change file extensions to avoid TBB-related errors

CMakeLists.txt

@@ -183,8 +183,8 @@ if(BUILD_WITH_CUDA)
     src/gtsam_points/cuda/cuda_graph.cu
     src/gtsam_points/cuda/cuda_graph_exec.cu
     # src/gtsam_points/cuda/gl_buffer_map.cu
-    src/gtsam_points/cuda/nonlinear_factor_set_gpu.cu
-    src/gtsam_points/cuda/nonlinear_factor_set_gpu_create.cu
+    src/gtsam_points/cuda/nonlinear_factor_set_gpu.cpp
+    src/gtsam_points/cuda/nonlinear_factor_set_gpu_create.cpp
     src/gtsam_points/cuda/stream_roundrobin.cu
     src/gtsam_points/cuda/stream_temp_buffer_roundrobin.cu
     # types
@@ -197,7 +197,7 @@ if(BUILD_WITH_CUDA)
     src/gtsam_points/factors/integrated_vgicp_derivatives_inliers.cu
     src/gtsam_points/factors/integrated_vgicp_derivatives_compute.cu
     src/gtsam_points/factors/integrated_vgicp_derivatives_linearize.cu
-    src/gtsam_points/factors/integrated_vgicp_factor_gpu.cu
+    src/gtsam_points/factors/integrated_vgicp_factor_gpu.cpp
     # util
     src/gtsam_points/util/easy_profiler_cuda.cu
   )

include/gtsam_points/factors/impl/integrated_gicp_factor_impl.hpp

@@ -105,6 +105,10 @@ void IntegratedGICPFactor_<TargetFrame, SourceFrame>::update_correspondences(con
     }
   }
 
+  if (do_update) {
+    last_correspondence_point = delta;
+  }
+
   correspondences.resize(frame::size(*source));
   mahalanobis.resize(frame::size(*source));
 
@@ -130,8 +134,6 @@ void IntegratedGICPFactor_<TargetFrame, SourceFrame>::update_correspondences(con
       mahalanobis[i](3, 3) = 0.0;
     }
   }
-
-  last_correspondence_point = delta;
 }
 
 template <typename TargetFrame, typename SourceFrame>

src/gtsam_points/cuda/nonlinear_factor_set_gpu.cu → src/gtsam_points/cuda/nonlinear_factor_set_gpu.cpp

@@ -12,14 +12,14 @@ namespace gtsam_points {
 NonlinearFactorSetGPU::DeviceBuffer::DeviceBuffer() : size(0), buffer(nullptr) {}
 
 NonlinearFactorSetGPU::DeviceBuffer::~DeviceBuffer() {
-  if(buffer) {
+  if (buffer) {
     check_error << cudaFreeAsync(buffer, 0);
   }
 }
 
 void NonlinearFactorSetGPU::DeviceBuffer::resize(size_t size, CUstream_st* stream) {
-  if(this->size < size) {
-    if(buffer) {
+  if (this->size < size) {
+    if (buffer) {
       check_error << cudaFreeAsync(buffer, stream);
     }
     check_error << cudaMallocAsync(&buffer, size, stream);
@@ -163,12 +163,8 @@ void NonlinearFactorSetGPU::error(const gtsam::Values& values) {
   }
 
   // copy input buffer from cpu to gpu
-  check_error << cudaMemcpyAsync(
-    evaluation_input_buffer_gpu->data(),
-    evaluation_input_buffer_cpu.data(),
-    input_buffer_size,
-    cudaMemcpyHostToDevice,
-    stream);
+  check_error
+    << cudaMemcpyAsync(evaluation_input_buffer_gpu->data(), evaluation_input_buffer_cpu.data(), input_buffer_size, cudaMemcpyHostToDevice, stream);
   check_error << cudaStreamSynchronize(stream);
 
   // issue error computation
@@ -195,12 +191,8 @@ void NonlinearFactorSetGPU::error(const gtsam::Values& values) {
   }
 
   // copy output buffer from gpu to cpu
-  check_error << cudaMemcpyAsync(
-    evaluation_output_buffer_cpu.data(),
-    evaluation_output_buffer_gpu->data(),
-    output_buffer_size,
-    cudaMemcpyDeviceToHost,
-    stream);
+  check_error
+    << cudaMemcpyAsync(evaluation_output_buffer_cpu.data(), evaluation_output_buffer_gpu->data(), output_buffer_size, cudaMemcpyDeviceToHost, stream);
   check_error << cudaStreamSynchronize(stream);
 
   // store computed results