Test out
This commit is contained in:
nums11
Родитель
76e247cb9a
Коммит
19fb88ccc7
|
|
@ -242,10 +242,10 @@ void D3D12Quad::LoadAssets()
|
|||
// Define the geometry for a quad.
|
||||
Vertex quadVertices[] =
|
||||
{
|
||||
{ { -0.25f, 0.25f * m_aspectRatio, 0.0f }, { 0.0f, 0.0f } }, //top left
|
||||
{ { 0.25f, 0.25f * m_aspectRatio, 0.0f }, { 1.0f, 0.0f } }, //top right
|
||||
{ { -0.25f, -0.25f * m_aspectRatio, 0.0f }, { 0.0f, 1.0f } }, //bottom left
|
||||
{ { 0.25f, -0.25f * m_aspectRatio, 0.0f }, { 1.0f, 1.0f } }, //bottom right
|
||||
{ { -0.75f, 0.75f * m_aspectRatio, 0.0f }, { 0.0f, 0.0f } }, //top left
|
||||
{ { 0.75f, 0.75f * m_aspectRatio, 0.0f }, { 1.0f, 0.0f } }, //top right
|
||||
{ { -0.75f, -0.75f * m_aspectRatio, 0.0f }, { 0.0f, 1.0f } }, //bottom left
|
||||
{ { 0.75f, -0.75f * m_aspectRatio, 0.0f }, { 1.0f, 1.0f } }, //bottom right
|
||||
};
|
||||
|
||||
const UINT vertexBufferSize = sizeof(quadVertices);
|
||||
|
|
|
|||
|
|
@ -88,7 +88,7 @@ Ort::Value CreateTensorValueFromRTVResource(
|
|||
deleting_unique_ptr<void> dmlAllocatorResourceCleanup(dmlAllocatorResource, deleter);
|
||||
|
||||
size_t tensorByteSize = static_cast<size_t>(d3dResource->GetDesc().Width * d3dResource->GetDesc().Height
|
||||
* 3 * 1 * 4);
|
||||
* 3 * 4);
|
||||
Ort::Value newValue(
|
||||
Ort::Value::CreateTensor(
|
||||
memoryInformation,
|
||||
|
|
@ -269,62 +269,83 @@ std::vector<float> Eval(Ort::Session& session, const Ort::Value& prev_input) {
|
|||
|
||||
////////////////////////////////////////
|
||||
// Bind the tensor inputs to the model, and run it.
|
||||
ioBinding.BindInput(modelInputTensorName, prev_input);
|
||||
ioBinding.BindOutput(modelOutputTensorName, outputTensor);
|
||||
ioBinding.SynchronizeInputs();
|
||||
QueryPerformanceCounter(&bindingTime);
|
||||
//ioBinding.BindInput(modelInputTensorName, prev_input);
|
||||
//ioBinding.BindOutput(modelOutputTensorName, outputTensor);
|
||||
//ioBinding.SynchronizeInputs();
|
||||
//QueryPerformanceCounter(&bindingTime);
|
||||
|
||||
Ort::RunOptions runOptions;
|
||||
|
||||
// TODO: Upload inputTensorValues to GPU inputTensor.
|
||||
|
||||
session.Run(runOptions, ioBinding);
|
||||
OutputDebugString(L"Done evaluating inference session");
|
||||
QueryPerformanceCounter(&runTime);
|
||||
printf("Synchronizing outputs.\n");
|
||||
ioBinding.SynchronizeOutputs();
|
||||
QueryPerformanceCounter(&synchronizeOutputsTime);
|
||||
printf("Finished execution.\n");
|
||||
std::vector<const char*> input_node_names;
|
||||
input_node_names.push_back(modelInputTensorName);
|
||||
std::vector<const char*> output_node_names;
|
||||
output_node_names.push_back(modelOutputTensorName);
|
||||
|
||||
for (int i = 0; i <= std::min(outputTensorValues.size(), size_t(10)); ++i)
|
||||
{
|
||||
std::wstring value = std::to_wstring(outputTensorValues[i]);
|
||||
auto output_tensors = session.Run(Ort::RunOptions{ nullptr }, input_node_names.data(), &prev_input, 1, output_node_names.data(), 1);
|
||||
// Get pointer to output tensor float values
|
||||
float* floatarr = output_tensors.front().GetTensorMutableData<float>();
|
||||
std::vector<float> final_results;
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
float score = floatarr[i];
|
||||
std::wstring value = std::to_wstring(floatarr[i]);
|
||||
std::wstring i_w_str = std::to_wstring(i);
|
||||
OutputDebugString(L"Output[");
|
||||
OutputDebugString(i_w_str.c_str());
|
||||
OutputDebugString(L"]: ");
|
||||
OutputDebugString(value.c_str());
|
||||
OutputDebugString(L"\n");
|
||||
//OutputDebugString(L"Output[");
|
||||
//OutputDebugString(i_w_str.c_str());
|
||||
//OutputDebugString(L"]: ");
|
||||
//OutputDebugString(value.c_str());
|
||||
//OutputDebugString(L"\n");
|
||||
final_results.push_back(floatarr[i]);
|
||||
}
|
||||
std::vector<uint32_t> indices(outputTensorValues.size(), 0);
|
||||
std::iota(indices.begin(), indices.end(), 0);
|
||||
sort(
|
||||
indices.begin(),
|
||||
indices.end(),
|
||||
[&](uint32_t a, uint32_t b)
|
||||
{
|
||||
return (outputTensorValues[a] > outputTensorValues[b]);
|
||||
}
|
||||
);
|
||||
OutputDebugString(L"Top 10:");
|
||||
OutputDebugString(L"\n");
|
||||
|
||||
std::vector<float> top_10;
|
||||
for (int i = 0; i <= std::min(indices.size(), size_t(10)); ++i)
|
||||
{
|
||||
std::wstring first = std::to_wstring(indices[i]);
|
||||
std::wstring second = std::to_wstring(outputTensorValues[indices[i]]);
|
||||
top_10.push_back(outputTensorValues[indices[i]]);
|
||||
//session.Run(runOptions, ioBinding);
|
||||
//OutputDebugString(L"Done evaluating inference session");
|
||||
//QueryPerformanceCounter(&runTime);
|
||||
//printf("Synchronizing outputs.\n");
|
||||
////ioBinding.SynchronizeOutputs();
|
||||
//QueryPerformanceCounter(&synchronizeOutputsTime);
|
||||
//printf("Finished execution.\n");
|
||||
|
||||
printf("output[%d] = %f\n", indices[i], outputTensorValues[indices[i]]);
|
||||
OutputDebugString(L"Output[");
|
||||
OutputDebugString(first.c_str());
|
||||
OutputDebugString(L"]: ");
|
||||
OutputDebugString(second.c_str());
|
||||
OutputDebugString(L"\n");
|
||||
}
|
||||
//for (int i = 0; i <= std::min(outputTensorValues.size(), size_t(10)); ++i)
|
||||
//{
|
||||
// std::wstring value = std::to_wstring(outputTensorValues[i]);
|
||||
// std::wstring i_w_str = std::to_wstring(i);
|
||||
// OutputDebugString(L"Output[");
|
||||
// OutputDebugString(i_w_str.c_str());
|
||||
// OutputDebugString(L"]: ");
|
||||
// OutputDebugString(value.c_str());
|
||||
// OutputDebugString(L"\n");
|
||||
//}
|
||||
//std::vector<uint32_t> indices(outputTensorValues.size(), 0);
|
||||
//std::iota(indices.begin(), indices.end(), 0);
|
||||
//sort(
|
||||
// indices.begin(),
|
||||
// indices.end(),
|
||||
// [&](uint32_t a, uint32_t b)
|
||||
// {
|
||||
// return (outputTensorValues[a] > outputTensorValues[b]);
|
||||
// }
|
||||
//);
|
||||
//OutputDebugString(L"Top 10:");
|
||||
//OutputDebugString(L"\n");
|
||||
|
||||
//std::vector<float> top_10;
|
||||
//for (int i = 0; i <= std::min(indices.size(), size_t(10)); ++i)
|
||||
//{
|
||||
// std::wstring first = std::to_wstring(indices[i]);
|
||||
// std::wstring second = std::to_wstring(outputTensorValues[indices[i]]);
|
||||
// top_10.push_back(outputTensorValues[indices[i]]);
|
||||
|
||||
// printf("output[%d] = %f\n", indices[i], outputTensorValues[indices[i]]);
|
||||
// OutputDebugString(L"Output[");
|
||||
// OutputDebugString(first.c_str());
|
||||
// OutputDebugString(L"]: ");
|
||||
// OutputDebugString(second.c_str());
|
||||
// OutputDebugString(L"\n");
|
||||
//}
|
||||
//return top_10;
|
||||
return outputTensorValues;
|
||||
return final_results;
|
||||
}
|
||||
catch (Ort::Exception const& exception)
|
||||
{
|
||||
|
|
@ -380,6 +401,35 @@ winrt::com_array<float> Preprocess(Ort::Session& session,
|
|||
const std::array<int64_t, 4> preprocessInputShape = { 1, 512, 512, 4 };
|
||||
const std::array<int64_t, 4> preprocessOutputShape = { 1, 224, 224, 3 };
|
||||
|
||||
//std::vector<std::vector<std::vector<std::vector<float>>>> test;
|
||||
//for (int a = 0; a < 1; a++) {
|
||||
// std::vector<std::vector<std::vector<float>>> first;
|
||||
// for (int b = 0; b < 512; b++) {
|
||||
// std::vector<std::vector<float>> second;
|
||||
// for (int c = 0; c < 512; c++) {
|
||||
// std::vector<float> third;
|
||||
// for (int d = 0; d < 4; d++) {
|
||||
// third.push_back(1.0);
|
||||
// }
|
||||
// second.push_back(third);
|
||||
// }
|
||||
// first.push_back(second);
|
||||
// }
|
||||
// test.push_back(first);
|
||||
//}
|
||||
|
||||
std::vector<float> test;
|
||||
for (int a = 0; a < 1; a++) {
|
||||
for (int b = 0; b < 512; b++) {
|
||||
for (int c = 0; c < 512; c++) {
|
||||
for (int d = 0; d < 4; d++) {
|
||||
test.push_back(0.5);
|
||||
}
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
|
||||
auto resize_op = LearningModelOperator(L"Resize")
|
||||
.SetInput(L"X", L"Input")
|
||||
.SetConstant(L"roi", TensorFloat::CreateFromIterable({ 8 }, { 0, 0, 0, 0, 1, 1, 1, 1 }))
|
||||
|
|
@ -428,7 +478,7 @@ winrt::com_array<float> Preprocess(Ort::Session& session,
|
|||
D3D12_RESOURCE_DESC resourceDesc = {
|
||||
D3D12_RESOURCE_DIMENSION_BUFFER,
|
||||
0,
|
||||
static_cast<uint64_t>(800 * 600 * 3 * 4),
|
||||
static_cast<uint64_t>(800 * 600 * 4),
|
||||
1,
|
||||
1,
|
||||
1,
|
||||
|
|
@ -456,6 +506,8 @@ winrt::com_array<float> Preprocess(Ort::Session& session,
|
|||
hr = swapChain->GetBuffer(frameIndex, IID_PPV_ARGS(¤t_buffer));
|
||||
auto buffer_desc = current_buffer->GetDesc();
|
||||
|
||||
auto float_size = sizeof(float);
|
||||
|
||||
if (FAILED(hr))
|
||||
{
|
||||
OutputDebugString(L"Failed to get buffer");
|
||||
|
|
@ -538,28 +590,35 @@ winrt::com_array<float> Preprocess(Ort::Session& session,
|
|||
/*out*/ IID_PPV_ARGS_Helper(inputTensorEpWrapper.GetAddressOf())
|
||||
);
|
||||
|
||||
Ort::Value outputTensor(nullptr);
|
||||
std::vector<float> outputTensorValues(static_cast<size_t>(GetElementCount(preprocessOutputShape)), 0.0f);
|
||||
ComPtr<IUnknown> outputTensorEpWrapper;
|
||||
//Ort::Value outputTensor(nullptr);
|
||||
//std::vector<float> outputTensorValues(static_cast<size_t>(GetElementCount(preprocessOutputShape)), 0.0f);
|
||||
//ComPtr<IUnknown> outputTensorEpWrapper;
|
||||
|
||||
outputTensor = CreateTensorValueUsingD3DResource(
|
||||
device,
|
||||
*ortDmlApi,
|
||||
memoryInformation,
|
||||
preprocessOutputShape,
|
||||
ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT,
|
||||
sizeof(float),
|
||||
/*out*/ IID_PPV_ARGS_Helper(outputTensorEpWrapper.GetAddressOf())
|
||||
);
|
||||
//outputTensor = CreateTensorValueUsingD3DResource(
|
||||
// device,
|
||||
// *ortDmlApi,
|
||||
// memoryInformation,
|
||||
// preprocessOutputShape,
|
||||
// ONNX_TENSOR_ELEMENT_DATA_TYPE_FLOAT,
|
||||
// sizeof(float),
|
||||
// /*out*/ IID_PPV_ARGS_Helper(outputTensorEpWrapper.GetAddressOf())
|
||||
//);
|
||||
|
||||
QueryPerformanceCounter(&tensorCreationTime);
|
||||
// Create output tensor on device memory.
|
||||
//Ort::Value outputTensor(nullptr);
|
||||
//std::vector<float> outputTensorValues(static_cast<size_t>(GetElementCount(preprocessOutputShape)), 0.0f);
|
||||
//Microsoft::WRL::ComPtr<IUnknown> outputTensorEpWrapper;
|
||||
|
||||
//const char* otMemoryInformationName = "Cpu";
|
||||
//Ort::MemoryInfo otMemoryInformation(otMemoryInformationName, OrtAllocatorType::OrtDeviceAllocator, 0, OrtMemType::OrtMemTypeDefault);
|
||||
//outputTensor = Ort::Value::CreateTensor<float>(otMemoryInformation, outputTensorValues.data(), outputTensorValues.size(), preprocessOutputShape.data(), 4);
|
||||
|
||||
////////////////////////////////////////
|
||||
// Bind the tensor inputs to the model, and run it.
|
||||
ioBinding.BindInput(preprocessModelInputTensorName, inputTensor);
|
||||
ioBinding.BindOutput(preprocessModelOutputTensorName, outputTensor);
|
||||
ioBinding.SynchronizeInputs();
|
||||
QueryPerformanceCounter(&bindingTime);
|
||||
//ioBinding.BindInput(preprocessModelInputTensorName, inputTensor);
|
||||
//ioBinding.BindOutput(preprocessModelOutputTensorName, outputTensor);
|
||||
//ioBinding.SynchronizeInputs();
|
||||
//QueryPerformanceCounter(&bindingTime);
|
||||
|
||||
Ort::RunOptions runOptions;
|
||||
|
||||
|
|
@ -567,11 +626,62 @@ winrt::com_array<float> Preprocess(Ort::Session& session,
|
|||
|
||||
printf("Beginning execution.\n");
|
||||
printf("Running Session.\n");
|
||||
session.Run(runOptions, ioBinding);
|
||||
|
||||
std::vector<const char*> input_node_names;
|
||||
input_node_names.push_back(preprocessModelInputTensorName);
|
||||
std::vector<const char*> output_node_names;
|
||||
output_node_names.push_back(preprocessModelOutputTensorName);
|
||||
|
||||
//std::vector<Ort::Value> output_tensor_ort_values = session.Run(Ort::RunOptions{ nullptr }, input_node_names.data(),
|
||||
// &inputTensor, 1, output_node_names.data(), 1);
|
||||
//auto outputTensor = output_tensor_ort_values[0];
|
||||
//auto output_tensor_float_values = output_tensor_ort_values.front().GetTensorMutableData<Ort::Value>();
|
||||
//auto outputTensor = output_tensor_float_values[0];
|
||||
|
||||
//const float* floatarr = inputTensor.GetTensorData<float>();
|
||||
//for (int i = 0; i < 20000; i++) {
|
||||
// float score = floatarr[i];
|
||||
// std::wstring value = std::to_wstring(floatarr[i]);
|
||||
// std::wstring i_w_str = std::to_wstring(i);
|
||||
// OutputDebugString(L"Input[");
|
||||
// OutputDebugString(i_w_str.c_str());
|
||||
// OutputDebugString(L"]: ");
|
||||
// OutputDebugString(value.c_str());
|
||||
// OutputDebugString(L"\n");
|
||||
//}
|
||||
|
||||
auto memory_info = Ort::MemoryInfo::CreateCpu(OrtArenaAllocator, OrtMemTypeDefault);
|
||||
auto input_tensor_size = 1 * 512 * 512 * 4;
|
||||
Ort::Value test_input_tensor = Ort::Value::CreateTensor<float>(memory_info, test.data(), input_tensor_size, preprocessInputShape.data(), 4);
|
||||
Ort::Value outputTensor(nullptr);
|
||||
session.Run(Ort::RunOptions{ nullptr }, input_node_names.data(),
|
||||
&inputTensor, 1, output_node_names.data(), &outputTensor, 1);
|
||||
//session.Run(Ort::RunOptions{ nullptr }, input_node_names.data(),
|
||||
// &test_input_tensor, 1, output_node_names.data(), &outputTensor, 1);
|
||||
|
||||
//auto some_len = output_tensor_ort_values.size();
|
||||
//auto output_tensor_float_values = output_tensor_ort_values.front().GetTensorMutableData<float>();
|
||||
|
||||
//auto memory_info = Ort::MemoryInfo::CreateCpu(OrtArenaAllocator, OrtMemTypeDefault);
|
||||
//size_t output_tensor_size = 224 * 224 * 3;
|
||||
//Ort::Value outputTensor = Ort::Value::CreateTensor<float>(memory_info, output_tensor_float_values, output_tensor_size, preprocessOutputShape.data(), 4);
|
||||
|
||||
//session.Run(runOptions, ioBinding);
|
||||
OutputDebugString(L"Done evaluating preprocessing session");
|
||||
//ioBinding.SynchronizeOutputs();
|
||||
QueryPerformanceCounter(&synchronizeOutputsTime);
|
||||
|
||||
float* opfloatarr = outputTensor.GetTensorMutableData<float>();
|
||||
for (int i = 0; i < 100000; i++) {
|
||||
std::wstring value = std::to_wstring(opfloatarr[i]);
|
||||
std::wstring i_w_str = std::to_wstring(i);
|
||||
OutputDebugString(L"Output[");
|
||||
OutputDebugString(i_w_str.c_str());
|
||||
OutputDebugString(L"]: ");
|
||||
OutputDebugString(value.c_str());
|
||||
OutputDebugString(L"\n");
|
||||
}
|
||||
|
||||
auto eval_results_std = Eval(inferenceSession, outputTensor);
|
||||
winrt::com_array<float> eval_results(1000);
|
||||
for (int i = 0; i < 1000; i++) {
|
||||
|
|
|
|||
Двоичный файл не отображается.
|
До Ширина: | Высота: | Размер: 20 KiB После Ширина: | Высота: | Размер: 81 KiB |
Загрузка…
Ссылка в новой задаче