polydat 0.3.1

Polydat — a variates construction engine
Documentation 
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
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211

// Copyright 2024-2026 Jonathan Shook
// SPDX-License-Identifier: Apache-2.0

//! Cursors on every engine (engine_parity.md, step 3). A cursor's
//! `over` clause resolves to partitions at build when the clause and
//! the extent are constant; a clause that denotes one partition seeds
//! the cursor so the program runs on every engine with no host call,
//! and a clause that denotes several is narrowed through the same
//! `set_cursor` on the interpreter, the closure tier, and the hybrid
//! kernel. The partition family reads the cursor through its `Ext`
//! slot and the scalar projections through their own.

#![cfg(feature = "jit")]

use polydat::ast::Value;
use polydat::dsl::compile::compile_polydat_to_assembler;
use polydat::iteration::cursor_partition::Partition;

/// The partition family and the scalar projections over one cursor.
const CONSUMERS: &str = "\
n := cardinality(q.cursor)
s := start_of(q.cursor)
e := end_of(q.cursor)
i := idx_of(q.cursor)
c := count_of(q.cursor)
m := mod_in(cycle, q.cursor)
k := clamp_in(cycle, q.cursor)
r := random_in(q.cursor, cycle)
a := at(q.cursor, u64_mod(cycle, cardinality(q.cursor)))
h := partition_count(subdivide(q.cursor, 2))
pi := q.cursor.idx
ps := q.cursor.start_ordinal
pe := q.cursor.end_ordinal
pc := q.cursor.partition_count
";

const OUTPUTS: [&str; 14] = [
    "n", "s", "e", "i", "c", "m", "k", "r", "a", "h", "pi", "ps", "pe", "pc",
];

fn program(over: &str) -> String {
    format!("input cycle: u64\ncursor q = range(0, 1000) over \"{over}\"\n{CONSUMERS}")
}

/// Every output on every engine, for a run of cycles, after `narrow`
/// has been applied to each kernel.
fn agree(src: &str, narrow: Option<&Partition>) {
    let mut p1 = compile_polydat_to_assembler(src)
        .unwrap()
        .compile()
        .unwrap();
    let mut p2 = compile_polydat_to_assembler(src)
        .unwrap()
        .try_compile_raw()
        .unwrap_or_else(|_| panic!("the partition family has a closure form\n{src}"));
    let mut p2pp = compile_polydat_to_assembler(src)
        .unwrap()
        .try_compile()
        .unwrap_or_else(|_| panic!("P2 push-pull\n{src}"));
    let mut hybrid = compile_polydat_to_assembler(src)
        .unwrap()
        .compile_hybrid()
        .unwrap_or_else(|e| panic!("hybrid: {e}\n{src}"));
    if let Some(p) = narrow {
        p1.set_cursor("q", p).unwrap();
        p2.set_cursor("q", p).unwrap();
        p2pp.set_cursor("q", p).unwrap();
        hybrid.set_cursor("q", p).unwrap();
    }
    for c in [0u64, 1, 7, 250, 999, 1000, 4096] {
        p1.set_inputs(&[c]);
        let want: Vec<Value> = OUTPUTS.iter().map(|o| p1.pull(o).clone()).collect();
        p2.eval(&[c]);
        let got_p2: Vec<Value> = OUTPUTS.iter().map(|o| p2.get_value(o)).collect();
        p2pp.eval(&[c]);
        let got_p2pp: Vec<Value> = OUTPUTS.iter().map(|o| p2pp.get_value(o)).collect();
        hybrid.eval(&[c]);
        let got_hybrid: Vec<Value> = OUTPUTS.iter().map(|o| hybrid.get_value(o)).collect();
        for (i, out) in OUTPUTS.iter().enumerate() {
            for (tier, got) in [
                ("P2", &got_p2[i]),
                ("P2 push-pull", &got_p2pp[i]),
                ("hybrid", &got_hybrid[i]),
            ] {
                assert_eq!(
                    want[i].port_type(),
                    got.port_type(),
                    "{tier}: `{out}` at cycle {c}: type\n{src}"
                );
                assert_eq!(
                    want[i].to_display_string(),
                    got.to_display_string(),
                    "{tier}: `{out}` at cycle {c}\n{src}"
                );
            }
        }
    }
}

#[test]
fn a_single_partition_clause_needs_no_host_call_on_any_engine() {
    for over in ["0..50%", "25%..75%", "*"] {
        let src = program(over);
        let asm = compile_polydat_to_assembler(&src).unwrap();
        let schema = &asm.cursor_schemas()[0];
        let parts = schema.partitions.as_ref().expect("resolved at build");
        assert_eq!(parts.len(), 1, "`over \"{over}\"` denotes one partition");
        agree(&src, None);
        // What the interpreter reads is the one partition.
        let mut p1 = compile_polydat_to_assembler(&src)
            .unwrap()
            .compile()
            .unwrap();
        p1.set_inputs(&[0]);
        assert_eq!(p1.pull("n").as_u64(), parts[0].end_ord - parts[0].start_ord);
        assert_eq!(p1.pull("ps").as_u64(), parts[0].start_ord);
    }
}

#[test]
fn every_partition_of_a_list_narrows_every_engine_alike() {
    for over in ["*/4", "20%,30%,*", "10%,*"] {
        let src = program(over);
        let asm = compile_polydat_to_assembler(&src).unwrap();
        let parts = asm.cursor_schemas()[0]
            .partitions
            .clone()
            .expect("resolved at build");
        assert!(
            parts.len() > 1,
            "`over \"{over}\"` denotes several partitions"
        );
        for p in &parts {
            agree(&src, Some(p));
        }
    }
}

#[test]
fn the_compiled_kernels_report_the_cursors_the_interpreter_does() {
    let src = program("*/4");
    let p1 = compile_polydat_to_assembler(&src)
        .unwrap()
        .compile()
        .unwrap();
    let p2 = compile_polydat_to_assembler(&src)
        .unwrap()
        .try_compile_raw()
        .ok()
        .unwrap();
    let hybrid = compile_polydat_to_assembler(&src)
        .unwrap()
        .compile_hybrid()
        .unwrap();
    let names = |s: &[polydat::iteration::source::SourceSchema]| -> Vec<(String, usize)> {
        s.iter()
            .map(|c| (c.name.clone(), c.partitions.as_ref().map_or(0, Vec::len)))
            .collect()
    };
    assert_eq!(
        names(p1.program().cursor_schemas()),
        vec![("q".to_string(), 4)]
    );
    assert_eq!(names(p2.cursor_schemas()), vec![("q".to_string(), 4)]);
    assert_eq!(names(hybrid.cursor_schemas()), vec![("q".to_string(), 4)]);
    // An unknown cursor is refused by name on every engine.
    let p = p2.cursor_schemas()[0].partitions.as_ref().unwrap()[0];
    let mut p2 = p2;
    let err = p2.set_cursor("zz", &p).unwrap_err();
    assert!(err.contains("no cursor named 'zz'"), "{err}");
    let mut p1 = p1;
    let err = p1.set_cursor("zz", &p).unwrap_err();
    assert!(err.contains("no cursor named 'zz'"), "{err}");
}

#[test]
fn a_clause_denoting_several_partitions_is_unset_until_narrowed() {
    // The cursor is `None` until the host narrows it, and every
    // consumer reads `None` through it, on the interpreter and on the
    // closure tier alike (engine_parity.md, A3 and A12). Native code
    // cannot carry `None` and refuses to run.
    let src = program("*/4");
    let mut p1 = compile_polydat_to_assembler(&src)
        .unwrap()
        .compile()
        .unwrap();
    p1.set_inputs(&[0]);
    assert_eq!(*p1.pull("n"), Value::None);
    let mut p2 = compile_polydat_to_assembler(&src)
        .unwrap()
        .try_compile_raw()
        .ok()
        .unwrap();
    p2.eval(&[0]);
    for o in OUTPUTS {
        // The partition consumers read `None`; the scalar projections
        // read their declared defaults, on both engines.
        assert_eq!(p2.get_value(o), *p1.pull(o), "`{o}` on the closure tier");
    }
    // Narrowed, the same kernel computes; the outputs stop being `None`.
    let parts = compile_polydat_to_assembler(&src).unwrap().cursor_schemas()[0]
        .partitions
        .clone()
        .unwrap();
    p2.set_cursor("q", &parts[1]).unwrap();
    p2.eval(&[0]);
    assert_eq!(
        p2.get_value("n").as_u64(),
        parts[1].end_ord - parts[1].start_ord
    );
}