David Padua has made many contributions in the area  of
program  analysis and translation.  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  introduc-
tion 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  collision  and synchronization between parallel
threads in explicitly parallel  programs  which  enable  the
identification  of race conditions  and the correct applica-
tion of traditional compiler  techniques  to  parallel  pro-
grams.




























9

9