coeus-ops 0.14.0

Numerical tensor operators for Coeus.
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
use crate::ptr::{MutPtr, Ptr};
use coeus_core::{Backend, CpuAddressableStorage, CpuAddressableStorageMut, Scalar};
use coeus_sparse::CsrTensor;
use coeus_tensor::Tensor;

use super::validation::{
    csr_index, validate_csr, validate_csr_components, validate_csr_parts, validate_read,
};

/// Sparse Matrix-Vector multiplication (SpMV): y = A x
///
/// Computes multiplication of a sparse CSR matrix `A` and a dense vector `x`.
/// Returns a dense vector `y`.
#[inline]
pub fn spmv<T: Scalar, B: Backend>(
    a: &CsrTensor<T, B>,
    x: &Tensor<T, B>,
    backend: &B,
) -> Result<Tensor<T, B>, B::Error>
where
    B::DeviceBuffer<T>: CpuAddressableStorageMut<T>,
    B::DeviceBuffer<i64>: CpuAddressableStorage<i64>,
{
    validate_csr(a, "sparse matrix-vector multiplication")?;
    validate_read(
        x,
        "sparse matrix-vector multiplication",
        Some(&[a.shape()[1]]),
    )?;
    let rows = a.shape()[0];

    // alloc_on: every row r writes y[r] = sum via y_ptr.write — no zero-init needed.
    // SAFETY: `validate_csr` proves every CSR read is within initialized,
    // contiguous storage, and this kernel writes each output row exactly once.
    let mut y = unsafe { Tensor::<T, B>::alloc_on([rows], backend)? };

    let val_slice = a.values().as_slice();
    let col_slice = a.col_indices().as_slice();
    let row_slice = a.row_offsets().as_slice();

    let val_ptr = Ptr(val_slice.as_ptr());
    let col_ptr = Ptr(col_slice.as_ptr());
    let row_ptr = Ptr(row_slice.as_ptr());
    let y_ptr = MutPtr(y.as_mut_slice()?.as_mut_ptr());

    let x_slice = x.storage().as_slice();
    let x_ptr = Ptr(x_slice.as_ptr());
    let x_stride = x.layout().strides()[0];
    let x_offset = x.layout().offset();

    backend.parallel_for(0, rows, move |r| unsafe {
        let start = csr_index(row_ptr.read(r));
        let end = csr_index(row_ptr.read(r + 1));
        let mut sum = T::ZERO;
        for i in start..end {
            let col = csr_index(col_ptr.read(i));
            let val = val_ptr.read(i);
            let xv = x_ptr.read(x_offset + col * x_stride);
            sum += val * xv;
        }
        y_ptr.write(r, sum);
    });

    Ok(y)
}

/// Sparse-Dense Matrix multiplication (SpMM): C = A B
///
/// Multiplies a sparse CSR matrix `A` [M, K] by a dense matrix `B` [K, N].
/// Returns a dense matrix `C` [M, N].
#[inline]
pub fn spmm<T: Scalar, B: Backend>(
    a: &CsrTensor<T, B>,
    b: &Tensor<T, B>,
    backend: &B,
) -> Result<Tensor<T, B>, B::Error>
where
    B::DeviceBuffer<T>: CpuAddressableStorageMut<T>,
    B::DeviceBuffer<i64>: CpuAddressableStorage<i64>,
{
    validate_csr(a, "sparse-dense matrix multiplication")?;
    if b.ndim() != 2 {
        return Err(coeus_core::BackendError::UnsupportedRank {
            operation: "sparse-dense matrix multiplication",
            rank: b.ndim(),
            max_rank: 2,
        }
        .into());
    }
    let m = a.shape()[0];
    let k = a.shape()[1];
    let n = b.shape()[1];
    validate_read(b, "sparse-dense matrix multiplication", Some(&[k, n]))?;

    // alloc_on: parallel_for over rows writes every c[r,j] for j in 0..n — no zero-init needed.
    // SAFETY: `validate_csr` and `validate_read` prove all inputs are valid;
    // each row/column pair is written exactly once before success.
    let mut c = unsafe { Tensor::<T, B>::alloc_on([m, n], backend)? };

    let val_slice = a.values().as_slice();
    let col_slice = a.col_indices().as_slice();
    let row_slice = a.row_offsets().as_slice();

    let val_ptr = Ptr(val_slice.as_ptr());
    let col_ptr = Ptr(col_slice.as_ptr());
    let row_ptr = Ptr(row_slice.as_ptr());
    let c_ptr = MutPtr(c.as_mut_slice()?.as_mut_ptr());

    let b_slice = b.storage().as_slice();
    let b_ptr = Ptr(b_slice.as_ptr());
    let b_stride_row = b.layout().strides()[0];
    let b_stride_col = b.layout().strides()[1];
    let b_offset = b.layout().offset();

    let c_stride_row = c.layout().strides()[0];
    let c_stride_col = c.layout().strides()[1];
    let c_offset = c.layout().offset();

    backend.parallel_for(0, m, move |r| unsafe {
        let start = csr_index(row_ptr.read(r));
        let end = csr_index(row_ptr.read(r + 1));
        let mut row_accumulator = smallvec::SmallVec::<[T; 256]>::from_elem(T::ZERO, n);
        for i in start..end {
            let col = csr_index(col_ptr.read(i));
            let val = val_ptr.read(i);
            let b_col_offset = b_offset + col * b_stride_row;
            for j in 0..n {
                let bv = b_ptr.read(b_col_offset + j * b_stride_col);
                row_accumulator[j] += val * bv;
            }
        }
        for j in 0..n {
            c_ptr.write(
                c_offset + r * c_stride_row + j * c_stride_col,
                row_accumulator[j],
            );
        }
    });

    Ok(c)
}

/// Sparse-Dense Matrix multiplication values backward pass.
/// Computes the gradient with respect to sparse matrix A values.
#[inline]
pub fn spmm_backward_values<T: Scalar, B: Backend>(
    a_col_indices: &Tensor<i64, B>,
    a_row_offsets: &Tensor<i64, B>,
    a_shape: &[usize],
    b: &Tensor<T, B>,
    grad_out: &Tensor<T, B>,
    backend: &B,
) -> Result<Tensor<T, B>, B::Error>
where
    B::DeviceBuffer<T>: CpuAddressableStorageMut<T>,
    B::DeviceBuffer<i64>: CpuAddressableStorage<i64>,
{
    if a_shape.len() != 2 {
        return Err(coeus_core::BackendError::UnsupportedRank {
            operation: "sparse-dense values backward",
            rank: a_shape.len(),
            max_rank: 2,
        }
        .into());
    }
    if a_col_indices.ndim() != 1 || a_row_offsets.ndim() != 1 {
        return Err(coeus_core::BackendError::Storage {
            operation: "sparse-dense values backward",
            reason: "CSR indices and row offsets must be one-dimensional".to_owned(),
        }
        .into());
    }
    let nnz = a_col_indices.shape()[0];
    let rows_plus_one = a_shape[0]
        .checked_add(1)
        .ok_or(coeus_core::BackendError::Overflow {
            operation: "sparse-dense values backward",
            reason: "CSR row count",
        })?;
    if a_row_offsets.shape() != [rows_plus_one] {
        return Err(coeus_core::BackendError::ShapeMismatch {
            operation: "sparse-dense values backward",
            lhs: a_row_offsets.shape().to_vec(),
            rhs: vec![rows_plus_one],
        }
        .into());
    }
    validate_read(a_col_indices, "sparse-dense values backward", Some(&[nnz]))?;
    validate_read(
        a_row_offsets,
        "sparse-dense values backward",
        Some(&[rows_plus_one]),
    )?;
    if b.ndim() != 2 {
        return Err(coeus_core::BackendError::UnsupportedRank {
            operation: "sparse-dense values backward",
            rank: b.ndim(),
            max_rank: 2,
        }
        .into());
    }
    validate_read(b, "sparse-dense values backward", None)?;
    if grad_out.ndim() != 2 {
        return Err(coeus_core::BackendError::UnsupportedRank {
            operation: "sparse-dense values backward",
            rank: grad_out.ndim(),
            max_rank: 2,
        }
        .into());
    }
    validate_read(
        grad_out,
        "sparse-dense values backward",
        Some(&[a_shape[0], b.shape()[1]]),
    )?;
    validate_csr_parts(
        a_col_indices,
        a_row_offsets,
        a_shape[1],
        "sparse-dense values backward",
    )?;
    // alloc_on: every i in 0..nnz is written via grad_values_ptr.write(i, sum) — no zero-init needed.
    // SAFETY: the validated CSR row ranges partition `[0, nnz)`, so every
    // gradient slot is written exactly once before the kernel returns.
    let mut grad_values = unsafe { Tensor::<T, B>::alloc_on([nnz], backend)? };
    let m = a_shape[0];
    let n = b.shape()[1];

    let col_slice = a_col_indices.as_slice();
    let row_slice = a_row_offsets.as_slice();
    let b_slice = b.storage().as_slice();
    let grad_out_slice = grad_out.storage().as_slice();

    let col_ptr = Ptr(col_slice.as_ptr());
    let row_ptr = Ptr(row_slice.as_ptr());
    let b_ptr = Ptr(b_slice.as_ptr());
    let grad_out_ptr = Ptr(grad_out_slice.as_ptr());
    let grad_values_ptr = MutPtr(grad_values.as_mut_slice()?.as_mut_ptr());

    let b_stride_row = b.layout().strides()[0];
    let b_stride_col = b.layout().strides()[1];
    let b_offset = b.layout().offset();

    let go_stride_row = grad_out.layout().strides()[0];
    let go_stride_col = grad_out.layout().strides()[1];
    let go_offset = grad_out.layout().offset();

    backend.parallel_for(0, m, move |r| {
        // SAFETY: The raw pointers `row_ptr`, `col_ptr`, `b_ptr`, `grad_out_ptr`, and `grad_values_ptr`
        // point to valid memory buffers allocated by the tensor library. The parallel execution index `r`
        // is guaranteed to be within [0, m), which is safe to read. All offset math is bounded by the shapes
        // of B, grad_out, and A.
        unsafe {
            let start = csr_index(row_ptr.read(r));
            let end = csr_index(row_ptr.read(r + 1));
            let go_row_offset = go_offset + r * go_stride_row;
            for i in start..end {
                let col = csr_index(col_ptr.read(i));
                let b_col_offset = b_offset + col * b_stride_row;
                let mut sum = T::ZERO;
                for j in 0..n {
                    let go_v = grad_out_ptr.read(go_row_offset + j * go_stride_col);
                    let b_v = b_ptr.read(b_col_offset + j * b_stride_col);
                    sum += go_v * b_v;
                }
                grad_values_ptr.write(i, sum);
            }
        }
    });

    Ok(grad_values)
}

/// Sparse-Dense Matrix multiplication dense matrix backward pass.
/// Computes the gradient with respect to dense matrix B.
#[inline]
pub fn spmm_backward_dense<T: Scalar, B: Backend>(
    a_values: &Tensor<T, B>,
    a_col_indices: &Tensor<i64, B>,
    a_row_offsets: &Tensor<i64, B>,
    a_shape: &[usize],
    grad_out: &Tensor<T, B>,
    backend: &B,
) -> Result<Tensor<T, B>, B::Error>
where
    B::DeviceBuffer<T>: CpuAddressableStorageMut<T>,
    B::DeviceBuffer<i64>: CpuAddressableStorage<i64>,
{
    if a_shape.len() != 2 {
        return Err(coeus_core::BackendError::UnsupportedRank {
            operation: "sparse-dense backward",
            rank: a_shape.len(),
            max_rank: 2,
        }
        .into());
    }
    let m = a_shape[0];
    let k = a_shape[1];
    if grad_out.ndim() != 2 {
        return Err(coeus_core::BackendError::UnsupportedRank {
            operation: "sparse-dense backward",
            rank: grad_out.ndim(),
            max_rank: 2,
        }
        .into());
    }
    let n = grad_out.shape()[1];
    validate_csr_components(
        a_values,
        a_col_indices,
        a_row_offsets,
        a_shape,
        "sparse-dense backward",
    )?;
    validate_read(grad_out, "sparse-dense backward", Some(&[m, n]))?;
    // alloc_on: parallel_for over j writes every grad_b[col,j] for col in 0..k — no zero-init needed.
    // SAFETY: each output column `j` writes every `[col, j]` exactly once;
    // validated CSR metadata bounds all reads and accumulation indices.
    let mut grad_b = unsafe { Tensor::<T, B>::alloc_on([k, n], backend)? };

    let val_slice = a_values.as_slice();
    let col_slice = a_col_indices.as_slice();
    let row_slice = a_row_offsets.as_slice();
    let grad_out_slice = grad_out.storage().as_slice();

    let val_ptr = Ptr(val_slice.as_ptr());
    let col_ptr = Ptr(col_slice.as_ptr());
    let row_ptr = Ptr(row_slice.as_ptr());
    let grad_out_ptr = Ptr(grad_out_slice.as_ptr());
    let grad_b_ptr = MutPtr(grad_b.as_mut_slice()?.as_mut_ptr());

    let gb_stride_row = grad_b.layout().strides()[0];
    let gb_stride_col = grad_b.layout().strides()[1];
    let gb_offset = grad_b.layout().offset();

    let go_stride_row = grad_out.layout().strides()[0];
    let go_stride_col = grad_out.layout().strides()[1];
    let go_offset = grad_out.layout().offset();

    backend.parallel_for(0, n, move |j| {
        // SAFETY: The raw pointers `row_ptr`, `col_ptr`, `val_ptr`, `grad_out_ptr`, and `grad_b_ptr`
        // point to valid memory buffers allocated by the tensor library. The parallel execution index `j`
        // is guaranteed to be within [0, n), which is safe to read/write. Since each worker thread processes
        // a unique column index `j`, there are no data race write conflicts on `grad_b`.
        unsafe {
            let mut col_accumulator = smallvec::SmallVec::<[T; 1024]>::from_elem(T::ZERO, k);
            for r in 0..m {
                let start = csr_index(row_ptr.read(r));
                let end = csr_index(row_ptr.read(r + 1));
                let go_idx = go_offset + r * go_stride_row + j * go_stride_col;
                let go_v = grad_out_ptr.read(go_idx);
                if go_v == T::ZERO {
                    continue;
                }
                for i in start..end {
                    let col = csr_index(col_ptr.read(i));
                    let val = val_ptr.read(i);
                    col_accumulator[col] += val * go_v;
                }
            }
            for col in 0..k {
                let gb_idx = gb_offset + col * gb_stride_row + j * gb_stride_col;
                grad_b_ptr.write(gb_idx, col_accumulator[col]);
            }
        }
    });

    Ok(grad_b)
}

#[cfg(test)]
mod tests {
    use super::*;
    use coeus_core::{BackendError, SequentialBackend, Shape};

    fn csr(
        values: &[f32],
        columns: &[i64],
        offsets: &[i64],
        shape: [usize; 2],
    ) -> CsrTensor<f32, SequentialBackend> {
        let backend = SequentialBackend::new();
        CsrTensor::new(
            Shape::from(shape.to_vec()),
            Tensor::from_slice_on(vec![values.len()], values, &backend)
                .expect("invariant: test values allocate"),
            Tensor::from_slice_on(vec![columns.len()], columns, &backend)
                .expect("invariant: test columns allocate"),
            Tensor::from_slice_on(vec![offsets.len()], offsets, &backend)
                .expect("invariant: test offsets allocate"),
        )
    }

    #[test]
    fn malformed_row_offsets_are_rejected_before_pointer_access() {
        let backend = SequentialBackend::new();
        let matrix = csr(&[2.0], &[0], &[0, 2], [1, 1]);
        let x = Tensor::from_slice_on(vec![1], &[3.0], &backend)
            .expect("invariant: test vector allocates");
        let result = spmv(&matrix, &x, &backend);
        assert!(matches!(result, Err(BackendError::Storage { .. })));
    }

    #[test]
    fn invalid_column_indices_are_rejected() {
        let backend = SequentialBackend::new();
        let x = Tensor::from_slice_on(vec![2], &[3.0, 4.0], &backend)
            .expect("invariant: test vector allocates");
        let negative = csr(&[2.0], &[-1], &[0, 1], [1, 2]);
        assert!(matches!(
            spmv(&negative, &x, &backend),
            Err(BackendError::InvalidNumericInput { .. })
        ));
        let out_of_range = csr(&[2.0], &[2], &[0, 1], [1, 2]);
        assert!(matches!(
            spmv(&out_of_range, &x, &backend),
            Err(BackendError::IndexOutOfRange { .. })
        ));
    }

    #[test]
    fn empty_rows_and_valid_multiplication_keep_values() {
        let backend = SequentialBackend::new();
        let matrix = csr(&[2.0, 3.0], &[0, 2], &[0, 1, 1, 2], [3, 3]);
        let x = Tensor::from_slice_on(vec![3], &[5.0, 7.0, 11.0], &backend)
            .expect("invariant: test vector allocates");
        let y = spmv(&matrix, &x, &backend).expect("invariant: valid CSR multiplies");
        assert_eq!(y.as_slice(), &[10.0, 0.0, 33.0]);

        let dense = Tensor::from_slice_on(vec![3, 2], &[1.0, 2.0, 3.0, 4.0, 5.0, 6.0], &backend)
            .expect("invariant: test matrix allocates");
        let product = spmm(&matrix, &dense, &backend).expect("invariant: valid SpMM");
        assert_eq!(product.as_slice(), &[2.0, 4.0, 0.0, 0.0, 15.0, 18.0]);
    }
}