melinoe 0.9.0

Zero-sized, branded, multi-token phantom capabilities for compile-time data-access and thread-synchronization proofs (a generalized evolution of GhostCell) for the Mnemosyne memory ecosystem.
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

use core::num::NonZeroUsize;

/// Shard sizing policy for partitioned scoped-thread execution.
///
/// The plan controls only how a region is tiled into non-empty
/// [`WriterShard`](crate::region::WriterShard)s. It does not introduce locks,
/// atomics, queues, or worker pools; each shard is still moved into one worker
/// and joined before the call returns.
#[derive(Clone, Copy, Debug, PartialEq, Eq)]
pub enum PartitionPlan {
    /// Split into at most this many non-empty shards.
    Parts(NonZeroUsize),
    /// Split into at most `std::thread::available_parallelism()` non-empty
    /// shards, falling back to one shard if the platform cannot report it.
    AvailableParallelism,
    /// Split into non-empty shards containing at most this many cells.
    ChunkSize(NonZeroUsize),
}

impl PartitionPlan {
    /// Create a fixed-part plan, clamping zero to one.
    #[inline]
    #[must_use]
    pub const fn parts(parts: usize) -> Self {
        Self::Parts(nonzero_or_one(parts))
    }

    /// Create a plan based on the process's reported hardware parallelism.
    #[inline]
    #[must_use]
    pub const fn available_parallelism() -> Self {
        Self::AvailableParallelism
    }

    /// Create a fixed-chunk-size plan, clamping zero to one.
    #[inline]
    #[must_use]
    pub const fn chunk_size(chunk_size: usize) -> Self {
        Self::ChunkSize(nonzero_or_one(chunk_size))
    }

    /// Resolve the plan to a concrete per-shard chunk size for a region of
    /// length `len`.
    ///
    /// Empty regions resolve to `1`, so callers can pass the result directly to
    /// slice `chunks`/Melinoe shard chunking without risking a zero chunk size.
    #[inline]
    #[must_use]
    pub fn chunk_len_for(self, len: usize) -> usize {
        match self {
            Self::Parts(parts) => chunk_for_parts(len, parts.get()),
            Self::AvailableParallelism => {
                let parts = std::thread::available_parallelism().map_or(1, NonZeroUsize::get);
                chunk_for_parts(len, parts)
            }
            Self::ChunkSize(chunk_size) => chunk_size.get(),
        }
    }

    /// Resolve the plan for the partition driver.
    ///
    /// The shard *count* is intentionally not computed here: it is derived once,
    /// at the single source of truth, from the [`ShardChunks`](crate::region)
    /// iterator's exact size when the driver reserves worker capacity.
    #[inline]
    pub(super) fn resolve(self, len: usize) -> usize {
        self.chunk_len_for(len)
    }
}

#[inline]
const fn nonzero_or_one(value: usize) -> NonZeroUsize {
    if let Some(nz) = NonZeroUsize::new(value) {
        nz
    } else {
        // SAFETY: 1 is non-zero
        unsafe { NonZeroUsize::new_unchecked(1) }
    }
}

#[inline]
fn chunk_for_parts(len: usize, parts: usize) -> usize {
    if len == 0 {
        1
    } else {
        // ceil(len / parts), clamped to >= 1. Written as
        // `1 + (len - 1) / parts` because `usize::div_ceil` is not in the MSRV
        // and `len + parts - 1` can overflow for adversarial inputs.
        1 + (len - 1) / parts
    }
}