Many-core programming company Caps has added support for AMD's GPU technology to its HMPP directive-based compiler, a product based on the OpenACC and OpenHMPP programming model. The HMPP compiler integrates a data-parallel backend for the OpenCL heterogeneous computing framework and will use the computing power of AMD GPU and APU devices.
The initial GPUs supported in this release are AMD FirePro W9000, FirePro S8000, and FirePro W8000 series.
Note: HMPP stands for Hybrid Multicore Parallel Programming. HMPP directives are meta-information added in the application source code that do not change the semantic of the original code. They address the remote execution (RPC) of functions or regions of code on GPUs and many-core accelerators as well as the transfer of data to and from the target device memory.
"Supporting an industry standard such as OpenCL in our products is critical to reach the larger developer community that is using GPU compute," said Eric Courtois, director of software development at Caps. "When designing our products, we focus on key tools used by programmers with various levels of expertise from beginners to advanced users. With this release, we are enabling our customers to transition their applications to the latest generation of AMD hardware to fully leverage the underlying computing power of the platform."
The initial performance results are described by the company as "extremely promising" and may now illustrate how the compiler directives can be used to leverage AMD FirePro capabilities while preserving C or FORTRAN codes.
Caps technology also helps users take advantage of AMD Accelerated Parallel Processing Math Libraries (APPML), the software libraries containing FFT and BLAS functions designed to run on AMD GPUs.
AMD for its part argues that directive-based programming offers a simple way to extract the benefits of heterogeneous compute platforms and the compilers such as Caps HMPP can help build GPU accelerated applications faster and increase overall developer productivity.
In a perfect world, if all these technologies gel as promised, a larger community of developers may now be able to take advantage of industry standard directives, including OpenACC now and OpenMP accelerators in the future.