The concept of barrier synchronization as we use it at Purdue began with Hank Dietz and Tom Schwederski one afternoon in Spring 1987.
Tom was the principal architect of the PASM prototype hardware, which was just becoming operational; Hank was a new faculty member working on compilers and interested in VLIW code scheduling. Both were interested in extending the PASM hardware to support a VLIW mode as well as SIMD and MIMD. After about half an hour, it was realized that VLIW couldn't be supported directly, but that the SIMD instruction fetch interlock and enable masking hardware could actually be used to implement something better... a barrier MIMD.
Thus, we began writing this paper in the summer of 1987. However, publication was delayed while Purdue considered patenting the mechanism.
After Purdue decided not to pursue a patent, the paper was submitted to IEEE Supercomputing 1988. At that time we also released the work as Purdue School of Electrical Engineering Technical Report TR-EE 88-25, which is reproduced here as both HTML and Postscript.
Although the paper was rejected by IEEE Supercomputing in 1988, word about the barrier MIMD technical report spread quickly, and we had a number of very positive reactions from both researchers and companies. So we submitted the exact same paper to IEEE Supercomputing 1989. This time, it was accepted. In fact, the paper was even invited to appear in a journal collecting the few best papers from the conference. We decided not to do that journal publication -- after all, the paper was nearly three years old....
Since then, we have had little difficulty publishing a variety of conference and journal papers about various aspects of barrier MIMD. Still, it is interesting to realize that barrier MIMD almost didn't get past the initial barrier of being too different and too new.