𝔖 Bobbio Scriptorium
✦   LIBER   ✦

Parallel programming in MIMD type parallel systems using transputer and i860 in physical simulations

✍ Scribed by S. Ido; S. Hikosaka


Book ID
104744657
Publisher
Springer
Year
1992
Tongue
English
Weight
596 KB
Volume
10-10
Category
Article
ISSN
0178-7675

No coin nor oath required. For personal study only.

✦ Synopsis


Parallel programming and calculation performance were examined by using two types of MIMD parallel systems, that is, a transputer (T800) network and iPSC/860. Some interface subroutines were developed to apply the programs paralMized by using a transputer network to iPSC/860. Compatibility and performance of paralMized programs are discussed.

Receiving

call f77_chan_in_message (A, B, C) A: Data length (bytes) B: The buffer which contalns message C: The address of the channel call csend (T, B, A, D, E) T: Defining message type B: The buffer which contains the message A: The number of bytes to be transmitted D: The identifier of the destination node E: The identifier of the destination process call crecv (T, B, A) T: Defining message type B: The buffer which contains the message A: The number of types to be transmitted Sending chan out_message (A, B, C) A: Data length (bytes) B: The buffer which contains message C: The address of the channel Receiving chan in message (A, B, C) A. Data length (bytes) B: The buffer which contains message C: The address of the channel csend (T, B, A, D, E) T: Defining message type B: The buffer which contains the message A: The number of bytes to be transmitted D: The identifier of the destination node E: The identifier of the destination process crecv (T, B, A) T: Defining message type B: The buffer which contains the message A: The number of types to be transmitted Appendix II: Speed of floating point operations FORTRAN programs in double precision or single precision were carried out to measure the calculation time for 10 types of floating operations by using a single processing element of a transputer (T800 of 20 MHz) or i860 of 40 MHz. The results are summarized in

Table A.3 (double precision) and Table A.4 (single precision).