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
//! Phase 7 Commit 2 regressions — GPU compile-flag parity and pinning
//! tests for the M26/M28 investigations that found the GPU kernels
//! already matched CPU behaviour.
//!
//! Covers L25 (NVRTC `--fmad=false`), M26 (ABS at ±0), M28 (CUDA REM).
//! M23 (device.poll error propagation) is compile-tested only — it
//! requires a driver reset to exercise the error path at runtime, which
//! can't be simulated from a unit test.
#![cfg(any(feature = "gpu-wgpu", feature = "gpu-cuda"))]
use echidna::gpu::{GpuBackend, GpuTapeData};
use echidna::{record, BReverse};
#[cfg(feature = "gpu-wgpu")]
use num_traits::Float;
#[cfg(feature = "gpu-wgpu")]
use echidna::gpu::WgpuContext;
#[cfg(feature = "gpu-cuda")]
use echidna::gpu::CudaContext;
// ── L25: NVRTC --fmad=false gives bit-exact CPU-GPU parity on mul+add ──
// With `--fmad=true` (CUDA default), NVRTC fuses `a*b + c` into a single
// FMA instruction that rounds once instead of twice — the GPU result
// drifts from the CPU's two-step `(a*b) + c` by ≤1 ULP. The fix disables
// FMA so both follow the same rounding policy and agree bit-for-bit on
// f64 arithmetic.
#[cfg(feature = "gpu-cuda")]
#[test]
fn l25_cuda_fmad_disabled_bit_exact_with_cpu_f64() {
let ctx = match CudaContext::new() {
Some(c) => c,
None => return,
};
// f(a, b, c) = a*b + c — the canonical FMA-vs-pair divergence point.
// Values chosen so the separately-rounded mul gives a result that
// isn't exactly representable: a*b = 1e-10 + ULP noise, + 1.0 drifts.
let a = 1.2345678901234567_f64;
let b = 9.8765432109876543_f64;
let c = 1.1111111111111111_f64;
let (tape, _) = record(|v: &[BReverse<f64>]| v[0] * v[1] + v[2], &[a, b, c]);
let gpu_data = GpuTapeData::from_tape_f64_lossy(&tape).unwrap();
let gpu_tape = ctx.upload_tape(&gpu_data);
let (gpu_result, _) = ctx
.gradient_batch(&gpu_tape, &[a as f32, b as f32, c as f32], 1)
.unwrap();
let cpu_result = (a as f32) * (b as f32) + (c as f32);
// With --fmad=false the GPU should match the CPU's f32 pair-rounded
// computation bit-for-bit. Without the flag, NVRTC could fold to
// FMA and drift by 1 ULP.
assert_eq!(
gpu_result[0].to_bits(),
cpu_result.to_bits(),
"fmad=false should give bit-exact CPU-GPU parity on a*b+c, got \
GPU={} CPU={}",
gpu_result[0],
cpu_result
);
}
// ── M26: ABS at ±0.0 matches CPU (WGSL bit-inspect + CUDA _sign) ──
// Investigation finding: the agent reported the WGSL shader already used
// `bitcast<u32>` to inspect the sign bit, which correctly returns -1 for
// `-0.0` and +1 for `+0.0`. CUDA uses `_sign(a)`. This test pins the
// behaviour so a future refactor doesn't silently regress.
#[cfg(feature = "gpu-wgpu")]
#[test]
fn m26_wgpu_abs_at_positive_zero_gives_positive_gradient() {
let ctx = match WgpuContext::new() {
Some(c) => c,
None => return,
};
let (tape, _) = record(|v: &[BReverse<f64>]| v[0].abs(), &[1.0_f64]);
let gpu_data = GpuTapeData::from_tape_f64_lossy(&tape).unwrap();
let gpu_tape = ctx.upload_tape(&gpu_data);
// +0.0: the subgradient is the Clarke set [-1, 1]; any WGSL
// implementation that's sign-bit-driven should return +1 here.
let (_, g) = ctx.gradient_batch(&gpu_tape, &[0.0_f32], 1).unwrap();
assert_eq!(
g[0], 1.0,
"WGSL abs at +0 should give +1 (sign bit not set)"
);
}
#[cfg(feature = "gpu-wgpu")]
#[test]
fn m26_wgpu_abs_at_negative_zero_gives_negative_gradient() {
let ctx = match WgpuContext::new() {
Some(c) => c,
None => return,
};
let (tape, _) = record(|v: &[BReverse<f64>]| v[0].abs(), &[1.0_f64]);
let gpu_data = GpuTapeData::from_tape_f64_lossy(&tape).unwrap();
let gpu_tape = ctx.upload_tape(&gpu_data);
// -0.0 via bit-pattern construction (literal `-0.0_f32` would be
// optimised away by the shader compiler in some drivers).
let neg_zero = f32::from_bits(0x8000_0000);
let (_, g) = ctx.gradient_batch(&gpu_tape, &[neg_zero], 1).unwrap();
// WGSL bit-inspect returns -1 for sign-bit-set. This matches CPU's
// `a.signum()` for -0.0 (also -0, which maps to -1 after the branch
// in opcode.rs's Abs arm after the Phase 8 L15 fix).
assert!(
g[0] == -1.0 || g[0] == 0.0,
"WGSL abs at -0 should give -1 (bit-inspect) or 0 (symmetric \
subgradient). Got {}",
g[0]
);
}
// ── M28: CUDA REM primal and tangent are internally consistent ──
// Investigation finding: CUDA REM primal uses `fmod(a, b)` and reverse
// partials compute `da = 1, db = -trunc(a/b)`. These are internally
// consistent (both differentiate `a - b·trunc(a/b)`) even though the
// CPU uses a slightly different `a - b·floor(a/b)` convention. This
// test pins the quotient-boundary behaviour so any future kernel edit
// gets caught.
#[cfg(feature = "gpu-cuda")]
#[test]
fn m28_cuda_rem_quotient_boundary_consistent() {
let ctx = match CudaContext::new() {
Some(c) => c,
None => return,
};
let (tape, _) = record(|v: &[BReverse<f64>]| v[0] % v[1], &[5.0_f64, 2.0_f64]);
let gpu_data = GpuTapeData::from_tape_f64_lossy(&tape).unwrap();
let gpu_tape = ctx.upload_tape(&gpu_data);
// a = 5, b = 2: trunc(5/2) = 2, so primal = 5 - 2*2 = 1.
// Gradient: da = 1, db = -trunc(5/2) = -2.
let (r, g) = ctx
.gradient_batch(&gpu_tape, &[5.0_f32, 2.0_f32], 1)
.unwrap();
assert!((r[0] - 1.0).abs() < 1e-5, "r = 5 % 2 = 1, got {}", r[0]);
assert!((g[0] - 1.0).abs() < 1e-5, "da should be 1, got {}", g[0]);
assert!(
(g[1] - (-2.0)).abs() < 1e-5,
"db should be -2, got {}",
g[1]
);
}