I got my first dev job after Uni at the sister company to Allinea, so I know and have had a beer or three with most of the people involved.
Will admit to having worked near them for nearly 2 years and still not understanding fully what they do (a cloud distributed debugger for HPC workloads, iirc) but I can tell you this isn't a unicorn growth startup -- they've been working hard on this for over a decade.
One of the less obvious benefits of joining ARM is having extra sites to recruit developers at. Adding Manchester and potentially Cambridge to the existing two sites is cool.
Their web site might give a clue that they produce a parallel debugger and profiler integrated with an (Eclipse PPT-like?) GUI. The debugger is based on launching instances of gdb (which isn't trivial at scale).
More to the point, can someone tell me why I should use the profiler rather than the free/gratis tools, like TAU? Their salesman couldn't.
While a full parallel debugger is clearly useful in difficult cases, easily the most useful debugging tool is stack traces, perhaps from openmpi's automatic backtrace or LLNL's STAT. I don't know whether Totalview (proprietary alternative to DDT) was ever seriously used when we had it available to a multi-site project.
From what I recall of prices, if you're going to buy the tools for more cores than on the sort of single node we have, you're talking something like my salary. Then, in practice most university users won't measure or debug anyway, and may not accept deadlock as a concept when systems people do.
Reminds me a lot of Nvidia buying Portland Group, although that was probably a better match. I don't really see this making much difference - what AMD needs in HPC is tooling for their GPU/APU that covers a wide range of usecases, is stable and performant. Publications and then sales pretty much follow automatically, but IMO so far the tooling was the missing element. E.g. where's my Fortran API for OpenCL and Nvidia? How do I debug and profile this? What about device memory aware MPI wrappers? BLAS on device? Etc.
Allinea AFAIK only helps with the debugging/profiling part - but first you need a sensible programming model.
I think they know this, e.g. they opened a new office in Manchester (UK), and LinkedIn suggests the location is for HPC and tooling.
I'm guessing Allinea makes sense as they're also in the UK. It might be an acquihire thing, instead of specific tooling. The Manchester location took a while to set up, it's still growing and still looking for people, meanwhile I think they're moving to new offices. Talent can be a bit limited in the UK, especially outside of London (and after Brexit).
Will admit to having worked near them for nearly 2 years and still not understanding fully what they do (a cloud distributed debugger for HPC workloads, iirc) but I can tell you this isn't a unicorn growth startup -- they've been working hard on this for over a decade.
Congrats all!