2021-11-11 18:23:20 +00:00
|
|
|
#include <ATen/Utils.h>
|
|
|
|
|
#include <c10/core/TensorImpl.h>
|
2021-02-26 19:51:29 +00:00
|
|
|
#include <torch/csrc/jit/backends/backend.h>
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
#include <torch/csrc/jit/backends/backend_exception.h>
|
2022-02-24 17:30:14 +00:00
|
|
|
|
|
|
|
|
#ifndef NO_PROFILING
|
2021-10-12 17:49:55 +00:00
|
|
|
#include <torch/csrc/jit/mobile/profiler_edge.h>
|
2022-02-24 17:30:14 +00:00
|
|
|
#endif
|
2021-02-26 19:51:29 +00:00
|
|
|
|
|
|
|
|
namespace torch {
|
|
|
|
|
namespace jit {
|
|
|
|
|
|
|
|
|
|
// Implementation of a PyTorch Backend that can process, compile and execute
|
|
|
|
|
// TorchScript Modules composed of 'add' and 'sub' operators. It just supports
|
|
|
|
|
// for modules that implement a sum or subtraction of 2 inputs (i.e. in1 + in2
|
|
|
|
|
// or in1 - in2). Hence the methods of the models expect exactly 2 inputs of
|
|
|
|
|
// type Tensor. This backend is used to demonstrate the flow of compilation and
|
|
|
|
|
// execution with minimum amount of work. It's not intended to a practical
|
|
|
|
|
// backend that can be used for actual inference.
|
|
|
|
|
|
|
|
|
|
// Implementation details:
|
|
|
|
|
//
|
|
|
|
|
// Compilation
|
|
|
|
|
// 1. A backend with minimum compilation features, "backend_with_compiler_demo"
|
|
|
|
|
// is added.
|
|
|
|
|
// 2. The compilation happens AOT in the preprocess function registered to this
|
|
|
|
|
// backend.
|
|
|
|
|
// 3. Compiled results are stored in a string blob for each method. They are
|
|
|
|
|
// serialized to the lowered module with __getstate__ function.
|
|
|
|
|
// 4. Error message with model source code is thrown, for features not handled
|
|
|
|
|
// by the backend compiler.
|
|
|
|
|
//
|
|
|
|
|
// Runtime
|
|
|
|
|
// 1. The compiled blob is loaded in __setstate__ method.
|
|
|
|
|
// 2. The compile function of the backend: parse the preprocessed blob to the
|
|
|
|
|
// format (a list of tokens) that the backend can understand.
|
|
|
|
|
// 3. The execute function of the backend executes the specified method
|
|
|
|
|
// (handle).
|
|
|
|
|
|
|
|
|
|
namespace {
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
std::vector<std::tuple<std::string, int64_t>> parseMethodHandle(
|
|
|
|
|
const std::string& blob) {
|
|
|
|
|
std::vector<std::tuple<std::string, int64_t>> result;
|
2021-02-26 19:51:29 +00:00
|
|
|
std::stringstream s_stream(blob);
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
constexpr char debug_handle_token[] = "<debug_handle>";
|
2021-02-26 19:51:29 +00:00
|
|
|
while (s_stream.good()) {
|
|
|
|
|
std::string substr;
|
|
|
|
|
getline(s_stream, substr, ',');
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
auto debug_handle_pos = substr.find(debug_handle_token);
|
|
|
|
|
int64_t debug_handle{-1};
|
|
|
|
|
auto instruction = substr.substr(0);
|
|
|
|
|
if (debug_handle_pos != std::string::npos) {
|
|
|
|
|
instruction = substr.substr(0, debug_handle_pos);
|
|
|
|
|
debug_handle = stoi(substr.substr(debug_handle_pos + 14));
|
|
|
|
|
}
|
|
|
|
|
result.push_back(std::make_tuple(instruction, debug_handle));
|
2021-02-26 19:51:29 +00:00
|
|
|
}
|
|
|
|
|
return result;
|
|
|
|
|
}
|
2021-11-11 18:23:20 +00:00
|
|
|
|
|
|
|
|
float* float_data_ptr(const at::Tensor& t) {
|
|
|
|
|
return t.unsafeGetTensorImpl()->data_ptr_impl<float>();
|
|
|
|
|
}
|
2021-02-26 19:51:29 +00:00
|
|
|
} // namespace
|
|
|
|
|
|
|
|
|
|
class BackendWithCompiler : public PyTorchBackendInterface {
|
|
|
|
|
public:
|
|
|
|
|
// Constructor.
|
Make PyTorch code-base clang-tidy compliant (#56892)
Summary:
This is an automatic change generated by the following script:
```
#!/usr/bin/env python3
from subprocess import check_output, check_call
import os
def get_compiled_files_list():
import json
with open("build/compile_commands.json") as f:
data = json.load(f)
files = [os.path.relpath(node['file']) for node in data]
for idx, fname in enumerate(files):
if fname.startswith('build/') and fname.endswith('.DEFAULT.cpp'):
files[idx] = fname[len('build/'):-len('.DEFAULT.cpp')]
return files
def run_clang_tidy(fname):
check_call(["python3", "tools/clang_tidy.py", "-c", "build", "-x", fname,"-s"])
changes = check_output(["git", "ls-files", "-m"])
if len(changes) == 0:
return
check_call(["git", "commit","--all", "-m", f"NOLINT stubs for {fname}"])
def main():
git_files = check_output(["git", "ls-files"]).decode("ascii").split("\n")
compiled_files = get_compiled_files_list()
for idx, fname in enumerate(git_files):
if fname not in compiled_files:
continue
if fname.startswith("caffe2/contrib/aten/"):
continue
print(f"[{idx}/{len(git_files)}] Processing {fname}")
run_clang_tidy(fname)
if __name__ == "__main__":
main()
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56892
Reviewed By: H-Huang
Differential Revision: D27991944
Pulled By: malfet
fbshipit-source-id: 5415e1eb2c1b34319a4f03024bfaa087007d7179
2021-04-28 21:09:06 +00:00
|
|
|
// NOLINTNEXTLINE(modernize-use-equals-default)
|
2021-02-26 19:51:29 +00:00
|
|
|
explicit BackendWithCompiler() {}
|
Make PyTorch code-base clang-tidy compliant (#56892)
Summary:
This is an automatic change generated by the following script:
```
#!/usr/bin/env python3
from subprocess import check_output, check_call
import os
def get_compiled_files_list():
import json
with open("build/compile_commands.json") as f:
data = json.load(f)
files = [os.path.relpath(node['file']) for node in data]
for idx, fname in enumerate(files):
if fname.startswith('build/') and fname.endswith('.DEFAULT.cpp'):
files[idx] = fname[len('build/'):-len('.DEFAULT.cpp')]
return files
def run_clang_tidy(fname):
check_call(["python3", "tools/clang_tidy.py", "-c", "build", "-x", fname,"-s"])
changes = check_output(["git", "ls-files", "-m"])
if len(changes) == 0:
return
check_call(["git", "commit","--all", "-m", f"NOLINT stubs for {fname}"])
def main():
git_files = check_output(["git", "ls-files"]).decode("ascii").split("\n")
compiled_files = get_compiled_files_list()
for idx, fname in enumerate(git_files):
if fname not in compiled_files:
continue
if fname.startswith("caffe2/contrib/aten/"):
continue
print(f"[{idx}/{len(git_files)}] Processing {fname}")
run_clang_tidy(fname)
if __name__ == "__main__":
main()
```
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56892
Reviewed By: H-Huang
Differential Revision: D27991944
Pulled By: malfet
fbshipit-source-id: 5415e1eb2c1b34319a4f03024bfaa087007d7179
2021-04-28 21:09:06 +00:00
|
|
|
// NOLINTNEXTLINE(modernize-use-override)
|
2021-02-26 19:51:29 +00:00
|
|
|
virtual ~BackendWithCompiler() = default;
|
|
|
|
|
|
Adds a bool is_available() method to the backend contract (#53068)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/53068
Adds a ```bool is_available()``` method to the backend contract: it returns ```true``` if ```compile()``` and ```execute()``` can be called; ```false``` otherwise.
It is used to implement the following changes in the ```LoweredModule```:
* ```compile()``` in ```__setstate__``` will run if ```is_available()```, else ```__setstate__``` throws an exception (“Backend not available.”).
* ```compile()``` at ```LoweredModule``` creation will run if ```is_available()```, else a WARNING will be thrown.
* ```execute()``` will only be executed if ```is_available()``` returns true; else throws an exception (“Backend not available.”).
The goal of these changes is to ensure we have a well defined behaviour for the different combinations of backend availability on-host and on-target.
More specifically, backends may have different capabilities to compile and/or execute the Module, depending whether this happens on-host (i.e. where the program is being written) or on-target (where the program is being executed).
First of all, we know that "preprocess" always takes place, and that only happens on-host at creation time. So, we can assume that any compilation is needed/possible on-host then all of it could be pushed here.
Overall, we want to ensure the following:
**On host**
| compile | execute | Outcome |
| -- | -- | -- |
| No | No | On module creation, LoweredModule is generated, with a warning (since compilation and execution can still take place on-target). On module load, throws an exception (since execution is not possible). |
| No | Yes | This configuration should not be possible. This assumes the full compiler is not available, even if some work was done in preprocess the program cannot be finalized for execution. |
| Yes | No | In this case, the expectation would be for is_available() to return false, and compilation logic to move into preprocess. |
| Yes | Yes | All good. This is the only case that is_available() should return true. |
**On target**
| compile | execute | Outcome |
| -- | -- | -- |
| No | No | Loading the LoweredModule throws an exception. Since execution is not possible. |
| No | Yes | Basically this is another instance of Yes/Yes: compilation per se may not be possible on device, which means compile() can be called without issue but it is a no-op, and thus is_available should return true. Consequently, loading the LoweredModule: Succeeds, if the preprocessed module is ready for execution. Fails with exception otherwise. |
| Yes | No | This configuration should not be possible. Just putting here for completeness. |
| Yes | Yes | All good. This, along with No/Yes case (because compilation is assumed to have happened on-host, so it's just another instance of Yes/Yes), are the cases where is_available() should return true. |
**Refactoring existing code**
This change also updates other backends (Glow) code, to implement the is_available() method to have the same behaviour as before this change (i.e. always available).
This should not cause backward incompatibilities with already saved models since we're adding a new method to the PyTorchBackendInterface.
Models saved with the old interface that didn't have is_available() will still find the other 2 methods in the bound object (i.e. compile and execute), and the saved LoweredModule logic will be the old one.
**Future**
We plan to use is_available() to implement support for fallback to the PyTorch interpreter.
ghstack-source-id: 123498571
Test Plan: Added C++ (test_backend.cpp) and Python (test_backends.py) tests to validate the exceptions.
Reviewed By: jackm321, spaugh, iseeyuan
Differential Revision: D26615833
fbshipit-source-id: 562e8b11db25784348b5f86bbc4179aedf15e0d3
2021-03-10 08:21:34 +00:00
|
|
|
bool is_available() override {
|
|
|
|
|
return true;
|
|
|
|
|
}
|
|
|
|
|
|
2022-02-14 23:39:56 +00:00
|
|
|
// Since the actual compilation is done AOT for this backend, compile just
|
|
|
|
|
// forwards everything along. In a non toy setup this could grab information
|
|
|
|
|
// from that runtime that might be relevant to execute, such as build flags
|
|
|
|
|
// the resolution of the devices camera, or basically any runtime specific
|
|
|
|
|
// information that wouldnt be available server side where preprocess is
|
|
|
|
|
// called.
|
2021-02-26 19:51:29 +00:00
|
|
|
c10::impl::GenericDict compile(
|
|
|
|
|
c10::IValue processed,
|
|
|
|
|
c10::impl::GenericDict method_compile_spec) override {
|
|
|
|
|
auto dict = processed.toGenericDict();
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
auto handles =
|
|
|
|
|
c10::Dict<std::string, std::vector<std::tuple<std::string, int64_t>>>();
|
2021-02-26 19:51:29 +00:00
|
|
|
for (const auto& kv : dict) {
|
|
|
|
|
auto tokens = parseMethodHandle(kv.value().toStringRef());
|
|
|
|
|
handles.insert(kv.key().toStringRef(), tokens);
|
|
|
|
|
}
|
|
|
|
|
return c10::impl::toGenericDict(handles);
|
|
|
|
|
}
|
|
|
|
|
|
2022-02-14 23:39:56 +00:00
|
|
|
// Function that actually executes the model in the backend. Here there is
|
|
|
|
|
// nothing to dispatch to, so the backend is implemented locally within
|
|
|
|
|
// execute and it only supports add, subtract, and constant. In a non toy
|
|
|
|
|
// backend you can imagine how this function could be used to actually
|
|
|
|
|
// dispatch the inputs to the relevant backend/device.
|
2021-02-26 19:51:29 +00:00
|
|
|
c10::impl::GenericList execute(
|
2022-02-14 23:39:56 +00:00
|
|
|
c10::IValue
|
|
|
|
|
handle, // example: [('prim::Constant#1', 14), ('aten::add', 15)]
|
2021-02-26 19:51:29 +00:00
|
|
|
c10::impl::GenericList inputs) override {
|
|
|
|
|
TORCH_INTERNAL_ASSERT(inputs.size() == 2);
|
|
|
|
|
c10::IValue val0 = inputs[0];
|
|
|
|
|
at::Tensor x = val0.toTensor();
|
|
|
|
|
c10::IValue val1 = inputs[1];
|
|
|
|
|
at::Tensor h = val1.toTensor();
|
2021-10-12 17:49:55 +00:00
|
|
|
std::vector<std::tuple<int64_t, int64_t, std::string>> op_runtimes_us;
|
|
|
|
|
op_runtimes_us.reserve(handle.toList().size());
|
2021-02-26 19:51:29 +00:00
|
|
|
|
|
|
|
|
c10::List<at::Tensor> output_list;
|
2022-02-24 17:30:14 +00:00
|
|
|
#ifndef NO_PROFILING
|
2021-12-15 20:48:27 +00:00
|
|
|
auto start_us = torch::profiler::impl::getTime() / 1000;
|
2022-02-24 17:30:14 +00:00
|
|
|
#endif
|
2021-02-26 19:51:29 +00:00
|
|
|
for (const auto& token : handle.toList()) {
|
|
|
|
|
IValue val = token;
|
2021-11-02 17:13:02 +00:00
|
|
|
auto instruction = val.toTupleRef().elements()[0].toStringRef();
|
|
|
|
|
auto debug_handle = val.toTupleRef().elements()[1].toInt();
|
2022-02-24 17:30:14 +00:00
|
|
|
#ifndef NO_PROFILING
|
2021-12-15 20:48:27 +00:00
|
|
|
auto start_time_us = torch::profiler::impl::getTime() / 1000;
|
2022-02-24 17:30:14 +00:00
|
|
|
#endif
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
try {
|
|
|
|
|
if (instruction.rfind("prim::Constant", 0) == 0) {
|
2022-02-14 23:39:56 +00:00
|
|
|
// 15 is the length of 'prim::Constant#' the constant val comes after
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
TORCH_CHECK(
|
|
|
|
|
instruction.size() > 15,
|
|
|
|
|
"Constant value is expected in ",
|
|
|
|
|
instruction);
|
|
|
|
|
// NOLINTNEXTLINE(cppcoreguidelines-avoid-magic-numbers)
|
|
|
|
|
auto sub = instruction.substr(15);
|
2021-11-11 18:23:20 +00:00
|
|
|
} else if (instruction == "aten::add" || instruction == "aten::sub") {
|
|
|
|
|
TORCH_CHECK(x.sizes() == h.sizes());
|
|
|
|
|
if (x.dim() > 1 || (x.dim() == 1 && x.size(0) > 1)) {
|
|
|
|
|
TORCH_WARN(
|
|
|
|
|
"Only the first elements of the tensors are added or subbed.");
|
|
|
|
|
}
|
|
|
|
|
TORCH_CHECK(
|
|
|
|
|
(x.scalar_type() == c10::ScalarType::Float &&
|
|
|
|
|
h.scalar_type() == c10::ScalarType::Float),
|
|
|
|
|
"Only float tensors are compatible for add and sub.");
|
2022-01-18 23:57:30 +00:00
|
|
|
at::Tensor y = at::detail::empty_cpu(x.sizes(), at::kFloat);
|
2021-11-11 18:23:20 +00:00
|
|
|
auto x_ptr = float_data_ptr(x);
|
|
|
|
|
auto h_ptr = float_data_ptr(h);
|
|
|
|
|
auto y_ptr = float_data_ptr(y);
|
2022-07-28 23:08:52 +00:00
|
|
|
#ifndef NO_PROFILING
|
|
|
|
|
RECORD_BACKEND_MEMORY_EVENT_TO_EDGE_PROFILER(
|
|
|
|
|
x_ptr,
|
|
|
|
|
x.numel() * sizeof(float),
|
|
|
|
|
x.numel() * sizeof(float),
|
|
|
|
|
x.numel() * sizeof(float) + y.numel() * sizeof(float) +
|
|
|
|
|
h.numel() * sizeof(float),
|
|
|
|
|
c10::Device(c10::kCPU));
|
|
|
|
|
#endif
|
2021-11-11 18:23:20 +00:00
|
|
|
if (instruction == "aten::add") {
|
|
|
|
|
y_ptr[0] = x_ptr[0] + h_ptr[0];
|
|
|
|
|
} else {
|
|
|
|
|
y_ptr[0] = x_ptr[0] - h_ptr[0];
|
|
|
|
|
}
|
|
|
|
|
output_list.emplace_back(y);
|
[Pytorch Backend delegation] Add api for backend lowering to query debug (#55462)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/55462
handles and symbolicate exception callstack thrown from backend.
Objective of this diff is to achieve improve error reporting when
exceptions are raised from lowered backend. We would effectively like to
get the same model level stack trace that you would get without having
lowered some module to backend.
For example:
```
class AA(nn.Module):
def forward(self, x, y):
return x + y
class A(nn.Module):
def __init__(...):
self.AA0 = AA()
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
class B(nn.Module):
def forward(self, x):
return x + 2
class C(nn.Module):
def __init__(...):
self.A0 = A()
self.B0 = B()
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
```
If the we then do C().forward(torch.rand((2,3)), torch.rand(14,2))) we
will likely see error stack like:
```
C++ exception with description "The following operation failed in the TorchScript interpreter.
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in forward
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in forward
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
We would like to see the same error stack if we lowered C.A0 to some
backend.
With this diff we get something like:
```
Module hierarchy:top(C).A0(backend_with_compiler_demoLoweredModule).AA0(AA)
Traceback of TorchScript (most recent call last):
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.A0.forward(x, y) + self.B0.forward(x)
~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 5, in FunctionName_UNKNOWN
typed_inputs: List[Any] = [x, y, ]
if self.__backend.is_available() :
_0, = self.__backend.execute(self.__handles["forward"], typed_inputs)
~~~~~~~~~~~~~~~~~~~~~~ <--- HERE
assert isinstance(_0, Tensor)
return _0
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return self.AA0.forward(x, y) + 3
~~~~~~~~~~~~~~~~ <--- HERE
File "<string>", line 3, in FunctionName_UNKNOWN
def forward(self, x, y):
return x + y
~~~~~ <--- HERE
```
This is achieved in 3 parts:
Part 1:
A. BackendDebugInfoRecorder:
During backend lowering, in `to_backend`, before calling the preprocess
function corresponding to the backend. This will facilitate recording of
debug info (such as source range + inlined callstack) for the lowered module.
B. Instantiate WithBackendDebugInfoRecorder with BackendDebugInfoRecorder.
This initializes thread local pointer to BackendDebugInfoRecorder.
C. generate_debug_handles:
In preprocess function, the backend will call generate_debug_handles
for each method being lowered separately. generate_debug_handles
takes `Graph` of the method being lowered and returns a map
of Node*-to-debug_handles. Backend is responsible for storing debug
handles appropriately so as to raise exception (and later profiling)
using debug handles when the exception being raised corresponds to
particular Node that was lowered.
Inside generate_debug_handles, we will query the current
BackendDebugHandleInfoRecorder, that is issuing debug handles. This debug
handle manager will issue debug handles as well as record
debug_handles-to-<source range, inlined callstack> map.
D. Back in `to_backend`, once the preprocess function is has finished
lowering the module, we will call `stopRecord` on
BackendDebugInfoRecorder. This will return the debug info map. This
debug info is then stored inside the lowered module.
Part 2:
Serialization:
During serialization for bytecode (lite interpreter), we will do two
things:
1. Extract all the source ranges that are contained inside
debug_handles-to-<source range, inlined callstack> map for lowered
module. This will be source range corresponding to debug handles,
including what is there is inlined callstack. Since we replaced original
module with lowered module, we wont be serializing code for the original
module and thus no source range. That is why the source range will have
to be stored separately. We will lump all the source ranges for all the
lowered modules in one single debug_pkl file.
2. Then we will serialize debug_handles-to-<source range, inlined
callstack> map.
Now during deserialization we will be able to reconstruct
debug_handles-to-<source range, inlined callstack> map. Given all
debug_handles are unique we would not need any module information.
Test Plan:
Tests are added in test_backend.cpp
Tests are added in test_backend.cpp
Imported from OSS
Differential Revision:
D27621330
D27621330
Reviewed By: raziel
Pulled By: kimishpatel
fbshipit-source-id: 0650ec68cda0df0a945864658cab226a97ba1890
2021-05-22 15:31:46 +00:00
|
|
|
} else {
|
|
|
|
|
TORCH_CHECK(
|
|
|
|
|
false,
|
|
|
|
|
"Instruction, ",
|
|
|
|
|
instruction,
|
|
|
|
|
" is not supported. ",
|
|
|
|
|
"Contact the backend POC for details. ");
|
|
|
|
|
}
|
|
|
|
|
} catch (c10::Error& e) {
|
|
|
|
|
TORCH_DELEGATED_BACKEND_THROW(false, e.what(), debug_handle);
|
2021-02-26 19:51:29 +00:00
|
|
|
}
|
2022-02-24 17:30:14 +00:00
|
|
|
#ifndef NO_PROFILING
|
2021-12-15 20:48:27 +00:00
|
|
|
auto end_time_us = torch::profiler::impl::getTime() / 1000;
|
2021-10-12 17:49:55 +00:00
|
|
|
auto duration = end_time_us - start_time_us;
|
|
|
|
|
op_runtimes_us.emplace_back(duration, debug_handle, instruction);
|
2022-02-24 17:30:14 +00:00
|
|
|
#endif
|
2021-10-12 17:49:55 +00:00
|
|
|
}
|
2022-02-24 17:30:14 +00:00
|
|
|
#ifndef NO_PROFILING
|
2021-10-12 17:49:55 +00:00
|
|
|
for (const auto& tup : op_runtimes_us) {
|
|
|
|
|
RECORD_BACKEND_EVENT_TO_EDGE_PROFILER(
|
|
|
|
|
start_us,
|
|
|
|
|
start_us + std::get<0>(tup),
|
|
|
|
|
std::get<1>(tup),
|
|
|
|
|
std::get<2>(tup),
|
|
|
|
|
"test_backend");
|
|
|
|
|
start_us = start_us + std::get<0>(tup);
|
2021-02-26 19:51:29 +00:00
|
|
|
}
|
2022-02-24 17:30:14 +00:00
|
|
|
#endif
|
2021-02-26 19:51:29 +00:00
|
|
|
return c10::impl::toList(output_list);
|
|
|
|
|
}
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
namespace {
|
2021-04-06 07:53:59 +00:00
|
|
|
constexpr auto backend_name = "backend_with_compiler_demo";
|
|
|
|
|
static auto cls = torch::jit::backend<BackendWithCompiler>(backend_name);
|
2021-02-26 19:51:29 +00:00
|
|
|
} // namespace
|
|
|
|
|
|
|
|
|
|
} // namespace jit
|
|
|
|
|
} // namespace torch
|