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
//! Sparse matrix-vector and sparse matrix-matrix multiplication
use super::super::{CudaClient, CudaRuntime};
use crate::dtype::Element;
use crate::error::{Error, Result};
use crate::runtime::cuda::kernels::{
launch_csr_spmm, launch_csr_spmv, launch_csr_spmv_warp, should_use_warp_kernel,
};
use crate::tensor::Tensor;
impl CudaClient {
pub(crate) fn spmv_csr_impl<T: Element>(
&self,
row_ptrs: &Tensor<CudaRuntime>,
col_indices: &Tensor<CudaRuntime>,
values: &Tensor<CudaRuntime>,
x: &Tensor<CudaRuntime>,
shape: [usize; 2],
) -> Result<Tensor<CudaRuntime>> {
let [nrows, ncols] = shape;
let device = values.device();
let dtype = values.dtype();
// Validate input shapes
if x.numel() != ncols {
return Err(Error::ShapeMismatch {
expected: vec![ncols],
got: vec![x.numel()],
});
}
// Allocate output on GPU
let y = Tensor::<CudaRuntime>::zeros(&[nrows], dtype, device);
// Get device pointers (no data transfer!)
let row_ptrs_ptr = row_ptrs.ptr();
let col_indices_ptr = col_indices.ptr();
let values_ptr = values.ptr();
let x_ptr = x.ptr();
let y_ptr = y.ptr();
// Choose optimal kernel based on sparsity
let nnz = values.numel();
let avg_nnz_per_row = nnz as f32 / nrows.max(1) as f32;
// Dispatch based on dtype (only F32/F64/F16/BF16 supported on CUDA)
use crate::dtype::DType;
match dtype {
DType::F32 => unsafe {
if should_use_warp_kernel(avg_nnz_per_row) {
launch_csr_spmv_warp::<f32>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
} else {
launch_csr_spmv::<f32>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
}
},
DType::F64 => unsafe {
if should_use_warp_kernel(avg_nnz_per_row) {
launch_csr_spmv_warp::<f64>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
} else {
launch_csr_spmv::<f64>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
}
},
#[cfg(feature = "f16")]
DType::F16 => unsafe {
use half::f16;
if should_use_warp_kernel(avg_nnz_per_row) {
launch_csr_spmv_warp::<f16>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
} else {
launch_csr_spmv::<f16>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
}
},
#[cfg(feature = "f16")]
DType::BF16 => unsafe {
use half::bf16;
if should_use_warp_kernel(avg_nnz_per_row) {
launch_csr_spmv_warp::<bf16>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
} else {
launch_csr_spmv::<bf16>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
x_ptr,
y_ptr,
nrows,
)?;
}
},
_ => {
return Err(Error::Internal(format!(
"Unsupported dtype for CUDA sparse SpMV: {:?}",
dtype
)));
}
}
// Return GPU tensor (no transfer back!)
Ok(y)
}
pub(crate) fn spmm_csr_impl<T: Element>(
&self,
row_ptrs: &Tensor<CudaRuntime>,
col_indices: &Tensor<CudaRuntime>,
values: &Tensor<CudaRuntime>,
b: &Tensor<CudaRuntime>,
shape: [usize; 2],
) -> Result<Tensor<CudaRuntime>> {
let [m, k] = shape;
let device = values.device();
let dtype = values.dtype();
// Validate B is 2D
if b.ndim() != 2 {
return Err(Error::Internal(format!(
"Expected 2D tensor for SpMM, got {}D",
b.ndim()
)));
}
let b_shape = b.shape();
let b_k = b_shape[0];
let n = b_shape[1];
// Validate dimensions match
if b_k != k {
return Err(Error::ShapeMismatch {
expected: vec![k],
got: vec![b_k],
});
}
// Allocate output on GPU
let c = Tensor::<CudaRuntime>::zeros(&[m, n], dtype, device);
// Get device pointers (no data transfer!)
let row_ptrs_ptr = row_ptrs.ptr();
let col_indices_ptr = col_indices.ptr();
let values_ptr = values.ptr();
let b_ptr = b.ptr();
let c_ptr = c.ptr();
// Dispatch based on dtype (only F32/F64/F16/BF16 supported on CUDA)
use crate::dtype::DType;
match dtype {
DType::F32 => unsafe {
launch_csr_spmm::<f32>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
b_ptr,
c_ptr,
m,
n,
)?;
},
DType::F64 => unsafe {
launch_csr_spmm::<f64>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
b_ptr,
c_ptr,
m,
n,
)?;
},
#[cfg(feature = "f16")]
DType::F16 => unsafe {
use half::f16;
launch_csr_spmm::<f16>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
b_ptr,
c_ptr,
m,
n,
)?;
},
#[cfg(feature = "f16")]
DType::BF16 => unsafe {
use half::bf16;
launch_csr_spmm::<bf16>(
&self.context,
&self.stream,
self.device.index,
row_ptrs_ptr,
col_indices_ptr,
values_ptr,
b_ptr,
c_ptr,
m,
n,
)?;
},
_ => {
return Err(Error::Internal(format!(
"Unsupported dtype for CUDA sparse SpMM: {:?}",
dtype
)));
}
}
// Return GPU tensor (no transfer back!)
Ok(c)
}
}