primal-barracuda 0.4.0

barraCuda — BARrier-free Rust Abstracted Cross-platform Unified Dimensional Algebra. Vendor-agnostic GPU compute via WGSL.
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
// SPDX-License-Identifier: AGPL-3.0-or-later

//! Backbone update, torsion angles, FAPE loss, structure violation, pLDDT, confidence.

use crate::device::WgpuDevice;
use crate::device::compute_pipeline::ComputeDispatch;
use crate::error::Result;
use bytemuck::{Pod, Zeroable};
use std::sync::Arc;

use super::WG_64;

const SHADER_BACKBONE: &str = include_str!("../../shaders/misc/backbone_update_f64.wgsl");
const SHADER_TORSION: &str = include_str!("../../shaders/misc/torsion_angles_f64.wgsl");
const SHADER_FAPE: &str = include_str!("../../shaders/misc/fape_loss_f64.wgsl");
const SHADER_PLDDT: &str = include_str!("../../shaders/misc/plddt_f64.wgsl");
const SHADER_STRUCTURE_VIOLATION: &str =
    include_str!("../../shaders/misc/structure_violation_f64.wgsl");
const SHADER_CONFIDENCE: &str = include_str!("../../shaders/misc/confidence_f64.wgsl");

#[repr(C)]
#[derive(Copy, Clone, Pod, Zeroable)]
struct BackboneParams {
    n: u32,
    _pad: [u32; 3],
}

#[repr(C)]
#[derive(Copy, Clone, Pod, Zeroable)]
struct TorsionParams {
    n: u32,
    m: u32,
}

#[repr(C)]
#[derive(Copy, Clone, Pod, Zeroable)]
struct FapeParams {
    n_residues: u32,
    n_atoms: u32,
    _pad: [u32; 2],
    d_clamp: f64,
}

#[repr(C)]
#[derive(Copy, Clone, Pod, Zeroable)]
struct PlddtParams {
    n_residues: u32,
    _pad: [u32; 3],
    cutoff: f64,
}

#[repr(C)]
#[derive(Copy, Clone, Pod, Zeroable)]
struct StructureViolationParams {
    n_atoms: u32,
    n_bonds: u32,
    _pad: [u32; 2],
    d_vdw: f64,
}

#[repr(C)]
#[derive(Copy, Clone, Pod, Zeroable)]
struct ConfidenceParams {
    n_residues: u32,
    n_bins: u32,
    _pad: [u32; 2],
}

/// SE(3) backbone frame update: compose delta rotations/translations onto current frames.
///
/// Quaternion-based rigid-body update for structure module iteration.
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation fails, the device is lost, or buffer
/// readback fails.
pub fn backbone_update(
    device: &Arc<WgpuDevice>,
    quaternions: &[f64],
    translations: &[f64],
    delta_quat: &[f64],
    delta_trans: &[f64],
    n: u32,
) -> Result<(Vec<f64>, Vec<f64>)> {
    let quat_len = (n * 4) as usize;
    let trans_len = (n * 3) as usize;

    let q_buf = device.create_buffer_f64_init("backbone:quat", quaternions);
    let t_buf = device.create_buffer_f64_init("backbone:trans", translations);
    let dq_buf = device.create_buffer_f64_init("backbone:dq", delta_quat);
    let dt_buf = device.create_buffer_f64_init("backbone:dt", delta_trans);
    let oq_buf = device.create_buffer_f64(quat_len)?;
    let ot_buf = device.create_buffer_f64(trans_len)?;
    let params = BackboneParams { n, _pad: [0; 3] };
    let params_buf = device.create_uniform_buffer("backbone:params", &params);

    ComputeDispatch::new(device, "backbone_update")
        .shader(SHADER_BACKBONE, "main")
        .f64()
        .storage_read(0, &q_buf)
        .storage_read(1, &t_buf)
        .storage_read(2, &dq_buf)
        .storage_read(3, &dt_buf)
        .storage_rw(4, &oq_buf)
        .storage_rw(5, &ot_buf)
        .uniform(6, &params_buf)
        .dispatch((n as usize).div_ceil(WG_64 as usize) as u32, 1, 1)
        .submit()?;

    let out_q = device.read_f64_buffer(&oq_buf, quat_len)?;
    let out_t = device.read_f64_buffer(&ot_buf, trans_len)?;
    Ok((out_q, out_t))
}

/// Extract dihedral (torsion) angles from atomic coordinates.
///
/// For each group of 4 consecutive atoms, computes the dihedral angle using
/// the atan2 formula on cross-product vectors.
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation fails, the device is lost, or buffer
/// readback fails.
pub fn torsion_angles(
    device: &Arc<WgpuDevice>,
    positions: &[f64],
    n: u32,
    m: u32,
) -> Result<Vec<f64>> {
    let out_len = (n * m) as usize;
    let pos_buf = device.create_buffer_f64_init("torsion:pos", positions);
    let out_buf = device.create_buffer_f64(out_len)?;
    let params = TorsionParams { n, m };
    let params_buf = device.create_uniform_buffer("torsion:params", &params);

    ComputeDispatch::new(device, "torsion_angles")
        .shader(SHADER_TORSION, "main")
        .f64()
        .storage_read(0, &pos_buf)
        .storage_rw(1, &out_buf)
        .uniform(2, &params_buf)
        .dispatch(out_len.div_ceil(WG_64 as usize) as u32, 1, 1)
        .submit()?;

    device.read_f64_buffer(&out_buf, out_len)
}

/// Frame Aligned Point Error loss: per-residue FAPE.
///
/// Compares predicted vs true atom positions in local frames.
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation fails, the device is lost, or buffer
/// readback fails.
pub fn fape_loss(
    device: &Arc<WgpuDevice>,
    pred_pos: &[f64],
    true_pos: &[f64],
    pred_frames: &[f64],
    true_frames: &[f64],
    n_residues: u32,
    n_atoms: u32,
    d_clamp: f64,
) -> Result<Vec<f64>> {
    let out_len = n_residues as usize;
    let pred_pos_buf = device.create_buffer_f64_init("fape:pred_pos", pred_pos);
    let true_pos_buf = device.create_buffer_f64_init("fape:true_pos", true_pos);
    let pred_frames_buf = device.create_buffer_f64_init("fape:pred_frames", pred_frames);
    let true_frames_buf = device.create_buffer_f64_init("fape:true_frames", true_frames);
    let out_buf = device.create_buffer_f64(out_len)?;
    let params = FapeParams {
        n_residues,
        n_atoms,
        _pad: [0; 2],
        d_clamp,
    };
    let params_buf = device.create_uniform_buffer("fape:params", &params);

    ComputeDispatch::new(device, "fape_loss")
        .shader(SHADER_FAPE, "main")
        .f64()
        .storage_read(0, &pred_pos_buf)
        .storage_read(1, &true_pos_buf)
        .storage_read(2, &pred_frames_buf)
        .storage_read(3, &true_frames_buf)
        .storage_rw(4, &out_buf)
        .uniform(5, &params_buf)
        .dispatch(out_len.div_ceil(WG_64 as usize) as u32, 1, 1)
        .submit()?;

    device.read_f64_buffer(&out_buf, out_len)
}

/// Predicted LDDT confidence score per residue.
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation fails, the device is lost, or buffer
/// readback fails.
pub fn plddt(
    device: &Arc<WgpuDevice>,
    pred_pos: &[f64],
    true_pos: &[f64],
    n_residues: u32,
    cutoff: f64,
) -> Result<Vec<f64>> {
    let out_len = n_residues as usize;
    let pred_pos_buf = device.create_buffer_f64_init("plddt:pred_pos", pred_pos);
    let true_pos_buf = device.create_buffer_f64_init("plddt:true_pos", true_pos);
    let out_buf = device.create_buffer_f64(out_len)?;
    let params = PlddtParams {
        n_residues,
        _pad: [0; 3],
        cutoff,
    };
    let params_buf = device.create_uniform_buffer("plddt:params", &params);

    ComputeDispatch::new(device, "plddt")
        .shader(SHADER_PLDDT, "main")
        .f64()
        .storage_read(0, &pred_pos_buf)
        .storage_read(1, &true_pos_buf)
        .storage_rw(2, &out_buf)
        .uniform(3, &params_buf)
        .dispatch(out_len.div_ceil(WG_64 as usize) as u32, 1, 1)
        .submit()?;

    device.read_f64_buffer(&out_buf, out_len)
}

/// Steric clash + bond geometry violations.
///
/// Returns (`out_clash`, `out_bond`).
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation fails, the device is lost, or buffer
/// readback fails.
pub fn structure_violation(
    device: &Arc<WgpuDevice>,
    positions: &[f64],
    bond_pairs: &[u32],
    n_atoms: u32,
    n_bonds: u32,
    d_vdw: f64,
) -> Result<(Vec<f64>, Vec<f64>)> {
    let clash_len = n_atoms as usize;
    let bond_len = n_bonds as usize;
    let total_threads = clash_len + bond_len;

    let pos_buf = device.create_buffer_f64_init("struct_viol:pos", positions);
    let bond_buf = device.create_buffer_u32_init("struct_viol:bonds", bond_pairs);
    let clash_buf = device.create_buffer_f64(clash_len)?;
    let bond_out_buf = device.create_buffer_f64(bond_len)?;
    let params = StructureViolationParams {
        n_atoms,
        n_bonds,
        _pad: [0; 2],
        d_vdw,
    };
    let params_buf = device.create_uniform_buffer("struct_viol:params", &params);

    ComputeDispatch::new(device, "structure_violation")
        .shader(SHADER_STRUCTURE_VIOLATION, "main")
        .f64()
        .storage_read(0, &pos_buf)
        .storage_read(1, &bond_buf)
        .storage_rw(2, &clash_buf)
        .storage_rw(3, &bond_out_buf)
        .uniform(4, &params_buf)
        .dispatch(total_threads.div_ceil(WG_64 as usize) as u32, 1, 1)
        .submit()?;

    let out_clash = device.read_f64_buffer(&clash_buf, clash_len)?;
    let out_bond = device.read_f64_buffer(&bond_out_buf, bond_len)?;
    Ok((out_clash, out_bond))
}

/// Per-residue confidence from logits: `conf[i] = sum(softmax(logits[i,:]) * bin_centers)`.
///
/// # Errors
///
/// Returns [`Err`] if buffer allocation fails, the device is lost, or buffer
/// readback fails.
pub fn confidence(
    device: &Arc<WgpuDevice>,
    logits: &[f64],
    bin_centers: &[f64],
    n_residues: u32,
    n_bins: u32,
) -> Result<Vec<f64>> {
    let out_len = n_residues as usize;
    let logits_buf = device.create_buffer_f64_init("confidence:logits", logits);
    let bin_buf = device.create_buffer_f64_init("confidence:bin_centers", bin_centers);
    let out_buf = device.create_buffer_f64(out_len)?;
    let params = ConfidenceParams {
        n_residues,
        n_bins,
        _pad: [0; 2],
    };
    let params_buf = device.create_uniform_buffer("confidence:params", &params);

    ComputeDispatch::new(device, "confidence")
        .shader(SHADER_CONFIDENCE, "main")
        .f64()
        .storage_read(0, &logits_buf)
        .storage_read(1, &bin_buf)
        .storage_rw(2, &out_buf)
        .uniform(3, &params_buf)
        .dispatch(out_len.div_ceil(WG_64 as usize) as u32, 1, 1)
        .submit()?;

    device.read_f64_buffer(&out_buf, out_len)
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn params_layout_backbone() {
        assert_eq!(std::mem::size_of::<BackboneParams>(), 16);
    }

    #[test]
    fn params_layout_torsion() {
        assert_eq!(std::mem::size_of::<TorsionParams>(), 8);
    }

    #[test]
    fn params_layout_fape() {
        assert_eq!(std::mem::size_of::<FapeParams>(), 24);
    }

    #[test]
    fn params_layout_plddt() {
        assert_eq!(std::mem::size_of::<PlddtParams>(), 24);
    }

    #[test]
    fn params_layout_structure_violation() {
        assert_eq!(std::mem::size_of::<StructureViolationParams>(), 24);
    }

    #[test]
    fn params_layout_confidence() {
        assert_eq!(std::mem::size_of::<ConfidenceParams>(), 16);
    }

    #[cfg(feature = "gpu")]
    fn shader_compiles_via_naga(source: &str) {
        let full = crate::shaders::precision::ShaderTemplate::for_driver_auto(source, false);
        let normalized = format!(
            "enable f64;\n\n{}",
            full.replace("enable f64;\n\n", "")
                .replace("enable f64;\n", "")
        );
        if let Ok(module) = naga::front::wgsl::parse_str(&normalized) {
            naga::valid::Validator::new(
                naga::valid::ValidationFlags::all(),
                naga::valid::Capabilities::all(),
            )
            .validate(&module)
            .expect("WGSL validation failed");
            assert!(!module.entry_points.is_empty());
        } else {
            assert!(!full.is_empty());
            assert!(full.contains("fn main"), "shader must contain fn main");
            assert!(
                full.contains("@compute") || full.contains("main"),
                "expected compute entry"
            );
        }
    }

    #[test]
    #[cfg(feature = "gpu")]
    fn backbone_update_f64_compiles_via_naga() {
        let source = include_str!("../../shaders/misc/backbone_update_f64.wgsl");
        shader_compiles_via_naga(source);
    }

    #[test]
    #[cfg(feature = "gpu")]
    fn torsion_angles_f64_compiles_via_naga() {
        let source = include_str!("../../shaders/misc/torsion_angles_f64.wgsl");
        shader_compiles_via_naga(source);
    }

    #[test]
    #[cfg(feature = "gpu")]
    fn fape_loss_f64_compiles_via_naga() {
        let source = include_str!("../../shaders/misc/fape_loss_f64.wgsl");
        shader_compiles_via_naga(source);
    }

    #[test]
    #[cfg(feature = "gpu")]
    fn plddt_f64_compiles_via_naga() {
        let source = include_str!("../../shaders/misc/plddt_f64.wgsl");
        shader_compiles_via_naga(source);
    }

    #[test]
    #[cfg(feature = "gpu")]
    fn structure_violation_f64_compiles_via_naga() {
        let source = include_str!("../../shaders/misc/structure_violation_f64.wgsl");
        shader_compiles_via_naga(source);
    }

    #[test]
    #[cfg(feature = "gpu")]
    fn confidence_f64_compiles_via_naga() {
        let source = include_str!("../../shaders/misc/confidence_f64.wgsl");
        shader_compiles_via_naga(source);
    }
}