + golden trace baselines

[project]
name = "trueno"
version = "0.7.0"

# Performance budgets from golden traces (empirically validated)
[performance.budgets]
# SIMD operations should complete in <2ms with <200 syscalls
backend_detection = { max_time_ms = 2.0, max_syscalls = 200 }
matrix_operations = { max_time_ms = 2.0, max_syscalls = 200 }
activation_functions = { max_time_ms = 2.0, max_syscalls = 200 }
performance_demo = { max_time_ms = 2.0, max_syscalls = 200 }
ml_similarity = { max_time_ms = 2.0, max_syscalls = 200 }

# Syscall pattern assertions (from golden traces)
[[assertions]]
name = "minimal_io_overhead"
description = "SIMD operations should have minimal I/O overhead"
check = "syscall_percentage"
syscall = "write"
max_percentage = 35.0  # From golden traces: write is 30.66% max

[[assertions]]
name = "memory_allocation_budget"
description = "Memory allocation should be controlled"
check = "syscall_count"
syscall = "mmap"
max_count = 30  # From golden traces: 13-28 mmap calls

[[assertions]]
name = "no_excessive_reads"
description = "Minimal disk I/O for in-memory SIMD operations"
check = "syscall_count"
syscall = "read"
max_count = 10  # From golden traces: 5-6 read calls

[[assertions]]
name = "fast_execution"
description = "Total runtime should be sub-2ms for SIMD benchmarks"
check = "total_time"
max_time_ms = 2.0  # From golden traces: all examples <2ms

# GPU-specific assertions (RTX 4090 empirical data)
[[assertions]]
name = "gpu_transfer_overhead"
description = "GPU operations have ~3.5ms fixed overhead"
check = "min_time"
min_time_ms = 3.0  # GPU operations cannot be faster than transfer time
applies_to = ["gpu_*"]

# Regression detection thresholds
[regression]
# Alert if performance degrades by more than 10%
time_threshold_percent = 10.0
syscall_threshold_percent = 15.0

# Golden trace comparison
[golden_traces]
directory = "golden_traces"
format = "renacer-json-v1"

# Compare against these baseline traces
[[golden_traces.baselines]]
name = "backend_detection"
file = "golden_traces/backend_detection.json"
max_deviation_percent = 10.0

[[golden_traces.baselines]]
name = "performance_demo"
file = "golden_traces/performance_demo.json"
max_deviation_percent = 10.0

[[golden_traces.baselines]]
name = "matrix_operations"
file = "golden_traces/matrix_operations.json"
max_deviation_percent = 10.0

[[golden_traces.baselines]]
name = "activation_functions"
file = "golden_traces/activation_functions.json"
max_deviation_percent = 10.0

[[golden_traces.baselines]]
name = "ml_similarity"
file = "golden_traces/ml_similarity.json"
max_deviation_percent = 10.0

# Anti-pattern detection
[anti_patterns]
# Detect PCIe bottleneck for GPU workloads
[[anti_patterns.rules]]
name = "pcie_bottleneck"
description = "Excessive GPU memory transfers relative to compute time"
enabled = true
check = "ratio"
syscalls = ["write", "read"]  # GPU device I/O
min_ratio = 0.5  # If >50% time in I/O, flag as bottleneck
severity = "warning"

# Detect memory thrashing
[[anti_patterns.rules]]
name = "memory_thrashing"
description = "Excessive mmap/munmap calls indicate memory pressure"
enabled = true
check = "syscall_count"
syscall = "mmap"
max_count = 50
severity = "error"

# CI/CD integration
[ci]
# Fail build if performance budgets exceeded
fail_on_budget_exceeded = true
fail_on_regression = true
fail_on_anti_patterns = ["error"]  # Only fail on error severity

# Reporting
[reporting]
# Output formats for CI integration
formats = ["json", "markdown", "junit"]
output_dir = "target/renacer-reports"

# Generate flamegraphs for performance investigation
[profiling]
enable_flamegraphs = false  # Set to true for deep performance analysis
flamegraph_dir = "target/flamegraphs"

# Comparison with baseline
compare_with_baseline = true
baseline_branch = "main"