>> Tuesday, December 20, 2011
I've received a handful of e-mails over the last month and it's about time I responded:
- Global Barrier - One e-mail discusses the shortcomings of my beloved global barrier. My kernel used to execute properly every time I tested it, but with AMD's latest SDK, it fails on my Linux/AMD platform. And when I say it fails, I mean it locks the GPU and freezes my computer. I've added a warning to the blog post and I've removed the barrier from the book's example code.
- Altera - I've received two e-mails concerning Altera's initiative to convert OpenCL code into FPGA designs. OpenCL is easier for traditional programmers to work with than VHDL or Verilog, and I hope this broadens the usage of FPGAs. But it would be even better if Xilinx and Altera developed a customizable soft processor (similar to Xilinx's MicroBlaze) capable of meeting OpenCL's compliance criteria. Then developers could take advantage of FPGAs without having to deal with the FPGA toolchain.
- ARM - In my opinion, the most exciting news comes from ARM, whose latest embedded GPU is OpenCL-compliant. CUDA dominates GPGPU development on servers and desktop systems, but I predict OpenCL will take the lead in embedded and mobile devices. And for smartphones, OpenCL isn't just nice to have--it's ability to accelerate graphics will make it essential for development of immersive 3-D applications.