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
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
|
#include <utility>
#include <unordered_map>
#include <vector>
#include "pybind11/pybind11.h"
#include "pybind11/cast.h"
#include "pybind11/stl.h"
#include "cudnn_frontend.h"
namespace py = pybind11;
using namespace pybind11::literals;
namespace cudnn_frontend::python_bindings {
// This class is only meant direct pythonic API calls to c++ Graph class.
class PyGraph {
public:
template <cudnn_frontend::PointwiseMode_t MODE>
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
pointwise_ternary(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& a,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& b,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& c,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
template <cudnn_frontend::PointwiseMode_t MODE>
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
pointwise_binary(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& a,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& b,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
template <cudnn_frontend::PointwiseMode_t MODE>
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
pointwise_unary(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& a,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
// This Graph class is the sole structure which implicitly makes PyGraph own all tensors, nodes, and cudnn
// descriptors.
cudnn_frontend::graph::Graph graph;
cudnnHandle_t handle;
bool is_handle_owner = false;
PyGraph(std::string const&,
cudnn_frontend::DataType_t io_data_type,
cudnn_frontend::DataType_t intermediate_data_type,
cudnn_frontend::DataType_t compute_data_type,
std::optional<std::intptr_t> handle_,
py::object sm_count,
std::shared_ptr<KernelCache> kernel_cache) {
graph.set_compute_data_type(compute_data_type)
.set_intermediate_data_type(intermediate_data_type)
.set_io_data_type(io_data_type);
if (handle_.has_value()) {
handle = static_cast<cudnnHandle_t>((void*)(handle_.value()));
} else {
detail::create_handle(&handle);
is_handle_owner = true;
}
if (sm_count.is(py::none()) == false) {
graph.set_sm_count(sm_count.cast<int32_t>());
}
if (kernel_cache) {
graph.set_kernel_cache(kernel_cache);
graph.set_dynamic_shape_enabled(true);
}
}
~PyGraph() {
if (is_handle_owner) {
detail::destroy_handle(handle);
}
}
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
tensor(std::vector<int64_t> const& dim,
std::vector<int64_t> const& stride,
cudnn_frontend::DataType_t const& data_type,
bool const& is_virtual,
bool const& is_pass_by_value,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& ragged_offset,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
tensor_like(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& pyobj, std::string const&);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
tensor_like(py::object const& pyobj);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
batchnorm(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& bias,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& in_running_mean,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& in_running_var,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& epsilon,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& momentum,
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>& peer_stats,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
layernorm(cudnn_frontend::NormFwdPhase_t const forward_phase,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& bias,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& epsilon,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
batchnorm_inference(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& mean,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& inv_variance,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& bias,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
layernorm_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& dy,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& mean,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& inv_variance,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
batchnorm_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& dy,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& mean,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& inv_variance,
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>& peer_stats,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
slice(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
std::vector<py::slice> const& slices,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
conv_fprop(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& image,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& weight,
std::vector<int64_t> const& pre_padding,
std::vector<int64_t> const& post_padding,
std::vector<int64_t> const& stride,
std::vector<int64_t> const& dilation,
cudnn_frontend::ConvolutionMode_t const& conv_mode,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
conv_dgrad(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& loss,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& filter,
std::vector<int64_t> const& pre_padding,
std::vector<int64_t> const& post_padding,
std::vector<int64_t> const& stride,
std::vector<int64_t> const& dilation,
cudnn_frontend::ConvolutionMode_t const& conv_mode,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
conv_wgrad(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& image,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& loss,
std::vector<int64_t> const& pre_padding,
std::vector<int64_t> const& post_padding,
std::vector<int64_t> const& stride,
std::vector<int64_t> const& dilation,
cudnn_frontend::ConvolutionMode_t const& conv_mode,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
matmul(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& A,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& B,
cudnn_frontend::DataType_t const& compute_data_type,
double const padding,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
relu(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
std::optional<float> const& negative_slope,
std::optional<float> const& lower_clip,
std::optional<float> const& upper_clip,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
gen_index(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
int64_t const axis,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
relu_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& loss,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
std::optional<float> const& negative_slope,
std::optional<float> const& lower_clip,
std::optional<float> const& upper_clip,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
leaky_relu_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& loss,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
float const negative_slope,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
leaky_relu(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
float const negative_slope,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::array<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>, 2UL>
genstats(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
reduction(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input,
cudnn_frontend::ReductionMode_t const mode,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
reshape(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& input, std::string const& name);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
rmsnorm(cudnn_frontend::NormFwdPhase_t const forward_phase,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& bias,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& epsilon,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
rmsnorm_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& dy,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& inv_variance,
bool const has_dbias,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
instancenorm(cudnn_frontend::NormFwdPhase_t const forward_phase,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& bias,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& epsilon,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
std::vector<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>>
instancenorm_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& dy,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& x,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& mean,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes> const& inv_variance,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
// return [o, stats]
std::array<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>, 2>
sdpa(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& k,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& v,
bool const is_inference,
py::object const& attn_scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& bias,
bool const use_alibi_mask,
bool const use_padding_mask,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_kv,
bool const use_causal_mask,
bool const use_causal_mask_bottom_right,
py::object const& sliding_window_length,
py::object const& dropout,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& rng_dump,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& paged_attention_k_table,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& paged_attention_v_table,
py::object const& paged_attention_max_seq_len_kv,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
// return [dQ, dK, dV]
std::array<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>, 3>
sdpa_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& k,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& v,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& o,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& dO,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& stats,
py::object const& attn_scale,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& bias,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& dBias,
bool const use_alibi_mask,
bool const use_padding_mask,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_kv,
py::object const& max_total_seq_len_q,
py::object const& max_total_seq_len_kv,
bool const use_causal_mask,
bool const use_causal_mask_bottom_right,
py::object const& sliding_window_length,
py::object const& dropout,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& rng_dump,
bool const use_deterministic_algorithm,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
// return [o, stats, amax_s, amax_o]
std::array<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>, 4>
sdpa_fp8(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& k,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& v,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_k,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_v,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_s,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale_s,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale_o,
bool const is_inference,
py::object const& attn_scale,
bool const use_padding_mask,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_kv,
bool const use_causal_mask,
py::object const& dropout,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
// return [dQ, dK, dV, amax_dQ, amax_dK, amax_dV, amax_dP]
std::array<std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>, 7>
sdpa_fp8_backward(std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& k,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& v,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& o,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& dO,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& stats,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_k,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_v,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_o,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_dO,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_s,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& descale_dP,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale_s,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale_dQ,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale_dK,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale_dV,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& scale_dP,
py::object const& attn_scale,
bool const use_padding_mask,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_q,
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>& seq_len_kv,
bool const use_causal_mask,
py::object const& dropout,
cudnn_frontend::DataType_t const& compute_data_type,
std::string const& name);
void
validate();
size_t
key();
void
build_operation_graph();
void
create_execution_plans(std::vector<cudnn_frontend::HeurMode_t> const&);
void
build_plans(BuildPlanPolicy_t const);
void
build_plan_at_index(int64_t const index);
void
check_support();
void
build(std::vector<cudnn_frontend::HeurMode_t> const&);
int64_t
get_workspace_size();
void
populate_cuda_graph(std::intptr_t handle,
std::unordered_map<cudnn_frontend::graph::Tensor_attributes::uid_t, int64_t> var_pack,
std::intptr_t workspace,
std::intptr_t cuda_graph);
void
update_cuda_graph(std::intptr_t handle,
std::unordered_map<cudnn_frontend::graph::Tensor_attributes::uid_t, int64_t> var_pack,
std::intptr_t workspace,
std::intptr_t cuda_graph);
void
execute(std::unordered_map<int64_t, int64_t> var_pack, int64_t workspace, std::optional<std::intptr_t>);
void
execute_plan_at_index(std::unordered_map<int64_t, int64_t> var_pack,
int64_t workspace,
int64_t index,
std::optional<std::intptr_t>);
void
select_numeric_notes(std::vector<NumericalNote_t> const& notes) {
graph.select_numeric_notes(notes);
return;
}
void
select_behavior_notes(std::vector<BehaviorNote_t> const& notes) {
graph.select_behavior_notes(notes);
return;
}
void
deselect_engines(std::vector<std::string> const& engine_names) {
graph.deselect_engines(engine_names);
return;
}
void
deselect_numeric_notes(std::vector<NumericalNote_t> const& notes) {
graph.deselect_numeric_notes(notes);
return;
}
void
deselect_behavior_notes(std::vector<BehaviorNote_t> const& notes) {
graph.deselect_behavior_notes(notes);
return;
}
void
deselect_workspace_greater_than(int64_t const workspace) {
graph.deselect_workspace_greater_than(workspace);
return;
}
std::vector<uint8_t>
serialize() const;
void
deserialize(py::object const& pyobj);
int64_t
get_execution_plan_count() const {
return graph.get_execution_plan_count();
}
int64_t
get_workspace_size_plan_at_index(int64_t index);
std::shared_ptr<cudnn_frontend::graph::Tensor_attributes>
query_tensor_attributes_of_uid(int64_t const uid) const;
};
} // namespace cudnn_frontend::python_bindings
|
