Remove API interface, cleanup and restore test

mconcas · Sep 9, 2024 · 09fdad9 · 09fdad9
1 parent 4c56558
commit 09fdad9
Show file tree

Hide file tree

Showing 8 changed files with 87 additions and 325 deletions.
diff --git a/Common/DCAFitter/GPU/cuda/CMakeLists.txt b/Common/DCAFitter/GPU/cuda/CMakeLists.txt
@@ -9,8 +9,6 @@
 # granted to it by virtue of its status as an Intergovernmental Organization
 # or submit itself to any jurisdiction.
 
-SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -O0")
-
 o2_add_library(DCAFitterCUDA
                TARGETVARNAME targetName
                SOURCES DCAFitterN.cu
@@ -21,35 +19,14 @@ o2_add_library(DCAFitterCUDA
                PRIVATE_LINK_LIBRARIES O2::GPUTrackingCUDAExternalProvider)
 set_property(TARGET ${targetName} PROPERTY CUDA_SEPARABLE_COMPILATION ON)
 
-o2_add_library(DCAFitterAPICUDA
-              # TARGETVARNAME targetName2
-               SOURCES DCAFitterGPUAPI.cu
-               PUBLIC_INCLUDE_DIRECTORIES ${CMAKE_SOURCE_DIR}/Common/DCAFitter/GPU
-               PUBLIC_LINK_LIBRARIES O2::GPUCommon
-                                     O2::ReconstructionDataFormats
-               PRIVATE_LINK_LIBRARIES O2::DCAFitterCUDA
-               # O2::GPUTrackingCUDAExternalProvider
-                )
-# set_property(TARGET ${targetName2} PROPERTY CUDA_SEPARABLE_COMPILATION ON)
-
 o2_add_test(DCAFitterNCUDA
             SOURCES test/testDCAFitterNGPU.cxx
             PUBLIC_LINK_LIBRARIES O2::ReconstructionDataFormats
-                                  O2::DCAFitterAPICUDA
+                                  O2::DCAFitterCUDA
                                   O2::DCAFitter
                                   ROOT::Core
                                   ROOT::Physics
             COMPONENT_NAME gpu
             LABELS vertexing
             ENVIRONMENT O2_ROOT=${CMAKE_BINARY_DIR}/stage
-            VMCWORKDIR=${CMAKE_BINARY_DIR}/stage/${CMAKE_INSTALL_DATADIR})
-
-o2_add_executable(dca-fitter-on-gpu-cuda
-                  TARGETVARNAME targetName3
-                  SOURCES test/main_prog.cxx
-                  PUBLIC_LINK_LIBRARIES O2::ReconstructionDataFormats
-                                        O2::DCAFitterCUDA
-                                        O2::DCAFitter
-                                        ROOT::Core
-                                        ROOT::Physics)
-# add_compile_options("-g -O0")
+            VMCWORKDIR=${CMAKE_BINARY_DIR}/stage/${CMAKE_INSTALL_DATADIR})
diff --git a/Common/DCAFitter/GPU/cuda/DCAFitterGPUAPI.cu b/Common/DCAFitter/GPU/cuda/DCAFitterGPUAPI.cu
diff --git a/Common/DCAFitter/GPU/cuda/DCAFitterGPUAPI.h b/Common/DCAFitter/GPU/cuda/DCAFitterGPUAPI.h
diff --git a/Common/DCAFitter/GPU/cuda/DCAFitterN.cu b/Common/DCAFitter/GPU/cuda/DCAFitterN.cu
@@ -49,18 +49,58 @@ GPUg() void processKernel(o2::vertexing::DCAFitterN<2>* ft, o2::track::TrackParC
 {
   *res = ft->process(*t1, *t2);
 }
+} // namespace kernel
 
-void printKHost(o2::vertexing::DCAFitterN<2>* ft, int th, int bl)
+void printOnDevice(o2::vertexing::DCAFitterN<2>* ft,
+                   const int nBlocks,
+                   const int nThreads)
 {
   DCAFitterN<2>* ft_device;
   gpuCheckError(cudaMalloc(reinterpret_cast<void**>(&ft_device), sizeof(o2::vertexing::DCAFitterN<2>)));
   gpuCheckError(cudaMemcpy(ft_device, ft, sizeof(o2::vertexing::DCAFitterN<2>), cudaMemcpyHostToDevice));
-  LOGP(info, "ft: {} ft_device: {} size: {}", (void*)ft, (void*)ft_device, sizeof(o2::vertexing::DCAFitterN<2>));
-  printKernel<<<bl, th>>>(ft);
+
+  kernel::printKernel<<<nBlocks, nThreads>>>(ft_device);
+
   gpuCheckError(cudaPeekAtLastError());
   gpuCheckError(cudaDeviceSynchronize());
-  // static_assert(false);
 }
-} // namespace kernel
 
-} // namespace o2::vertexing::gpu
+int processOnDevice(o2::vertexing::DCAFitterN<2>* fitter,
+                    o2::track::TrackParCov& track1,
+                    o2::track::TrackParCov& track2,
+                    const int nBlocks,
+                    const int nThreads)
+{
+  DCAFitterN<2>* ft_device;
+  o2::track::TrackParCov* t1_device;
+  o2::track::TrackParCov* t2_device;
+  int result, *result_device;
+
+  gpuCheckError(cudaMalloc(reinterpret_cast<void**>(&ft_device), sizeof(o2::vertexing::DCAFitterN<2>)));
+  gpuCheckError(cudaMalloc(reinterpret_cast<void**>(&t1_device), sizeof(o2::track::TrackParCov)));
+  gpuCheckError(cudaMalloc(reinterpret_cast<void**>(&t2_device), sizeof(o2::track::TrackParCov)));
+  gpuCheckError(cudaMalloc(reinterpret_cast<void**>(&result_device), sizeof(int)));
+
+  gpuCheckError(cudaMemcpy(ft_device, fitter, sizeof(o2::vertexing::DCAFitterN<2>), cudaMemcpyHostToDevice));
+  gpuCheckError(cudaMemcpy(t1_device, &track1, sizeof(o2::track::TrackParCov), cudaMemcpyHostToDevice));
+  gpuCheckError(cudaMemcpy(t2_device, &track2, sizeof(o2::track::TrackParCov), cudaMemcpyHostToDevice));
+
+  kernel::processKernel<<<nBlocks, nThreads>>>(ft_device, t1_device, t2_device, result_device);
+
+  gpuCheckError(cudaPeekAtLastError());
+  gpuCheckError(cudaDeviceSynchronize());
+
+  gpuCheckError(cudaMemcpy(&result, result_device, sizeof(int), cudaMemcpyDeviceToHost));
+  gpuCheckError(cudaMemcpy(fitter, ft_device, sizeof(o2::vertexing::DCAFitterN<2>), cudaMemcpyDeviceToHost));
+  gpuCheckError(cudaMemcpy(&track1, t1_device, sizeof(o2::track::TrackParCov), cudaMemcpyDeviceToHost));
+  gpuCheckError(cudaMemcpy(&track2, t2_device, sizeof(o2::track::TrackParCov), cudaMemcpyDeviceToHost));
+  gpuCheckError(cudaFree(ft_device));
+  gpuCheckError(cudaFree(t1_device));
+  gpuCheckError(cudaFree(t2_device));
+
+  gpuCheckError(cudaFree(result_device));
+
+  return result;
+}
+
+} // namespace o2::vertexing::gpu
diff --git a/Common/DCAFitter/GPU/cuda/test/main_prog.cxx b/Common/DCAFitter/GPU/cuda/test/main_prog.cxx