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

use super::{
    execution::{ExecutionMode, Processor, StreamSegment},
    store::{ExecutionPlanId, ExecutionPlanStore},
    Operation, OperationDescription, OperationQueue, StreamId,
};
use crate::{FusionBackend, HandleContainer};
use std::collections::HashMap;

/// Keep track of multiple concurrent streams of operations.
pub struct MultiStream<B: FusionBackend> {
    streams: HashMap<StreamId, Stream<B>>,
    optimizations: ExecutionPlanStore<B::Optimization>,
    device: B::FusionDevice,
}

impl<B: FusionBackend> MultiStream<B> {
    pub(crate) fn new(device: B::FusionDevice) -> Self {
        Self {
            streams: HashMap::new(),
            optimizations: ExecutionPlanStore::new(),
            device,
        }
    }

    /// Register a new tensor operation.
    pub(crate) fn register(
        &mut self,
        streams: Vec<StreamId>,
        desc: OperationDescription,
        operation: Box<dyn Operation<B>>,
        handles: &mut HandleContainer<B>,
    ) {
        let id = self.maybe_drain(streams, handles);

        let stream = match self.streams.get_mut(&id) {
            Some(stream) => stream,
            None => {
                let stream = Stream::new(self.device.clone());
                self.streams.insert(id, stream);
                self.streams
                    .get_mut(&id)
                    .expect("Just added, so should be included in the hashmap.")
            }
        };

        stream.queue.add(desc, operation);

        let size_before = stream.queue.len();
        stream.processor.process(
            Segment::new(&mut stream.queue, handles),
            &mut self.optimizations,
            ExecutionMode::Lazy,
        );
        let size_after = stream.queue.len();

        if size_after != size_before {
            self.free_orphans(handles);
        }

        if size_after == 0 {
            self.streams.remove(&id);
        }
    }

    /// Drain the streams.
    pub fn drain(&mut self, handles: &mut HandleContainer<B>, id: StreamId) {
        if let Some(mut stream) = self.streams.remove(&id) {
            stream.processor.process(
                Segment::new(&mut stream.queue, handles),
                &mut self.optimizations,
                ExecutionMode::Sync,
            );
            self.free_orphans(handles);
        }
    }

    /// When one of the provided streams is different from the current stream, we drain them.
    fn maybe_drain(
        &mut self,
        streams: Vec<StreamId>,
        handles: &mut HandleContainer<B>,
    ) -> StreamId {
        let streams = Self::remove_duplicate(streams);
        let current = StreamId::current();

        if streams.len() == 1 {
            // The only case where we don't need to drain.
            if streams[0] == current {
                return current;
            }
        }

        for id in streams {
            self.drain(handles, id);
        }

        current
    }

    fn remove_duplicate(items: Vec<StreamId>) -> Vec<StreamId> {
        if items.len() == 1 {
            return items;
        }

        let mut output = Vec::with_capacity(items.len());
        for item in items {
            if !output.contains(&item) {
                output.push(item);
            }
        }
        output
    }

    fn free_orphans(&self, handles: &mut HandleContainer<B>) {
        let nodes = self
            .streams
            .values()
            .flat_map(|a| a.queue.global.iter())
            .flat_map(|a| a.nodes())
            .map(|tensor| &tensor.id)
            .collect::<Vec<_>>();

        handles.free_orphans(&nodes);
    }
}

struct Stream<B: FusionBackend> {
    queue: OperationQueue<B>,
    processor: Processor<B::Optimization>,
}

#[derive(new)]
struct Segment<'a, B: FusionBackend> {
    queue: &'a mut OperationQueue<B>,
    handles: &'a mut HandleContainer<B>,
}

impl<'i, B: FusionBackend> StreamSegment<B::Optimization> for Segment<'i, B> {
    fn operations(&self) -> &[OperationDescription] {
        &self.queue.relative
    }

    fn execute(&mut self, id: ExecutionPlanId, store: &mut ExecutionPlanStore<B::Optimization>) {
        self.queue.execute(id, self.handles, store)
    }
}

impl<B: FusionBackend> Stream<B> {
    fn new(device: B::FusionDevice) -> Self {
        Self {
            processor: Processor::new(B::optimizations(device.into())),
            queue: OperationQueue::new(),
        }
    }
}