Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Enable python debug runtime for exported network libraries #8793

Merged
merged 7 commits into from
Sep 2, 2021
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
20 changes: 18 additions & 2 deletions docs/dev/debugger.rst
Original file line number Diff line number Diff line change
Expand Up @@ -123,12 +123,12 @@ Example of loading the parameters
How to use Debugger?
***************************************

1. In ``config.cmake`` set the ``USE_GRAPH_EXECUTOR_DEBUG`` flag to ``ON``
1. In ``config.cmake`` set the ``USE_PROFILER`` flag to ``ON``

::

# Whether enable additional graph debug functions
set(USE_GRAPH_EXECUTOR_DEBUG ON)
set(USE_PROFILER ON)

2. Do 'make' tvm, so that it will make the ``libtvm_runtime.so``

Expand All @@ -148,6 +148,22 @@ How to use Debugger?
m.run()
tvm_out = m.get_output(0, tvm.nd.empty(out_shape, dtype)).numpy()

4. If network previously was exported to external libray using ``lib.export_library("network.so")``
like shared object file/dynamic linked library, the initialization
of debug runtime will be slightly different

::

lib = tvm.runtime.load_module("network.so")
m = graph_executor.create(lib["get_graph_json"](), lib, dev, dump_root="/tmp/tvmdbg")
# set inputs
m.set_input('data', tvm.nd.array(data.astype(dtype)))
m.set_input(**params)
# execute
m.run()
tvm_out = m.get_output(0, tvm.nd.empty(out_shape, dtype)).numpy()


The outputs are dumped to a temporary folder in ``/tmp`` folder or the
folder specified while creating the runtime.

Expand Down
4 changes: 4 additions & 0 deletions src/runtime/graph_executor/graph_executor_factory.cc
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,10 @@ PackedFunc GraphExecutorFactory::GetFunction(
}
*rv = this->ExecutorCreate(devices);
});
} else if (name == "get_graph_json") {
return PackedFunc(
[sptr_to_self, this](TVMArgs args, TVMRetValue* rv) { *rv = this->graph_json_; });

} else if (name == "debug_create") {
return PackedFunc([sptr_to_self, this](TVMArgs args, TVMRetValue* rv) {
ICHECK_GE(args.size(), 2);
Expand Down
45 changes: 24 additions & 21 deletions tests/python/unittest/test_runtime_module_based_interface.py
Original file line number Diff line number Diff line change
Expand Up @@ -46,10 +46,8 @@ def verify(data):
return out


@tvm.testing.requires_llvm
def test_legacy_compatibility():
if not tvm.testing.device_enabled("llvm"):
print("Skip because llvm is not enabled")
return
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
graph, lib, graph_params = relay.build_module.build(mod, "llvm", params=params)
Expand All @@ -63,10 +61,8 @@ def test_legacy_compatibility():
tvm.testing.assert_allclose(out, verify(data), atol=1e-5)


@tvm.testing.requires_llvm
def test_cpu():
if not tvm.testing.device_enabled("llvm"):
print("Skip because llvm is not enabled")
return
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
complied_graph_lib = relay.build_module.build(mod, "llvm", params=params)
Expand All @@ -90,6 +86,23 @@ def test_cpu():
tvm.testing.assert_allclose(out, verify(data), atol=1e-5)


@tvm.testing.requires_llvm
def test_cpu_get_graph_json():
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
complied_graph_lib = relay.build_module.build(mod, "llvm", params=params)
from tvm.contrib import utils

temp = utils.tempdir()
file_name = "deploy_lib.so"
path_lib = temp.relpath(file_name)
complied_graph_lib.export_library(path_lib)
loaded_lib = tvm.runtime.load_module(path_lib)
json = loaded_lib["get_graph_json"]()
assert isinstance(json, str) == True
assert json.find("tvmgen_default_fused_nn_softmax1") > -1


@tvm.testing.requires_cuda
@tvm.testing.requires_gpu
def test_gpu():
Expand Down Expand Up @@ -120,9 +133,6 @@ def test_gpu():
@tvm.testing.uses_gpu
def test_mod_export():
def verify_cpu_export(obj_format):
if not tvm.testing.device_enabled("llvm"):
print("Skip because llvm is not enabled")
return
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
complied_graph_lib = relay.build_module.build(mod, "llvm", params=params)
Expand Down Expand Up @@ -210,10 +220,8 @@ def setup_gmod():
out = gmod.get_output(0).numpy()
tvm.testing.assert_allclose(out, verify(data), atol=1e-5)

@tvm.testing.requires_llvm
def verify_rpc_cpu_export(obj_format):
if not tvm.testing.device_enabled("llvm"):
print("Skip because llvm is not enabled")
return
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
complied_graph_lib = relay.build_module.build(mod, "llvm", params=params)
Expand Down Expand Up @@ -308,12 +316,10 @@ def check_remote(server):
verify_rpc_gpu_export(obj_format)


@tvm.testing.requires_llvm
@tvm.testing.uses_gpu
def test_remove_package_params():
def verify_cpu_remove_package_params(obj_format):
if not tvm.testing.device_enabled("llvm"):
print("Skip because llvm is not enabled")
return
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
complied_graph_lib = relay.build_module.build(mod, "llvm", params=params)
Expand Down Expand Up @@ -404,10 +410,8 @@ def verify_gpu_remove_package_params(obj_format):
out = gmod.get_output(0).numpy()
tvm.testing.assert_allclose(out, verify(data), atol=1e-5)

@tvm.testing.requires_llvm
def verify_rpc_cpu_remove_package_params(obj_format):
if not tvm.testing.device_enabled("llvm"):
print("Skip because llvm is not enabled")
return
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
complied_graph_lib = relay.build_module.build(mod, "llvm", params=params)
Expand Down Expand Up @@ -517,10 +521,8 @@ def verify_rpc_gpu_remove_package_params(obj_format):
verify_rpc_gpu_remove_package_params(obj_format)


@tvm.testing.requires_llvm
def test_debug_graph_executor():
if not tvm.testing.device_enabled("llvm"):
print("Skip because llvm is not enabled")
return
mod, params = relay.testing.synthetic.get_workload()
with relay.build_config(opt_level=3):
complied_graph_lib = relay.build_module.build(mod, "llvm", params=params)
Expand Down Expand Up @@ -619,3 +621,4 @@ def make_module(mod):
test_remove_package_params()
test_debug_graph_executor()
test_multiple_imported_modules()
test_cpu_get_graph_json()