Does Portland Goup intend to extend their product range to allow ATI Firestream to be used as a GPU device (i.e. somehow covering openCL as well as CUDA?)…
The PGI Accelerator Model was designed to be device independent. You the programmer would only need to recompile to target a new device. Better yet, you can create a unified binary where your code could run multiple types of devices depending upon what was installed on the system (important if you’re and ISV and only want one binary to ship to customers).
Currently we only target NVIDIA devices. This was largely a market driven choice since NVIDIA is way ahead of ATI in terms of tools and HPC market penetration. Though, we would like to expand support to other devices (we’re planning on targeting multi-core x86 next).
For ATI support, we need to wait for their tool chain to be more developed and see if the HPC market begins to adopt them.
Thanks Mat - it’s useful to know this.