foxhunt/tests/rdtsc_performance_validation.rs

//! RDTSC Performance Validation Tests
//!
//! This module implements REAL hardware-level performance testing using RDTSC (Read Time-Stamp Counter)
//! to validate the Foxhunt HFT system meets its <14ns latency requirements.
//!
//! ## Test Coverage:
//! 1. **RDTSC Precision Testing**: Validate timing accuracy to nanosecond level
//! 2. **Lock-free Operations**: Test atomic operations and lock-free data structures
//! 3. **SIMD Performance**: Validate vectorized operations meet performance targets
//! 4. **CPU Cache Performance**: Test memory access patterns and cache efficiency
//! 5. **System Call Overhead**: Measure OS interaction costs
//! 6. **Thread Context Switching**: Validate multi-threading performance
//! 7. **Memory Allocation**: Test allocation patterns and performance
//! 8. **Network I/O Latency**: Measure network stack performance
//!
//! ## Performance Targets:
//! - RDTSC timing resolution: <14ns
//! - Lock-free queue operations: <100ns
//! - SIMD operations: <1μs
//! - Cache miss penalty: <50ns
//! - System call overhead: <1μs
//! - Context switch time: <10μs
//! - Memory allocation: <100ns
//! - Network round-trip: <50μs

#![warn(missing_docs)]
#![warn(clippy::all)]
#![allow(clippy::too_many_arguments)]

use std::sync::atomic::{AtomicU64, Ordering};
use std::sync::Arc;
use std::time::{Duration, Instant};
use std::arch::x86_64::*;
use criterion::{black_box, Criterion, BenchmarkId};
use trading_engine::prelude::*;
use tracing::{info, warn, error};

/// RDTSC performance validation test suite
pub struct RdtscPerformanceValidator {
    /// Number of iterations for statistical significance
    iterations: usize,
    /// TSC frequency for time conversion
    tsc_frequency: u64,
    /// Test results storage
    results: Vec<PerformanceTestResult>,
}

/// Individual performance test result
#[derive(Debug, Clone)]
pub struct PerformanceTestResult {
    /// Test name
    pub name: String,
    /// Measured latency in nanoseconds
    pub latency_ns: u64,
    /// Whether test passed the target
    pub passed: bool,
    /// Target latency in nanoseconds
    pub target_ns: u64,
    /// Number of samples taken
    pub samples: usize,
    /// Standard deviation of measurements
    pub std_dev_ns: f64,
    /// Percentile measurements
    pub percentiles: PerformancePercentiles,
}

/// Performance percentile measurements
#[derive(Debug, Clone)]
pub struct PerformancePercentiles {
    /// 50th percentile (median)
    pub p50_ns: u64,
    /// 95th percentile
    pub p95_ns: u64,
    /// 99th percentile
    pub p99_ns: u64,
    /// 99.9th percentile
    pub p999_ns: u64,
    /// Maximum measurement
    pub max_ns: u64,
    /// Minimum measurement
    pub min_ns: u64,
}

impl RdtscPerformanceValidator {
    /// Create a new RDTSC performance validator
    pub fn new() -> Result<Self, Box<dyn std::error::Error>> {
        let iterations = 100_000; // Large sample size for statistical significance
        let tsc_frequency = Self::calibrate_tsc_frequency()?;
        
        info!("Initialized RDTSC Performance Validator");
        info!("TSC Frequency: {} Hz", tsc_frequency);
        info!("Test iterations: {}", iterations);

        Ok(Self {
            iterations,
            tsc_frequency,
            results: Vec::new(),
        })
    }

    /// Run comprehensive RDTSC performance validation
    pub async fn run_comprehensive_validation(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        info!("🚀 Starting comprehensive RDTSC performance validation");

        // Warm up the CPU and stabilize frequency
        self.cpu_warmup().await?;

        // Test 1: RDTSC timing precision
        info!("Test 1/8: RDTSC timing precision");
        self.test_rdtsc_precision().await?;

        // Test 2: Lock-free atomic operations
        info!("Test 2/8: Lock-free atomic operations");
        self.test_lockfree_operations().await?;

        // Test 3: SIMD vectorized operations
        info!("Test 3/8: SIMD vectorized operations");
        self.test_simd_operations().await?;

        // Test 4: CPU cache performance
        info!("Test 4/8: CPU cache performance");
        self.test_cache_performance().await?;

        // Test 5: System call overhead
        info!("Test 5/8: System call overhead");
        self.test_syscall_overhead().await?;

        // Test 6: Thread context switching
        info!("Test 6/8: Thread context switching");
        self.test_context_switching().await?;

        // Test 7: Memory allocation performance
        info!("Test 7/8: Memory allocation performance");
        self.test_memory_allocation().await?;

        // Test 8: Network I/O latency
        info!("Test 8/8: Network I/O latency");
        self.test_network_io().await?;

        // Generate comprehensive report
        self.generate_performance_report().await?;

        Ok(())
    }

    /// Test RDTSC timing precision
    async fn test_rdtsc_precision(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        let mut measurements = Vec::with_capacity(self.iterations);
        
        for _ in 0..self.iterations {
            let start_tsc = unsafe { _rdtsc() };
            
            // Minimal operation to measure timing resolution
            black_box(1 + 1);
            
            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);
            
            measurements.push(nanoseconds);
        }

        let result = self.analyze_measurements("RDTSC Precision", &measurements, 14)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ RDTSC precision: {}ns (target: <14ns)", result.latency_ns);
        } else {
            warn!("⚠️  RDTSC precision: {}ns (target: <14ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Test lock-free atomic operations
    async fn test_lockfree_operations(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        let counter = Arc::new(AtomicU64::new(0));
        let mut measurements = Vec::with_capacity(self.iterations);

        for _ in 0..self.iterations {
            let start_tsc = unsafe { _rdtsc() };
            
            // Atomic increment operation
            counter.fetch_add(1, Ordering::Relaxed);
            
            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);
            
            measurements.push(nanoseconds);
        }

        let result = self.analyze_measurements("Lock-free Atomic Operations", &measurements, 100)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ Lock-free operations: {}ns (target: <100ns)", result.latency_ns);
        } else {
            warn!("⚠️  Lock-free operations: {}ns (target: <100ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Test SIMD vectorized operations
    async fn test_simd_operations(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        // Check for AVX2 support
        if !is_x86_feature_detected!("avx2") {
            warn!("AVX2 not supported, skipping SIMD tests");
            return Ok(());
        }

        let data = vec![1.0f32; 256]; // 256 floats for AVX2 processing
        let mut measurements = Vec::with_capacity(self.iterations);

        for _ in 0..self.iterations {
            let start_tsc = unsafe { _rdtsc() };
            
            // SIMD vector addition
            unsafe {
                self.simd_vector_add(&data);
            }
            
            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);
            
            measurements.push(nanoseconds);
        }

        let result = self.analyze_measurements("SIMD Operations", &measurements, 1000)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ SIMD operations: {}ns (target: <1000ns)", result.latency_ns);
        } else {
            warn!("⚠️  SIMD operations: {}ns (target: <1000ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Test CPU cache performance
    async fn test_cache_performance(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        // Test L1 cache performance with sequential access
        let cache_line_size = 64;
        let l1_cache_size = 32 * 1024; // 32KB typical L1 cache
        let data_size = l1_cache_size / std::mem::size_of::<u64>();
        let data = vec![0u64; data_size];
        let mut measurements = Vec::with_capacity(self.iterations);

        for _ in 0..self.iterations {
            let start_tsc = unsafe { _rdtsc() };
            
            // Sequential memory access pattern (cache-friendly)
            let mut sum = 0u64;
            for i in (0..data.len()).step_by(cache_line_size / std::mem::size_of::<u64>()) {
                sum = sum.wrapping_add(unsafe { *data.get_unchecked(i) });
            }
            black_box(sum);
            
            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);
            
            measurements.push(nanoseconds);
        }

        let result = self.analyze_measurements("Cache Performance", &measurements, 50)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ Cache performance: {}ns (target: <50ns)", result.latency_ns);
        } else {
            warn!("⚠️  Cache performance: {}ns (target: <50ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Test system call overhead
    async fn test_syscall_overhead(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        let mut measurements = Vec::with_capacity(self.iterations);

        for _ in 0..self.iterations {
            let start_tsc = unsafe { _rdtsc() };
            
            // Minimal system call - get process ID
            unsafe {
                libc::getpid();
            }
            
            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);
            
            measurements.push(nanoseconds);
        }

        let result = self.analyze_measurements("System Call Overhead", &measurements, 1000)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ System call overhead: {}ns (target: <1000ns)", result.latency_ns);
        } else {
            warn!("⚠️  System call overhead: {}ns (target: <1000ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Test thread context switching
    async fn test_context_switching(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        use std::sync::mpsc;
        use std::thread;

        let mut measurements = Vec::with_capacity(100); // Fewer iterations due to overhead

        for _ in 0..100 {
            let (tx, rx) = mpsc::channel();
            let (tx_back, rx_back) = mpsc::channel();

            let start_tsc = unsafe { _rdtsc() };

            // Create thread and measure round-trip time
            let handle = thread::spawn(move || {
                let _msg = rx.recv().unwrap();
                tx_back.send(()).unwrap();
            });

            tx.send(()).unwrap();
            rx_back.recv().unwrap();

            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);

            measurements.push(nanoseconds);
            handle.join().unwrap();
        }

        let result = self.analyze_measurements("Context Switching", &measurements, 10000)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ Context switching: {}ns (target: <10000ns)", result.latency_ns);
        } else {
            warn!("⚠️  Context switching: {}ns (target: <10000ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Test memory allocation performance
    async fn test_memory_allocation(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        let mut measurements = Vec::with_capacity(self.iterations);

        for _ in 0..self.iterations {
            let start_tsc = unsafe { _rdtsc() };
            
            // Small allocation typical for HFT operations
            let _data: Vec<u64> = Vec::with_capacity(64);
            
            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);
            
            measurements.push(nanoseconds);
        }

        let result = self.analyze_measurements("Memory Allocation", &measurements, 100)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ Memory allocation: {}ns (target: <100ns)", result.latency_ns);
        } else {
            warn!("⚠️  Memory allocation: {}ns (target: <100ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Test network I/O latency (localhost loopback)
    async fn test_network_io(&mut self) -> Result<(), Box<dyn std::error::Error>> {
        use tokio::net::{TcpListener, TcpStream};
        use tokio::io::{AsyncReadExt, AsyncWriteExt};

        // Start a simple echo server
        let listener = TcpListener::bind("127.0.0.1:0").await?;
        let addr = listener.local_addr()?;

        let server_handle = tokio::spawn(async move {
            while let Ok((mut socket, _)) = listener.accept().await {
                tokio::spawn(async move {
                    let mut buf = [0; 8];
                    if socket.read_exact(&mut buf).await.is_ok() {
                        let _ = socket.write_all(&buf).await;
                    }
                });
            }
        });

        // Give server time to start
        tokio::time::sleep(Duration::from_millis(10)).await;

        let mut measurements = Vec::with_capacity(1000); // Fewer iterations for network I/O

        for _ in 0..1000 {
            let start_tsc = unsafe { _rdtsc() };
            
            // Connect and send/receive data
            if let Ok(mut stream) = TcpStream::connect(addr).await {
                let test_data = [0x42u8; 8];
                if stream.write_all(&test_data).await.is_ok() {
                    let mut response = [0u8; 8];
                    let _ = stream.read_exact(&mut response).await;
                }
            }
            
            let end_tsc = unsafe { _rdtsc() };
            let cycles = end_tsc - start_tsc;
            let nanoseconds = self.cycles_to_nanoseconds(cycles);
            
            measurements.push(nanoseconds);
        }

        server_handle.abort();

        let result = self.analyze_measurements("Network I/O", &measurements, 50000)?;
        self.results.push(result.clone());

        if result.passed {
            info!("✅ Network I/O: {}ns (target: <50000ns)", result.latency_ns);
        } else {
            warn!("⚠️  Network I/O: {}ns (target: <50000ns)", result.latency_ns);
        }

        Ok(())
    }

    /// Analyze measurement data and calculate statistics
    fn analyze_measurements(
        &self,
        name: &str,
        measurements: &[u64],
        target_ns: u64,
    ) -> Result<PerformanceTestResult, Box<dyn std::error::Error>> {
        if measurements.is_empty() {
            return Err("No measurements available".into());
        }

        let mut sorted = measurements.to_vec();
        sorted.sort_unstable();

        let sum: u64 = measurements.iter().sum();
        let mean = sum / measurements.len() as u64;
        
        let variance = measurements
            .iter()
            .map(|&x| {
                let diff = x as f64 - mean as f64;
                diff * diff
            })
            .sum::<f64>()
            / measurements.len() as f64;
        
        let std_dev = variance.sqrt();

        let percentiles = PerformancePercentiles {
            p50_ns: sorted[sorted.len() * 50 / 100],
            p95_ns: sorted[sorted.len() * 95 / 100],
            p99_ns: sorted[sorted.len() * 99 / 100],
            p999_ns: sorted[sorted.len() * 999 / 1000],
            max_ns: sorted[sorted.len() - 1],
            min_ns: sorted[0],
        };

        let result = PerformanceTestResult {
            name: name.to_string(),
            latency_ns: percentiles.p95_ns, // Use P95 for pass/fail determination
            passed: percentiles.p95_ns <= target_ns,
            target_ns,
            samples: measurements.len(),
            std_dev_ns: std_dev,
            percentiles,
        };

        Ok(result)
    }

    /// Generate comprehensive performance report
    async fn generate_performance_report(&self) -> Result<(), Box<dyn std::error::Error>> {
        info!("📊 RDTSC Performance Validation Report");
        info!("=====================================");

        let mut total_tests = 0;
        let mut passed_tests = 0;

        for result in &self.results {
            total_tests += 1;
            if result.passed {
                passed_tests += 1;
            }

            let status = if result.passed { "✅ PASS" } else { "❌ FAIL" };
            info!("{} - {}", status, result.name);
            info!("  P50: {}ns, P95: {}ns, P99: {}ns, P99.9: {}ns", 
                result.percentiles.p50_ns,
                result.percentiles.p95_ns,
                result.percentiles.p99_ns,
                result.percentiles.p999_ns
            );
            info!("  Target: {}ns, Samples: {}, StdDev: {:.2}ns",
                result.target_ns,
                result.samples,
                result.std_dev_ns
            );
            info!("");
        }

        let success_rate = (passed_tests as f64 / total_tests as f64) * 100.0;
        info!("Overall Results: {}/{} tests passed ({:.1}%)", passed_tests, total_tests, success_rate);

        if passed_tests == total_tests {
            info!("🎉 All performance tests PASSED - System meets HFT requirements!");
        } else {
            warn!("⚠️  Some performance tests FAILED - Review system configuration");
        }

        Ok(())
    }

    /// CPU warmup to stabilize frequency and cache state
    async fn cpu_warmup(&self) -> Result<(), Box<dyn std::error::Error>> {
        info!("Warming up CPU and stabilizing frequency...");
        
        let warmup_start = Instant::now();
        let warmup_duration = Duration::from_secs(2);

        while warmup_start.elapsed() < warmup_duration {
            // CPU-intensive work to boost frequency
            let mut sum = 0u64;
            for i in 0..10000 {
                sum = sum.wrapping_add(i);
            }
            black_box(sum);
        }

        info!("CPU warmup completed");
        Ok(())
    }

    /// Calibrate TSC frequency
    fn calibrate_tsc_frequency() -> Result<u64, Box<dyn std::error::Error>> {
        let start_time = Instant::now();
        let start_tsc = unsafe { _rdtsc() };
        
        std::thread::sleep(Duration::from_millis(100));
        
        let end_time = Instant::now();
        let end_tsc = unsafe { _rdtsc() };
        
        let elapsed_ns = (end_time - start_time).as_nanos() as u64;
        let elapsed_cycles = end_tsc - start_tsc;
        
        let frequency = (elapsed_cycles * 1_000_000_000) / elapsed_ns;
        Ok(frequency)
    }

    /// Convert TSC cycles to nanoseconds
    fn cycles_to_nanoseconds(&self, cycles: u64) -> u64 {
        (cycles * 1_000_000_000) / self.tsc_frequency
    }

    /// SIMD vector addition using AVX2
    #[target_feature(enable = "avx2")]
    unsafe fn simd_vector_add(&self, data: &[f32]) {
        let chunks = data.chunks_exact(8);
        for chunk in chunks {
            let a = _mm256_loadu_ps(chunk.as_ptr());
            let b = _mm256_set1_ps(1.0);
            let result = _mm256_add_ps(a, b);
            black_box(result);
        }
    }
}

/// Run RDTSC performance validation tests
#[cfg(test)]
mod tests {
    use super::*;

    #[tokio::test]
    async fn test_rdtsc_performance_validation() {
        let mut validator = RdtscPerformanceValidator::new()
            .expect("Failed to create RDTSC validator");

        validator.run_comprehensive_validation().await
            .expect("RDTSC performance validation failed");

        // Check that critical tests pass
        let critical_tests = ["RDTSC Precision", "Lock-free Atomic Operations"];
        for test_name in &critical_tests {
            let result = validator.results.iter()
                .find(|r| r.name == *test_name)
                .expect(&format!("Test '{}' not found", test_name));
            
            assert!(result.passed, 
                "Critical test '{}' failed: {}ns > {}ns target", 
                test_name, result.latency_ns, result.target_ns
            );
        }
    }

    #[tokio::test]
    async fn test_individual_rdtsc_precision() {
        let mut validator = RdtscPerformanceValidator::new()
            .expect("Failed to create RDTSC validator");

        validator.test_rdtsc_precision().await
            .expect("RDTSC precision test failed");

        let result = &validator.results[0];
        assert_eq!(result.name, "RDTSC Precision");
        
        // RDTSC should provide sub-nanosecond precision on modern CPUs
        // but we allow up to 14ns due to system variability
        info!("RDTSC precision: {}ns (target: <14ns)", result.latency_ns);
    }

    #[tokio::test]
    async fn test_simd_operations() {
        if !is_x86_feature_detected!("avx2") {
            println!("Skipping SIMD test - AVX2 not supported");
            return;
        }

        let mut validator = RdtscPerformanceValidator::new()
            .expect("Failed to create RDTSC validator");

        validator.test_simd_operations().await
            .expect("SIMD operations test failed");

        let result = &validator.results[0];
        assert_eq!(result.name, "SIMD Operations");
        
        info!("SIMD operations: {}ns (target: <1000ns)", result.latency_ns);
    }
}