j2k-cuda 0.8.0

CUDA adapter for resident HTJ2K decode/encode and shared JPEG 2000 stages
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
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249

// SPDX-License-Identifier: MIT OR Apache-2.0

use std::sync::Arc;

use j2k::{
    prepare_batch, BatchDecodeOptions, CpuBatchDecoder, CpuBatchSamples, DecodeRequest, Downscale,
    EncodedImage, PreparedBatchGroup, Rect,
};
use j2k_cuda::{CudaBatchDecoder, CudaExternalBatchTryFinish, CudaSession};
use j2k_cuda_runtime::{CudaContext, CudaExternalDeviceBufferViewMut};

use super::support::submit_external_for_test;

#[test]
fn external_batch_handles_roi_reduction_and_drop_safe_session_reuse_when_runtime_required() {
    if !j2k_test_support::cuda_runtime_gate(module_path!()) {
        return;
    }

    let pixels = (0_u8..=255).collect::<Vec<_>>();
    let encoded = Arc::<[u8]>::from(
        j2k_native::encode_htj2k(
            &pixels,
            16,
            16,
            1,
            8,
            false,
            &j2k_native::EncodeOptions {
                reversible: true,
                num_decomposition_levels: 2,
                ..j2k_native::EncodeOptions::default()
            },
        )
        .expect("encode HTJ2K ROI/reduction fixture"),
    );
    let inputs = vec![
        EncodedImage::new(
            Arc::clone(&encoded),
            DecodeRequest::Region {
                roi: Rect {
                    x: 4,
                    y: 4,
                    w: 8,
                    h: 8,
                },
            },
        ),
        EncodedImage::new(
            encoded,
            DecodeRequest::Reduced {
                scale: Downscale::Half,
            },
        ),
    ];
    let options = BatchDecodeOptions::default();
    let prepared = prepare_batch(inputs, options).expect("prepare shared batch");
    assert!(prepared.errors().is_empty());
    assert_eq!(prepared.groups().len(), 2);
    assert_eq!(prepared.groups()[0].source_indices(), &[0]);
    assert_eq!(prepared.groups()[1].source_indices(), &[1]);

    let mut cpu = CpuBatchDecoder::new(options);
    let cpu_output = cpu
        .decode_prepared(&prepared)
        .expect("CPU oracle batch decode");
    assert_eq!(cpu_output.groups().len(), prepared.groups().len());

    let context = CudaContext::system_default().expect("CUDA context");
    let session = CudaSession::with_context(context.clone());
    let mut decoder = CudaBatchDecoder::with_session_and_options(session, options);
    for (prepared_group, oracle_group) in prepared.groups().iter().zip(cpu_output.groups().iter()) {
        assert_eq!(
            prepared_group.source_indices(),
            oracle_group.source_indices()
        );
        let expected = match oracle_group.samples() {
            CpuBatchSamples::U8(samples) => samples,
            samples => panic!("expected U8 CPU oracle, got {samples:?}"),
        };
        assert_drop_retirement_and_reuse(&context, &mut decoder, prepared_group, expected);
    }
}

fn assert_drop_retirement_and_reuse(
    context: &CudaContext,
    decoder: &mut CudaBatchDecoder,
    prepared_group: &PreparedBatchGroup,
    expected: &[u8],
) {
    let mut allocation = context
        .allocate(expected.len())
        .expect("external destination allocation");

    // The first submission is intentionally dropped unfinished. Drop must
    // retire its resources before the persistent session is reused.
    {
        let ptr = allocation.device_ptr();
        let len = allocation.byte_len();
        // SAFETY: `allocation` is exclusively borrowed through submission.
        let mut destination = unsafe {
            CudaExternalDeviceBufferViewMut::from_raw_parts(
                context,
                ptr,
                len,
                std::mem::align_of::<u8>(),
                &mut allocation,
            )
        }
        .expect("external destination view");
        // SAFETY: dropping the completion owner retires work before reuse.
        let submitted = unsafe { decoder.submit_batch_into(prepared_group, &mut destination) }
            .expect("submit external batch for drop retirement");
        drop(submitted);
    }

    let submitted = {
        let ptr = allocation.device_ptr();
        let len = allocation.byte_len();
        // SAFETY: the allocation remains live and inaccessible until the
        // returned completion owner is retired below.
        let mut destination = unsafe {
            CudaExternalDeviceBufferViewMut::from_raw_parts(
                context,
                ptr,
                len,
                std::mem::align_of::<u8>(),
                &mut allocation,
            )
        }
        .expect("reused external destination view");
        // SAFETY: there is no overlapping access before completion.
        unsafe { decoder.submit_batch_into(prepared_group, &mut destination) }
            .expect("submit external batch after drop retirement")
    };
    let group = match submitted.try_finish().expect("poll external batch") {
        CudaExternalBatchTryFinish::Pending(submitted) => {
            submitted.wait().expect("wait external batch")
        }
        CudaExternalBatchTryFinish::Complete(group) => group,
    };
    assert_eq!(group.source_indices(), prepared_group.source_indices());
    assert_eq!(group.ranges().len(), 1);

    let mut actual = vec![0_u8; expected.len()];
    allocation
        .copy_to_host(&mut actual)
        .expect("download completed external output");
    assert_eq!(actual, expected);
}

#[test]
fn mixed_decomposition_shapes_keep_async_pool_resources_live_when_runtime_required() {
    if !j2k_test_support::cuda_runtime_gate(module_path!()) {
        return;
    }

    let pixels_a = (0_u8..=255).collect::<Vec<_>>();
    let pixels_b = pixels_a
        .iter()
        .map(|value| value.wrapping_mul(29).wrapping_add(7))
        .collect::<Vec<_>>();
    let encode = |pixels: &[u8], levels| {
        Arc::<[u8]>::from(
            j2k_native::encode_htj2k(
                pixels,
                16,
                16,
                1,
                8,
                false,
                &j2k_native::EncodeOptions {
                    reversible: true,
                    num_decomposition_levels: levels,
                    ..j2k_native::EncodeOptions::default()
                },
            )
            .expect("encode mixed-decomposition HTJ2K fixture"),
        )
    };
    let prepared = prepare_batch(
        vec![
            EncodedImage::full(encode(&pixels_a, 1)),
            EncodedImage::full(encode(&pixels_b, 2)),
        ],
        BatchDecodeOptions::default(),
    )
    .expect("prepare mixed-decomposition batch");
    assert_eq!(prepared.groups().len(), 2);
    assert_eq!(prepared.groups()[0].source_indices(), [0]);
    assert_eq!(prepared.groups()[1].source_indices(), [1]);

    let mut cpu = CpuBatchDecoder::new(BatchDecodeOptions::default());
    let oracle = cpu
        .decode_prepared(&prepared)
        .expect("CPU mixed-decomposition oracle");
    let expected = oracle
        .groups()
        .iter()
        .map(|group| match group.samples() {
            CpuBatchSamples::U8(samples) => samples,
            samples => panic!("expected U8 CPU oracle, got {samples:?}"),
        })
        .collect::<Vec<_>>();

    let context = CudaContext::system_default().expect("CUDA context");
    let session = CudaSession::with_context(context.clone());
    let mut decoder = CudaBatchDecoder::with_session(session);
    let mut first_output = context
        .allocate(expected[0].len())
        .expect("first mixed-shape output");
    let mut second_output = context
        .allocate(expected[1].len())
        .expect("second mixed-shape output");
    // SAFETY: both allocations remain live and untouched until both pending
    // owners are waited below. Distinct destinations do not overlap.
    let first = unsafe {
        submit_external_for_test(
            &mut decoder,
            &prepared.groups()[0],
            &context,
            &mut first_output,
        )
    };
    // Submit again before waiting the first batch. This catches recycling of
    // IDWT scratch still referenced by the first default-stream graph.
    // SAFETY: same lifetime guarantee as the first distinct allocation.
    let second = unsafe {
        submit_external_for_test(
            &mut decoder,
            &prepared.groups()[1],
            &context,
            &mut second_output,
        )
    };
    second.wait().expect("wait second mixed-shape batch");
    first.wait().expect("wait first mixed-shape batch");

    let mut first_actual = vec![0_u8; expected[0].len()];
    let mut second_actual = vec![0_u8; expected[1].len()];
    first_output
        .copy_to_host(&mut first_actual)
        .expect("download first mixed-shape batch");
    second_output
        .copy_to_host(&mut second_actual)
        .expect("download second mixed-shape batch");
    assert_eq!(first_actual, *expected[0]);
    assert_eq!(second_actual, *expected[1]);
}