Module gpu_bench 

Performance benchmarks comparing CPU, wgpu, and CUDA backends.

This module measures throughput for the core compute kernels (SPH density, LBM collision, parallel scan) across available backends and reports wall-clock timing and effective GFLOP/s estimates.

Quick usage

use oxiphysics_gpu::gpu_bench::{GpuBenchHarness, BackendKind};

let mut h = GpuBenchHarness::new();

// Benchmark SPH density summation for 256 particles
let reports = h.bench_sph_density(256);
for r in &reports {
    println!("{}", r);
}

// Compare available backends
let available = GpuBenchHarness::available_backends();
assert!(available.contains(&BackendKind::Cpu));

Structs

CudaSpeedupReport

Computed speedup between a CPU and a CUDA benchmark report.

GpuBenchHarness

Timing harness for GPU/CPU backend comparison benchmarks.

GpuBenchReport

Result of a single GPU benchmark run.

SpeedupReport

Computed speedup between a CPU and a wgpu benchmark report.

Enums

BackendKind

Identifies a compute backend for benchmark reporting.

Functions

compute_cuda_speedup

Compute a speedup ratio from a pair of bench reports produced by GpuBenchHarness::cpu_vs_cuda_sph.

compute_speedup

Compute a speedup ratio from a pair of bench reports.

inclusive_scan_cpu

Sequential inclusive prefix scan (Σ) on f64 elements.