-
Notifications
You must be signed in to change notification settings - Fork 42
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Add regression tests for kernel convolution #92
- Loading branch information
Showing
8 changed files
with
193 additions
and
23 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -1 +1 @@ | ||
| 336 | ||
| 337 |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,168 @@ | ||
| #include "reg_test_common.h" | ||
| #include "CudaContent.h" | ||
| #include "CudaKernelConvolution.hpp" | ||
|
|
||
| /** | ||
| * Kernel convolution regression test to ensure the CPU and CUDA versions yield the same output | ||
| **/ | ||
|
|
||
| class KernelConvolutionTest { | ||
| protected: | ||
| using TestData = std::tuple<std::string, NiftiImage&, vector<float>, int, bool*, bool*>; | ||
| using TestCase = std::tuple<std::string, NiftiImage, NiftiImage>; | ||
|
|
||
| inline static vector<TestCase> testCases; | ||
|
|
||
| public: | ||
| KernelConvolutionTest() { | ||
| if (!testCases.empty()) | ||
| return; | ||
|
|
||
| // Create a random number generator | ||
| std::mt19937 gen(0); | ||
| std::uniform_real_distribution<float> distr(0, 1); | ||
|
|
||
| // Create images | ||
| constexpr int imageCount = 8; | ||
| constexpr NiftiImage::dim_t size = 16; | ||
| vector<NiftiImage::dim_t> dims[imageCount]{ { size, size }, | ||
| { size, size, 1, 1, 2 }, | ||
| { size, size, 1, 1, 3 }, | ||
| { size, size, 1, 2, 2 }, | ||
| { size, size, size }, | ||
| { size, size, size, 2, 1 }, | ||
| { size, size, size, 3, 1 }, | ||
| { size, size, size, 2, 2 } }; | ||
| NiftiImage images[imageCount]; | ||
|
|
||
| // Fill images with random values | ||
| for (int i = 0; i < imageCount; i++) { | ||
| images[i] = NiftiImage(dims[i], NIFTI_TYPE_FLOAT32); | ||
| auto imagePtr = images[i].data(); | ||
| for (size_t j = 0; j < images[i].nVoxels(); j++) | ||
| imagePtr[j] = distr(gen); | ||
| } | ||
|
|
||
| // Create a lambda to concatenate strings for std::accumulate | ||
| auto strConcat = [](const std::string& str, const auto& val) { return str + " "s + std::to_string(val); }; | ||
|
|
||
| // Create the data container for the regression test | ||
| constexpr int kernelTypeCount = 4; | ||
| distr.param(std::uniform_real_distribution<float>::param_type(1, 10)); // Change the range of the distribution | ||
| vector<TestData> testData; | ||
| for (int i = 0; i < imageCount; i++) { | ||
| for (int kernelType = 0; kernelType < kernelTypeCount; kernelType++) { | ||
| vector<float> sigmaValues(images[i]->nt * images[i]->nu); | ||
| std::generate(sigmaValues.begin(), sigmaValues.end(), [&]() { return distr(gen); }); | ||
| const std::string sigmaStr = std::accumulate(sigmaValues.begin(), sigmaValues.end(), ""s, strConcat); | ||
| const std::string dimsStr = std::accumulate(dims[i].begin(), dims[i].end(), ""s, strConcat); | ||
| testData.emplace_back(TestData( | ||
| "Kernel: "s + std::to_string(kernelType) + " Sigma:"s + sigmaStr + " Dims:"s + dimsStr, | ||
| images[i], | ||
| std::move(sigmaValues), | ||
| kernelType, | ||
| nullptr, | ||
| nullptr | ||
| )); | ||
| } | ||
| } | ||
|
|
||
| // Define time points and axes to smooth | ||
| constexpr auto timePointCount = 4; | ||
| bool timePoints[timePointCount][4]{ { true, false, false, false }, | ||
| { false, true, false, false }, | ||
| { false, false, true, false }, | ||
| { false, false, false, true } }; | ||
| bool axes[timePointCount][3]{ { true, false, false }, | ||
| { false, true, false }, | ||
| { false, false, true }, | ||
| { true, true, true } }; | ||
|
|
||
| // Add the time points and axes to the latest test data | ||
| auto latestTestData = testData.end() - timePointCount; | ||
| for (int i = 0; i < timePointCount; i++) { | ||
| auto&& [testName, image, sigmaValues, kernelType, activeTimePoints, activeAxes] = latestTestData[i]; | ||
| const std::string timePointsStr = std::accumulate(timePoints[i], timePoints[i] + 4, ""s, strConcat); | ||
| const std::string axesStr = std::accumulate(axes[i], axes[i] + 3, ""s, strConcat); | ||
| testData.emplace_back(TestData( | ||
| testName + " TimePoints:"s + timePointsStr + " Axes:"s + axesStr, | ||
| image, | ||
| sigmaValues, | ||
| kernelType, | ||
| timePoints[i], | ||
| axes[i] | ||
| )); | ||
| } | ||
|
|
||
| // Create the platforms | ||
| Platform platformCpu(PlatformType::Cpu); | ||
| Platform platformCuda(PlatformType::Cuda); | ||
|
|
||
| for (auto&& testData : testData) { | ||
| // Get the test data | ||
| auto&& [testName, image, sigmaValues, kernelType, activeTimePoints, activeAxes] = testData; | ||
|
|
||
| // Create images | ||
| NiftiImage imageCpu(image), imageCuda(image); | ||
|
|
||
| // Create the contents | ||
| unique_ptr<Content> contentCpu{ new Content( | ||
| imageCpu, | ||
| imageCpu, | ||
| nullptr, | ||
| nullptr, | ||
| sizeof(float) | ||
| ) }; | ||
| unique_ptr<CudaContent> contentCuda{ new CudaContent( | ||
| imageCuda, | ||
| imageCuda, | ||
| nullptr, | ||
| nullptr, | ||
| sizeof(float) | ||
| ) }; | ||
|
|
||
| // Use deformation fields to store images | ||
| contentCpu->SetDeformationField(imageCpu.disown()); | ||
| contentCuda->SetDeformationField(imageCuda.disown()); | ||
|
|
||
| // Compute the kernel convolution for CPU and CUDA | ||
| reg_tools_kernelConvolution(contentCpu->GetDeformationField(), sigmaValues.data(), kernelType, nullptr, activeTimePoints, activeAxes); | ||
| Cuda::KernelConvolution(contentCuda->Content::GetDeformationField(), contentCuda->GetDeformationFieldCuda(), sigmaValues.data(), kernelType, activeTimePoints, activeAxes); | ||
|
|
||
| // Get the images | ||
| imageCpu = NiftiImage(contentCpu->GetDeformationField(), NiftiImage::Copy::Image); | ||
| imageCuda = NiftiImage(contentCuda->GetDeformationField(), NiftiImage::Copy::Image); | ||
|
|
||
| // Save for testing | ||
| testCases.push_back({ testName, std::move(imageCpu), std::move(imageCuda) }); | ||
| } | ||
| } | ||
| }; | ||
|
|
||
| TEST_CASE_METHOD(KernelConvolutionTest, "Regression Kernel Convolution", "[regression]") { | ||
| // Loop over all generated test cases | ||
| for (auto&& testCase : testCases) { | ||
| // Retrieve test information | ||
| auto&& [testName, imageCpu, imageCuda] = testCase; | ||
|
|
||
| SECTION(testName) { | ||
| NR_COUT << "\n**************** Section " << testName << " ****************" << std::endl; | ||
|
|
||
| // Increase the precision for the output | ||
| NR_COUT << std::fixed << std::setprecision(10); | ||
|
|
||
| // Check the images | ||
| const auto imageCpuPtr = imageCpu.data(); | ||
| const auto imageCudaPtr = imageCuda.data(); | ||
| for (size_t i = 0; i < imageCpu.nVoxels(); ++i) { | ||
| const float cpuVal = imageCpuPtr[i]; | ||
| const float cudaVal = imageCudaPtr[i]; | ||
| if (cpuVal != cpuVal && cudaVal != cudaVal) continue; // Skip NaN values | ||
| const float diff = fabs(cpuVal - cudaVal); | ||
| if (diff > EPS) | ||
| NR_COUT << i << " " << cpuVal << " " << cudaVal << std::endl; | ||
| REQUIRE(diff < EPS); | ||
| } | ||
| } | ||
| } | ||
| } |