Altera has announced progress on its efforts to use OpenCL to program FPGAs.
The company has in a prototype stage a software tool with "a complete front to back flow that takes OpenCL code and compiles and implements it on our FPGAs with adequate performance," said Udi Landon, Altera’s vice president of software and IP.
Altera is at various stages of introducing and testing the software with companies in vertical markets such as high-performance computing, climate and financial modeling, radar and medical. Engineers at one video processing company "wrote a few hundred lines of OpenCL algorithms, and they were running in on our FPGA in a few hours, then downloaded the same program to CPUs and GPUs," said Landon.
To date, graphics processors have gained more acceptance in vertical markets with highly parallel algorithms. But Altera's work and announcements at industry conferences suggest FPGAs are starting to get traction as an alternative to GPUs in some markets.
The Altera software implements a data path on the FPGA and OpenCL C-language kernels on external or embedded CPUs. Altera currently supports ARM and x86 CPUs and plans to announce support for another embedded processor architecture next year.
The Altera software is still in a definition phase as the company understands the needs of the diverse vertical markets and tunes its code for performance. "We're just announcing a program, not a product at this point," said Jordon Inkeles, Altera’s senior manager of software.
The company is expanding its work on OpenCL with universities. It has been a member of OpenCL since January 2010. Archrival Xilinx joined the OpenCL group this spring, Trevett said.
Altera hopes to add to a future version of the OpenCL spec support for the streaming memory interfaces used by FPGAs, Inkeles said.