weirflow 0.1.0

GPU-first dataflow analysis primitives for Vyre and Santh compiler pipelines.
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
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228

#![allow(clippy::too_many_arguments)]

use super::ifds_reach_step;
use crate::dispatch_input_cache::OwnedDispatchInputs;
use vyre_primitives::graph::program_graph::ProgramGraphShape;

pub fn ifds_reach_closure_via<F>(
    dispatch: &F,
    node_count: u32,
    edge_offsets: &[u32],
    edge_targets: &[u32],
    edge_kind_mask: &[u32],
    seed_bits: &[u32],
    max_iterations: u32,
) -> Result<Vec<u32>, String>
where
    F: Fn(&vyre::ir::Program, &[Vec<u8>], Option<[u32; 3]>) -> Result<Vec<Vec<u8>>, String>,
{
    let owned_inputs = OwnedDispatchInputs::new();
    ifds_reach_closure_borrowed_via(
        &|program, inputs, grid| {
            owned_inputs.dispatch_refreshing_frontier(program, inputs, grid, dispatch)
        },
        node_count,
        edge_offsets,
        edge_targets,
        edge_kind_mask,
        seed_bits,
        max_iterations,
    )
}

/// GPU IFDS reachability closure using borrowed dispatch inputs.
///
/// The graph buffers are immutable across fixpoint iterations and are packed
/// once; only the frontier buffer changes between dispatches.
pub fn ifds_reach_closure_borrowed_via<F>(
    dispatch: &F,
    node_count: u32,
    edge_offsets: &[u32],
    edge_targets: &[u32],
    edge_kind_mask: &[u32],
    seed_bits: &[u32],
    max_iterations: u32,
) -> Result<Vec<u32>, String>
where
    F: Fn(&vyre::ir::Program, &[&[u8]], Option<[u32; 3]>) -> Result<Vec<Vec<u8>>, String>,
{
    crate::dispatch_decode::require_positive_iterations("ifds_reach_closure", max_iterations)?;
    crate::dispatch_decode::require_bitset_tail_clear("ifds_reach_closure", seed_bits, node_count)?;
    let out = ifds_reach_closure_borrowed_into_via(
        &|program, inputs, grid, outputs| {
            let result = dispatch(program, inputs, grid)?;
            crate::output_scratch::replace_outputs_preserving_slots(outputs, result);
            Ok(())
        },
        node_count,
        edge_offsets,
        edge_targets,
        edge_kind_mask,
        seed_bits,
        max_iterations,
    )?;
    crate::dispatch_decode::require_bitset_tail_clear(
        "ifds_reach_closure output",
        &out,
        node_count,
    )?;
    Ok(out)
}

/// GPU IFDS reachability closure using borrowed dispatch inputs and
/// caller-owned output storage.
pub fn ifds_reach_closure_borrowed_into_via<F>(
    dispatch: &F,
    node_count: u32,
    edge_offsets: &[u32],
    edge_targets: &[u32],
    edge_kind_mask: &[u32],
    seed_bits: &[u32],
    max_iterations: u32,
) -> Result<Vec<u32>, String>
where
    F: Fn(&vyre::ir::Program, &[&[u8]], Option<[u32; 3]>, &mut Vec<Vec<u8>>) -> Result<(), String>,
{
    let mut scratch = crate::fixed_point_scratch::FixedPointScratch::default();
    ifds_reach_closure_borrowed_into_with_scratch_via(
        dispatch,
        node_count,
        edge_offsets,
        edge_targets,
        edge_kind_mask,
        seed_bits,
        max_iterations,
        &mut scratch,
    )
}

/// GPU IFDS reachability closure using borrowed dispatch inputs and
/// caller-owned fixed-point scratch.
pub fn ifds_reach_closure_borrowed_into_with_scratch_via<F>(
    dispatch: &F,
    node_count: u32,
    edge_offsets: &[u32],
    edge_targets: &[u32],
    edge_kind_mask: &[u32],
    seed_bits: &[u32],
    max_iterations: u32,
    scratch: &mut crate::fixed_point_scratch::FixedPointScratch,
) -> Result<Vec<u32>, String>
where
    F: Fn(&vyre::ir::Program, &[&[u8]], Option<[u32; 3]>, &mut Vec<Vec<u8>>) -> Result<(), String>,
{
    let mut result = crate::staging_reserve::reserved_fixed_point_result(
        "ifds_reach_closure result",
        node_count,
    )?;
    ifds_reach_closure_borrowed_into_result_with_scratch_via(
        dispatch,
        node_count,
        edge_offsets,
        edge_targets,
        edge_kind_mask,
        seed_bits,
        max_iterations,
        scratch,
        &mut result,
    )?;
    Ok(result)
}

/// GPU IFDS reachability closure using borrowed dispatch inputs,
/// caller-owned fixed-point scratch, and caller-owned result storage.
pub fn ifds_reach_closure_borrowed_into_result_with_scratch_via<F>(
    dispatch: &F,
    node_count: u32,
    edge_offsets: &[u32],
    edge_targets: &[u32],
    edge_kind_mask: &[u32],
    seed_bits: &[u32],
    max_iterations: u32,
    scratch: &mut crate::fixed_point_scratch::FixedPointScratch,
    result: &mut Vec<u32>,
) -> Result<(), String>
where
    F: Fn(&vyre::ir::Program, &[&[u8]], Option<[u32; 3]>, &mut Vec<Vec<u8>>) -> Result<(), String>,
{
    crate::dispatch_decode::require_positive_iterations("ifds_reach_closure", max_iterations)?;
    let words = crate::dispatch_decode::bitset_word_capacity("ifds_reach_closure", node_count)?;
    crate::dispatch_decode::require_bitset_words("ifds_reach_closure", seed_bits, words)?;
    crate::dispatch_decode::require_bitset_tail_clear("ifds_reach_closure", seed_bits, node_count)?;
    crate::dispatch_decode::require_csr_shape(
        "ifds_reach_closure",
        node_count,
        edge_offsets,
        edge_targets,
        edge_kind_mask,
    )?;
    scratch.prepare_frontier_words(words, seed_bits)?;
    let edge_count = u32::try_from(edge_targets.len()).map_err(|error| {
        format!(
            "ifds_reach_closure edge target count does not fit u32: {error}. Fix: shard the ProgramGraph before GPU dispatch."
        )
    })?;
    let program = ifds_reach_step(
        ProgramGraphShape::new(node_count, edge_count),
        "fin",
        "fout",
    );
    let node_count_usize =
        crate::dispatch_decode::u32_to_usize(node_count, "ifds_reach_closure node count")?;
    scratch.prepare_static_input_byte_slots(5)?;
    crate::dispatch_decode::pack_repeated_u32_into(
        0,
        node_count_usize,
        &mut scratch.static_input_bytes[0],
    )?;
    crate::dispatch_decode::try_pack_u32_into(edge_offsets, &mut scratch.static_input_bytes[1], "weir u32 byte staging")?;
    crate::dispatch_decode::try_pack_u32_into(edge_targets, &mut scratch.static_input_bytes[2], "weir u32 byte staging")?;
    crate::dispatch_decode::try_pack_u32_into(edge_kind_mask, &mut scratch.static_input_bytes[3], "weir u32 byte staging")?;
    crate::dispatch_decode::pack_repeated_u32_into(
        0,
        node_count_usize,
        &mut scratch.static_input_bytes[4],
    )?;
    scratch.prepare_next_words(words)?;
    for _ in 0..max_iterations {
        crate::dispatch_decode::try_pack_u32_into(&scratch.frontier_words, &mut scratch.frontier_bytes, "weir u32 byte staging")?;
        let inputs = [
            scratch.static_input_bytes[0].as_slice(),
            scratch.static_input_bytes[1].as_slice(),
            scratch.static_input_bytes[2].as_slice(),
            scratch.static_input_bytes[3].as_slice(),
            scratch.static_input_bytes[4].as_slice(),
            scratch.frontier_bytes.as_slice(),
            scratch.frontier_bytes.as_slice(),
        ];
        dispatch(
            &program,
            &inputs,
            Some([node_count.max(1), 1, 1]),
            &mut scratch.outputs,
        )?;
        crate::dispatch_decode::unpack_only_exact_u32_into(
            &scratch.outputs,
            "ifds_reach_closure",
            "frontier",
            words,
            &mut scratch.next,
        )?;
        crate::dispatch_decode::require_bitset_tail_clear(
            "ifds_reach_closure output",
            &scratch.next,
            node_count,
        )?;
        if crate::fixed_point_scratch::copy_if_converged(
            &scratch.next,
            &scratch.frontier_words,
            result,
        )? {
            return Ok(());
        }
        std::mem::swap(&mut scratch.frontier_words, &mut scratch.next);
    }
    Err(format!(
        "ifds_reach_closure did not converge within {max_iterations} iterations"
    ))
}