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
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
|
//! Dispatch-strategy matrix benchmark: linear list vs char trie vs perfect
//! hash, plus the "auto" strategy (which picks one of the three from the
//! command table — the summary checks it against the per-cell minimum).
//!
//! Build & run (workspace-internal, `-O`):
//! ```bash
//! cargo run --release --manifest-path dev/bench/dispatch/Cargo.toml
//! ```
#![allow(dead_code, clippy::all)]
use prettytable::format::Alignment;
use prettytable::{Cell, Row, Table};
/// Minimal dispatch surface implemented by every generated cell so the four
/// strategies can be benchmarked through a uniform call.
pub trait BenchDispatch {
type Enum;
fn dispatch_args(
raw: &[String],
) -> Result<::mingling::AnyOutput<Self::Enum>, ::mingling::error::ProgramInternalExecuteError>;
fn build_entry_fallback(args: Vec<String>) -> ::mingling::AnyOutput<Self::Enum>;
}
include!(concat!(env!("OUT_DIR"), "/generated.rs"));
const N_STRATS: usize = 4; // linear, trie, phf, auto
fn right(v: f64) -> Cell {
Cell::new_align(&format!("{v:.1}"), Alignment::RIGHT)
}
fn right2(v: f64) -> Cell {
Cell::new_align(&format!("{v:.2}"), Alignment::RIGHT)
}
fn left(s: &str) -> Cell {
Cell::new(s)
}
fn main() {
println!("dispatch matrix — ns/op (lower is better), best of 5×50k, rustc -O");
println!();
let results: Vec<(String, String, f64, f64)> = CELL_META
.iter()
.enumerate()
.map(|(id, (label, strat, names))| {
let hits: Vec<Vec<String>> = names
.iter()
.map(|name| {
let mut v: Vec<String> = name.split(' ').map(str::to_string).collect();
v.push("argA".into());
v.push("argB".into());
v
})
.collect();
let misses: Vec<Vec<String>> = vec![
vec!["qqq".into(), "zzz".into()],
vec!["qqq".into(), "zzz".into(), "argA".into()],
vec!["qqq".into()],
];
let (h, m) = run_cell(id, &hits, &misses);
(label.to_string(), strat.to_string(), h, m)
})
.collect();
let mut table = Table::new();
table.set_format(*prettytable::format::consts::FORMAT_BOX_CHARS);
table.set_titles(Row::new(vec![
left("cell"),
left("── hit ──"),
left(""),
left(""),
left(""),
left("── miss ──"),
left(""),
left(""),
left(""),
]));
table.add_row(Row::new(vec![
left(""),
left("lin"),
left("trie"),
left("phf"),
left("auto"),
left("lin"),
left("trie"),
left("phf"),
left("auto"),
]));
let mut geo_hit = [1.0f64; N_STRATS];
let mut geo_miss = [1.0f64; N_STRATS];
let mut wins = [0usize; N_STRATS];
let mut auto_optimal = 0usize;
let mut auto_within5 = 0usize;
let mut cells = 0usize;
let mut auto_picks = [0usize; N_STRATS]; // auto's own selection counts
for (chunk_idx, chunk) in results.chunks(N_STRATS).enumerate() {
let label = &chunk[0].0;
let hit: Vec<f64> = chunk.iter().map(|c| c.2).collect();
let miss: Vec<f64> = chunk.iter().map(|c| c.3).collect();
table.add_row(Row::new(vec![
left(label),
right(hit[0]),
right(hit[1]),
right(hit[2]),
right(hit[3]),
right(miss[0]),
right(miss[1]),
right(miss[2]),
right(miss[3]),
]));
for s in 0..N_STRATS {
geo_hit[s] *= hit[s];
geo_miss[s] *= miss[s];
}
// Per-cell winner by hit×miss geomean among the three explicit
// strategies.
let gm: Vec<f64> = (0..N_STRATS).map(|s| (hit[s] * miss[s]).sqrt()).collect();
let mut best_idx = 0usize;
for s in 1..3 {
if gm[s] < gm[best_idx] {
best_idx = s;
}
}
let best_idx_gm = gm[best_idx];
wins[best_idx] += 1;
// The auto cell (last of the four) exposes its chosen strategy; check
// it against the measured minimum.
let auto_id = chunk_idx * N_STRATS + (N_STRATS - 1);
let pick_idx = match run_cell_pick(auto_id) {
"dispatch_linear" => 0,
"dispatch_tree" => 1,
_ => 2,
};
auto_picks[pick_idx] += 1;
if pick_idx == best_idx {
auto_optimal += 1;
}
// Identical generated code measures up to ~10 ns apart across cells,
// so also count selections within 5% of the measured minimum.
if gm[3] <= best_idx_gm * 1.05 {
auto_within5 += 1;
}
cells += 1;
}
let n = cells as f64;
table.add_row(Row::new(vec![
left("geomean"),
right2(geo_hit[0].powf(1.0 / n)),
right2(geo_hit[1].powf(1.0 / n)),
right2(geo_hit[2].powf(1.0 / n)),
right2(geo_hit[3].powf(1.0 / n)),
right2(geo_miss[0].powf(1.0 / n)),
right2(geo_miss[1].powf(1.0 / n)),
right2(geo_miss[2].powf(1.0 / n)),
right2(geo_miss[3].powf(1.0 / n)),
]));
table.add_row(Row::new(vec![
left("wins (hit×miss geomean)"),
left(&wins[0].to_string()),
left(&wins[1].to_string()),
left(&wins[2].to_string()),
]));
table.add_row(Row::new(vec![
left("auto picks"),
left(&auto_picks[0].to_string()),
left(&auto_picks[1].to_string()),
left(&auto_picks[2].to_string()),
]));
table.add_row(Row::new(vec![left(
format!("auto == per-cell best: {auto_optimal}/{cells}").as_str(),
)]));
table.add_row(Row::new(vec![left(
format!("auto within 5% of best: {auto_within5}/{cells}").as_str(),
)]));
table.printstd();
}
|
