j2k-jpeg-cuda 0.6.1

NVIDIA CUDA GPU JPEG decode adapter for j2k-jpeg
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

// SPDX-License-Identifier: Apache-2.0

use j2k_core::{BackendKind, BackendRequest, PixelFormat};
#[cfg(feature = "cuda-runtime")]
use j2k_cuda_runtime::CudaError;

use crate::surface::Storage;
use crate::{CudaSession, CudaSurfaceStats, Error, Surface};

pub(crate) fn wrap_surface(
    bytes: Vec<u8>,
    dimensions: (u32, u32),
    fmt: PixelFormat,
    backend: BackendRequest,
    session: &mut CudaSession,
) -> Result<Surface, Error> {
    validate_surface_request(backend)?;
    let pitch_bytes = dimensions.0 as usize * fmt.bytes_per_pixel();
    match backend {
        BackendRequest::Cpu | BackendRequest::Auto => {
            if j2k_profile::gpu_route_profile_enabled() {
                let request_s = format!("{backend:?}");
                let fmt_s = format!("{fmt:?}");
                let width_s = dimensions.0.to_string();
                let height_s = dimensions.1.to_string();
                j2k_profile::emit_gpu_route_profile(
                    "jpeg",
                    "cuda",
                    &[
                        ("op", "wrap_surface"),
                        ("request", request_s.as_str()),
                        ("fmt", fmt_s.as_str()),
                        ("width", width_s.as_str()),
                        ("height", height_s.as_str()),
                        ("decision", "host_surface"),
                    ],
                );
            }
            Ok(Surface {
                backend: BackendKind::Cpu,
                dimensions,
                fmt,
                pitch_bytes,
                stats: CudaSurfaceStats::default(),
                storage: Storage::Host(bytes),
            })
        }
        BackendRequest::Cuda => wrap_cuda_surface(&bytes, dimensions, fmt, pitch_bytes, session),
        BackendRequest::Metal => Err(Error::UnsupportedBackend { request: backend }),
    }
}

pub(crate) fn validate_surface_request(backend: BackendRequest) -> Result<(), Error> {
    j2k_core::validate_cuda_surface_backend_request(backend)
        .map_err(|request| Error::UnsupportedBackend { request })
}

#[cfg(feature = "cuda-runtime")]
fn wrap_cuda_surface(
    bytes: &[u8],
    dimensions: (u32, u32),
    fmt: PixelFormat,
    pitch_bytes: usize,
    session: &mut CudaSession,
) -> Result<Surface, Error> {
    let context = session.cuda_context()?;
    let output = context.copy_with_kernel(bytes).map_err(cuda_error)?;
    let (buffer, stats) = output.into_parts();
    if j2k_profile::gpu_route_profile_enabled() {
        let fmt_s = format!("{fmt:?}");
        let width_s = dimensions.0.to_string();
        let height_s = dimensions.1.to_string();
        let kernel_dispatches_s = stats.kernel_dispatches().to_string();
        let copy_dispatches_s = stats.copy_kernel_dispatches().to_string();
        j2k_profile::emit_gpu_route_profile(
            "jpeg",
            "cuda",
            &[
                ("op", "wrap_surface"),
                ("request", "Cuda"),
                ("fmt", fmt_s.as_str()),
                ("width", width_s.as_str()),
                ("height", height_s.as_str()),
                ("decision", "cuda_upload"),
                ("kernel_dispatches", kernel_dispatches_s.as_str()),
                ("copy_kernel_dispatches", copy_dispatches_s.as_str()),
            ],
        );
    }
    Ok(Surface {
        backend: BackendKind::Cuda,
        dimensions,
        fmt,
        pitch_bytes,
        stats: CudaSurfaceStats {
            kernel_dispatches: stats.kernel_dispatches(),
            copy_kernel_dispatches: stats.copy_kernel_dispatches(),
            decode_kernel_dispatches: stats.decode_kernel_dispatches(),
            hardware_decode: stats.used_hardware_decode(),
            decode_path: crate::surface::CudaJpegDecodePath::None,
        },
        storage: Storage::Cuda(buffer),
    })
}

#[cfg(not(feature = "cuda-runtime"))]
fn wrap_cuda_surface(
    _bytes: &[u8],
    dimensions: (u32, u32),
    fmt: PixelFormat,
    _pitch_bytes: usize,
    _session: &mut CudaSession,
) -> Result<Surface, Error> {
    if j2k_profile::gpu_route_profile_enabled() {
        let fmt_s = format!("{fmt:?}");
        let width_s = dimensions.0.to_string();
        let height_s = dimensions.1.to_string();
        j2k_profile::emit_gpu_route_profile(
            "jpeg",
            "cuda",
            &[
                ("op", "wrap_surface"),
                ("request", "Cuda"),
                ("fmt", fmt_s.as_str()),
                ("width", width_s.as_str()),
                ("height", height_s.as_str()),
                ("decision", "cuda_unavailable"),
            ],
        );
    }
    Err(Error::CudaUnavailable)
}

#[cfg(feature = "cuda-runtime")]
#[allow(clippy::needless_pass_by_value)]
pub(crate) fn cuda_error(error: CudaError) -> Error {
    if error.is_unavailable() {
        Error::CudaUnavailable
    } else {
        Error::CudaRuntime {
            message: error.to_string(),
        }
    }
}