vanity-core 0.8.1

Multi-chain vanity address grinding core library
Documentation
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
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
//! OpenCL runtime: compile kernels, batch grind, CPU re-verify matches.

use crate::chain::{ChainGrinder, GrindAttempt, KeypairResult};
use crate::gpu::pattern_pack::GpuPattern;
use crate::grinder::{CancelToken, GrindResult};
use crate::pattern::Pattern;
use opencl3::command_queue::CommandQueue;
use opencl3::context::Context;
use opencl3::device::{get_all_devices, Device, CL_DEVICE_TYPE_ALL, CL_DEVICE_TYPE_GPU};
use opencl3::kernel::{ExecuteKernel, Kernel};
use opencl3::memory::{
    Buffer, CL_MEM_COPY_HOST_PTR, CL_MEM_READ_ONLY, CL_MEM_READ_WRITE, CL_MEM_WRITE_ONLY,
};
use opencl3::program::Program;
use opencl3::types::{cl_int, cl_ulong, CL_BLOCKING};
use rand::rngs::OsRng;
use rand::RngCore;
use rayon::prelude::*;
use secp256k1::{PublicKey, Scalar, Secp256k1, SecretKey};
use std::ptr;
use std::time::Instant;

const EVM_SRC: &str = include_str!("kernels/evm.cl");
const SOL_SRC: &str = include_str!("kernels/solana.cl");
/// Points sharing one field inversion (Montgomery batch inversion) in the EVM kernel.
const EVM_GROUP: usize = 64;
/// Groups each work-item walks per launch.
const EVM_GROUPS: usize = 4;
/// Work-items; each owns a random start point that persists on the device between launches.
const EVM_ITEMS: usize = 8 * 1024;
/// Scalar distance every work-item advances per launch.
const EVM_STEPS: u64 = (EVM_GROUP * EVM_GROUPS) as u64;
/// Candidates tested per launch.
const EVM_PER_LAUNCH: u64 = EVM_ITEMS as u64 * EVM_STEPS;
const BATCH_SOL: usize = 16 * 1024; // ed25519 is heavier per work-item

fn device_ids() -> Result<Vec<opencl3::types::cl_device_id>, String> {
    // Prefer discrete/integrated GPUs; fall back to any OpenCL device (e.g. Apple).
    match get_all_devices(CL_DEVICE_TYPE_GPU) {
        Ok(ids) if !ids.is_empty() => Ok(ids),
        _ => get_all_devices(CL_DEVICE_TYPE_ALL).map_err(|e| format!("OpenCL devices: {e}")),
    }
}

pub fn enumerate_devices() -> Vec<crate::gpu::device::GpuDeviceInfo> {
    let Ok(ids) = device_ids() else {
        return Vec::new();
    };
    ids.into_iter()
        .enumerate()
        .filter_map(|(index, id)| {
            let d = Device::new(id);
            Some(crate::gpu::device::GpuDeviceInfo {
                index,
                name: d.name().ok()?,
                vendor: d.vendor().unwrap_or_default(),
                device_type: "GPU".into(),
                compute_units: d.max_compute_units().unwrap_or(0),
                max_clock_mhz: d.max_clock_frequency().unwrap_or(0),
            })
        })
        .collect()
}

fn first_gpu() -> Result<(Device, Context, CommandQueue), String> {
    let ids = device_ids()?;
    let id = *ids.first().ok_or("no OpenCL GPU devices found")?;
    let device = Device::new(id);
    let context = Context::from_device(&device).map_err(|e| format!("OpenCL context: {e}"))?;
    // Apple's OpenCL 1.2 loader lacks clCreateCommandQueueWithProperties.
    let queue = unsafe { CommandQueue::create(&context, device.id(), 0) }
        .map_err(|e| format!("OpenCL queue: {e}"))?;
    Ok((device, context, queue))
}

fn build_program(context: &Context, device: &Device, src: &str) -> Result<Program, String> {
    let mut program = Program::create_from_source(context, src)
        .map_err(|e| format!("OpenCL create program: {e}"))?;
    let devices = [device.id()];
    // Apple OpenCL is picky — empty options, then surface build log on failure.
    if let Err(e) = program.build(&devices, "") {
        let log = program
            .get_build_log(device.id())
            .unwrap_or_else(|_| "(no build log)".into());
        return Err(format!("OpenCL build failed: {e}\n{log}"));
    }
    Ok(program)
}

pub fn benchmark(chain_id: &str, duration_secs: f64) -> Result<(f64, String), String> {
    let id = normalize_id(chain_id)?;
    if id == "evm" {
        return evm_benchmark(duration_secs);
    }
    let (device, context, queue) = first_gpu()?;
    let name = device.name().unwrap_or_else(|_| "OpenCL GPU".into());
    let src = if id == "evm" { EVM_SRC } else { SOL_SRC };
    let program = build_program(&context, &device, src)?;
    let kernel_name = if id == "evm" {
        "evm_grind"
    } else {
        "sol_grind"
    };
    let kernel =
        Kernel::create(&program, kernel_name).map_err(|e| format!("OpenCL kernel: {e}"))?;

    let dummy = never_pattern(id)?;
    let start = Instant::now();
    let mut attempts = 0u64;
    let mut seed = [0u8; 32];
    OsRng.fill_bytes(&mut seed);
    let mut batch_base: u64 = 0;
    while start.elapsed().as_secs_f64() < duration_secs {
        let batch = BATCH_SOL;
        run_batch(&context, &queue, &kernel, &seed, batch_base, &dummy, batch)?;
        attempts += batch as u64;
        batch_base = batch_base.wrapping_add(batch as u64);
        OsRng.fill_bytes(&mut seed);
    }
    let secs = start.elapsed().as_secs_f64().max(0.001);
    Ok((attempts as f64 / secs, name))
}

pub fn grind<G: ChainGrinder + Clone>(
    grinder: G,
    patterns: &[Pattern],
    cancel: &CancelToken,
    on_progress: impl Fn(u64, f64, f64) + Sync,
) -> Result<(GrindResult, crate::gpu::engine::GpuGrindReport), String> {
    let id = normalize_id(grinder.id())?;
    if patterns.is_empty() {
        return Err("at least one pattern is required".into());
    }
    if id == "evm" {
        return grind_evm(grinder, patterns, cancel, on_progress);
    }
    let batch = BATCH_SOL;
    // GPU MVP: single OR-alternative, no wildcards (enforced in GpuPattern).
    let packed = GpuPattern::for_solana(&patterns[0])?;

    let (device, context, queue) = first_gpu()?;
    let name = device.name().unwrap_or_else(|_| "OpenCL GPU".into());
    let program = build_program(&context, &device, SOL_SRC)?;
    let kernel =
        Kernel::create(&program, "sol_grind").map_err(|e| format!("OpenCL kernel: {e}"))?;

    let expected = crate::pattern::expected_attempts_any(
        &patterns
            .iter()
            .map(|p| grinder.expected_attempts(p))
            .collect::<Vec<_>>(),
    );

    let start = Instant::now();
    let mut attempts = 0u64;
    let mut seed = [0u8; 32];
    let mut batch_base: u64 = 0;

    loop {
        if cancel.is_cancelled() {
            return Err("cancelled".into());
        }
        OsRng.fill_bytes(&mut seed);
        if let Some(sk) = run_batch(&context, &queue, &kernel, &seed, batch_base, &packed, batch)? {
            attempts += batch as u64;
            // CPU re-verify — never trust the GPU alone.
            let keypair = grinder.finalize(GrindAttempt::Secret32(sk));
            if patterns
                .iter()
                .any(|p| grinder.matches(&keypair.address, p))
            {
                let elapsed = start.elapsed().as_secs_f64();
                return Ok((
                    GrindResult {
                        keypair,
                        attempts,
                        elapsed_secs: elapsed,
                    },
                    crate::gpu::engine::GpuGrindReport {
                        device_name: name,
                        backend: "opencl",
                        attempts,
                        elapsed_secs: elapsed,
                    },
                ));
            }
            // False positive — keep going.
        } else {
            attempts += batch as u64;
        }
        batch_base = batch_base.wrapping_add(batch as u64);
        let secs = start.elapsed().as_secs_f64().max(0.001);
        let rate = attempts as f64 / secs;
        let eta_min = (expected - attempts as f64).max(0.0) / rate / 60.0;
        on_progress(attempts, rate, eta_min);
    }
}

// ---------------------------------------------------------------------------
// EVM: incremental search with persistent per-work-item start points
// ---------------------------------------------------------------------------

fn evm_source() -> String {
    format!("#define GROUP {EVM_GROUP}\n#define GROUPS {EVM_GROUPS}\n{EVM_SRC}")
}

/// 32 big-endian bytes → eight little-endian u32 limbs (the kernel's `fe`).
fn fe_limbs(be: &[u8]) -> [u32; 8] {
    let mut d = [0u32; 8];
    for (i, limb) in d.iter_mut().enumerate() {
        let o = 28 - 4 * i;
        *limb = u32::from_be_bytes([be[o], be[o + 1], be[o + 2], be[o + 3]]);
    }
    d
}

fn point_limbs(pk: &PublicKey) -> ([u32; 8], [u32; 8]) {
    let raw = pk.serialize_uncompressed(); // 0x04 ‖ X ‖ Y
    (fe_limbs(&raw[1..33]), fe_limbs(&raw[33..65]))
}

struct EvmSession {
    queue: CommandQueue,
    context: Context,
    kernel: Kernel,
    pat_buf: Buffer<u8>,
    found_buf: Buffer<cl_int>,
    off_buf: Buffer<u32>,
    bx: Buffer<u32>,
    by: Buffer<u32>,
    tabx: Buffer<u32>,
    taby: Buffer<u32>,
    /// Start scalar of each work-item (the device only holds the current point).
    k0: Vec<[u8; 32]>,
    launches: u64,
}

impl EvmSession {
    fn new(pat: &GpuPattern) -> Result<(Self, String), String> {
        let (device, context, queue) = first_gpu()?;
        let name = device.name().unwrap_or_else(|_| "OpenCL GPU".into());
        let program = build_program(&context, &device, &evm_source())?;
        let kernel =
            Kernel::create(&program, "evm_grind").map_err(|e| format!("OpenCL kernel: {e}"))?;

        let secp = Secp256k1::new();
        // Table of j·G, j = 1..=GROUP.
        let mut tabx = Vec::with_capacity(EVM_GROUP * 8);
        let mut taby = Vec::with_capacity(EVM_GROUP * 8);
        for j in 1..=EVM_GROUP {
            let mut be = [0u8; 32];
            be[24..].copy_from_slice(&(j as u64).to_be_bytes());
            let sk = SecretKey::from_slice(&be).map_err(|e| format!("table scalar: {e}"))?;
            let (x, y) = point_limbs(&PublicKey::from_secret_key(&secp, &sk));
            tabx.extend_from_slice(&x);
            taby.extend_from_slice(&y);
        }

        let mut session = Self {
            pat_buf: make_buf(&context, CL_MEM_READ_ONLY, pat_bytes(pat))?,
            found_buf: make_buf(&context, CL_MEM_READ_WRITE, vec![-1 as cl_int])?,
            off_buf: make_buf(&context, CL_MEM_READ_WRITE, vec![0u32; 2])?,
            bx: make_buf(&context, CL_MEM_READ_WRITE, vec![0u32; EVM_ITEMS * 8])?,
            by: make_buf(&context, CL_MEM_READ_WRITE, vec![0u32; EVM_ITEMS * 8])?,
            tabx: make_buf(&context, CL_MEM_READ_ONLY, tabx)?,
            taby: make_buf(&context, CL_MEM_READ_ONLY, taby)?,
            k0: Vec::new(),
            launches: 0,
            queue,
            context,
            kernel,
        };
        session.reseed()?;
        Ok((session, name))
    }

    /// Fresh random start point for every work-item (also used after a rejected hit).
    fn reseed(&mut self) -> Result<(), String> {
        let secp = Secp256k1::new();
        let starts: Vec<([u8; 32], [u32; 8], [u32; 8])> = (0..EVM_ITEMS)
            .into_par_iter()
            .map(|_| {
                let sk = SecretKey::new(&mut OsRng);
                let (x, y) = point_limbs(&PublicKey::from_secret_key(&secp, &sk));
                (sk.secret_bytes(), x, y)
            })
            .collect();
        let mut xs = Vec::with_capacity(EVM_ITEMS * 8);
        let mut ys = Vec::with_capacity(EVM_ITEMS * 8);
        self.k0.clear();
        for (k, x, y) in starts {
            self.k0.push(k);
            xs.extend_from_slice(&x);
            ys.extend_from_slice(&y);
        }
        write_buf(&self.queue, &mut self.bx, &xs)?;
        write_buf(&self.queue, &mut self.by, &ys)?;
        self.launches = 0;
        Ok(())
    }

    /// Run one launch; `Some(sk)` is a GPU hit (still needs CPU re-verify).
    fn launch(&mut self) -> Result<Option<[u8; 32]>, String> {
        write_buf(&self.queue, &mut self.found_buf, &[-1 as cl_int])?;
        write_buf(&self.queue, &mut self.off_buf, &[0u32; 2])?;
        unsafe {
            ExecuteKernel::new(&self.kernel)
                .set_arg(&self.pat_buf)
                .set_arg(&self.found_buf)
                .set_arg(&self.off_buf)
                .set_arg(&self.bx)
                .set_arg(&self.by)
                .set_arg(&self.tabx)
                .set_arg(&self.taby)
                .set_global_work_size(EVM_ITEMS)
                .enqueue_nd_range(&self.queue)
                .map_err(|e| format!("enqueue: {e}"))?;
        }
        self.queue.finish().map_err(|e| format!("finish: {e}"))?;

        let mut found: [cl_int; 1] = [-1];
        read_buf(&self.queue, &self.found_buf, &mut found)?;
        let launch_start = self.launches * EVM_STEPS;
        self.launches += 1;
        let mut off = [0u32; 2];
        read_buf(&self.queue, &self.off_buf, &mut off)?;
        if found[0] < 0 {
            if off[1] != 0 {
                // A work-item hit B == ±j·G (astronomically rare); its state is invalid.
                self.reseed()?;
            }
            return Ok(None);
        }
        let item = found[0] as usize;
        let k0 = self
            .k0
            .get(item)
            .ok_or("GPU returned an out-of-range work-item")?;
        let mut tweak = [0u8; 32];
        tweak[24..].copy_from_slice(&(launch_start + u64::from(off[0])).to_be_bytes());
        let tweak = Scalar::from_be_bytes(tweak).map_err(|_| "tweak out of range".to_string())?;
        let sk = SecretKey::from_slice(k0)
            .and_then(|k| k.add_tweak(&tweak))
            .map_err(|e| format!("secret reconstruction: {e}"))?;
        Ok(Some(sk.secret_bytes()))
    }

    #[allow(dead_code)]
    fn context(&self) -> &Context {
        &self.context
    }
}

fn pat_bytes(pat: &GpuPattern) -> Vec<u8> {
    let n = std::mem::size_of::<GpuPattern>();
    unsafe { std::slice::from_raw_parts(pat as *const GpuPattern as *const u8, n) }.to_vec()
}

fn make_buf<T>(context: &Context, flags: u64, mut data: Vec<T>) -> Result<Buffer<T>, String> {
    unsafe {
        Buffer::<T>::create(
            context,
            flags | CL_MEM_COPY_HOST_PTR,
            data.len(),
            data.as_mut_ptr() as *mut _,
        )
    }
    .map_err(|e| format!("buffer: {e}"))
}

fn write_buf<T>(queue: &CommandQueue, buf: &mut Buffer<T>, data: &[T]) -> Result<(), String> {
    unsafe { queue.enqueue_write_buffer(buf, CL_BLOCKING, 0, data, &[]) }
        .map(|_| ())
        .map_err(|e| format!("write buffer: {e}"))
}

fn read_buf<T>(queue: &CommandQueue, buf: &Buffer<T>, out: &mut [T]) -> Result<(), String> {
    unsafe { queue.enqueue_read_buffer(buf, CL_BLOCKING, 0, out, &[]) }
        .map(|_| ())
        .map_err(|e| format!("read buffer: {e}"))
}

fn evm_benchmark(duration_secs: f64) -> Result<(f64, String), String> {
    let (mut session, name) = EvmSession::new(&never_pattern("evm")?)?;
    let start = Instant::now();
    let mut attempts = 0u64;
    while start.elapsed().as_secs_f64() < duration_secs {
        session.launch()?;
        attempts += EVM_PER_LAUNCH;
    }
    Ok((
        attempts as f64 / start.elapsed().as_secs_f64().max(0.001),
        name,
    ))
}

fn grind_evm<G: ChainGrinder + Clone>(
    grinder: G,
    patterns: &[Pattern],
    cancel: &CancelToken,
    on_progress: impl Fn(u64, f64, f64) + Sync,
) -> Result<(GrindResult, crate::gpu::engine::GpuGrindReport), String> {
    // GPU MVP: single OR-alternative, no wildcards (enforced in GpuPattern).
    let packed = GpuPattern::for_evm(&patterns[0])?;
    let (mut session, name) = EvmSession::new(&packed)?;
    let expected = crate::pattern::expected_attempts_any(
        &patterns
            .iter()
            .map(|p| grinder.expected_attempts(p))
            .collect::<Vec<_>>(),
    );

    let start = Instant::now();
    let mut attempts = 0u64;
    let mut rejected = 0u32;
    loop {
        if cancel.is_cancelled() {
            return Err("cancelled".into());
        }
        let hit = session.launch()?;
        attempts += EVM_PER_LAUNCH;
        if let Some(sk) = hit {
            // CPU re-verify — never trust the GPU alone.
            let keypair = grinder.finalize(GrindAttempt::Secret32(sk));
            if patterns
                .iter()
                .any(|p| grinder.matches(&keypair.address, p))
            {
                let elapsed = start.elapsed().as_secs_f64();
                return Ok((
                    GrindResult {
                        keypair,
                        attempts,
                        elapsed_secs: elapsed,
                    },
                    crate::gpu::engine::GpuGrindReport {
                        device_name: name,
                        backend: "opencl",
                        attempts,
                        elapsed_secs: elapsed,
                    },
                ));
            }
            rejected += 1;
            if rejected > 8 {
                return Err("GPU kernel produced repeated invalid hits; use --gpu off".into());
            }
            // Device state is stale after an early exit — start from fresh points.
            session.reseed()?;
        }
        let secs = start.elapsed().as_secs_f64().max(0.001);
        let rate = attempts as f64 / secs;
        let eta_min = (expected - attempts as f64).max(0.0) / rate / 60.0;
        on_progress(attempts, rate, eta_min);
    }
}

fn normalize_id(id: &str) -> Result<&'static str, String> {
    let id = id.to_ascii_lowercase();
    match id.as_str() {
        "evm" | "eth" | "ethereum" => Ok("evm"),
        "sol" | "solana" => Ok("sol"),
        other => Err(format!(
            "OpenCL GPU MVP supports sol and evm only (got '{other}')"
        )),
    }
}

/// A pattern no address can match, so benchmarks measure real work (an empty
/// pattern matches instantly and makes every other work-item exit early).
fn never_pattern(id: &str) -> Result<GpuPattern, String> {
    let (fill, ignore_case) = if id == "evm" {
        ("f", true)
    } else {
        ("z", false)
    };
    let never = fill.repeat(32);
    let p = Pattern {
        prefix: never.clone(),
        suffix: String::new(),
        contains: String::new(),
        prefix_match: never,
        suffix_match: String::new(),
        contains_match: String::new(),
        ignore_case,
        fixed_prefix_len: 0,
    };
    if id == "evm" {
        GpuPattern::for_evm(&p)
    } else {
        GpuPattern::for_solana(&p)
    }
}

fn run_batch(
    context: &Context,
    queue: &CommandQueue,
    kernel: &Kernel,
    seed: &[u8; 32],
    batch_base: u64,
    pat: &GpuPattern,
    global: usize,
) -> Result<Option<[u8; 32]>, String> {
    let mut found: cl_int = -1;
    let mut out_sk = [0u8; 32];
    let batch_base_arg = batch_base as cl_ulong;

    // Pass GpuPattern as one buffer — Apple OpenCL mis-bound interleaved scalar +
    // __constant pointer args (prefix_len was read as 0 → every key "matched").
    let seed_buf = unsafe {
        Buffer::<u8>::create(
            context,
            CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
            32,
            seed.as_ptr() as *mut _,
        )
    }
    .map_err(|e| format!("seed buf: {e}"))?;
    let pat_buf = unsafe {
        Buffer::<u8>::create(
            context,
            CL_MEM_READ_ONLY | CL_MEM_COPY_HOST_PTR,
            std::mem::size_of::<GpuPattern>(),
            (pat as *const GpuPattern as *mut u8) as *mut _,
        )
    }
    .map_err(|e| format!("pat buf: {e}"))?;
    let found_buf = unsafe {
        Buffer::<cl_int>::create(
            context,
            CL_MEM_READ_WRITE | CL_MEM_COPY_HOST_PTR,
            1,
            &mut found as *mut cl_int as *mut _,
        )
    }
    .map_err(|e| format!("found buf: {e}"))?;
    let sk_buf = unsafe { Buffer::<u8>::create(context, CL_MEM_WRITE_ONLY, 32, ptr::null_mut()) }
        .map_err(|e| format!("sk buf: {e}"))?;

    unsafe {
        ExecuteKernel::new(kernel)
            .set_arg(&seed_buf)
            .set_arg(&batch_base_arg)
            .set_arg(&pat_buf)
            .set_arg(&found_buf)
            .set_arg(&sk_buf)
            .set_global_work_size(global)
            .enqueue_nd_range(queue)
            .map_err(|e| format!("enqueue: {e}"))?;
    }
    queue.finish().map_err(|e| format!("finish: {e}"))?;

    unsafe {
        queue
            .enqueue_read_buffer(
                &found_buf,
                CL_BLOCKING,
                0,
                std::slice::from_mut(&mut found),
                &[],
            )
            .map_err(|e| format!("read found: {e}"))?;
    }
    if found < 0 {
        return Ok(None);
    }
    unsafe {
        queue
            .enqueue_read_buffer(&sk_buf, CL_BLOCKING, 0, &mut out_sk, &[])
            .map_err(|e| format!("read sk: {e}"))?;
    }
    Ok(Some(out_sk))
}

// Silence unused import if KeypairResult unused in some cfgs.
#[allow(dead_code)]
fn _touch(_: &KeypairResult) {}

#[cfg(test)]
mod tests {
    use super::*;
    use crate::chain::ChainGrinder;
    use crate::grinder::CancelToken;

    /// Linux CI / headless hosts often have OpenCL headers but no usable device.
    /// Skip rather than fail so `cargo test -p vanity-address` stays green there.
    fn require_gpu() -> bool {
        match first_gpu() {
            Ok(_) => true,
            Err(e) => {
                eprintln!("skipping OpenCL GPU test: {e}");
                false
            }
        }
    }

    #[test]
    fn builds_evm_kernel() {
        if !require_gpu() {
            return;
        }
        let (device, context, _q) = first_gpu().expect("gpu");
        build_program(&context, &device, &evm_source()).expect("evm build");
    }

    #[test]
    fn builds_sol_kernel() {
        if !require_gpu() {
            return;
        }
        let (device, context, _q) = first_gpu().expect("gpu");
        build_program(&context, &device, SOL_SRC).expect("sol build");
    }

    #[test]
    fn evm_finds_short_prefix() {
        if !require_gpu() {
            return;
        }
        use crate::chains::EvmGrinder;
        let g = EvmGrinder;
        let pat = g.build_pattern(Some("a"), None, None, false).unwrap();
        let cancel = CancelToken::new();
        let (result, report) = grind(g, &[pat], &cancel, |_, _, _| {}).expect("gpu grind");
        assert!(
            result
                .keypair
                .address
                .to_ascii_lowercase()
                .starts_with("0xa"),
            "{}",
            result.keypair.address
        );
        assert!(report.attempts > 0);
        println!(
            "found {} on {} in {:.2}s ({} attempts)",
            result.keypair.address, report.device_name, result.elapsed_secs, result.attempts
        );
    }

    #[test]
    fn evm_finds_three_char_prefix() {
        if !require_gpu() {
            return;
        }
        use crate::chains::EvmGrinder;
        let g = EvmGrinder;
        let pat = g.build_pattern(Some("abc"), None, None, false).unwrap();
        assert_eq!(GpuPattern::for_evm(&pat).unwrap().prefix_len, 3);
        let cancel = CancelToken::new();
        let (result, report) = grind(g, &[pat], &cancel, |_, _, _| {}).expect("gpu grind abc");
        let addr = result.keypair.address.to_ascii_lowercase();
        assert!(
            addr.starts_with("0xabc"),
            "expected 0xabc… got {addr} after {} attempts",
            result.attempts
        );
        assert!(
            result.attempts < 5_000_000,
            "took {} attempts — pattern match likely broken again",
            result.attempts
        );
        println!(
            "found {} on {} in {:.2}s ({} attempts)",
            result.keypair.address, report.device_name, result.elapsed_secs, result.attempts
        );
    }

    #[test]
    fn evm_hit_after_many_launches_reconstructs_key() {
        if !require_gpu() {
            return;
        }
        use crate::chains::EvmGrinder;
        let g = EvmGrinder;
        // ~268M expected candidates = dozens of launches, so the launch offset math is exercised.
        let pat = g.build_pattern(Some("abcdef1"), None, None, false).unwrap();
        let cancel = CancelToken::new();
        let (result, _) = grind(g, &[pat], &cancel, |_, _, _| {}).expect("gpu grind");
        assert!(
            result.attempts > EVM_PER_LAUNCH,
            "needs more than one launch to prove offset math"
        );
        assert!(result
            .keypair
            .address
            .to_ascii_lowercase()
            .starts_with("0xabcdef1"));
    }

    #[test]
    fn evm_benchmark_reports_real_throughput() {
        if !require_gpu() {
            return;
        }
        let (rate, _) = benchmark("evm", 1.0).expect("bench");
        assert!(rate > 1_000_000.0, "implausibly slow: {rate}");
    }

    #[test]
    fn sol_finds_short_prefix() {
        if !require_gpu() {
            return;
        }
        use crate::chains::SolanaGrinder;
        let g = SolanaGrinder;
        let pat = g.build_pattern(Some("A"), None, None, true).unwrap();
        let cancel = CancelToken::new();
        let (result, report) = grind(g, &[pat], &cancel, |_, _, _| {}).expect("sol gpu grind");
        assert!(
            result.keypair.address.starts_with('A'),
            "{}",
            result.keypair.address
        );
        assert!(report.attempts > 0);
        println!(
            "found {} on {} in {:.2}s ({} attempts)",
            result.keypair.address, report.device_name, result.elapsed_secs, result.attempts
        );
    }
}