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
|
#if !defined(C10_MOBILE) && !defined(ANDROID)
#include <ATen/DynamicLibrary.h>
#include <torch/csrc/inductor/aoti_runner/model_container_runner.h>
#include <torch/csrc/inductor/aoti_torch/oss_proxy_executor.h>
#include <torch/csrc/inductor/aoti_torch/tensor_converter.h>
#ifndef _WIN32
#include <sys/stat.h>
#else
#include <filesystem>
namespace fs = std::filesystem;
#endif
namespace {
bool file_exists(std::string& path) {
#ifdef _WIN32
return fs::exists(path);
#else
struct stat rc {};
return lstat(path.c_str(), &rc) == 0;
#endif
}
} // namespace
namespace torch::inductor {
AOTIModelContainerRunner::AOTIModelContainerRunner(
const std::string& model_so_path,
size_t num_models,
const std::string& device_str,
const std::string& cubin_dir) {
model_so_ = std::make_unique<at::DynamicLibrary>(model_so_path.c_str());
TORCH_CHECK(model_so_, "Failed to load model: ", model_so_path);
create_func_ = reinterpret_cast<decltype(create_func_)>(
model_so_->sym("AOTInductorModelContainerCreateWithDevice"));
delete_func_ = reinterpret_cast<decltype(delete_func_)>(
model_so_->sym("AOTInductorModelContainerDelete"));
get_num_outputs_func_ = reinterpret_cast<decltype(get_num_outputs_func_)>(
model_so_->sym("AOTInductorModelContainerGetNumOutputs"));
run_func_ = reinterpret_cast<decltype(run_func_)>(
model_so_->sym("AOTInductorModelContainerRun"));
get_num_constants_func_ = reinterpret_cast<decltype(get_num_constants_func_)>(
model_so_->sym("AOTInductorModelContainerGetNumConstants"));
get_constant_name_func_ = reinterpret_cast<decltype(get_constant_name_func_)>(
model_so_->sym("AOTInductorModelContainerGetConstantName"));
get_constant_original_fqn_func_ =
reinterpret_cast<decltype(get_constant_original_fqn_func_)>(
model_so_->sym("AOTInductorModelContainerGetConstantOriginalFQN"));
get_constant_dtype_func_ =
reinterpret_cast<decltype(get_constant_dtype_func_)>(
model_so_->sym("AOTInductorModelContainerGetConstantDtype"));
update_constant_buffer_func_ =
reinterpret_cast<decltype(update_constant_buffer_func_)>(
model_so_->sym("AOTInductorModelContainerUpdateConstantBuffer"));
update_inactive_constant_buffer_func_ =
reinterpret_cast<decltype(update_inactive_constant_buffer_func_)>(
model_so_->sym(
"AOTInductorModelContainerUpdateInactiveConstantBuffer"));
run_const_fold_func_ = reinterpret_cast<decltype(run_const_fold_func_)>(
model_so_->sym("AOTInductorModelContainerRunConstantFolding"));
swap_constant_buffer_func_ =
reinterpret_cast<decltype(swap_constant_buffer_func_)>(
model_so_->sym("AOTInductorModelContainerSwapConstantBuffer"));
get_call_spec_func_ = reinterpret_cast<decltype(get_call_spec_func_)>(
model_so_->sym("AOTInductorModelContainerGetCallSpec"));
// Hack to find the json file name from the model so file
size_t lastindex = model_so_path.find_last_of('.');
std::string json_filename = model_so_path.substr(0, lastindex) + ".json";
if (file_exists(json_filename)) {
proxy_executor_ = std::make_unique<torch::aot_inductor::OSSProxyExecutor>(
json_filename, device_str == "cpu");
proxy_executor_handle_ =
reinterpret_cast<AOTIProxyExecutorHandle>(proxy_executor_.get());
} else {
proxy_executor_handle_ = nullptr;
}
AOTI_RUNTIME_ERROR_CODE_CHECK(create_func_(
&container_handle_,
num_models,
device_str.c_str(),
cubin_dir.empty() ? nullptr : cubin_dir.c_str()));
}
AOTIModelContainerRunner::~AOTIModelContainerRunner() {
AOTIRuntimeError result = delete_func_(container_handle_);
TORCH_CHECK(
result == AOTI_RUNTIME_SUCCESS, "AOTInductorModelContainerDelete failed");
}
std::vector<at::Tensor> AOTIModelContainerRunner::run(
const std::vector<at::Tensor>& inputs,
void* stream_handle) {
auto input_handles =
torch::aot_inductor::unsafe_alloc_new_handles_from_tensors(inputs);
// For outputs, we only allocate a vector to hold returned tensor handles,
// not allocating the actual output tensor storage here
size_t num_outputs = 0;
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_num_outputs_func_(container_handle_, &num_outputs));
std::vector<AtenTensorHandle> output_handles(num_outputs);
AOTI_RUNTIME_ERROR_CODE_CHECK(run_func_(
container_handle_,
input_handles.data(),
input_handles.size(),
output_handles.data(),
output_handles.size(),
reinterpret_cast<AOTInductorStreamHandle>(stream_handle),
proxy_executor_handle_));
return torch::aot_inductor::alloc_tensors_by_stealing_from_handles(
output_handles.data(), output_handles.size());
}
std::unordered_map<std::string, std::string> AOTIModelContainerRunner::
getConstantNamesToOriginalFQNs() const {
std::unordered_map<std::string, std::string> result;
size_t num_constants{0};
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_num_constants_func_(container_handle_, &num_constants));
for (size_t i = 0; i < num_constants; ++i) {
const char* name{nullptr};
const char* original_fqn{nullptr};
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_constant_name_func_(container_handle_, i, &name));
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_constant_original_fqn_func_(container_handle_, i, &original_fqn));
result.emplace(name, original_fqn);
}
return result;
}
std::unordered_map<std::string, int32_t> AOTIModelContainerRunner::
getConstantNamesToDtypes() const {
std::unordered_map<std::string, int32_t> result;
size_t num_constants{0};
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_num_constants_func_(container_handle_, &num_constants));
for (size_t i = 0; i < num_constants; ++i) {
const char* name{nullptr};
int32_t dtype{0};
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_constant_name_func_(container_handle_, i, &name));
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_constant_dtype_func_(container_handle_, i, &dtype));
result.emplace(name, dtype);
}
return result;
}
void AOTIModelContainerRunner::update_constant_buffer(
const TensorConstantMap& const_map,
bool use_inactive,
bool check_full_update) {
AOTI_RUNTIME_ERROR_CODE_CHECK(update_constant_buffer_func_(
container_handle_,
(AOTInductorConstantMapHandle)&const_map,
use_inactive,
check_full_update));
}
void AOTIModelContainerRunner::update_constant_buffer(
std::unordered_map<std::string, at::Tensor>& tensor_map,
bool use_inactive,
bool check_full_update) {
TensorConstantMap const_map;
for (auto& [k, v] : tensor_map) {
const_map.emplace(k, &v);
}
AOTI_RUNTIME_ERROR_CODE_CHECK(update_constant_buffer_func_(
container_handle_,
(AOTInductorConstantMapHandle)&const_map,
use_inactive,
check_full_update));
}
void AOTIModelContainerRunner::update_inactive_constant_buffer(
const TensorConstantMap& const_map) {
AOTI_RUNTIME_ERROR_CODE_CHECK(update_inactive_constant_buffer_func_(
container_handle_, (AOTInductorConstantMapHandle)&const_map));
}
void AOTIModelContainerRunner::run_const_fold(
bool use_inactive,
AOTInductorStreamHandle cuda_stream_handle) {
AOTI_RUNTIME_ERROR_CODE_CHECK(run_const_fold_func_(
container_handle_,
use_inactive,
cuda_stream_handle,
proxy_executor_handle_));
}
void AOTIModelContainerRunner::swap_constant_buffer() {
AOTI_RUNTIME_ERROR_CODE_CHECK(swap_constant_buffer_func_(container_handle_));
}
std::vector<std::string> AOTIModelContainerRunner::get_call_spec() {
const char* in_spec = nullptr;
const char* out_spec = nullptr;
AOTI_RUNTIME_ERROR_CODE_CHECK(
get_call_spec_func_(container_handle_, &in_spec, &out_spec));
return {in_spec, out_spec};
}
std::unordered_map<std::string, CreateAOTIModelRunnerFunc>&
getAOTIModelRunnerRegistry() {
static std::unordered_map<std::string, CreateAOTIModelRunnerFunc>
aoti_model_runner_registry_;
return aoti_model_runner_registry_;
}
} // namespace torch::inductor
#endif
|
