David Padua has made many contributions in the area of program analysis
and compiling.  These include techniques for autoparallelization 
of Fortran programs, compilation of explicitly parallel programs,
compilation of interactive languages like MATLAB,
and detection of race conditions in parallel programs. 
His extensive work in autoparallelization include the development of techniques
to translate loops into "cascade"  parallel form
which anticipated some of the techniques widely used today to compile  
for fine-grain parallelism and influenced the design of parallel language directives
and the organization of some multiprocessors.  
During the last few years he and his students have improved the effectiveness
of autoparallelization with the introduction of new algorithms
for array privatization, compile-time analysis of complex subscript expressions,
and run-time speculation.
His work on MATLAB compilation programs led to performance improvements
of up to three orders of magnitude over interpreted code and,
in some cases, of over a hundred over commercial compilers.
He has also developed techniques to analyze collission and syncrhonization between
parallel threads in explicitly parallel programs which can
be used to identify race conditions  and to  constrain the application
of traditional compiler techniques to ensure  correctness.