numr 0.5.2

High-performance numerical computing with multi-backend GPU acceleration (CPU/CUDA/WebGPU)
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

//! Advanced PRNGs for reproducible parallel random number generation.
//!
//! Counter-based PRNGs enable reproducible parallel generation: same (key, counter) → same output.
//!
//! # Algorithms
//! - **Philox4x32-10**: JAX/TensorFlow default, excellent GPU performance
//! - **ThreeFry4x64-20**: Cryptographic quality, slower but secure
//! - **PCG64**: NumPy default, CPU-optimized
//! - **Xoshiro256++**: Rust `rand` default, excellent quality/speed
//!
//! # WebGPU Limitations
//! WGSL lacks native u64: uses 32-bit keys for Philox/ThreeFry, emulated u64 for PCG64/Xoshiro.
//! Only F32 supported on WebGPU.

use crate::error::{Error, Result};
use crate::runtime::Runtime;
use crate::tensor::Tensor;

/// Advanced PRNG operations for reproducible parallel random number generation.
///
/// # Reproducibility
/// Same (key, counter) or (seed, stream) → identical output.
/// Different counter/stream → independent sequences.
///
/// # Example
/// ```
/// # use numr::prelude::*;
/// # let device = CpuDevice::new();
/// # let client = CpuRuntime::default_client(&device);
/// use numr::ops::AdvancedRandomOps;
///
/// let samples = client.philox_randn(&[1000], 42, 0, DType::F32)?;
/// let same = client.philox_randn(&[1000], 42, 0, DType::F32)?;  // identical
/// let diff = client.philox_randn(&[1000], 42, 1, DType::F32)?;  // different
/// # Ok::<(), numr::error::Error>(())
/// ```
pub trait AdvancedRandomOps<R: Runtime> {
    /// Generate N(0,1) samples using Philox4x32-10.
    /// Args: shape, key (seed), counter, dtype (F32/F64; WebGPU: F32 only)
    fn philox_randn(
        &self,
        shape: &[usize],
        key: u64,
        counter: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, key, counter, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::philox_randn",
        })
    }

    /// Generate uniform [0,1) samples using Philox4x32-10.
    fn philox_uniform(
        &self,
        shape: &[usize],
        key: u64,
        counter: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, key, counter, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::philox_uniform",
        })
    }

    /// Generate N(0,1) samples using ThreeFry4x64-20 (cryptographic quality).
    fn threefry_randn(
        &self,
        shape: &[usize],
        key: u64,
        counter: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, key, counter, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::threefry_randn",
        })
    }

    /// Generate uniform [0,1) samples using ThreeFry4x64-20.
    fn threefry_uniform(
        &self,
        shape: &[usize],
        key: u64,
        counter: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, key, counter, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::threefry_uniform",
        })
    }

    /// Generate N(0,1) samples using PCG64.
    /// Args: shape, seed, stream (for parallel generation), dtype
    fn pcg64_randn(
        &self,
        shape: &[usize],
        seed: u64,
        stream: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, seed, stream, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::pcg64_randn",
        })
    }

    /// Generate uniform [0,1) samples using PCG64.
    fn pcg64_uniform(
        &self,
        shape: &[usize],
        seed: u64,
        stream: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, seed, stream, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::pcg64_uniform",
        })
    }

    /// Generate N(0,1) samples using Xoshiro256++.
    /// Args: shape, seed, dtype
    fn xoshiro256_randn(
        &self,
        shape: &[usize],
        seed: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, seed, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::xoshiro256_randn",
        })
    }

    /// Generate uniform [0,1) samples using Xoshiro256++.
    fn xoshiro256_uniform(
        &self,
        shape: &[usize],
        seed: u64,
        dtype: crate::dtype::DType,
    ) -> Result<Tensor<R>> {
        let _ = (shape, seed, dtype);
        Err(Error::NotImplemented {
            feature: "AdvancedRandomOps::xoshiro256_uniform",
        })
    }
}