saymrwulf/onnxruntime
1
0
Fork 0
mirror of https://github.com/saymrwulf/onnxruntime.git synced 2026-05-23 22:13:38 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

enable more unit tests for ROCM EP (#7307)

Browse source
This commit is contained in:
Weixing Zhang 2021-04-09 15:15:13 -07:00 committed by GitHub
parent f27f5afd8a
commit 75c0192e4f
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
8 changed files with 73 additions and 30 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

63
onnxruntime/test/framework/inference_session_test.cc
View file

@ -31,6 +31,8 @@
#include "core/providers/cpu/math/element_wise_ops.h"
#ifdef USE_CUDA
#include "core/providers/cuda/gpu_data_transfer.h"
#elif USE_ROCM
#include "core/providers/rocm/gpu_data_transfer.h"
#endif
#include "core/session/environment.h"
#include "core/session/IOBinding.h"
@ -293,6 +295,11 @@ void RunModelWithBindingMatMul(InferenceSession& session_object,
#ifdef USE_CUDA
AllocateMLValue<float>(TestCudaExecutionProvider()->GetAllocator(0, OrtMemTypeDefault), expected_output_dims,
&output_ml_value);
#endif
} else if (allocation_provider == kRocmExecutionProvider) {
#ifdef USE_ROCM
AllocateMLValue<float>(TestRocmExecutionProvider()->GetAllocator(0, OrtMemTypeDefault), expected_output_dims,
&output_ml_value);
#endif
} else {
ORT_THROW("Unsupported provider");
@ -317,9 +324,9 @@ void RunModelWithBindingMatMul(InferenceSession& session_object,
std::cout << "Run returned status: " << st.ErrorMessage() << std::endl;
ASSERT_TRUE(st.IsOK());
if ((is_preallocate_output_vec && allocation_provider == kCudaExecutionProvider) ||
if ((is_preallocate_output_vec && (allocation_provider == kCudaExecutionProvider || allocation_provider == kRocmExecutionProvider)) ||
(output_device && output_device->Type() == OrtDevice::GPU)) {
#ifdef USE_CUDA
#if defined(USE_CUDA) || defined(USE_ROCM)
// in this case we need to copy the tensor from cuda to cpu
vector<OrtValue>& outputs = io_binding->GetOutputs();
ASSERT_EQ(1, outputs.size());
@ -330,7 +337,11 @@ void RunModelWithBindingMatMul(InferenceSession& session_object,
std::unique_ptr<Tensor> cpu_tensor = onnxruntime::make_unique<Tensor>(element_type,
shape,
cpu_allocator);
#ifdef USE_CUDA
cudaStream_t stream = static_cast<cudaStream_t>(static_cast<const onnxruntime::CUDAExecutionProvider*>(TestCudaExecutionProvider())->GetComputeStream());
#elif USE_ROCM
hipStream_t stream = static_cast<hipStream_t>(static_cast<const onnxruntime::ROCMExecutionProvider*>(TestRocmExecutionProvider())->GetComputeStream());
#endif
st = GPUDataTransfer(stream).CopyTensor(rtensor, *cpu_tensor.get(), 0);
ASSERT_TRUE(st.IsOK());
OrtValue ml_value;
@ -343,6 +354,10 @@ void RunModelWithBindingMatMul(InferenceSession& session_object,
if (allocation_provider == kCudaExecutionProvider) {
#ifdef USE_CUDA
TestCudaExecutionProvider()->Sync();
#endif
} else if (allocation_provider == kRocmExecutionProvider) {
#ifdef USE_ROCM
TestRocmExecutionProvider()->Sync();
#endif
}
VerifyOutputs(io_binding->GetOutputs(), expected_output_dims, expected_values_mul_y);
@ -832,11 +847,15 @@ static void TestBindHelper(const std::string& log_str,
InferenceSession session_object{so, GetEnvironment()};
if (bind_provider_type == kCudaExecutionProvider || run_provider_type == kCudaExecutionProvider) {
if (bind_provider_type == kCudaExecutionProvider || bind_provider_type == kRocmExecutionProvider) {
#ifdef USE_CUDA
CUDAExecutionProviderInfo epi;
epi.device_id = 0;
EXPECT_TRUE(session_object.RegisterExecutionProvider(onnxruntime::make_unique<CUDAExecutionProvider>(epi)).IsOK());
#elif USE_ROCM
ROCMExecutionProviderInfo epi;
epi.device_id = 0;
EXPECT_TRUE(session_object.RegisterExecutionProvider(onnxruntime::make_unique<ROCMExecutionProvider>(epi)).IsOK());
#endif
}
@ -944,34 +963,38 @@ TEST(InferenceSessionTests, InvalidInputTypeOfTensorElement) {
ASSERT_TRUE(!st.IsOK());
}
#ifdef USE_CUDA
#if defined(USE_CUDA) || defined(USE_ROCM)
#if USE_CUDA
constexpr const char* kGpuExecutionProvider = kCudaExecutionProvider;
#elif USE_ROCM
constexpr const char* kGpuExecutionProvider = kRocmExecutionProvider;
#endif
TEST(InferenceSessionTests, TestBindCuda) {
TestBindHelper("TestBindCuda",
kCudaExecutionProvider,
kCudaExecutionProvider,
kGpuExecutionProvider,
kGpuExecutionProvider,
false /* don't preallocate output */);
}
TEST(InferenceSessionTests, TestBindCudaPreallocateOutputOnCuda) {
TestBindHelper("TestBindCudaPreallocateOutputOnCuda",
kCudaExecutionProvider,
kCudaExecutionProvider,
kGpuExecutionProvider,
kGpuExecutionProvider,
true /* preallocate output on GPU */,
kCudaExecutionProvider);
kGpuExecutionProvider);
}
TEST(InferenceSessionTests, TestBindCudaPreallocateOutputOnCpu) {
TestBindHelper("TestBindCudaPreallocateOutputOnCpu",
kCudaExecutionProvider,
kCudaExecutionProvider,
kGpuExecutionProvider,
kGpuExecutionProvider,
true /* preallocate output on CPU */,
kCpuExecutionProvider);
}
TEST(InferenceSessionTests, TestBindCudaPreallocateOutputOnCpu2) {
TestBindHelper("TestBindCudaPreallocateOutputOnCpu2",
kCudaExecutionProvider,
kGpuExecutionProvider,
kCpuExecutionProvider,
true /* preallocate output on CPU */,
kCpuExecutionProvider);
@ -981,10 +1004,10 @@ TEST(InferenceSessionTests, TestBindCudaSpecifyOutputDeviceOnCuda) {
OrtDevice device(OrtDevice::GPU, OrtDevice::MemType::DEFAULT, 0);
TestBindHelper("TestBindCudaPreallocateOutputOnCuda",
kCudaExecutionProvider,
kCudaExecutionProvider,
kGpuExecutionProvider,
kGpuExecutionProvider,
false /* preallocate output on GPU */,
kCudaExecutionProvider,
kGpuExecutionProvider,
&device /* specify output device */);
}
@ -1390,6 +1413,10 @@ TEST(InferenceSessionTests, Test3LayerNestedSubgraph) {
CUDAExecutionProviderInfo epi;
epi.device_id = 0;
EXPECT_TRUE(session_object.RegisterExecutionProvider(onnxruntime::make_unique<CUDAExecutionProvider>(epi)).IsOK());
#elif USE_ROCM
ROCMExecutionProviderInfo epi;
epi.device_id = 0;
EXPECT_TRUE(session_object.RegisterExecutionProvider(onnxruntime::make_unique<ROCMExecutionProvider>(epi)).IsOK());
#endif
status = session_object.Load(model_file_name);
@ -1526,6 +1553,10 @@ TEST(InferenceSessionTests, Test2LayerNestedSubgraph) {
CUDAExecutionProviderInfo epi;
epi.device_id = 0;
EXPECT_TRUE(session_object.RegisterExecutionProvider(onnxruntime::make_unique<CUDAExecutionProvider>(epi)).IsOK());
#elif USE_ROCM
ROCMExecutionProviderInfo epi;
epi.device_id = 0;
EXPECT_TRUE(session_object.RegisterExecutionProvider(onnxruntime::make_unique<ROCMExecutionProvider>(epi)).IsOK());
#endif
status = session_object.Load(model_file_name);

8
onnxruntime/test/framework/test_utils.cc
View file

@ -20,6 +20,14 @@ IExecutionProvider* TestCudaExecutionProvider() {
}
#endif
#ifdef USE_ROCM
IExecutionProvider* TestRocmExecutionProvider() {
static ROCMExecutionProviderInfo info;
static ROCMExecutionProvider rocm_provider(info);
return &rocm_provider;
}
#endif
#ifdef USE_TENSORRT
#if 0 // TODO: TensorRT is shared, can't access these directly anymore
IExecutionProvider* TestTensorrtExecutionProvider() {

7
onnxruntime/test/framework/test_utils.h
View file

@ -15,6 +15,9 @@
#ifdef USE_CUDA
#include "core/providers/cuda/cuda_execution_provider.h"
#endif
#ifdef USE_ROCM
#include "core/providers/rocm/rocm_execution_provider.h"
#endif
#ifdef USE_NNAPI
#include "core/providers/nnapi/nnapi_builtin/nnapi_execution_provider.h"
#endif
@ -37,6 +40,10 @@ IExecutionProvider* TestCPUExecutionProvider();
IExecutionProvider* TestCudaExecutionProvider();
#endif
#ifdef USE_ROCM
IExecutionProvider* TestRocmExecutionProvider();
#endif
#ifdef USE_TENSORRT
// Doesn't work with ExecutionProviders class and KernelRegistryManager
IExecutionProvider* TestTensorrtExecutionProvider();

4
orttraining/orttraining/test/session/training_session_test.cc
View file

@ -17,6 +17,8 @@
#ifdef USE_CUDA
#include "core/providers/cuda/cuda_execution_provider.h"
#elif USE_ROCM
#include "core/providers/rocm/rocm_execution_provider.h"
#endif
using namespace onnxruntime::logging;
@ -62,7 +64,7 @@ TEST(TrainingSessionTest, LoadOptimState_FullPrecision_FP32Moments_Adam) {
RunTrainingSessionLoadOptimTests(k_adam_optimizer_op_name, false, false);
}
#ifdef USE_CUDA
#if defined(USE_CUDA) || defined(USE_ROCM)
TEST(TrainingSessionTest, LoadOptimState_MixedPrecision_FP32Moments_Adam) {
RunTrainingSessionLoadOptimTests(k_adam_optimizer_op_name, true, false);
}

5
orttraining/orttraining/test/session/training_session_test_utils.cc
View file

@ -99,7 +99,7 @@ void VerifyState(const DataTransferManager& data_transfer_mgr, const NameMLValMa
const auto& e_state_it = expected_state.find(key);
ORT_ENFORCE(e_state_it != expected_state.end());
auto& expected_tensor = e_state_it->second.Get<Tensor>();
#ifdef USE_CUDA
#if defined(USE_CUDA) || defined(USE_ROCM)
auto& actual_gpu_tensor = a_state_it.second.Get<Tensor>();
// Copying tensor to CPU when cuda is enabled.
@ -182,6 +182,9 @@ std::unique_ptr<TrainingSession> BuildAndRunTrainingSessionWithChecks(
#ifdef USE_CUDA
CUDAExecutionProviderInfo xp_info;
ORT_THROW_IF_ERROR(training_session->RegisterExecutionProvider(onnxruntime::make_unique<CUDAExecutionProvider>(xp_info)));
#elif USE_ROCM
ROCMExecutionProviderInfo xp_info;
ORT_THROW_IF_ERROR(training_session->RegisterExecutionProvider(onnxruntime::make_unique<ROCMExecutionProvider>(xp_info)));
#endif
ORT_THROW_IF_ERROR(training_session->Initialize());

2
orttraining/orttraining/test/session/training_session_test_utils.h
View file

@ -18,6 +18,8 @@
#ifdef USE_CUDA
#include "core/providers/cuda/cuda_execution_provider.h"
#elif USE_ROCM
#include "core/providers/rocm/rocm_execution_provider.h"
#endif
namespace onnxruntime {

2
orttraining/orttraining/training_ops/cuda/optimizer/lamb.cu
View file

@ -322,7 +322,7 @@ __global__ void LambMultiTensorComputeDirectionImpl(
const float alpha,
const float beta,
const float epsilon,
const T1 max_norm,
const float max_norm,
const float alpha_correction,
const float beta_correction) {
const int group_index = chunk_group.block_index_to_tensor_group_index[blockIdx.x];

12
tools/ci_build/github/pai/pai-excluded-tests.txt
View file

@ -39,23 +39,13 @@ ReductionOpTest.ReduceMin_int32
ReductionOpTest.ReduceMin_int8
ReductionOpTest.ReduceSum_double
ReductionOpTest.ReduceSumSquare_double
ReductionOpTest.ReduceInfMax
ReductionOpTest.ReduceInfMax_double
ReductionOpTest.ReduceInfMin
ReductionOpTest.ReduceInfMin_double
ReductionOpTest.ReduceInfSum
ReductionOpTest.ReduceInfLogSum
ReductionOpTest.ReduceInfLogSumExp
ReductionOpTest.ReduceInfLogSumExp_double
GatherOpTest.Gather_invalid_index_cpu
Scatter.InvalidIndex
LogSoftmaxOperator.LargeNumber
MathOpTest.Pow_int64_float
MathOpTest.Pow_int32_float
MathOpTest.Pow_int64_double
Scatter.InvalidIndex
GradientCheckerTest.AddGrad
GradientCheckerTest.SubGrad
GradientCheckerTest.MulGrad
GradientCheckerTest.ReduceMeanGrad
GradientCheckerTest.ReduceL2Grad
GradientCheckerTest.DivGrad