MaJIC: Compiling MATLAB for speed and responsiveness

1
MaJIC Compiling MATLAB for speed and
responsiveness

• George Almasi and David Padua

2
MaJIC Overview

• MAJIC (MAtlab Just-In-Time Compiler)
• interactive and fast
• - combination interpreter/JIT compiler
• - speculative ahead-of-time compilation
• - builds on top of FALCON techniques

3
Compiler Techniques in MaJIC

• Analysis
• Compile only code that takes time to execute
  (loops)
• type analysis and value/limit propagation
• recompile only when source has changed

• Code Generation
• naïve (per AST node) JIT code generation
• uses built-in MATLAB functions where possible
• average compile time 20ms per line of MATLAB
  source

4
MaJIC Compiler
5
MaJIC Software Architecture
6
Events in the Code Repository
7
Performance
8
Conclusion

• Provided high performance to the Matlab
  integrated environment
• gt JIT Compilation
• gt Speculative Compilation

9
Discussion

• What are the downfalls of this approach?
• Can it be applied to C/Scale compiler?
• Future optimizations?