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
mod cudnn;
pub use cudnn::*;
#[cfg(test)]
mod tests {
use super::*;
use cuda11_cudart_sys::{self, cudaMalloc, cudaStreamCreate, cudaMemcpy, cudaStreamSynchronize, cudaFree, cudaStreamDestroy, cudaMemcpyKind};
fn checkCudaStatus(status: cuda11_cudart_sys::cudaError_t ) {
if status != cuda11_cudart_sys::cudaError::cudaSuccess {
print!("cuda API failed with status \n");
panic!();
}
}
fn checkCudnnStatus(status: cudnnStatus_t) {
if status != cudnnStatus_t_CUDNN_STATUS_SUCCESS {
print!("cudnn API failed with status \n");
panic!();
}
}
// not a test for get driver version, but a test for linking
#[test]
fn link_test() {
struct CudaTensor {
device_data: *mut f32,
dim: Vec<usize>,
mm_data: Vec<f32>,
}
impl CudaTensor {
fn new() -> CudaTensor {
CudaTensor {
device_data: std::ptr::null_mut(),
dim: Vec::new(),
mm_data: Vec::new(),
}
}
fn new_raw(data: &[f32], shape: &[usize]) -> CudaTensor {
let mut device_data: *mut f32 = std::ptr::null_mut();
let elems: usize = shape.iter().product();
if elems != data.len() {
panic!();
}
unsafe {
println!("cudaMalloc");
checkCudaStatus(cudaMalloc(&mut device_data as *mut _ as *mut _,
std::mem::size_of::<f32>()*elems));
println!("cudaMemcpy");
cudaMemcpy(device_data as *mut _,
data.as_ptr() as *mut _,
std::mem::size_of::<f32>()*elems,
cudaMemcpyKind::cudaMemcpyHostToDevice);
}
CudaTensor {
device_data: device_data,
dim: shape.to_vec(),
mm_data: data.to_vec(),
}
}
fn _sync(&mut self) {
let elems: usize = self.dim.iter().product();
unsafe {
cudaMemcpy(self.mm_data.as_mut_ptr() as *mut _,
self.device_data as *mut _,
std::mem::size_of::<f32>()*elems,
cudaMemcpyKind::cudaMemcpyDeviceToHost);
}
}
}
impl Drop for CudaTensor {
fn drop(&mut self) {
if self.device_data != std::ptr::null_mut() {
unsafe {
println!("cudaFree");
checkCudaStatus(cudaFree(self.device_data as _));
}
}
}
}
impl std::fmt::Debug for CudaTensor {
fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
write!(f, "{:?}\n", self.dim)?;
write!(f, "{:?}", self.mm_data)
}
}
struct CudaConv {
a: f32,
}
impl CudaConv {
fn new() -> CudaConv {
CudaConv {
a: 0.,
}
}
fn forward(&self,
input: &CudaTensor,
filter: &CudaTensor,
alpha: f32,
beta: f32,
) -> CudaTensor {
unsafe {
// create cudnn handle
let mut cudnnHandle: cudnnHandle_t = std::ptr::null_mut();
checkCudnnStatus(cudnnCreate(&mut cudnnHandle));
// descriptors
let mut srcTensorDesc: cudnnTensorDescriptor_t = std::ptr::null_mut();
let mut dstTensorDesc: cudnnTensorDescriptor_t = std::ptr::null_mut();
let mut biasTensorDesc: cudnnTensorDescriptor_t = std::ptr::null_mut();
let mut filterDesc: cudnnFilterDescriptor_t = std::ptr::null_mut();
let mut convDesc: cudnnConvolutionDescriptor_t = std::ptr::null_mut();
// init descriptors
checkCudnnStatus(cudnnCreateTensorDescriptor(&mut srcTensorDesc));
checkCudnnStatus(cudnnCreateTensorDescriptor(&mut dstTensorDesc));
checkCudnnStatus(cudnnCreateTensorDescriptor(&mut biasTensorDesc));
checkCudnnStatus(cudnnCreateFilterDescriptor(&mut filterDesc));
checkCudnnStatus(cudnnCreateConvolutionDescriptor(&mut convDesc));
// set descriptors
checkCudnnStatus(cudnnSetTensor4dDescriptor(srcTensorDesc,
cudnnTensorFormat_t_CUDNN_TENSOR_NCHW,
cudnnDataType_t_CUDNN_DATA_FLOAT,
input.dim[0] as _,
input.dim[1] as _,
input.dim[2] as _,
input.dim[3] as _));
checkCudnnStatus(cudnnSetFilterNdDescriptor(filterDesc,
cudnnDataType_t_CUDNN_DATA_FLOAT,
cudnnTensorFormat_t_CUDNN_TENSOR_NCHW,
4,
vec![1, 1, 3, 3].as_ptr()));
(cudnnSetConvolutionNdDescriptor(convDesc,
2,
vec![1, 1].as_ptr(),
vec![1, 1].as_ptr(),
vec![1, 1].as_ptr(),
cudnnConvolutionMode_t_CUDNN_CROSS_CORRELATION,
cudnnDataType_t_CUDNN_DATA_FLOAT));
let mut tensorOuputDimA: Vec<i32> = vec![1,1,1,1];
checkCudnnStatus(cudnnGetConvolutionNdForwardOutputDim(convDesc,
srcTensorDesc,
filterDesc,
4,
tensorOuputDimA.as_mut_ptr()));
println!("tensorOuputDimA shape: {:?}", tensorOuputDimA);
let mut dst_data: *mut f32 = std::ptr::null_mut();
let output_elem: i32 = (&tensorOuputDimA).iter().product();
checkCudaStatus(cudaMalloc(&mut dst_data as *mut _ as *mut _,
std::mem::size_of::<f32>()*(output_elem as usize)));
//setTensorDesc(dstTensorDesc, tensorFormat, dataType, n, c, h, w);
//setTensorDesc(dstTensorDesc, tensorFormat, dataType, n, c, h, w);
checkCudnnStatus(cudnnSetTensor4dDescriptor(dstTensorDesc,
cudnnTensorFormat_t_CUDNN_TENSOR_NCHW,
cudnnDataType_t_CUDNN_DATA_FLOAT,
tensorOuputDimA[0],
tensorOuputDimA[1],
tensorOuputDimA[2],
tensorOuputDimA[3],));
let mut returnedAlgoCount: i32 = 0;
let mut one_algo: cudnnConvolutionFwdAlgoPerf_t = std::mem::uninitialized();
let mut results: Vec<cudnnConvolutionFwdAlgoPerf_t> = vec![one_algo; (cudnnConvolutionFwdAlgo_t_CUDNN_CONVOLUTION_FWD_ALGO_COUNT as usize) * 2];
checkCudnnStatus(cudnnFindConvolutionForwardAlgorithm(cudnnHandle,
srcTensorDesc,
filterDesc,
convDesc,
dstTensorDesc,
cudnnConvolutionFwdAlgo_t_CUDNN_CONVOLUTION_FWD_ALGO_COUNT as _,
&mut returnedAlgoCount,
results.as_mut_ptr()));
println!("returnedAlgoCount: {:?}", returnedAlgoCount);
let mut sizeInBytes: usize = 10;
checkCudnnStatus(cudnnGetConvolutionForwardWorkspaceSize(cudnnHandle,
srcTensorDesc,
filterDesc,
convDesc,
dstTensorDesc,
results[0].algo,
&mut sizeInBytes));
println!("sizeInbytes: {:?}", sizeInBytes);
let mut workspace: *mut f32 = std::ptr::null_mut();
cudaMalloc(&mut workspace as *mut _ as _, 128);
//
cudnnConvolutionForward(cudnnHandle,
&alpha as *const _ as _,
srcTensorDesc,
input.device_data as _,
filterDesc,
filter.device_data as _,
convDesc,
results[0].algo,
workspace as _,
sizeInBytes,
&beta as *const _ as _,
dstTensorDesc,
dst_data as _);
cudaFree(workspace as _);
cudaFree(dst_data as _);
cudnnDestroyConvolutionDescriptor(convDesc);
cudnnDestroyFilterDescriptor(filterDesc);
cudnnDestroyTensorDescriptor(srcTensorDesc);
cudnnDestroyTensorDescriptor(dstTensorDesc);
cudnnDestroyTensorDescriptor(biasTensorDesc);
cudnnDestroy(cudnnHandle);
}
CudaTensor::new()
}
}
unsafe {
println!("cudnn version {:?} compiled against cudart version {:?}",
cudnnGetVersion(),
cudnnGetCudartVersion());
let mut stream: cudaStream_t = std::ptr::null_mut();
checkCudaStatus(cudaStreamCreate(&mut stream as *mut _ as _));
let mut input = CudaTensor::new_raw(&vec![1., 2., 3., 4., 5., 6., 7., 8., 9.],
&vec![1, 1, 3, 3]);
let mut filter = CudaTensor::new_raw(&vec![1., 2., 3., 4., 5., 6., 7., 8., 9.],
&vec![1, 1, 3, 3]);
let mut conv = CudaConv::new();
let mut output = conv.forward(&input, &filter, 1., 0.);
input._sync();
println!("{:?}", input);
cudaStreamSynchronize(stream as _);
checkCudaStatus(cudaStreamDestroy(stream as _));
}
}
}