saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-07-24 19:43:35 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

[QNN EP] Update default QNN SDK to version 2.10.0 (#15818)

Browse source 
### Description
- Updates the default QNN SDK for CI pipelines to version 2.10.0.
- Disables convolution op tests that run on the QNN CPU backend due to a
potential bug with QNN SDK 2.10.0.

### Motivation and Context
Allows us to test the latest QNN SDK in default CI pipeline runs.
This commit is contained in:
Adrian Lizarraga 2023-05-05 13:01:21 -07:00 committed by GitHub
parent 2f6df99b89
commit 45f5c27632
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
7 changed files with 189 additions and 44 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

7
onnxruntime/test/providers/cpu/nn/conv_op_test.cc
View file

@ -59,10 +59,17 @@ void TestConvOp(const ConvOpAndTestAttributes& attributes,
std::unordered_set<std::string> excluded_providers(attributes.excluded_providers);
// Disable TensorRT because weight as input is not supported
excluded_providers.insert(kTensorrtExecutionProvider);
// QNN SDK 2.10.0 has a bug that breaks support for dynamic bias inputs.
excluded_providers.insert(kQnnExecutionProvider);
// TODO: Enable QNN EP when bug with QNN SDK 2.10.0 is fixed:
/*
// QNN have issue with dynamic weight, auto pad with SAME_UPPER, SAME_LOWER
if (!weight_is_initializer || attributes.auto_pad == "SAME_UPPER" || attributes.auto_pad == "SAME_LOWER") {
excluded_providers.insert(kQnnExecutionProvider);
}
*/
test.Run(expect_result, err_str, excluded_providers);
}

40
onnxruntime/test/providers/qnn/conv_htp_test.cc
View file

@ -1,40 +0,0 @@
// Copyright (c) Microsoft Corporation. All rights reserved.
// Licensed under the MIT License.
#if !defined(ORT_MINIMAL_BUILD)
#include <string>
#include "core/graph/graph.h"
#include "test/optimizer/qdq_test_utils.h"
#include "test/providers/qnn/qnn_test_utils.h"
#include "gtest/gtest.h"
namespace onnxruntime {
namespace test {
#if defined(__aarch64__) || defined(_M_ARM64) || defined(__linux__)
// Check that QNN compiles DQ -> Conv -> Q as a single unit.
TEST_F(QnnHTPBackendTests, TestQDQConvU8U8S32) {
ProviderOptions provider_options;
#if defined(_WIN32)
provider_options["backend_path"] = "QnnHtp.dll";
#else
provider_options["backend_path"] = "libQnnHtp.so";
#endif
RunQnnModelTest(BuildQDQConvTestCase<uint8_t, uint8_t, int32_t, uint8_t>({1, 1, 5, 5}, // Input shape
{1, 1, 3, 3}), // Weights shape
provider_options,
18, // Opset
ExpectedEPNodeAssignment::All, // All nodes assigned to QNN EP.
1, // Expect 1 fused node in graph.
"QDQConvU8U8S32");
}
#endif // defined(__aarch64__) || defined(_M_ARM64) || defined(__linux__)
} // namespace test
} // namespace onnxruntime
#endif

178
onnxruntime/test/providers/qnn/conv_test.cc Normal file
View file

@ -0,0 +1,178 @@
// Copyright (c) Microsoft Corporation. All rights reserved.
// Licensed under the MIT License.
#if !defined(ORT_MINIMAL_BUILD)
#include <string>
#include "core/graph/graph.h"
#include "test/providers/qnn/qnn_test_utils.h"
#include "gtest/gtest.h"
namespace onnxruntime {
namespace test {
// Creates a graph with a single Conv operator. Used for testing CPU backend.
static GetTestModelFn BuildConvTestCase(const std::vector<int64_t>& input_shape,
const std::vector<int64_t>& weights_shape,
bool is_bias_initializer) {
return [input_shape, weights_shape, is_bias_initializer](ModelTestBuilder& builder) {
auto* input = builder.MakeInput<float>(input_shape, 0.0f, 10.0f);
auto* output = builder.MakeOutput();
auto* weights = builder.MakeInitializer<float>(weights_shape, 0.0f, 1.0f);
onnxruntime::NodeArg* bias = nullptr;
if (is_bias_initializer) {
bias = builder.MakeInitializer<float>({weights_shape[0]}, -1.0f, 1.0f);
} else {
bias = builder.MakeInput<float>({weights_shape[0]}, -1.0f, 1.0f);
}
builder.AddNode("Conv", {input, weights, bias}, {output});
};
}
// Runs a Conv model on the QNN CPU backend. Checks the graph node assignment, and that inference
// outputs for QNN EP and CPU EP match.
static void RunCPUConvOpTest(const std::vector<int64_t>& input_shape,
const std::vector<int64_t>& weights_shape,
bool is_bias_initializer,
ExpectedEPNodeAssignment expected_ep_assignment, const char* test_description,
int opset = 13) {
ProviderOptions provider_options;
#if defined(_WIN32)
provider_options["backend_path"] = "QnnCpu.dll";
#else
provider_options["backend_path"] = "libQnnCpu.so";
#endif
constexpr int expected_nodes_in_partition = 1;
RunQnnModelTest(BuildConvTestCase(input_shape, weights_shape, is_bias_initializer),
provider_options,
opset,
expected_ep_assignment,
expected_nodes_in_partition,
test_description);
}
// Creates a graph with a single Q/DQ Conv operator. Used for testing HTP backend.
template <typename InputType, typename WeightType, typename BiasType, typename OutputType>
GetTestModelFn BuildQDQConvTestCase(const std::vector<int64_t>& input_shape,
const std::vector<int64_t>& weights_shape,
bool is_bias_initializer = true) {
return [input_shape, weights_shape, is_bias_initializer](ModelTestBuilder& builder) {
auto* input_arg = builder.MakeInput<float>(input_shape, -1.f, 1.f);
auto* output_arg = builder.MakeOutput();
using InputLimits = std::numeric_limits<InputType>;
using WeightLimits = std::numeric_limits<WeightType>;
using OutputLimits = std::numeric_limits<OutputType>;
InputType input_min_value = InputLimits::min();
InputType input_max_value = InputLimits::max();
WeightType weight_min_value = WeightLimits::min();
WeightType weight_max_value = WeightLimits::max();
auto* dq_w_output = builder.MakeIntermediate();
auto* weight = builder.MakeInitializer<WeightType>(weights_shape, weight_min_value, weight_max_value);
builder.AddDequantizeLinearNode<WeightType>(weight, .03f,
(weight_min_value + weight_max_value) / 2 + 1,
dq_w_output);
auto* dq_bias_output = builder.MakeIntermediate();
onnxruntime::NodeArg* bias = nullptr;
if (is_bias_initializer) {
bias = builder.MakeInitializer<BiasType>({weights_shape[0]}, static_cast<BiasType>(0),
static_cast<BiasType>(127));
} else {
bias = builder.MakeInput<BiasType>({weights_shape[0]}, static_cast<BiasType>(0),
static_cast<BiasType>(127));
}
builder.AddDequantizeLinearNode<BiasType>(bias, .0012f,
0,
dq_bias_output);
auto* conv_output = builder.MakeIntermediate();
auto* dq_output = AddQDQNodePair<InputType>(builder, input_arg, .04f,
(input_min_value + input_max_value) / 2 + 1);
builder.AddNode("Conv", {dq_output, dq_w_output, dq_bias_output}, {conv_output});
auto* q_output = builder.MakeIntermediate();
builder.AddQuantizeLinearNode<OutputType>(conv_output, .039f,
(OutputLimits::min() + OutputLimits::max()) / 2 + 1,
q_output);
builder.AddDequantizeLinearNode<OutputType>(q_output, .039f,
(OutputLimits::min() + OutputLimits::max()) / 2 + 1,
output_arg);
};
}
// Runs a Conv model on the QNN HTP backend. Checks the graph node assignment, and that inference
// outputs for QNN EP and CPU EP match.
template <typename InputType, typename WeightType, typename BiasType, typename OutputType>
static void RunHTPConvOpTest(const std::vector<int64_t>& input_shape,
const std::vector<int64_t>& weights_shape,
bool is_bias_initializer,
ExpectedEPNodeAssignment expected_ep_assignment, const char* test_description,
int opset = 13) {
ProviderOptions provider_options;
#if defined(_WIN32)
provider_options["backend_path"] = "QnnHtp.dll";
#else
provider_options["backend_path"] = "libQnnHtp.so";
#endif
constexpr int expected_nodes_in_partition = 1;
RunQnnModelTest(BuildQDQConvTestCase<InputType, WeightType, BiasType, OutputType>(input_shape, weights_shape,
is_bias_initializer),
provider_options,
opset,
expected_ep_assignment,
expected_nodes_in_partition,
test_description);
}
// Check that QNN compiles DQ -> Conv -> Q as a single unit.
// Tests bias as an input.
//
// TODO: Enable this test when QNN CPU backend (QNN sdk 2.10.0) fixes bug that causes graph finalization to
// throw a segfault when the bias is a non-static input.
TEST_F(QnnCPUBackendTests, DISABLED_TestCPUConvf32_bias_input) {
RunCPUConvOpTest({1, 1, 3, 3}, {2, 1, 2, 2}, false, ExpectedEPNodeAssignment::All, "TestCPUConvf32_bias_input");
}
// Check that QNN compiles DQ -> Conv -> Q as a single unit.
// Tests bias as an initializer.
TEST_F(QnnCPUBackendTests, TestCPUConvf32_bias_initializer) {
RunCPUConvOpTest({1, 1, 3, 3}, {2, 1, 2, 2}, true, ExpectedEPNodeAssignment::All, "TestCPUConvf32_bias_initializer");
}
#if defined(__aarch64__) || defined(_M_ARM64) || defined(__linux__)
// Check that QNN compiles DQ -> Conv -> Q as a single unit.
// Tests bias as an input.
TEST_F(QnnHTPBackendTests, TestQDQConvU8U8S32_bias_input) {
RunHTPConvOpTest<uint8_t, uint8_t, int32_t, uint8_t>({1, 1, 5, 5}, {1, 1, 3, 3}, false, ExpectedEPNodeAssignment::All,
"TestQDQConvU8U8S32_bias_input");
}
// Check that QNN compiles DQ -> Conv -> Q as a single unit.
// Tests bias as an initializer.
TEST_F(QnnHTPBackendTests, TestQDQConvU8U8S32_bias_initializer) {
RunHTPConvOpTest<uint8_t, uint8_t, int32_t, uint8_t>({1, 1, 5, 5}, {1, 1, 3, 3}, true, ExpectedEPNodeAssignment::All,
"TestQDQConvU8U8S32_bias_initializer");
}
#endif // defined(__aarch64__) || defined(_M_ARM64) || defined(__linux__)
} // namespace test
} // namespace onnxruntime
#endif

2
tools/ci_build/github/azure-pipelines/android-arm64-v8a-QNN-crosscompile-ci-pipeline.yml
View file

@ -3,7 +3,7 @@ parameters:
- name: QnnSdk
displayName: QNN SDK version
type: string
default: qnn-v2.9.0.230327191003_53330
default: qnn-v2.10.0.230425122932_54038
jobs:
- job: Build_QNN_EP

2
tools/ci_build/github/azure-pipelines/linux-qnn-ci-pipeline.yml
View file

@ -3,7 +3,7 @@ parameters:
- name: QnnSdk
displayName: QNN SDK version
type: string
default: qnn-v2.9.0.230327191003_53330
default: qnn-v2.10.0.230425122932_54038
jobs:
- job: Build_QNN_EP

2
tools/ci_build/github/azure-pipelines/win-qnn-arm64-ci-pipeline.yml
View file

@ -3,7 +3,7 @@ parameters:
- name: QnnSdk
displayName: QNN SDK version
type: string
default: qnn-v2.9.0.230327191003_53330_win
default: qnn-v2.10.0.230425122932_54038_win
jobs:
- job: 'build'

2
tools/ci_build/github/azure-pipelines/win-qnn-ci-pipeline.yml
View file

@ -3,7 +3,7 @@ parameters:
- name: QnnSdk
displayName: QNN SDK version
type: string
default: qnn-v2.9.0.230327191003_53330_win
default: qnn-v2.10.0.230425122932_54038_win
jobs:
- job: 'build'