-
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.
#92: block match & LTS CPU/CUDA regression tests
- Loading branch information
Showing
8 changed files
with
408 additions
and
19 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 @@ | ||
| 271 | ||
| 272 |
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 +1,175 @@ | ||
| #include "reg_test_common.h" | ||
| #include "_reg_blockMatching.h" | ||
| #include "CpuBlockMatchingKernel.h" | ||
| #include "CudaBlockMatchingKernel.h" | ||
|
|
||
| /* | ||
| This test file contains a regression test to ensure the CPU and GPU version yield the same output | ||
| */ | ||
|
|
||
| class BMTest { | ||
| /* | ||
| Class to call the block matching function | ||
| */ | ||
| protected: | ||
| using TestData = std::tuple<std::string, NiftiImage, NiftiImage, int>; | ||
| using TestCase = std::tuple<std::string, _reg_blockMatchingParam *, _reg_blockMatchingParam *>; | ||
| inline static vector<TestCase> testCases; | ||
| NiftiImage reference2d; | ||
| NiftiImage floating2d; | ||
| NiftiImage reference3d; | ||
| NiftiImage floating3d; | ||
| public: | ||
| ~BMTest() { | ||
| std::cout << "Calling destructor" << std::endl; | ||
| } | ||
| BMTest() { | ||
| std::cout << "Calling constructor" << std::endl; | ||
| if (!testCases.empty()) | ||
| return; | ||
|
|
||
| // Create a random number generator | ||
| std::random_device rd; | ||
| std::mt19937 gen(rd()); | ||
| std::uniform_real_distribution<float> distr(0, 1); | ||
|
|
||
| // Create a reference and floating 2D images | ||
| NiftiImage::dim_t size = 64; | ||
| vector<NiftiImage::dim_t> dim{ size, size }; | ||
| this->reference2d = NiftiImage(dim, NIFTI_TYPE_FLOAT32); | ||
| this->floating2d = NiftiImage(dim, NIFTI_TYPE_FLOAT32); | ||
|
|
||
| // Create a reference 3D image | ||
| dim.push_back(size); | ||
| this->reference3d = NiftiImage(dim, NIFTI_TYPE_FLOAT32); | ||
| this->floating3d = NiftiImage(dim, NIFTI_TYPE_FLOAT32); | ||
|
|
||
| // Fill images with random values | ||
| float *ref2dPtr = static_cast<float *>(reference2d->data); | ||
| float *flo2dPtr = static_cast<float *>(floating2d->data); | ||
| for (int y = 0; y < reference2d->ny; ++y) | ||
| for (int x = 0; x < reference2d->nx; ++x) { | ||
| *ref2dPtr++ = distr(gen); | ||
| *flo2dPtr++ = distr(gen); | ||
| } | ||
|
|
||
| // Fill images with random values | ||
| float *ref3dPtr = static_cast<float *>(reference3d->data); | ||
| float *flo3dPtr = static_cast<float *>(floating3d->data); | ||
| for (int z = 0; z < reference3d->nz; ++z) | ||
| for (int y = 0; y < reference3d->ny; ++y) | ||
| for (int x = 0; x < reference3d->nx; ++x) { | ||
| *ref3dPtr++ = distr(gen); | ||
| *flo3dPtr++ = distr(gen); | ||
| } | ||
|
|
||
|
|
||
| // Create the data container for the regression test | ||
| vector<TestData> testData; | ||
| for(int b=50; b<=100; b+=50){ | ||
| testData.emplace_back(TestData( | ||
| "BlockMatching 2D block " + std::to_string(b), | ||
| std::move(NiftiImage(this->reference2d)), | ||
| std::move(NiftiImage(this->floating2d)), | ||
| b | ||
| )); | ||
| testData.emplace_back(TestData( | ||
| "BlockMatching 3D block " + std::to_string(b), | ||
| std::move(NiftiImage(this->reference3d)), | ||
| std::move(NiftiImage(this->floating3d)), | ||
| b | ||
| )); | ||
| } | ||
|
|
||
| for (auto&& data : testData) { | ||
| unique_ptr<Platform> platformCPU{ new Platform(PlatformType::Cpu) }; | ||
| unique_ptr<Platform> platformCUDA{ new Platform(PlatformType::Cuda) }; | ||
| // Make a copy of the test data | ||
| auto&& [testName, reference, floating, block] = data; | ||
| // Create content creator | ||
| unique_ptr<AladinContentCreator> contentCreatorCPU{ | ||
| dynamic_cast<AladinContentCreator*>(platformCPU->CreateContentCreator(ContentType::Aladin)) | ||
| }; | ||
| unique_ptr<AladinContentCreator> contentCreatorCUDA{ | ||
| dynamic_cast<AladinContentCreator*>(platformCUDA->CreateContentCreator(ContentType::Aladin)) | ||
| }; | ||
| // Create the contents | ||
| unique_ptr<AladinContent> contentCPU{ contentCreatorCPU->Create( | ||
| NiftiImage(reference).disown(), | ||
| NiftiImage(floating).disown(), | ||
| nullptr, | ||
| nullptr, | ||
| sizeof(float), | ||
| 100, | ||
| block, | ||
| 1 | ||
| )}; | ||
| unique_ptr<AladinContent> contentCUDA{ contentCreatorCUDA->Create( | ||
| NiftiImage(reference).disown(), | ||
| NiftiImage(floating).disown(), | ||
| nullptr, | ||
| nullptr, | ||
| sizeof(float), | ||
| 100, | ||
| block, | ||
| 1 | ||
| )}; | ||
| // Initialise the warped image | ||
| contentCPU->SetWarped(NiftiImage(floating).disown()); | ||
| contentCUDA->SetWarped(NiftiImage(floating).disown()); | ||
| // Initialise the block matching | ||
| std::unique_ptr<Kernel> kernelCPU = nullptr; | ||
| kernelCPU.reset(platformCPU->CreateKernel(BlockMatchingKernel::GetName(), contentCPU.get())); | ||
| std::unique_ptr<Kernel> kernelCUDA = nullptr; | ||
| kernelCUDA.reset(platformCUDA->CreateKernel(BlockMatchingKernel::GetName(), contentCUDA.get())); | ||
|
|
||
| // run the computation | ||
| kernelCPU->template castTo<CpuBlockMatchingKernel>()->Calculate(); | ||
| kernelCUDA->template castTo<CudaBlockMatchingKernel>()->Calculate(); | ||
|
|
||
| // Retrieve the information | ||
| _reg_blockMatchingParam *blockMatchingParamsCPU = new _reg_blockMatchingParam(contentCPU->GetBlockMatchingParams()); | ||
| _reg_blockMatchingParam *blockMatchingParamsCUDA = new _reg_blockMatchingParam(contentCUDA->GetBlockMatchingParams()); | ||
|
|
||
| testCases.push_back({ | ||
| testName, | ||
| blockMatchingParamsCPU, | ||
| blockMatchingParamsCUDA | ||
| }); | ||
| contentCPU.reset(); | ||
| contentCUDA.reset(); | ||
| } | ||
| } | ||
| }; | ||
|
|
||
| TEST_CASE_METHOD(BMTest, "Regression BlockMatching", "[regression]") { | ||
| // Loop over all generated test cases | ||
| for (auto&& testCase : this->testCases) { | ||
| // Retrieve test information | ||
| auto&& [testName, blockMatchingParamsCPU, blockMatchingParamsCUDA] = testCase; | ||
|
|
||
| SECTION(testName) { | ||
|
|
||
| // Ensure both approaches retreive the same number of voxel | ||
| REQUIRE(blockMatchingParamsCPU->activeBlockNumber==blockMatchingParamsCUDA->activeBlockNumber); | ||
|
|
||
| // Loop over the block and ensure all values are identical | ||
| for(int b=0; b<blockMatchingParamsCPU->activeBlockNumber*blockMatchingParamsCPU->dim; ++b){ | ||
| float delta = blockMatchingParamsCPU->referencePosition[b] - blockMatchingParamsCUDA->referencePosition[b]; | ||
| if(fabs(delta) > EPS){ | ||
| std::cout << "HERE " << delta << std::endl; | ||
| std::cout.flush(); | ||
| } | ||
| REQUIRE(fabs(delta) < EPS); | ||
| delta = blockMatchingParamsCPU->warpedPosition[b] - blockMatchingParamsCUDA->warpedPosition[b]; | ||
| if(fabs(delta) > EPS){ | ||
| std::cout << "HERE " << delta << std::endl; | ||
| std::cout.flush(); | ||
| } | ||
| REQUIRE(fabs(delta) < EPS); | ||
| } | ||
| delete blockMatchingParamsCPU; | ||
| delete blockMatchingParamsCUDA; | ||
| } | ||
| } | ||
| } |
Oops, something went wrong.