How We Should Program GPGPUs
The future for GPU programming is getting brighter; these devices will become more convenient to program. There is no magic bullet; only appropriate algorithms written in a transparent style can be compiled for GPUs; users must understand and accept their advantages and limitations. These are not standard processor cores.
The industry can expect additional development of programmable accelerators, targeting different application markets. The cost of entering the accelerator market is much lower than for the CPU market, making a niche target market potentially attractive. The compiler method described here is robust enough to provide a consistent interface for a wide range of accelerators.
Michael Wolfe has been a compiler engineer at The Portland Group since joining in 1996, where his responsibilities and interests include deep compiler analysis and optimizations ranging from improving power consumption for embedded microcores to improving the efficiency of FORTRAN on parallel clusters. He has a PhD in Computer Science from the University of Illinois and authored High Performance Compilers for Parallel Computing, Optimizing Supercompilers for Supercomputers and many technical papers.