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

#![deny(clippy::pedantic)]
#![no_std]
#![feature(type_alias_impl_trait)]
#![feature(decl_macro)]
#![cfg_attr(target_os = "cuda", feature(abi_ptx))]
#![cfg_attr(target_os = "cuda", feature(asm_experimental_arch))]
#![cfg_attr(target_os = "cuda", feature(alloc_error_handler))]
#![allow(long_running_const_eval)]
#![recursion_limit = "1024"]

extern crate alloc;

#[cfg(target_os = "cuda")]
use core::ops::ControlFlow;
use core::sync::atomic::AtomicU64;

use necsim_core::{
    cogs::{
        CoalescenceSampler, DispersalSampler, EmigrationExit, Habitat, ImmigrationEntry,
        LineageStore, MathsCore, PrimeableRng, SpeciationProbability, TurnoverRate,
    },
    lineage::Lineage,
    reporter::boolean::Boolean,
    simulation::Simulation,
};
use necsim_core_bond::{NonNegativeF64, PositiveF64};

use necsim_impls_cuda::{event_buffer::EventBuffer, value_buffer::ValueBuffer};
use necsim_impls_no_std::cogs::{
    active_lineage_sampler::singular::SingularActiveLineageSampler,
    event_sampler::tracking::{MinSpeciationTrackingEventSampler, SpeciationSample},
};

use rust_cuda::{
    kernel::param::{DeepPerThreadBorrow, PerThreadShallowCopy, PtxJit, ShallowInteriorMutable},
    lend::RustToCuda,
};

#[rust_cuda::kernel::kernel(pub use link! for impl)]
#[kernel(
    allow(ptx::double_precision_use),
    allow(ptx::local_memory_use), // FIXME
    forbid(ptx::register_spills),
)]
#[allow(clippy::too_many_arguments)]
#[allow(clippy::type_complexity)]
pub fn simulate<
    M: MathsCore + Sync,
    H: Habitat<M> + RustToCuda + Sync,
    G: PrimeableRng<M> + RustToCuda + Sync,
    S: LineageStore<M, H> + RustToCuda + Sync,
    X: EmigrationExit<M, H, G, S> + RustToCuda + Sync,
    D: DispersalSampler<M, H, G> + RustToCuda + Sync,
    C: CoalescenceSampler<M, H, S> + RustToCuda + Sync,
    T: TurnoverRate<M, H> + RustToCuda + Sync,
    N: SpeciationProbability<M, H> + RustToCuda + Sync,
    E: MinSpeciationTrackingEventSampler<M, H, G, S, X, D, C, T, N> + RustToCuda + Sync,
    I: ImmigrationEntry<M> + RustToCuda + Sync,
    A: SingularActiveLineageSampler<M, H, G, S, X, D, C, T, N, E, I> + RustToCuda + Sync,
    ReportSpeciation: Boolean,
    ReportDispersal: Boolean,
>(
    simulation: &PtxJit<DeepPerThreadBorrow<Simulation<M, H, G, S, X, D, C, T, N, E, I, A>>>,
    task_list: &mut PtxJit<DeepPerThreadBorrow<ValueBuffer<Lineage, true, true>>>,
    event_buffer_reporter: &mut PtxJit<
        DeepPerThreadBorrow<EventBuffer<ReportSpeciation, ReportDispersal>>,
    >,
    min_spec_sample_buffer: &mut PtxJit<
        DeepPerThreadBorrow<ValueBuffer<SpeciationSample, false, true>>,
    >,
    next_event_time_buffer: &mut PtxJit<DeepPerThreadBorrow<ValueBuffer<PositiveF64, false, true>>>,
    total_time_max: &ShallowInteriorMutable<AtomicU64>,
    total_steps_sum: &ShallowInteriorMutable<AtomicU64>,
    max_steps: PerThreadShallowCopy<u64>,
    max_next_event_time: PerThreadShallowCopy<NonNegativeF64>,
) {
    // TODO: use simulation with non-allocating clone
    let mut simulation = unsafe { core::mem::ManuallyDrop::new(core::ptr::read(simulation)) };

    task_list.with_value_for_core(|task| {
        // Discard the prior task (the simulation is just a temporary local copy)
        core::mem::drop(
            simulation
                .active_lineage_sampler_mut()
                .replace_active_lineage(task),
        );

        // Discard the prior sample (the simulation is just a temporary local copy)
        simulation.event_sampler_mut().replace_min_speciation(None);

        let mut final_next_event_time = None;

        let (time, steps) = simulation.simulate_incremental_early_stop(
            |_, steps, next_event_time, reporter| {
                final_next_event_time = Some(next_event_time);

                if steps < max_steps
                    && next_event_time < max_next_event_time
                    && reporter.can_buffer_next_event()
                {
                    ControlFlow::Continue(())
                } else {
                    ControlFlow::Break(())
                }
            },
            event_buffer_reporter,
        );

        next_event_time_buffer.put_value_for_core(final_next_event_time);

        if steps > 0 {
            total_time_max.fetch_max(time.get().to_bits(), core::sync::atomic::Ordering::Relaxed);
            total_steps_sum.fetch_add(steps, core::sync::atomic::Ordering::Relaxed);
        }

        min_spec_sample_buffer
            .put_value_for_core(simulation.event_sampler_mut().replace_min_speciation(None));

        simulation
            .active_lineage_sampler_mut()
            .replace_active_lineage(None)
    });
}

#[cfg(target_os = "cuda")]
mod cuda_prelude {
    use rust_cuda::device::alloc::PTXAllocator;

    #[global_allocator]
    static _GLOBAL_ALLOCATOR: PTXAllocator = PTXAllocator;

    #[cfg(not(debug_assertions))]
    #[panic_handler]
    fn panic(_panic_info: &::core::panic::PanicInfo) -> ! {
        rust_cuda::device::utils::abort()
    }

    #[cfg(debug_assertions)]
    #[panic_handler]
    fn panic(info: &::core::panic::PanicInfo) -> ! {
        rust_cuda::device::utils::pretty_print_panic_info(info, true, true);
        rust_cuda::device::utils::abort()
    }

    #[cfg(not(debug_assertions))]
    #[alloc_error_handler]
    fn alloc_error_handler(_: core::alloc::Layout) -> ! {
        rust_cuda::device::utils::abort()
    }

    #[cfg(debug_assertions)]
    #[alloc_error_handler]
    fn alloc_error_handler(layout: core::alloc::Layout) -> ! {
        rust_cuda::device::utils::pretty_print_alloc_error(layout);
        rust_cuda::device::utils::abort()
    }
}