Altera Nios II Embedded Evaluation Kit Cyclone III Edition - Manual (Page 48)

Altera Corporation 5–12
July 2010 Nios II Embedded Evaluation Kit, Cyclone III Edition
Design Examples
The implementation is fixed point with all values pre-scaled by
0x10000000. The loop will continue until the number of iterations reaches
‘max_iter’ or x
2 + y 2 converges to the value of 4. This function is called for
each pixel so for this design that would be 384000 times since the screen
resolution is 800x480. The value of ‘iter’ is used as the index into the color
palette which picks the color of the pixel displayed on the screen. Even
though the main processor supports dynamic branch prediction and
contains cache memory, this operation of filling the screen can be very
time consuming.
The approach taken for the C2H accelerated version is to offload this
algorithm to pipelined and parallel hardware. Each Mandelbrot engine
contains dedicated multiply, addition, and subtraction logic to perform
multiple operations in parallel. Each Mandelbrot accelerator operates on
a quarter of the frame and is only called once per frame. The workload is
distributed on a pixel basis so each accelerator handles every fourth pixel.
Figure 5–9. Mandelbrot Engine Pixel Interleaving
While the hardware accelerators are calculating a frame, the coordinates
of the next frame are prepared by the main i.e. the Nios II processor. The
hardware accelerators contain a C pragma that instructs the compiler to
implement non-blocking accelerators which allows both the main
processor and Mandelbrot hardware accelerators to operate in parallel.
The main processor polls the hardware accelerator to determine if the
entire frame has been rendered.