OpenCL GPU vs FPGA implementation

[shaiko]

Hello,

When a program is written in OpenCL with a GPU as the target device - the host Cpu distributes kernels to the many GPU cores to achieve SIMD parallelism.
When the calculation results are ready, the program continues to run and the host distributes new kernels to the GPU...

I'm trying to understand how it works when the target device is an FPGA.
In this case, "Redistributing Kernels" would mean changing the content of the FPGAs logic...so does the host program the device on the fly?

 

[ads-ee]

I would think it would have to be done using partial reconfiguration, otherwise the entire infrastructure that allows access to these "kernels" would also vanish along with any kernels that are still doing something.

I would try to do some research on reconfigurable computing.