bgclang: Creating an Alternative, Customizable, Toolchain for the Blue Gene/Q

The IBM Blue Gene/Q (BG/Q) supercomputer, the third generation in the IBM Blue Gene line, has once again demonstrated the viability of using a non-commodity hardware to deliver scalability, power efficiency and good system reliability while providing high computational performance. One unfortunate disadvantage of this custom-hardware approach, however, is that the software ecosystem made available on these highly specialized machines pales in comparison to what is available to users of commodity hardware. This is especially true for software development tools: compilers, debuggers, profiling tools, etc. because these tools require nontrivial customization to work well on custom hardware. Not only does this restricted software ecosystem often make it difficult to compile, debug and profile applications, but it also limits the ability of researchers and engineers to experiment with alternative techniques across large codebases. This is because the usual open-source tools, which are not only widely supported by applications but also provide a solid platform for research, are often not available, or don’t work well, on noncommodity machines.