tenferro-ad 0.2.0

Eager runtime, eager tensors, and traced AD extension traits for tenferro.
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

use tenferro_ad::TracedTensorAdExt;
use tenferro_cpu::CpuBackend;
use tenferro_runtime::error::Error;
use tenferro_runtime::extension::{ExecInstruction, ExecOp, ExecProgram};
use tenferro_runtime::{DType, GraphCompiler, GraphExecutor, Tensor, TracedTensor};

#[test]
fn graph_executor_runs_compiled_single_output_program() {
    let x = TracedTensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
    let y = (&x + &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let program = compiler.compile(&y).unwrap();
    let mut executor = GraphExecutor::new(CpuBackend::new());

    let out = executor.run(&program).unwrap();

    assert_eq!(out.as_slice::<f64>().unwrap(), &[2.0, 4.0]);
}

#[test]
fn graph_executor_runs_compiled_multi_output_program() {
    let x = TracedTensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
    let sum = (&x + &x).unwrap();
    let product = (&x * &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let program = compiler.compile_many(&[&sum, &product]).unwrap();
    let mut executor = GraphExecutor::new(CpuBackend::new());

    let outputs = executor.run_many(&program).unwrap();

    assert_eq!(outputs.len(), 2);
    assert_eq!(outputs[0].as_slice::<f64>().unwrap(), &[2.0, 4.0]);
    assert_eq!(outputs[1].as_slice::<f64>().unwrap(), &[1.0, 4.0]);
}

#[test]
fn checkpoint_uses_explicit_compiler_and_executor() {
    let x = TracedTensor::from_vec_col_major(vec![], vec![3.0_f64]).unwrap();
    let mut y = (&x * &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let mut executor = GraphExecutor::new(CpuBackend::new());
    y.checkpoint(&mut compiler, &mut executor).unwrap();

    let program = compiler.compile(&y).unwrap();
    let out = executor.run(&program).unwrap();
    assert_eq!(out.as_slice::<f64>().unwrap(), &[9.0]);
}

#[test]
fn checkpoint_reuses_existing_cached_data_without_recompiling() {
    let x = TracedTensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
    let mut y = (&x + &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let mut executor = GraphExecutor::new(CpuBackend::new());
    y.checkpoint(&mut compiler, &mut executor).unwrap();

    y.checkpoint(&mut compiler, &mut executor).unwrap();

    let program = compiler.compile(&y).unwrap();
    let out = executor.run(&program).unwrap();
    assert_eq!(out.as_slice::<f64>().unwrap(), &[2.0, 4.0]);
}

#[test]
fn checkpoint_gradient_runs_through_graph_executor() {
    let x = TracedTensor::from_vec_col_major(vec![], vec![2.0_f64]).unwrap();
    let mut y = (&x * &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let mut executor = GraphExecutor::new(CpuBackend::new());
    y.checkpoint(&mut compiler, &mut executor).unwrap();

    let z = (&y * &y).unwrap();
    let grad = z.grad(&x).unwrap();
    let program = compiler.compile(&grad).unwrap();
    let out = executor.run(&program).unwrap();

    assert_eq!(out.as_slice::<f64>().unwrap(), &[32.0]);
}

#[test]
fn graph_executor_validates_runtime_bindings() {
    let x = TracedTensor::input_symbolic_shape(DType::F64, 1).unwrap();
    let y = (&x + &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let program = compiler
        .compile_with_input_specs(&y, &[(&x, DType::F64, &[3])])
        .unwrap();
    let mut executor = GraphExecutor::new(CpuBackend::new());

    let ok = Tensor::from_vec_col_major(vec![3], vec![1.0_f64, 2.0, 3.0]).unwrap();
    let out = executor.run_with_inputs(&program, &[(&x, &ok)]).unwrap();
    assert_eq!(out.as_slice::<f64>().unwrap(), &[2.0, 4.0, 6.0]);

    let wrong_shape = Tensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
    let err = executor
        .run_with_inputs(&program, &[(&x, &wrong_shape)])
        .unwrap_err();
    assert!(format!("{err}").contains("shape"));
}

#[test]
fn graph_executor_rejects_invalid_runtime_bindings() {
    let x = TracedTensor::input_symbolic_shape(DType::F64, 1).unwrap();
    let y = (&x + &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let program = compiler
        .compile_with_input_specs(&y, &[(&x, DType::F64, &[2])])
        .unwrap();
    let mut executor = GraphExecutor::new(CpuBackend::new());

    let bound = Tensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
    let err = executor
        .run_with_inputs(&program, &[(&x, &bound), (&x, &bound)])
        .unwrap_err();
    assert!(matches!(err, Error::DuplicateBinding { .. }), "got {err:?}");

    let data_leaf = TracedTensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
    let err = executor
        .run_with_inputs(&program, &[(&data_leaf, &bound)])
        .unwrap_err();
    assert!(
        matches!(err, Error::UnexpectedBinding { binding_index: 0 }),
        "got {err:?}"
    );

    let other = TracedTensor::input_symbolic_shape(DType::F64, 1).unwrap();
    let err = executor
        .run_with_inputs(&program, &[(&other, &bound)])
        .unwrap_err();
    assert!(
        matches!(err, Error::UnexpectedBinding { binding_index: 0 }),
        "got {err:?}"
    );

    let wrong_dtype = Tensor::from_vec_col_major(vec![2], vec![1.0_f32, 2.0]).unwrap();
    let err = executor
        .run_with_inputs(&program, &[(&x, &wrong_dtype)])
        .unwrap_err();
    assert!(
        matches!(
            err,
            Error::PlaceholderDtypeMismatch {
                expected: DType::F64,
                actual: DType::F32
            }
        ),
        "got {err:?}"
    );

    let err = executor.run_with_inputs(&program, &[]).unwrap_err();
    assert!(
        matches!(err, Error::UnboundPlaceholder { .. }),
        "got {err:?}"
    );
}

#[test]
fn graph_executor_eval_exec_ir_rejects_wrong_input_count() {
    let program = ExecProgram {
        instructions: vec![ExecInstruction {
            op: ExecOp::Add,
            input_slots: vec![0, 1],
            output_slots: vec![2],
            dtype: DType::F64,
            output_shapes: vec![vec![]].into(),
            output_extents: vec![vec![]].into(),
            last_use: vec![true, true],
        }],
        input_slots: vec![0, 1],
        output_slots: vec![2],
        n_slots: 3,
    };
    let inputs = vec![
        Tensor::from_vec_col_major(vec![], vec![2.0_f64]).unwrap(),
        Tensor::from_vec_col_major(vec![], vec![3.0_f64]).unwrap(),
        Tensor::from_vec_col_major(vec![], vec![5.0_f64]).unwrap(),
    ];

    let mut executor = GraphExecutor::new(CpuBackend::new());
    let err = executor.eval_exec_ir(&program, inputs).unwrap_err();

    assert!(
        format!("{err}").contains("expected 2 inputs"),
        "got {err:?}"
    );
}

#[test]
fn graph_executor_cache_stats_are_separate_from_compiler_stats() {
    let x = TracedTensor::from_vec_col_major(vec![2], vec![1.0_f64, 2.0]).unwrap();
    let y = (&x + &x).unwrap();

    let mut compiler = GraphCompiler::new();
    let program = compiler.compile(&y).unwrap();
    let mut executor = GraphExecutor::new(CpuBackend::new());
    let _ = executor.run(&program).unwrap();

    assert!(compiler.cache_stats().compile.entries > 0);
    assert_eq!(executor.cache_stats().extensions.entries, 0);
}

#[test]
fn graph_executor_runtime_cache_controls_are_available() {
    let mut executor = GraphExecutor::new(CpuBackend::new());

    let stats = executor.cache_stats();
    assert_eq!(stats.extensions.entries, 0);
    assert_eq!(stats.backend.entries, 0);

    executor.clear_backend_cache();
    assert_eq!(executor.cache_stats().backend.entries, 0);

    executor.clear_caches();
    assert_eq!(executor.cache_stats().extensions.entries, 0);
    assert_eq!(executor.cache_stats().backend.entries, 0);
}

#[test]
fn graph_executor_synthesizes_deferred_zero_tangents_from_primal_binding() {
    let x = TracedTensor::input_symbolic_shape(DType::F64, 1).unwrap();
    let loss = (&x * &x).unwrap().reduce_sum(&[0]).unwrap();
    let grad = loss.grad(&x).unwrap();

    let mut compiler = GraphCompiler::new();
    let program = compiler
        .compile_with_input_specs(&grad, &[(&x, DType::F64, &[4])])
        .unwrap();
    let mut executor = GraphExecutor::new(CpuBackend::new());

    let bound = Tensor::from_vec_col_major(vec![4], vec![1.0_f64, 2.0, 3.0, 4.0]).unwrap();
    let out = executor.run_with_inputs(&program, &[(&x, &bound)]).unwrap();

    assert_eq!(out.shape(), &[4]);
    assert_eq!(out.as_slice::<f64>().unwrap(), &[2.0, 4.0, 6.0, 8.0]);
}