candle_core/
dummy_cuda_backend.rs

1//! Implementation of the Cuda backend when Cuda support has not been compiled in.
2//!
3#![allow(dead_code)]
4use crate::op::{BinaryOpT, CmpOp, ReduceOp, UnaryOpT};
5use crate::{CpuStorage, DType, Error, Layout, Result, Shape};
6
7#[derive(Debug, Clone)]
8pub struct CudaDevice;
9
10#[derive(Debug)]
11pub struct CudaStorage;
12
13impl CudaStorage {
14    pub fn transfer_to_device(&self, _dst: &CudaDevice) -> Result<Self> {
15        Err(Error::NotCompiledWithCudaSupport)
16    }
17}
18
19macro_rules! fail {
20    () => {
21        unimplemented!("cuda support has not been enabled, add `cuda` feature to enable.")
22    };
23}
24#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
25pub struct DeviceId(usize);
26
27impl CudaDevice {
28    pub fn new_with_stream(_: usize) -> Result<Self> {
29        Err(Error::NotCompiledWithCudaSupport)
30    }
31    pub fn id(&self) -> DeviceId {
32        DeviceId(0)
33    }
34}
35
36impl crate::backend::BackendStorage for CudaStorage {
37    type Device = CudaDevice;
38
39    fn try_clone(&self, _: &Layout) -> Result<Self> {
40        Err(Error::NotCompiledWithCudaSupport)
41    }
42
43    fn dtype(&self) -> DType {
44        fail!()
45    }
46
47    fn device(&self) -> &Self::Device {
48        fail!()
49    }
50
51    fn const_set(&mut self, _: crate::scalar::Scalar, _: &Layout) -> Result<()> {
52        Err(Error::NotCompiledWithCudaSupport)
53    }
54
55    fn to_cpu_storage(&self) -> Result<CpuStorage> {
56        Err(Error::NotCompiledWithCudaSupport)
57    }
58
59    fn affine(&self, _: &Layout, _: f64, _: f64) -> Result<Self> {
60        Err(Error::NotCompiledWithCudaSupport)
61    }
62
63    fn powf(&self, _: &Layout, _: f64) -> Result<Self> {
64        Err(Error::NotCompiledWithCudaSupport)
65    }
66
67    fn elu(&self, _: &Layout, _: f64) -> Result<Self> {
68        Err(Error::NotCompiledWithCudaSupport)
69    }
70
71    fn reduce_op(&self, _: ReduceOp, _: &Layout, _: &[usize]) -> Result<Self> {
72        Err(Error::NotCompiledWithCudaSupport)
73    }
74
75    fn cmp(&self, _: CmpOp, _: &Self, _: &Layout, _: &Layout) -> Result<Self> {
76        Err(Error::NotCompiledWithCudaSupport)
77    }
78
79    fn to_dtype(&self, _: &Layout, _: DType) -> Result<Self> {
80        Err(Error::NotCompiledWithCudaSupport)
81    }
82
83    fn unary_impl<B: UnaryOpT>(&self, _: &Layout) -> Result<Self> {
84        Err(Error::NotCompiledWithCudaSupport)
85    }
86
87    fn binary_impl<B: BinaryOpT>(&self, _: &Self, _: &Layout, _: &Layout) -> Result<Self> {
88        Err(Error::NotCompiledWithCudaSupport)
89    }
90
91    fn where_cond(&self, _: &Layout, _: &Self, _: &Layout, _: &Self, _: &Layout) -> Result<Self> {
92        Err(Error::NotCompiledWithCudaSupport)
93    }
94
95    fn conv1d(
96        &self,
97        _: &Layout,
98        _: &Self,
99        _: &Layout,
100        _: &crate::conv::ParamsConv1D,
101    ) -> Result<Self> {
102        Err(Error::NotCompiledWithCudaSupport)
103    }
104
105    fn conv_transpose1d(
106        &self,
107        _: &Layout,
108        _: &Self,
109        _: &Layout,
110        _: &crate::conv::ParamsConvTranspose1D,
111    ) -> Result<Self> {
112        Err(Error::NotCompiledWithCudaSupport)
113    }
114
115    fn conv2d(
116        &self,
117        _: &Layout,
118        _: &Self,
119        _: &Layout,
120        _: &crate::conv::ParamsConv2D,
121    ) -> Result<Self> {
122        Err(Error::NotCompiledWithCudaSupport)
123    }
124
125    fn conv_transpose2d(
126        &self,
127        _l: &Layout,
128        _kernel: &Self,
129        _kernel_l: &Layout,
130        _params: &crate::conv::ParamsConvTranspose2D,
131    ) -> Result<Self> {
132        Err(Error::NotCompiledWithCudaSupport)
133    }
134
135    fn index_select(&self, _: &Self, _: &Layout, _: &Layout, _: usize) -> Result<Self> {
136        Err(Error::NotCompiledWithCudaSupport)
137    }
138    fn gather(&self, _: &Layout, _: &Self, _: &Layout, _: usize) -> Result<Self> {
139        Err(Error::NotCompiledWithCudaSupport)
140    }
141
142    fn scatter_set(
143        &mut self,
144        _: &Layout,
145        _: &Self,
146        _: &Layout,
147        _: &Self,
148        _: &Layout,
149        _: usize,
150    ) -> Result<()> {
151        Err(Error::NotCompiledWithCudaSupport)
152    }
153
154    fn scatter_add_set(
155        &mut self,
156        _: &Layout,
157        _: &Self,
158        _: &Layout,
159        _: &Self,
160        _: &Layout,
161        _: usize,
162    ) -> Result<()> {
163        Err(Error::NotCompiledWithCudaSupport)
164    }
165
166    fn index_add(
167        &self,
168        _: &Layout,
169        _: &Self,
170        _: &Layout,
171        _: &Self,
172        _: &Layout,
173        _: usize,
174    ) -> Result<Self> {
175        Err(Error::NotCompiledWithCudaSupport)
176    }
177
178    fn matmul(
179        &self,
180        _: &Self,
181        _: (usize, usize, usize, usize),
182        _: &Layout,
183        _: &Layout,
184    ) -> Result<Self> {
185        Err(Error::NotCompiledWithCudaSupport)
186    }
187
188    fn copy_strided_src(&self, _: &mut Self, _: usize, _: &Layout) -> Result<()> {
189        Err(Error::NotCompiledWithCudaSupport)
190    }
191
192    fn copy2d(
193        &self,
194        _: &mut Self,
195        _: usize,
196        _: usize,
197        _: usize,
198        _: usize,
199        _: usize,
200        _: usize,
201    ) -> Result<()> {
202        Err(Error::NotCompiledWithCudaSupport)
203    }
204
205    fn avg_pool2d(&self, _: &Layout, _: (usize, usize), _: (usize, usize)) -> Result<Self> {
206        Err(Error::NotCompiledWithCudaSupport)
207    }
208
209    fn max_pool2d(&self, _: &Layout, _: (usize, usize), _: (usize, usize)) -> Result<Self> {
210        Err(Error::NotCompiledWithCudaSupport)
211    }
212
213    fn upsample_nearest1d(&self, _: &Layout, _: usize) -> Result<Self> {
214        Err(Error::NotCompiledWithCudaSupport)
215    }
216
217    fn upsample_nearest2d(&self, _: &Layout, _: usize, _: usize) -> Result<Self> {
218        Err(Error::NotCompiledWithCudaSupport)
219    }
220
221    fn upsample_bilinear2d(
222        &self,
223        _: &Layout,
224        _: usize,
225        _: usize,
226        _: bool,
227        _: Option<f64>,
228        _: Option<f64>,
229    ) -> Result<Self> {
230        Err(Error::NotCompiledWithCudaSupport)
231    }
232}
233
234impl crate::backend::BackendDevice for CudaDevice {
235    type Storage = CudaStorage;
236    fn new(_: usize) -> Result<Self> {
237        Err(Error::NotCompiledWithCudaSupport)
238    }
239
240    fn set_seed(&self, _: u64) -> Result<()> {
241        Err(Error::NotCompiledWithCudaSupport)
242    }
243
244    fn get_current_seed(&self) -> Result<u64> {
245        Err(Error::NotCompiledWithCudaSupport)
246    }
247
248    fn location(&self) -> crate::DeviceLocation {
249        fail!()
250    }
251
252    fn same_device(&self, _: &Self) -> bool {
253        fail!()
254    }
255
256    fn zeros_impl(&self, _shape: &Shape, _dtype: DType) -> Result<Self::Storage> {
257        Err(Error::NotCompiledWithCudaSupport)
258    }
259
260    unsafe fn alloc_uninit(&self, _shape: &Shape, _dtype: DType) -> Result<Self::Storage> {
261        Err(Error::NotCompiledWithCudaSupport)
262    }
263
264    fn storage_from_slice<T: crate::WithDType>(&self, _: &[T]) -> Result<Self::Storage> {
265        Err(Error::NotCompiledWithCudaSupport)
266    }
267
268    fn storage_from_cpu_storage(&self, _: &CpuStorage) -> Result<Self::Storage> {
269        Err(Error::NotCompiledWithCudaSupport)
270    }
271
272    fn storage_from_cpu_storage_owned(&self, _: CpuStorage) -> Result<Self::Storage> {
273        Err(Error::NotCompiledWithCudaSupport)
274    }
275
276    fn rand_uniform(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage> {
277        Err(Error::NotCompiledWithCudaSupport)
278    }
279
280    fn rand_normal(&self, _: &Shape, _: DType, _: f64, _: f64) -> Result<Self::Storage> {
281        Err(Error::NotCompiledWithCudaSupport)
282    }
283
284    fn synchronize(&self) -> Result<()> {
285        Ok(())
286    }
287}
288
289/// This bool controls whether reduced precision reductions (e.g., with fp16 accumulation type) are
290/// allowed with f16 GEMMs.
291pub fn gemm_reduced_precision_f16() -> bool {
292    true
293}
294
295/// This bool controls whether reduced precision reductions (e.g., with fp16 accumulation type) are
296/// allowed with f16 GEMMs.
297pub fn set_gemm_reduced_precision_f16(_: bool) {}
298
299/// This bool controls whether reduced precision reductions (e.g., with fp16 accumulation type) are
300/// allowed with bf16 GEMMs.
301pub fn gemm_reduced_precision_bf16() -> bool {
302    true
303}
304
305/// This bool controls whether reduced precision reductions (e.g., with fp16 accumulation type) are
306/// allowed with bf16 GEMMs.
307pub fn set_gemm_reduced_precision_bf16(_: bool) {}
308
309/// This bool controls whether reduced precision reductions (e.g., with tf32 accumulation type) are
310/// allowed with f32 GEMMs.
311pub fn gemm_reduced_precision_f32() -> bool {
312    true
313}
314
315/// This bool controls whether reduced precision reductions (e.g., with tf32 accumulation type) are
316/// allowed with f32 GEMMs.
317pub fn set_gemm_reduced_precision_f32(_b: bool) {}
candle_core/dummy_cuda_backend.rs

candle_core/
dummy_cuda_backend.rs
