Memory management

Handles memory management of different types of host and device classes. More...

Modules
	Pitched memory: Image
	Memory management for Image classes.
	Linear memory
	Memory management for LinearMemory classes.
	Pitched Memory: Volume
	Memory management for Volume classes.

Detailed Description

Handles memory management of different types of host and device classes.

• LinearMemory
  • LinearHostMemory: linear memory with ND size layout
  • LinearDeviceMemory: linear memory with ND size layout
  • TensorCpu: Specialization with 1D size layout. Wrapper for special 4D tensor. Same access convention as python.
  • TensorGpu: Specialization with 1D size layout. Wrapper for special 4D tensor. Same access convention as python.
• Pitched memory: Image
  • ImageCpu
  • ImageGpu
• Pitched memory: Volume
  • VolumeCpu
  • VolumeGpu

The device memory classes can be easily passed to CUDA kernels using a special struct. This struct gives the possibility to not only access the data pointer of the image but also other useful information such as numel/size of the object.

• LinearDeviceMemory::KernelData
• TensorGpu::TensorKernelData
• ImageGpu::KernelData
• VolumeGpu::KernelData