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

// Copyright 2019-2022 Clemens Lutz
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
//     http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.

use crate::error::Result;
use crate::runtime::cpu_affinity::CpuAffinity;
use rayon::{ThreadPool, ThreadPoolBuilder};
use rustacuda::context::CurrentContext;
use rustacuda::error::CudaError;
use rustacuda::stream::{Stream, StreamFlags};
use std::default::Default;
use std::sync::Arc;

#[derive(Clone)]
pub struct MorselSpec {
    pub cpu_morsel_bytes: usize,
    pub gpu_morsel_bytes: usize,
}

#[derive(Clone, Default)]
pub struct WorkerCpuAffinity {
    pub cpu_workers: CpuAffinity,
    pub gpu_workers: CpuAffinity,
}

pub struct HetMorselExecutor {
    pub(super) morsel_spec: MorselSpec,
    pub(super) cpu_workers: usize,
    pub(super) gpu_workers: usize,
    pub(super) cpu_thread_pool: ThreadPool,
    pub(super) gpu_thread_pool: ThreadPool,
    pub(super) streams: Vec<Stream>,
}

pub struct HetMorselExecutorBuilder {
    morsel_spec: MorselSpec,
    cpu_threads: usize,
    cpu_worker_affinity: Arc<CpuAffinity>,
    gpu_worker_affinity: Arc<CpuAffinity>,
    gpu_ids: Vec<u16>,
}

impl HetMorselExecutorBuilder {
    pub fn new() -> Self {
        let cpu_morsel_bytes = 16_384;
        let gpu_morsel_bytes = 33_554_432;

        Self {
            morsel_spec: MorselSpec {
                cpu_morsel_bytes,
                gpu_morsel_bytes,
            },
            cpu_threads: 0,
            cpu_worker_affinity: Arc::new(CpuAffinity::default()),
            gpu_worker_affinity: Arc::new(CpuAffinity::default()),
            gpu_ids: Vec::new(),
        }
    }

    pub fn morsel_spec(mut self, morsel_spec: MorselSpec) -> Self {
        self.morsel_spec = morsel_spec;
        self
    }

    pub fn cpu_threads(mut self, threads: usize) -> Self {
        self.cpu_threads = threads;
        self
    }

    pub fn worker_cpu_affinity(
        mut self,
        WorkerCpuAffinity {
            cpu_workers,
            gpu_workers,
        }: WorkerCpuAffinity,
    ) -> Self {
        self.cpu_worker_affinity = Arc::new(cpu_workers);
        self.gpu_worker_affinity = Arc::new(gpu_workers);
        self
    }

    pub fn gpu_ids(mut self, gpu_ids: Vec<u16>) -> Self {
        self.gpu_ids = gpu_ids;
        self
    }

    pub fn build(self) -> Result<HetMorselExecutor> {
        let cpu_workers = self.cpu_threads;
        let gpu_workers = self.gpu_ids.len();
        let cpu_affinity = self.cpu_worker_affinity.clone();
        let gpu_affinity = self.gpu_worker_affinity.clone();

        let unowned_context_cpu_pool = CurrentContext::get_current()?;
        let unowned_context_gpu_pool = unowned_context_cpu_pool.clone();

        let cpu_thread_pool = ThreadPoolBuilder::new()
            .num_threads(cpu_workers)
            .start_handler(move |tid| {
                CurrentContext::set_current(&unowned_context_cpu_pool)
                    .expect("Failed to set CUDA context in CPU worker thread");

                cpu_affinity
                    .clone()
                    .set_affinity(tid as u16)
                    .expect("Couldn't set CPU core affinity");
            })
            .build()?;

        let gpu_thread_pool = ThreadPoolBuilder::new()
            .num_threads(gpu_workers)
            .start_handler(move |tid| {
                CurrentContext::set_current(&unowned_context_gpu_pool)
                    .expect("Failed to set CUDA context in GPU worker thread");

                gpu_affinity
                    .clone()
                    .set_affinity(tid as u16)
                    .expect("Couldn't set CPU core affinity");
            })
            .build()?;

        let streams = (0..gpu_workers)
            .map(|_| Stream::new(StreamFlags::NON_BLOCKING, None))
            .collect::<std::result::Result<Vec<_>, CudaError>>()?;

        Ok(HetMorselExecutor {
            morsel_spec: self.morsel_spec,
            cpu_workers,
            gpu_workers,
            cpu_thread_pool,
            gpu_thread_pool,
            streams,
        })
    }
}