FYI:
Brook+ source is included with the Brook+ packaging. usually installs to /usr/local/amdbrook/platform
Secondly, I am planning on supporting Brook+ to an extent as an optimized backend, to see differences in performance between it and the up and coming OpenCL.
I've been made aware OpenCL is more feasible for graphics interaction as well as cross-platform cross-hardware support. I'm sure the tools within the StreamSDK will prove to be more optimized than the newer cross-platform;/hardware solutions, and thusly even if just for my own sake, I'm going to try and allow for fair comparison in the areas (through detailed categorization of performance benchmarks) that each may have an advantage... For example without OpenCL in my application there may be some serious bottlenecks as I transfer things off the videocard and back again in the form of OpenGL objects. However I would not doubt the kernels to run faster using IL.
So at least for my own purposes as this topic becomes more interesting, I'm going to try and implement mechanisms to account for Brook+'s downfalls and allow for some benchmark comparisons.
Long story short, as long as the vendor specific optimized libraries work the fastest for pure computational purposes, there's little hope for the intricasies being released very soon, it would probably expose many things to the green guys about said optimizations... This is still a competitive industry 
.
At this point, because of the simplicity of the application of Brook+ (I think in the future it's only gonna be easier)... I whole heartedly feel it is a treasure for computation in professional and scientific studies of all sorts.
With CAL AND Brook+ APIs it should be much easier to maintain such projects, especially when the Stream SDK makes it out of beta status.
Note: we are in a time of gpgpu flavor overload, I expect over the years there'll be much battling... Vendor-specific implementations should have an advantage, and hopefully OpenCL will help cast a shadow toward DX11, because it's yet another way to entrap people within a closed community. That leaves us the developers to try and make a push with these tools we are given, towards the open community and cross-platform support, if not necessarily cross-hardware.
Also do note I've not had too much trouble using Ubuntu, even 8.10. All the kinks I used to experience have long since been worked out, besides mapping a kernel again after it leaves the scope it was first called in... That will be fixed in 1.3 as well (and is functional in 1.1).
