diffsol-la 0.3.0

Linear algebra traits and backends (vectors, matrices, linear solvers) for diffsol.
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

use cudarc::driver::{DeviceRepr, ValidAsZeroBits};

use super::Scalar;

/// `cuda_core::DeviceBuffer<T>` needs `T: DeviceCopy`; a no-op without `cuda-oxide`.
#[cfg(feature = "cuda-oxide")]
pub use cuda_core::DeviceCopy as OxideCopy;
#[cfg(not(feature = "cuda-oxide"))]
pub trait OxideCopy {}
#[cfg(not(feature = "cuda-oxide"))]
impl<T> OxideCopy for T {}

pub enum CudaType {
    F32,
    F64,
}

pub trait ScalarCuda: Scalar + ValidAsZeroBits + DeviceRepr + OxideCopy {
    fn as_enum() -> CudaType;
    fn as_f64(self) -> f64 {
        panic!("Unsupported type for as_f64");
    }
    fn as_str() -> &'static str {
        match Self::as_enum() {
            CudaType::F32 => "f32",
            CudaType::F64 => "f64",
        }
    }

    /// Device-only: `warp::shuffle_xor_*` for this type.
    #[cfg(feature = "cuda-oxide")]
    fn shuffle_xor(self, lane_mask: u32) -> Self;
    /// Device-only: full-warp sum, `warp::reduce_sum_*` for this type.
    #[cfg(feature = "cuda-oxide")]
    fn warp_reduce_sum(self) -> Self;
    /// Bit pattern whose unsigned order matches the value's order for `self >= 0`,
    /// read back on the host with `f64::from_bits`.
    #[cfg(feature = "cuda-oxide")]
    fn to_max_bits(self) -> u64;
}

impl ScalarCuda for f64 {
    fn as_enum() -> CudaType {
        CudaType::F64
    }
    fn as_f64(self) -> f64 {
        self
    }

    #[cfg(feature = "cuda-oxide")]
    #[inline(always)]
    fn shuffle_xor(self, lane_mask: u32) -> Self {
        cuda_device::warp::shuffle_xor_f64(self, lane_mask)
    }
    #[cfg(feature = "cuda-oxide")]
    #[inline(always)]
    fn warp_reduce_sum(self) -> Self {
        cuda_device::warp::reduce_sum_f64(self)
    }
    #[cfg(feature = "cuda-oxide")]
    #[inline(always)]
    fn to_max_bits(self) -> u64 {
        self.to_bits()
    }
}

impl ScalarCuda for f32 {
    fn as_enum() -> CudaType {
        CudaType::F32
    }
    fn as_f64(self) -> f64 {
        self as f64
    }

    #[cfg(feature = "cuda-oxide")]
    #[inline(always)]
    fn shuffle_xor(self, lane_mask: u32) -> Self {
        cuda_device::warp::shuffle_xor_f32(self, lane_mask)
    }
    #[cfg(feature = "cuda-oxide")]
    #[inline(always)]
    fn warp_reduce_sum(self) -> Self {
        cuda_device::warp::reduce_sum_f32(self)
    }
    #[cfg(feature = "cuda-oxide")]
    #[inline(always)]
    fn to_max_bits(self) -> u64 {
        (self as f64).to_bits()
    }
}