xtrain/crates/xtrain-tensor/tests/integration.rs

// GPU integration tests for the tensor abstraction. Both require nvcc + a GPU,
// so they are gated behind `not(no_cuda)`. On a GPU-less machine build.rs sets
// the `no_cuda` cfg and these compile out, keeping host `cargo check` green.
#![cfg(not(no_cuda))]

use xtrain_cuda::device;
use xtrain_tensor::{Device, Tensor};

/// (a) Host → device → host roundtrip preserves the data exactly.
#[test]
fn host_device_roundtrip() {
    assert!(
        device::device_count().expect("device count") > 0,
        "no CUDA device"
    );
    device::set_device(0).unwrap();

    let host: Vec<f32> = (0..1024).map(|i| i as f32 * 0.5).collect();
    let cpu = Tensor::from_slice(&host, &[1024]);

    let gpu = cpu.to_device(Device::Cuda(0));
    assert_eq!(gpu.device(), Device::Cuda(0));
    assert_eq!(gpu.shape(), &[1024]);

    let back = gpu.to_device(Device::Cpu);
    assert_eq!(back.device(), Device::Cpu);
    assert_eq!(back.as_slice::<f32>(), host.as_slice());
    println!("roundtrip OK: {} elems preserved", host.len());
}

/// (b) The elementwise `scale` kernel produces correct results.
#[test]
fn elementwise_scale_kernel() {
    assert!(
        device::device_count().expect("device count") > 0,
        "no CUDA device"
    );
    device::set_device(0).unwrap();

    let host: Vec<f32> = (0..2048).map(|i| i as f32).collect();
    let alpha = 3.0f32;
    let expected: Vec<f32> = host.iter().map(|x| x * alpha).collect();

    let gpu = Tensor::from_slice(&host, &[2048]).to_device(Device::Cuda(0));
    let scaled = gpu.scale(alpha);
    let result = scaled.to_device(Device::Cpu);

    assert_eq!(result.shape(), &[2048]);
    assert_eq!(result.as_slice::<f32>(), expected.as_slice());
    let r = result.as_slice::<f32>();
    println!(
        "scale OK (alpha={alpha}): first={} mid={} last={} ({} elems)",
        r[0],
        r[r.len() / 2],
        r[r.len() - 1],
        r.len()
    );
}

/// (c) `rope_at` (KV-cache decode RoPE at an absolute position) is bit-identical
/// to the full-sequence `rope`'s corresponding row. This is the invariant the
/// decode KV-cache relies on: a single new token RoPE'd at position `t` must equal
/// what the full-sequence forward would have produced at row `t` (so cached
/// post-RoPE K matches the full-recompute path → token-identical decode).
#[test]
fn rope_at_matches_full_rope_row() {
    assert!(
        device::device_count().expect("device count") > 0,
        "no CUDA device"
    );
    device::set_device(0).unwrap();

    let (n, heads, hd) = (7usize, 3usize, 8usize);
    let theta = 10000.0f32;
    // Deterministic pseudo-random fill in [-1, 1).
    let host: Vec<f32> = (0..n * heads * hd)
        .map(|i| ((i * 37 % 101) as f32 / 50.0) - 1.0)
        .collect();

    // Full-sequence rope (period = n → row r gets position r).
    let full = Tensor::from_slice(&host, &[n, heads, hd]).to_device(Device::Cuda(0));
    let roped_full = full
        .rope(theta, n)
        .to_device(Device::Cpu)
        .as_slice::<f32>()
        .to_vec();

    let row_len = heads * hd;
    for t in 0..n {
        let row = &host[t * row_len..(t + 1) * row_len];
        let roped_row = Tensor::from_slice(row, &[1, heads, hd])
            .to_device(Device::Cuda(0))
            .rope_at(theta, t)
            .to_device(Device::Cpu)
            .as_slice::<f32>()
            .to_vec();
        let expect = &roped_full[t * row_len..(t + 1) * row_len];
        assert_eq!(
            roped_row.as_slice(),
            expect,
            "rope_at(pos0={t}) != full rope row {t}"
        );
    }
    println!("rope_at OK: bit-identical to full rope across {n} positions");
}

/// (d) `decode_attention` (single query vs cached K/V, no mask) equals the LAST
/// query row of the full causal `attention`. This is the core decode-engine
/// invariant: the incremental path must reproduce what the full-recompute forward
/// computes for the final position, so KV-cache greedy decode is token-identical.
/// Tolerance is fp rounding (different softmax kernel + reduction order), not bits.
#[test]
fn decode_attention_matches_full_attention_last_row() {
    assert!(
        device::device_count().expect("device count") > 0,
        "no CUDA device"
    );
    device::set_device(0).unwrap();

    let (bh, t, hd) = (6usize, 5usize, 8usize);
    let scale = 1.0 / (hd as f32).sqrt();
    let n = bh * t * hd;
    let qh: Vec<f32> = (0..n).map(|i| ((i * 31 % 97) as f32 / 48.0) - 1.0).collect();
    let kh: Vec<f32> = (0..n).map(|i| ((i * 53 % 89) as f32 / 44.0) - 1.0).collect();
    let vh: Vec<f32> = (0..n).map(|i| ((i * 17 % 83) as f32 / 41.0) - 1.0).collect();
    let q = Tensor::from_slice(&qh, &[bh, t, hd]).to_device(Device::Cuda(0));
    let k = Tensor::from_slice(&kh, &[bh, t, hd]).to_device(Device::Cuda(0));
    let v = Tensor::from_slice(&vh, &[bh, t, hd]).to_device(Device::Cuda(0));

    // Reference: full causal attention, take each head's last query row.
    let (full, _) = q.attention(&k, &v, scale);
    let full_h = full.to_device(Device::Cpu).as_slice::<f32>().to_vec();

    // Decode: build Q_last [bh,1,hd] from each head's last row, attend to all K/V.
    let mut ql = vec![0f32; bh * hd];
    for b in 0..bh {
        let src = (b * t + (t - 1)) * hd;
        ql[b * hd..(b + 1) * hd].copy_from_slice(&qh[src..src + hd]);
    }
    let q_last = Tensor::from_slice(&ql, &[bh, 1, hd]).to_device(Device::Cuda(0));
    let dec = q_last
        .decode_attention(&k, &v, scale)
        .to_device(Device::Cpu)
        .as_slice::<f32>()
        .to_vec();
    assert_eq!(dec.len(), bh * hd, "decode out shape");

    let mut max_abs = 0f32;
    for b in 0..bh {
        for d in 0..hd {
            let got = dec[b * hd + d];
            let exp = full_h[(b * t + (t - 1)) * hd + d];
            max_abs = max_abs.max((got - exp).abs());
        }
    }
    assert!(
        max_abs < 1e-4,
        "decode_attention vs full last-row max abs diff {max_abs} exceeds 1e-4"
    );
    println!("decode_attention OK: matches full causal last row (bh={bh}, t={t}, max|Δ|={max_abs:.2e})");
}