Skip to content

Qualcomm AI Engine Direct - Add Unittest for Saver #12325

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.

Sign up for GitHub

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

Merged
cccclai merged 2 commits into from
Jul 16, 2025
Merged

Qualcomm AI Engine Direct - Add Unittest for Saver #12325

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
24ee9ef
Qualcomm AI Engine Direct - Add Unittest for Saver
DannyYuyang-quic Jul 9, 2025
ca4c955
Merge branch 'main' into dev1/danny/add_saver_ut
DannyYuyang-quic Jul 15, 2025
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
5 changes: 5 additions & 0 deletions backends/qualcomm/qnn_preprocess.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -178,6 +178,11 @@ def preprocess_multimethod(

if len(py_op_wrapper_list) == len(edge_programs.values()):
qnn_context_binary = qnn_manager.Compile(graph_name, py_op_wrapper_list)
if option.saver:
# TODO: Currently, only the first method is saved. Update this logic if saving multiple methods becomes necessary in the future.
Copy link
Contributor

@cccclai cccclai Jul 10, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Can you share what it means?

Copy link
Contributor Author

@DannyYuyang-quic DannyYuyang-quic Jul 11, 2025

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Consider the scenario for lowering a multi-method pte:

When the saver option is enabled in QNN, the program is configured to not generate context binary. So, the condition len(qnn_context_binary) != 0 will be False. In this case, we hope the program can exit immediately before reaching the assertion checks. As a result, only the result for method_0 will be saved. If support for saving multi-method with option.saver=True is required in the future, this will need to be extended.

exit(
f"Record all QNN API calls from saver backend at: {option.saver_output_dir}"
)
assert (
len(qnn_context_binary) != 0
), "Failed to generate Qnn context binary."
Expand Down
32 changes: 32 additions & 0 deletions backends/qualcomm/tests/test_qnn_delegate.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -3384,6 +3384,38 @@ def test_qnn_backend_rewrite_prepared_observer(self):
quantized_module = convert_pt2e(prepared)
self.lower_module_and_test_output(quantized_module, sample_input)

def test_qnn_backend_saver_backend(self):
backend_options = generate_htp_compiler_spec(use_fp16=False)
TestQNN.compiler_specs = generate_qnn_executorch_compiler_spec(
soc_model=self.chipset_table[TestQNN.model],
backend_options=backend_options,
saver=True,
)
module = Relu() # noqa: F405
sample_input = (torch.randn([2, 5, 1, 3]),)
module = self.get_qdq_module(module, sample_input)

from executorch.backends.qualcomm.serialization.qc_schema_serialize import (
flatbuffer_to_option,
option_to_flatbuffer,
)

with tempfile.TemporaryDirectory() as tmp_dir:
option = flatbuffer_to_option(TestQNN.compiler_specs[0].value)
option.saver_output_dir = f"{tmp_dir}/saver_output"
TestQNN.compiler_specs[0].value = option_to_flatbuffer(option)

with self.assertRaises(SystemExit):
self.lower_module_and_test_output(module, sample_input)
self.assertTrue(
os.path.isfile(f"{tmp_dir}/saver_output/params.bin"),
"failed to find params.bin",
)
self.assertTrue(
os.path.isfile(f"{tmp_dir}/saver_output/saver_output.c"),
"failed to find saver_output.c",
)

def test_qnn_backend_skip_node_id_partitioner(self):
module = SimpleModel() # noqa: F405
sample_input = (torch.ones(1, 32, 28, 28), torch.ones(1, 32, 28, 28))
Expand Down
Loading