tenferro-gpu 0.3.0

CubeCL-backed CUDA and WebGPU provider backends for tenferro tensors.
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

// Run with: cargo test --features cuda -- --ignored
use num_complex::{Complex32, Complex64};
use std::num::NonZeroUsize;

use crate::DotGeneralConfig;
use crate::Tensor;
use tenferro_tensor::TensorDot;

use super::super::gemm::cutensor_plan_cache_workspace_bytes;
use super::{
    assert_tensor_close, cpu_backend, download, gpu_backend, tensor_c32, tensor_c64, tensor_f32,
    tensor_f64, upload,
};

fn run_dot_general_case(lhs: Tensor, rhs: Tensor, config: DotGeneralConfig, tol: f64) {
    let mut cpu = cpu_backend();
    let mut gpu = gpu_backend();

    let expected = cpu.dot_general(&lhs, &rhs, &config).unwrap();
    let gpu_lhs = upload(&gpu, &lhs);
    let gpu_rhs = upload(&gpu, &rhs);
    let actual_gpu = gpu.dot_general(&gpu_lhs, &gpu_rhs, &config).unwrap();
    let actual = download(&gpu, &actual_gpu);

    assert_eq!(actual.shape(), expected.shape());
    assert_tensor_close(&actual, &expected, tol);
}

#[test]
#[ignore = "requires CUDA 12.8+ GPU"]
fn cuda_cutensor_cache_eviction_keeps_inflight_workspace_valid() {
    let mut cpu = cpu_backend();
    let mut gpu = gpu_backend();
    gpu.set_cutensor_plan_cache_max_entries(NonZeroUsize::new(1).unwrap())
        .unwrap();

    let lhs_a = tensor_f32(vec![64, 64], vec![1.0; 64 * 64]);
    let rhs_a = tensor_f32(vec![64, 64], vec![1.0; 64 * 64]);
    let lhs_b = tensor_f32(vec![65, 64], vec![1.0; 65 * 64]);
    let rhs_b = tensor_f32(vec![64, 65], vec![1.0; 64 * 65]);
    let expected_a = cpu.dot_general(&lhs_a, &rhs_a, &matmul_config()).unwrap();
    let expected_b = cpu.dot_general(&lhs_b, &rhs_b, &matmul_config()).unwrap();

    let gpu_lhs_a = upload(&gpu, &lhs_a);
    let gpu_rhs_a = upload(&gpu, &rhs_a);
    let gpu_lhs_b = upload(&gpu, &lhs_b);
    let gpu_rhs_b = upload(&gpu, &rhs_b);
    let actual_a = gpu
        .dot_general(&gpu_lhs_a, &gpu_rhs_a, &matmul_config())
        .unwrap();
    assert!(
        cutensor_plan_cache_workspace_bytes(&gpu).unwrap() > 0,
        "first contraction must retain a nonzero cuTENSOR workspace"
    );

    let actual_b = gpu
        .dot_general(&gpu_lhs_b, &gpu_rhs_b, &matmul_config())
        .unwrap();
    let cache_stats = gpu.cutensor_plan_cache_stats().unwrap();
    assert_eq!(cache_stats.entries, 1);
    assert!(
        cache_stats.evictions > 0,
        "second contraction must evict the first plan"
    );

    // Eviction drops the first plan while its launch may still be queued. The
    // pinned CubeCL CUDA allocator retires async frees on the owning stream.
    gpu.runtime().synchronize().unwrap();
    assert_tensor_close(&download(&gpu, &actual_a), &expected_a, 1e-4);
    assert_tensor_close(&download(&gpu, &actual_b), &expected_b, 1e-4);
}

fn matmul_config() -> DotGeneralConfig {
    DotGeneralConfig {
        lhs_contracting_dims: vec![1],
        rhs_contracting_dims: vec![0],
        lhs_batch_dims: vec![],
        rhs_batch_dims: vec![],
    }
}

#[test]
#[ignore]
fn test_dot_general_matmul_f32() {
    run_dot_general_case(
        tensor_f32(vec![2, 3], vec![1.0, 4.0, 2.0, 5.0, 3.0, 6.0]),
        tensor_f32(
            vec![3, 4],
            vec![
                1.0, 5.0, 9.0, 2.0, 6.0, 10.0, 3.0, 7.0, 11.0, 4.0, 8.0, 12.0,
            ],
        ),
        matmul_config(),
        1e-4,
    );
}

#[test]
#[ignore]
fn test_dot_general_matmul_f64() {
    run_dot_general_case(
        tensor_f64(vec![2, 3], vec![1.0, 4.0, 2.0, 5.0, 3.0, 6.0]),
        tensor_f64(
            vec![3, 4],
            vec![
                1.0, 5.0, 9.0, 2.0, 6.0, 10.0, 3.0, 7.0, 11.0, 4.0, 8.0, 12.0,
            ],
        ),
        matmul_config(),
        1e-9,
    );
}

#[test]
#[ignore]
fn test_dot_general_batched_matmul_f32() {
    run_dot_general_case(
        tensor_f32(
            vec![2, 3, 5],
            vec![
                1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 3.0, 4.0, 5.0, 6.0,
                7.0, 8.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0,
            ],
        ),
        tensor_f32(
            vec![3, 4, 5],
            vec![
                1.0, 2.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 2.0, 3.0, 4.0, 5.0,
                6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 3.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0,
                10.0, 11.0, 12.0, 13.0, 14.0, 4.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0,
                14.0, 15.0, 5.0, 6.0, 7.0, 8.0, 9.0, 10.0, 11.0, 12.0, 13.0, 14.0, 15.0, 16.0,
            ],
        ),
        DotGeneralConfig {
            lhs_contracting_dims: vec![1],
            rhs_contracting_dims: vec![0],
            lhs_batch_dims: vec![2],
            rhs_batch_dims: vec![2],
        },
        1e-4,
    );
}

#[test]
#[ignore]
fn test_dot_general_complex_matmul_c32() {
    run_dot_general_case(
        tensor_c32(
            vec![2, 3],
            vec![
                Complex32::new(1.0, 0.5),
                Complex32::new(2.0, -1.0),
                Complex32::new(3.0, 0.25),
                Complex32::new(4.0, 1.0),
                Complex32::new(5.0, -0.5),
                Complex32::new(6.0, 0.75),
            ],
        ),
        tensor_c32(
            vec![3, 4],
            vec![
                Complex32::new(1.0, -1.0),
                Complex32::new(2.0, 0.25),
                Complex32::new(3.0, 0.5),
                Complex32::new(4.0, -0.75),
                Complex32::new(5.0, 0.0),
                Complex32::new(6.0, 1.0),
                Complex32::new(7.0, -0.5),
                Complex32::new(8.0, 0.75),
                Complex32::new(9.0, -1.25),
                Complex32::new(10.0, 0.5),
                Complex32::new(11.0, 1.5),
                Complex32::new(12.0, -0.25),
            ],
        ),
        matmul_config(),
        1e-4,
    );
}

#[test]
#[ignore]
fn test_dot_general_complex_matmul_c64() {
    run_dot_general_case(
        tensor_c64(
            vec![2, 3],
            vec![
                Complex64::new(1.0, 0.5),
                Complex64::new(2.0, -1.0),
                Complex64::new(3.0, 0.25),
                Complex64::new(4.0, 1.0),
                Complex64::new(5.0, -0.5),
                Complex64::new(6.0, 0.75),
            ],
        ),
        tensor_c64(
            vec![3, 4],
            vec![
                Complex64::new(1.0, -1.0),
                Complex64::new(2.0, 0.25),
                Complex64::new(3.0, 0.5),
                Complex64::new(4.0, -0.75),
                Complex64::new(5.0, 0.0),
                Complex64::new(6.0, 1.0),
                Complex64::new(7.0, -0.5),
                Complex64::new(8.0, 0.75),
                Complex64::new(9.0, -1.25),
                Complex64::new(10.0, 0.5),
                Complex64::new(11.0, 1.5),
                Complex64::new(12.0, -0.25),
            ],
        ),
        matmul_config(),
        1e-9,
    );
}