France-based CAPS-Entreprise announced this week that they are expanding their flagship HMPP compiler to support NVIDIA’s next-generation architecture
“With several key features such as a true cache hierarchy, concurrent thread execution and ECC, Fermi brings real breakthrough in GPU computing that we are very excited to exploit within our HMPP directive-based CUDA compiler.” declares Dr François Bodin, CAPS Chief Technical Officer.
By generating CUDA code from C and Fortran, HMPP offers a high level abstraction of hybrid programming that fully leverages the computing power of NVIDA Tesla processors. HMPP enables developers to incrementally write hybrid applications, simply from accelerating regions of code, letting HMPP manage all the optimizations such as CPU-GPU data movements, to advanced tuning CUDA kernels. One of the key benefits of HMPP is to ensure portability and performance of CUDA applications across new versions of NVIDIA processors.