International organisation for standardisation

6.19Complexity assessment

6.19.1.1.1.1.1.1.1JCTVC-F043 Complexity assessment methodology [D. Alfonso (STM)]

This contribution presents a complexity assessment methodology based on three kinds of system modelling: algorithmic, software and hardware. It was proposed to adopt said methodology as standard practice in JCT-VC project management for the current HEVC standardization project.

It was proposed to consider:

• 
  Resources consumed statically, not related to the notion of time. This refers to resources needed to realize the system, like silicon footprint, memory size, number of elementary processing units, et cetera.
  
• 
  Resources consumed dynamically, depending on the time dimension. This refers to resources needed to operate the system, in particular computation (amount of elementary operations performed per time unit) and communication (amount of data transferred per time unit).
  

The proposed methodology is based on three classes of systems:

• 
  Level 1 is based on abstract modelling, i.e. the system is realized as one or more algorithms, ideally executed on a theoretical architecture. At this level, complexity assessment is made by algorithmic analysis.
  
• 
  Level 2 is based on software modelling, i.e. the system is a software implementation, running on a general-purpose hardware architecture such as a CPU. Practically, the HM software is the HEVC system implementation and the complexity is assessed by measuring HEVC software performance.
  
• 
  Level 3 is based on hardware modelling, i.e. the system is a hardware implementation realized over a certain technology. At this level, it is possible to measure some other complexity dimensions that it is not possible to consider at level 2.
  

It was proposed that this be considered at least a starting point toward an agreed assessment methodology for the work.

Profiling was suggested as an important element of complexity study.

As general guidelines to be considered, the contribution contains useful suggestions and presents a good description of things to be done in a proper complexity analysis.

However, strict application of a methodology may not be fully practical for routine proposal evaluations in our work, and we must rely somewhat on expert judgment to enable moving forward in a timely fashion.

Contributors are encouraged to consider this methodology and use it in our work. Further study of complexity analysis methodology is also encouraged.

6.19.1.1.1.1.1.1.2JCTVC-F487 Study of memory bandwidth to develop HEVC [Teruhiko Suzuki (Sony)] [late upload 07-11]

Presented Thu 07-21 2030.

The contribution suggests considering profile / level constraints based on memory bandwidth issues.

The contribution included an illustration of the memory bandwidth impact of restrictions, illustrating some cases where HEVC has higher bandwidth than AVC.

The concept was generally supported.