Programming Systems for a Digital Human

1
Programming Systems for a Digital Human

• Kathy Yelick
• EECS Department
• U.C. Berkeley

2
The 20 Year Vision

• Imagine a digital body double
• 3D image-based medical record
• Includes diagnostic, pathologic, and other
  information
• Used for
• Diagnosis
• Less invasive surgery-by-robot
• Experimental treatments
• Digital Human Effort
• Lead by the Federation of American Scientists

3
Organ Simulation

• Brain
• ElIisman

• Lung transport
• Vanderbilt

• Cochlea
• Caltech, UM

• Lung flow
• ORNL

• Cardiac flow
• NYU,

• Cardiac cells/muscles
• SDSC, Auckland, UW, Utah,

• Kidney mesh generation
• Dartmouth

• Electrocardiography
• Johns Hopkins,

Skeletal mesh generation
Just a few of the efforts at understanding and
simulating parts of the human body
4
Programming Systems Challenges

• Programmability of parallel machines
• Machines more complex than ever
• Titanium Java for high performance
• Performance on memory hierachies
• Automatic performance tuning
• BeBOP Berkeley Benchmarking Optimization
• Programming tools
• Race detection, load balancing, communication
• Scalable scientific algorithms

5
Heart Simulation

• Animation of lower portion of the heart

Source www.psc.org
6
Cochlea Simulation

• Embedded structures are
• Elastic membranes
• Shell model
• Variable elasticity
• Capture shape of cochlea
• Shown here is a plate
• Implemented in a few weeks within Titanium code

Source Titanium simulation on a T3E
7
Titanium Overview

• Object-oriented language based on Java with
• Scalable parallelism
• SPMD model with global address space
• Multidimensional arrays
• points and index sets as first-class values
• Immutable classes
• user-definable non-reference types for
  performance
• Operator overloading
• by demand from our user community
• Semi-automated memory management
• uses memory regions for high performance

8
Automatic Performance Tuning

• Problem
• Increasing memory/processor gap
• Increasing complexity of memory systems
• State of the art
• Hand tuning of important kernels
• Why because even when analysis can be done, the
  performance models are too complex for compilers
• BeBOP approach
• Identify and generate a space of implementations
• Instruction mixes, data structures, memory access
  patterns, and mathematical formulations
• Search for the fastest one, by running them
• Examples PHiPAC, ATLAS, Sparsity, FFTW, Spiral,