oxionnx 0.1.2

Pure Rust ONNX inference engine — zero C/C++ dependencies
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
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304

//! Benchmark-based execution path selection.
//!
//! Times CPU vs GPU execution for specific operators and recommends
//! the faster path based on input size.

use std::fmt;
use std::time::{Duration, Instant};

use oxionnx_core::Tensor;

/// Result of benchmarking a single operator.
#[derive(Debug, Clone)]
pub struct BenchmarkResult {
    /// The ONNX operator type (e.g. "MatMul", "Relu").
    pub op_type: String,
    /// Dimensions describing the benchmark input.
    pub input_shape: Vec<usize>,
    /// Time spent executing on CPU.
    pub cpu_time: Duration,
    /// Time spent executing on GPU, if available.
    pub gpu_time: Option<Duration>,
    /// Recommended execution path based on timing.
    pub recommended: ExecutionPath,
    /// Ratio of gpu_time / cpu_time.  Less than 1.0 means GPU is faster.
    /// `NaN` when GPU timing is unavailable.
    pub speedup: f64,
}

/// Which execution back-end is recommended.
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
pub enum ExecutionPath {
    /// CPU execution.
    Cpu,
    /// GPU execution.
    Gpu,
}

impl fmt::Display for ExecutionPath {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
        match self {
            Self::Cpu => write!(f, "CPU"),
            Self::Gpu => write!(f, "GPU"),
        }
    }
}

// ---------------------------------------------------------------------------
// Internal helpers
// ---------------------------------------------------------------------------

fn recommend(cpu_time: Duration, gpu_time: Option<Duration>) -> (ExecutionPath, f64) {
    match gpu_time {
        Some(gt) => {
            let ratio = gt.as_secs_f64() / cpu_time.as_secs_f64().max(1e-12);
            if ratio < 1.0 {
                (ExecutionPath::Gpu, ratio)
            } else {
                (ExecutionPath::Cpu, ratio)
            }
        }
        None => (ExecutionPath::Cpu, f64::NAN),
    }
}

// ---------------------------------------------------------------------------
// MatMul benchmark
// ---------------------------------------------------------------------------

/// Benchmark `MatMul` with matrices of shape `[m, k]` x `[k, n]`.
pub fn benchmark_matmul(m: usize, k: usize, n: usize) -> BenchmarkResult {
    let a_data: Vec<f32> = (0..m * k).map(|i| (i % 7) as f32 * 0.1).collect();
    let b_data: Vec<f32> = (0..k * n).map(|i| (i % 11) as f32 * 0.1).collect();
    let a = Tensor::new(a_data.clone(), vec![m, k]);
    let b = Tensor::new(b_data.clone(), vec![k, n]);

    // CPU benchmark
    let cpu_start = Instant::now();
    let _ = oxionnx_ops::math::matmul(&a, &b);
    let cpu_time = cpu_start.elapsed();

    // GPU benchmark (if available)
    #[cfg(feature = "gpu")]
    let gpu_time = {
        use oxionnx_gpu::GpuContext;
        match GpuContext::try_new() {
            Some(ctx) => {
                let start = Instant::now();
                let _ = oxionnx_gpu::gpu_matmul(&ctx, &a_data, &b_data, m, k, n);
                Some(start.elapsed())
            }
            None => None,
        }
    };
    #[cfg(not(feature = "gpu"))]
    let gpu_time: Option<Duration> = None;

    let (recommended, speedup) = recommend(cpu_time, gpu_time);

    BenchmarkResult {
        op_type: "MatMul".into(),
        input_shape: vec![m, k, n],
        cpu_time,
        gpu_time,
        recommended,
        speedup,
    }
}

// ---------------------------------------------------------------------------
// Element-wise benchmark
// ---------------------------------------------------------------------------

/// Benchmark an element-wise activation at a given tensor size.
///
/// Supported `op` values: `"Relu"`, `"Sigmoid"`, `"Tanh"`, `"Gelu"`.
pub fn benchmark_elementwise(op: &str, size: usize) -> BenchmarkResult {
    let data: Vec<f32> = (0..size).map(|i| (i % 100) as f32 * 0.01 - 0.5).collect();
    let tensor = Tensor::new(data.clone(), vec![size]);

    let cpu_start = Instant::now();
    match op {
        "Relu" => {
            let _ = oxionnx_ops::nn::relu(&tensor);
        }
        "Sigmoid" => {
            let _ = oxionnx_ops::nn::sigmoid(&tensor);
        }
        "Tanh" => {
            let _ = oxionnx_ops::nn::tanh_op(&tensor);
        }
        "Gelu" => {
            let _ = oxionnx_ops::nn::gelu(&tensor);
        }
        _ => {}
    }
    let cpu_time = cpu_start.elapsed();

    #[cfg(feature = "gpu")]
    let gpu_time = {
        use oxionnx_gpu::GpuContext;
        match GpuContext::try_new() {
            Some(ctx) => {
                let start = Instant::now();
                let result = match op {
                    "Relu" => oxionnx_gpu::gpu_relu(&ctx, &data),
                    "Sigmoid" => oxionnx_gpu::gpu_sigmoid(&ctx, &data),
                    "Gelu" => oxionnx_gpu::gpu_gelu(&ctx, &data),
                    _ => None,
                };
                if result.is_some() {
                    Some(start.elapsed())
                } else {
                    None
                }
            }
            None => None,
        }
    };
    #[cfg(not(feature = "gpu"))]
    let gpu_time: Option<Duration> = None;

    let (recommended, speedup) = recommend(cpu_time, gpu_time);

    BenchmarkResult {
        op_type: op.into(),
        input_shape: vec![size],
        cpu_time,
        gpu_time,
        recommended,
        speedup,
    }
}

// ---------------------------------------------------------------------------
// Suite runner
// ---------------------------------------------------------------------------

/// Run a suite of benchmarks across common sizes and return all results.
pub fn benchmark_suite() -> Vec<BenchmarkResult> {
    let mut results = Vec::new();

    // MatMul benchmarks at increasing sizes
    for &(m, k, n) in &[
        (64, 64, 64),
        (256, 256, 256),
        (512, 512, 512),
        (1024, 1024, 1024),
    ] {
        results.push(benchmark_matmul(m, k, n));
    }

    // Element-wise activation benchmarks
    for &size in &[1_000, 10_000, 100_000, 1_000_000] {
        for op in &["Relu", "Sigmoid", "Gelu"] {
            results.push(benchmark_elementwise(op, size));
        }
    }

    results
}

// ---------------------------------------------------------------------------
// Formatting
// ---------------------------------------------------------------------------

/// Format benchmark results as a human-readable table.
pub fn format_results(results: &[BenchmarkResult]) -> String {
    let mut out = String::new();
    out.push_str(&format!(
        "{:<12} {:<20} {:>12} {:>12} {:>8} {:>8}\n",
        "Op", "Shape", "CPU (us)", "GPU (us)", "Winner", "Speedup"
    ));
    out.push_str(&"-".repeat(76));
    out.push('\n');
    for r in results {
        let shape_str = format!("{:?}", r.input_shape);
        let cpu_us = r.cpu_time.as_micros();
        let gpu_us = r
            .gpu_time
            .map(|d| format!("{}", d.as_micros()))
            .unwrap_or_else(|| "N/A".into());
        let winner = r.recommended;
        let speedup = if r.speedup.is_nan() {
            "N/A".into()
        } else {
            format!("{:.2}x", r.speedup)
        };
        out.push_str(&format!(
            "{:<12} {:<20} {:>12} {:>12} {:>8} {:>8}\n",
            r.op_type, shape_str, cpu_us, gpu_us, winner, speedup
        ));
    }
    out
}

// ---------------------------------------------------------------------------
// Tests
// ---------------------------------------------------------------------------

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_benchmark_matmul() {
        let result = benchmark_matmul(16, 16, 16);
        assert_eq!(result.op_type, "MatMul");
        assert_eq!(result.input_shape, vec![16, 16, 16]);
        assert!(!result.cpu_time.is_zero() || result.cpu_time == Duration::ZERO);
        // On a CPU-only build, recommended should be CPU
        #[cfg(not(feature = "gpu"))]
        assert_eq!(result.recommended, ExecutionPath::Cpu);
    }

    #[test]
    fn test_benchmark_elementwise() {
        let result = benchmark_elementwise("Relu", 1000);
        assert_eq!(result.op_type, "Relu");
        assert_eq!(result.input_shape, vec![1000]);
        #[cfg(not(feature = "gpu"))]
        assert_eq!(result.recommended, ExecutionPath::Cpu);
    }

    #[test]
    fn test_benchmark_suite() {
        let results = benchmark_suite();
        // 4 MatMul sizes + 4 sizes * 3 ops = 16 total
        assert_eq!(results.len(), 16);
        assert_eq!(results[0].op_type, "MatMul");
        assert_eq!(results[4].op_type, "Relu");
    }

    #[test]
    fn test_format_results() {
        let results = vec![BenchmarkResult {
            op_type: "MatMul".into(),
            input_shape: vec![32, 32, 32],
            cpu_time: Duration::from_micros(100),
            gpu_time: None,
            recommended: ExecutionPath::Cpu,
            speedup: f64::NAN,
        }];
        let table = format_results(&results);
        assert!(table.contains("MatMul"));
        assert!(table.contains("CPU"));
        assert!(table.contains("N/A"));
        assert!(table.contains("[32, 32, 32]"));
    }

    #[test]
    fn test_execution_path_display() {
        assert_eq!(format!("{}", ExecutionPath::Cpu), "CPU");
        assert_eq!(format!("{}", ExecutionPath::Gpu), "GPU");
    }

    #[cfg(feature = "gpu")]
    #[test]
    fn test_gpu_supported_ops() {
        let ops = crate::session::GpuExecutionProvider::supported_ops();
        assert!(!ops.is_empty());
        assert!(ops.contains(&"MatMul"));
        assert!(ops.contains(&"Relu"));
    }
}