Detailed concepts

Here we present more detailed concepts of HTTomo’s framework, such as, Sections, Re-slicing, Data terminology (Chunks, Blocks), Method wrappers, gpu-memory aware processing and others.

• Sections
  • Example 1: Sections with re-slice
  • Example 2 : Sections with re-slice and data saving
  • Example 3 : Sections with side outputs
• Re-slicing
• Method wrappers

Data terminology

HTTomo’s framework deals with handling all the data involved in executing the pipeline. “Data handling” in HTTomo involves operations such as splitting data up into pieces, passing the pieces of data into methods, gathering data up and re-splitting into a different set of pieces, and more.

The concept of data being split into smaller pieces is a common theme in HTTomo, and there naturally arose two main “levels” of data splitting. These two levels have been given names, as an easy way to give some context to a piece of data when referring to it.

One level of data splitting produces a piece of data called a chunk (not to be confused with the term “chunk” used in the hdf5 data format), and the other level of data splitting produces a piece of data called a block.

• Chunks
  • Definition
  • Motivation: distributing data across MPI processes
  • How are chunk shapes calculated?
  • Example
• Blocks
  • Definition
  • Motivation
  • How are block shapes calculated?
  • Blocks as the fundamental data quantity in HTTomo
  • Example