Page 162 - My FlipBook
P. 162
研
人 Research Fellow
員 李丕榮 PeiZong Lee
Faculty Ph.D., Computer Science, New York University, United States
T +886-2-2788-3799 ext. 1812 E leepe@iis.sinica.edu.tw
F +886-2-2782-4814 W www.iis.sinica.edu.tw/pages/leepe/
・ Research Fellow, Academia Sinica, Taiwan (1998/1-present)
・ Adjunct Associate Professor, Department of Electrical Engineering, National Chung
Cheng University, Chiayi, Taiwan (1993-1996)
・ Associate Research Fellow, Academia Sinica, Taiwan (1989/10-1998/1)
Research Description
My research interests are in compilers for scienti c applications, parallel algorithm design, computer architectures, and the interplay among
architectures, algorithms, and compilers. The challenge of implementing large scienti c applications for current parallel computers is to
handle data locality among memory hierarchies, so that to avoid memory conflict on shared memory parallel computers and to avoid
communication overhead on distributed memory parallel computers. Compilers act as bridges connecting algorithms and architectures. I am
interested in studying this interdependence.
For running on distributed memory parallel computers, if a program does not use indirect memory accesses, such as subscript array of arrays
or pointers, compilers can nd data dependence relations among statements and data alignment relations among data arrays. Therefore,
compilers have enough information to determine data distribution, execution scheduling, and the generation of communication code using
e ective communication primitives. However, if a program does use subscript array of arrays or pointers to implement indirect memory
accesses for irregular computation, compilers at the compiling time cannot determine neither data dependence relations nor data alignment
relations; if compilers only provide naïve data distributions, generated code cannot avoid adopting expensive communication primitives
which de nitely lengthen the execution time and thus degrade the performance of parallel processing.
To understand the techniques for compiling irregular computation, we analyze real code for scienti c computation, in which we have to
conduct research into unstructured mesh generation; unstructured mesh partition; Euler equation and Navier- Stokes equation solvers for a
numerical wind tunnel platform, an engine combustion platform for computing reactive ows; visualization; and the challenge of using MPI
on multi-core workstation/PC clusters to accelerate irregular computation. Our target is to provide paradigms for designing di erent classes
of computation-intensive applications, so that compilers can automatically generated e cient code for parallel computation.
Publications
1. PeiZong Lee and Zvi M. Kedem. 1988. "Synthesizing Linear- Parallel and Distributed Systems, Vol. 8, No. 8, Aug. 1997, page
Array Algorithms from Nested For Loop Algorithms," in The 825-839.
Special Issue on Parallel and Distributed Algorithms, IEEE
Transactions on Computers, Vol. C-37, No. 12, Dec. 1988, page 7. P e i Z o n g L e e a n d We n - Ya o C h e n . 2 0 0 2 . " G e n e r a t i n g
1578-1598. Communication Sets of Array Assignment Statements for Block-
Cyclic Distribution on Distributed Memory Parallel Computers,"
2. PeiZong Lee and Zvi M. Kedem. 1990. "Mapping Nested Loop Parallel Computing, Vol. 28, No. 9, Sep. 2002, page 1329-1368.
Algorithms into Multidimensional Systolic Arrays," in IEEE
Transactions on Parallel and Distributed Systems, Vol. 1, No. 1, 8. PeiZong Lee and Zvi M. Kedem. 2002. "Automatic Data and
Jan. 1990, page 64-76. Computation Decomposition on Distributed Memory Parallel
Computers," ACM Transactions on Programming Languages and
3. PeiZong Lee and Fang-Yu Huang. 1994. "Restructured Recursive Systems, Vol. 24, No. 1, Jan. 2002, page 1-50.
DCT and DST Algorithms," in IEEE Transactions on Signal
Processing, Vol. 42, No. 7, July 1994, page 1600-1609. 9. PeiZong Lee, Chih-Hsueh Yang, and Jeng-Renn Yang. 2004.
"Fast Algorithms for Computing Self-Avoiding Walks and
4. PeiZong Lee and Fang-Yu Huang. 1994. "An Efficient Prime- Mesh Intersections over Unstructured Meshes," Advances in
Factor Algorithm for the Discrete Cosine Transform and Its Engineering Software, Vol. 35, No. 2, Feb. 2004, page 61-73.
Hardware Implementations," IEEE Transactions on Signal
Processing, Vol. 42, No. 8, Aug. 1994, page 1996-2005. 10. PeiZong Lee, Chien-Min Wang, and Jan-Jan Wu. 2006.
"Compiler and Run-time Parallelization Techniques for Scientific
5. PeiZong Lee. 1995. "Techniques for Compiling Programs on Computations on Distributed Memory Parallel Computers,"
Distributed Memory Multicomputers," Parallel Computing, Vol. included in the book High Performance Computing: Paradigm
21, No. 12, Dec. 1995, page 1895-1923. and Infrastructure, pp. 135-181, edited by Dr. Laurence T. Yang
and Dr. Minyi Guo, John Wiley & Sons, Inc., 2006.
6. PeiZong Lee. 1997. "Efficient Algorithms for Data Distribution
on Distributed Memory Multicomputers," IEEE Transactions on
160
人 Research Fellow
員 李丕榮 PeiZong Lee
Faculty Ph.D., Computer Science, New York University, United States
T +886-2-2788-3799 ext. 1812 E leepe@iis.sinica.edu.tw
F +886-2-2782-4814 W www.iis.sinica.edu.tw/pages/leepe/
・ Research Fellow, Academia Sinica, Taiwan (1998/1-present)
・ Adjunct Associate Professor, Department of Electrical Engineering, National Chung
Cheng University, Chiayi, Taiwan (1993-1996)
・ Associate Research Fellow, Academia Sinica, Taiwan (1989/10-1998/1)
Research Description
My research interests are in compilers for scienti c applications, parallel algorithm design, computer architectures, and the interplay among
architectures, algorithms, and compilers. The challenge of implementing large scienti c applications for current parallel computers is to
handle data locality among memory hierarchies, so that to avoid memory conflict on shared memory parallel computers and to avoid
communication overhead on distributed memory parallel computers. Compilers act as bridges connecting algorithms and architectures. I am
interested in studying this interdependence.
For running on distributed memory parallel computers, if a program does not use indirect memory accesses, such as subscript array of arrays
or pointers, compilers can nd data dependence relations among statements and data alignment relations among data arrays. Therefore,
compilers have enough information to determine data distribution, execution scheduling, and the generation of communication code using
e ective communication primitives. However, if a program does use subscript array of arrays or pointers to implement indirect memory
accesses for irregular computation, compilers at the compiling time cannot determine neither data dependence relations nor data alignment
relations; if compilers only provide naïve data distributions, generated code cannot avoid adopting expensive communication primitives
which de nitely lengthen the execution time and thus degrade the performance of parallel processing.
To understand the techniques for compiling irregular computation, we analyze real code for scienti c computation, in which we have to
conduct research into unstructured mesh generation; unstructured mesh partition; Euler equation and Navier- Stokes equation solvers for a
numerical wind tunnel platform, an engine combustion platform for computing reactive ows; visualization; and the challenge of using MPI
on multi-core workstation/PC clusters to accelerate irregular computation. Our target is to provide paradigms for designing di erent classes
of computation-intensive applications, so that compilers can automatically generated e cient code for parallel computation.
Publications
1. PeiZong Lee and Zvi M. Kedem. 1988. "Synthesizing Linear- Parallel and Distributed Systems, Vol. 8, No. 8, Aug. 1997, page
Array Algorithms from Nested For Loop Algorithms," in The 825-839.
Special Issue on Parallel and Distributed Algorithms, IEEE
Transactions on Computers, Vol. C-37, No. 12, Dec. 1988, page 7. P e i Z o n g L e e a n d We n - Ya o C h e n . 2 0 0 2 . " G e n e r a t i n g
1578-1598. Communication Sets of Array Assignment Statements for Block-
Cyclic Distribution on Distributed Memory Parallel Computers,"
2. PeiZong Lee and Zvi M. Kedem. 1990. "Mapping Nested Loop Parallel Computing, Vol. 28, No. 9, Sep. 2002, page 1329-1368.
Algorithms into Multidimensional Systolic Arrays," in IEEE
Transactions on Parallel and Distributed Systems, Vol. 1, No. 1, 8. PeiZong Lee and Zvi M. Kedem. 2002. "Automatic Data and
Jan. 1990, page 64-76. Computation Decomposition on Distributed Memory Parallel
Computers," ACM Transactions on Programming Languages and
3. PeiZong Lee and Fang-Yu Huang. 1994. "Restructured Recursive Systems, Vol. 24, No. 1, Jan. 2002, page 1-50.
DCT and DST Algorithms," in IEEE Transactions on Signal
Processing, Vol. 42, No. 7, July 1994, page 1600-1609. 9. PeiZong Lee, Chih-Hsueh Yang, and Jeng-Renn Yang. 2004.
"Fast Algorithms for Computing Self-Avoiding Walks and
4. PeiZong Lee and Fang-Yu Huang. 1994. "An Efficient Prime- Mesh Intersections over Unstructured Meshes," Advances in
Factor Algorithm for the Discrete Cosine Transform and Its Engineering Software, Vol. 35, No. 2, Feb. 2004, page 61-73.
Hardware Implementations," IEEE Transactions on Signal
Processing, Vol. 42, No. 8, Aug. 1994, page 1996-2005. 10. PeiZong Lee, Chien-Min Wang, and Jan-Jan Wu. 2006.
"Compiler and Run-time Parallelization Techniques for Scientific
5. PeiZong Lee. 1995. "Techniques for Compiling Programs on Computations on Distributed Memory Parallel Computers,"
Distributed Memory Multicomputers," Parallel Computing, Vol. included in the book High Performance Computing: Paradigm
21, No. 12, Dec. 1995, page 1895-1923. and Infrastructure, pp. 135-181, edited by Dr. Laurence T. Yang
and Dr. Minyi Guo, John Wiley & Sons, Inc., 2006.
6. PeiZong Lee. 1997. "Efficient Algorithms for Data Distribution
on Distributed Memory Multicomputers," IEEE Transactions on
160
