numr 0.5.1

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
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187

//! ILU(0) and IC(0) factorization kernel launchers

use cudarc::driver::PushKernelArg;
use cudarc::driver::safe::{CudaContext, CudaStream};
use std::sync::Arc;

use super::{
    BLOCK_SIZE, SPARSE_LINALG_MODULE, get_kernel_function, get_or_load_module, grid_size,
    launch_config, launch_error,
};
use crate::error::Result;

// ============================================================================
// ILU(0) Level Kernel Launchers
// ============================================================================

/// Launch ILU(0) factorization level kernel - f32
///
/// # Safety
///
/// - `level_rows`, `row_ptrs`, `col_indices`, `values`, and `diag_indices` must be valid device
///   memory pointers on the device associated with `context`.
/// - `level_rows` must have at least `level_size` elements with valid row indices in `[0, n)`.
/// - `row_ptrs` must have at least `n + 1` elements; `col_indices`, `values`, and `diag_indices`
///   must each have at least `nnz` elements (as encoded in `row_ptrs`).
/// - The stream must be from the same context and must not be destroyed while the kernel runs.
#[allow(clippy::too_many_arguments)]
pub unsafe fn launch_ilu0_level_f32(
    context: &Arc<CudaContext>,
    stream: &CudaStream,
    device_index: usize,
    level_rows: u64,
    level_size: i32,
    row_ptrs: u64,
    col_indices: u64,
    values: u64,
    diag_indices: u64,
    n: i32,
    diagonal_shift: f32,
) -> Result<()> {
    let module = get_or_load_module(context, device_index, SPARSE_LINALG_MODULE)?;
    let func = get_kernel_function(&module, "ilu0_level_f32")?;
    let cfg = launch_config((grid_size(level_size as u32), 1, 1), (BLOCK_SIZE, 1, 1), 0);

    let mut builder = stream.launch_builder(&func);
    builder.arg(&level_rows);
    builder.arg(&level_size);
    builder.arg(&row_ptrs);
    builder.arg(&col_indices);
    builder.arg(&values);
    builder.arg(&diag_indices);
    builder.arg(&n);
    builder.arg(&diagonal_shift);
    // SAFETY: All pointers are valid device pointers with correct sizes (ensured by caller)
    unsafe { builder.launch(cfg) }.map_err(|e| launch_error("ilu0_level_f32", e))?;
    Ok(())
}

/// Launch ILU(0) factorization level kernel - f64
///
/// # Safety
///
/// - `level_rows`, `row_ptrs`, `col_indices`, `values`, and `diag_indices` must be valid device
///   memory pointers on the device associated with `context`.
/// - `level_rows` must have at least `level_size` elements with valid row indices in `[0, n)`.
/// - `row_ptrs` must have at least `n + 1` elements; `col_indices`, `values`, and `diag_indices`
///   must each have at least `nnz` elements (as encoded in `row_ptrs`).
/// - The stream must be from the same context and must not be destroyed while the kernel runs.
#[allow(clippy::too_many_arguments)]
pub unsafe fn launch_ilu0_level_f64(
    context: &Arc<CudaContext>,
    stream: &CudaStream,
    device_index: usize,
    level_rows: u64,
    level_size: i32,
    row_ptrs: u64,
    col_indices: u64,
    values: u64,
    diag_indices: u64,
    n: i32,
    diagonal_shift: f64,
) -> Result<()> {
    let module = get_or_load_module(context, device_index, SPARSE_LINALG_MODULE)?;
    let func = get_kernel_function(&module, "ilu0_level_f64")?;
    let cfg = launch_config((grid_size(level_size as u32), 1, 1), (BLOCK_SIZE, 1, 1), 0);

    let mut builder = stream.launch_builder(&func);
    builder.arg(&level_rows);
    builder.arg(&level_size);
    builder.arg(&row_ptrs);
    builder.arg(&col_indices);
    builder.arg(&values);
    builder.arg(&diag_indices);
    builder.arg(&n);
    builder.arg(&diagonal_shift);
    // SAFETY: All pointers are valid device pointers with correct sizes (ensured by caller)
    unsafe { builder.launch(cfg) }.map_err(|e| launch_error("ilu0_level_f64", e))?;
    Ok(())
}

// ============================================================================
// IC(0) Level Kernel Launchers
// ============================================================================

/// Launch IC(0) factorization level kernel - f32
///
/// # Safety
///
/// - `level_rows`, `row_ptrs`, `col_indices`, `values`, and `diag_indices` must be valid device
///   memory pointers on the device associated with `context`.
/// - `level_rows` must have at least `level_size` elements with valid row indices in `[0, n)`.
/// - `row_ptrs` must have at least `n + 1` elements; `col_indices`, `values`, and `diag_indices`
///   must each have at least `nnz` elements (as encoded in `row_ptrs`).
/// - The stream must be from the same context and must not be destroyed while the kernel runs.
#[allow(clippy::too_many_arguments)]
pub unsafe fn launch_ic0_level_f32(
    context: &Arc<CudaContext>,
    stream: &CudaStream,
    device_index: usize,
    level_rows: u64,
    level_size: i32,
    row_ptrs: u64,
    col_indices: u64,
    values: u64,
    diag_indices: u64,
    n: i32,
    diagonal_shift: f32,
) -> Result<()> {
    let module = get_or_load_module(context, device_index, SPARSE_LINALG_MODULE)?;
    let func = get_kernel_function(&module, "ic0_level_f32")?;
    let cfg = launch_config((grid_size(level_size as u32), 1, 1), (BLOCK_SIZE, 1, 1), 0);

    let mut builder = stream.launch_builder(&func);
    builder.arg(&level_rows);
    builder.arg(&level_size);
    builder.arg(&row_ptrs);
    builder.arg(&col_indices);
    builder.arg(&values);
    builder.arg(&diag_indices);
    builder.arg(&n);
    builder.arg(&diagonal_shift);
    // SAFETY: All pointers are valid device pointers with correct sizes (ensured by caller)
    unsafe { builder.launch(cfg) }.map_err(|e| launch_error("ic0_level_f32", e))?;
    Ok(())
}

/// Launch IC(0) factorization level kernel - f64
///
/// # Safety
///
/// - `level_rows`, `row_ptrs`, `col_indices`, `values`, and `diag_indices` must be valid device
///   memory pointers on the device associated with `context`.
/// - `level_rows` must have at least `level_size` elements with valid row indices in `[0, n)`.
/// - `row_ptrs` must have at least `n + 1` elements; `col_indices`, `values`, and `diag_indices`
///   must each have at least `nnz` elements (as encoded in `row_ptrs`).
/// - The stream must be from the same context and must not be destroyed while the kernel runs.
#[allow(clippy::too_many_arguments)]
pub unsafe fn launch_ic0_level_f64(
    context: &Arc<CudaContext>,
    stream: &CudaStream,
    device_index: usize,
    level_rows: u64,
    level_size: i32,
    row_ptrs: u64,
    col_indices: u64,
    values: u64,
    diag_indices: u64,
    n: i32,
    diagonal_shift: f64,
) -> Result<()> {
    let module = get_or_load_module(context, device_index, SPARSE_LINALG_MODULE)?;
    let func = get_kernel_function(&module, "ic0_level_f64")?;
    let cfg = launch_config((grid_size(level_size as u32), 1, 1), (BLOCK_SIZE, 1, 1), 0);

    let mut builder = stream.launch_builder(&func);
    builder.arg(&level_rows);
    builder.arg(&level_size);
    builder.arg(&row_ptrs);
    builder.arg(&col_indices);
    builder.arg(&values);
    builder.arg(&diag_indices);
    builder.arg(&n);
    builder.arg(&diagonal_shift);
    // SAFETY: All pointers are valid device pointers with correct sizes (ensured by caller)
    unsafe { builder.launch(cfg) }.map_err(|e| launch_error("ic0_level_f64", e))?;
    Ok(())
}