rustytorch/demos/inference-profiler-shared/src/results.rs

//! Profile results and aggregation.

use crate::config::{DeviceType, ModelType};
use crate::metrics::{LatencyMetrics, MemoryMetrics, ThroughputMetrics};
use serde::{Deserialize, Serialize};

/// Single profiling result for a specific configuration.
#[derive(Debug, Clone, PartialEq, Serialize, Deserialize)]
pub struct ProfileResult {
    /// Model that was profiled
    pub model_type: ModelType,
    /// Device used for profiling
    pub device: DeviceType,
    /// Batch size tested
    pub batch_size: usize,
    /// Latency measurements
    pub latency: LatencyMetrics,
    /// Memory usage
    pub memory: MemoryMetrics,
    /// Throughput statistics
    pub throughput: ThroughputMetrics,
    /// Unix timestamp when profile was captured
    pub timestamp: u64,
}

impl ProfileResult {
    /// Creates a new profile result.
    #[must_use]
    pub fn new(
        model_type: ModelType,
        device: DeviceType,
        batch_size: usize,
        latency: LatencyMetrics,
        memory: MemoryMetrics,
        throughput: ThroughputMetrics,
        timestamp: u64,
    ) -> Self {
        Self {
            model_type,
            device,
            batch_size,
            latency,
            memory,
            throughput,
            timestamp,
        }
    }
}

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

    #[test]
    fn test_profile_result_creation() {
        let latency = LatencyMetrics::from_measurements(&[10.0, 20.0, 30.0]).unwrap();
        let memory = MemoryMetrics::new(100.0, 80.0, 120.0).unwrap();
        let throughput = ThroughputMetrics::new(50.0, None).unwrap();

        let result = ProfileResult::new(
            ModelType::ResNet18,
            DeviceType::CPU,
            8,
            latency,
            memory,
            throughput,
            1234567890,
        );

        assert_eq!(result.model_type, ModelType::ResNet18);
        assert_eq!(result.device, DeviceType::CPU);
        assert_eq!(result.batch_size, 8);
        assert_eq!(result.timestamp, 1234567890);
    }

    #[test]
    fn test_profile_result_serialization() {
        let latency = LatencyMetrics::from_measurements(&[10.0, 20.0, 30.0]).unwrap();
        let memory = MemoryMetrics::new(100.0, 80.0, 120.0).unwrap();
        let throughput = ThroughputMetrics::new(50.0, None).unwrap();

        let result = ProfileResult::new(
            ModelType::ViTB16,
            DeviceType::CUDA,
            16,
            latency,
            memory,
            throughput,
            1234567890,
        );

        let json = serde_json::to_string(&result).expect("serialization failed");
        assert!(json.contains("\"model_type\""));
        assert!(json.contains("\"vit-b16\""));
        assert!(json.contains("\"batch_size\""));
        assert!(json.contains("\"latency\""));
        assert!(json.contains("\"memory\""));
        assert!(json.contains("\"throughput\""));
    }

    #[test]
    fn test_profile_result_deserialization() {
        let json = r#"{
            "model_type": "resnet50",
            "device": "CUDA",
            "batch_size": 32,
            "latency": {
                "mean_ms": 20.0,
                "std_ms": 2.0,
                "min_ms": 18.0,
                "max_ms": 25.0,
                "p50_ms": 20.0,
                "p95_ms": 23.0,
                "p99_ms": 24.0
            },
            "memory": {
                "peak_memory_mb": 200.0,
                "allocated_mb": 180.0,
                "reserved_mb": 220.0
            },
            "throughput": {
                "samples_per_sec": 1600.0,
                "tokens_per_sec": null
            },
            "timestamp": 1234567890
        }"#;

        let result: ProfileResult = serde_json::from_str(json).expect("deserialization failed");
        assert_eq!(result.model_type, ModelType::ResNet50);
        assert_eq!(result.device, DeviceType::CUDA);
        assert_eq!(result.batch_size, 32);
    }

    #[test]
    fn test_profile_result_roundtrip() {
        let latency = LatencyMetrics::from_measurements(&[10.0, 20.0, 30.0]).unwrap();
        let memory = MemoryMetrics::new(100.0, 80.0, 120.0).unwrap();
        let throughput = ThroughputMetrics::new(50.0, Some(200.0)).unwrap();

        let original = ProfileResult::new(
            ModelType::ConvNeXtBase,
            DeviceType::Metal,
            4,
            latency,
            memory,
            throughput,
            9876543210,
        );

        let json = serde_json::to_string(&original).expect("serialization failed");
        let decoded: ProfileResult = serde_json::from_str(&json).expect("deserialization failed");

        assert_eq!(decoded.model_type, original.model_type);
        assert_eq!(decoded.device, original.device);
        assert_eq!(decoded.batch_size, original.batch_size);
        assert_eq!(decoded.timestamp, original.timestamp);
    }
}