Previous Results

These are previous results from earlier compiler versions and machines. (Mostly for internal purposes.)

Seconds (AMS Starfire)
Cycles (MClocks)
Logic Blocks
41.7 MHz

How does this compare to a Pentium? Hard to say. We wrote a version of the same program in C, using Intel's IPL whenever possible. The resulting program took 0.85 seconds on an 800MHz Pentium 3. On the other hand, Intel's OpenCV library has a hand-optimized assembly-coded version of Canny that includes the fourth (connected components) step. We set the high and low thresholds to be the same (so that connected components would not iterate) and timed the OpenCV routine at 0.135 seconds. So SA-C + FPGA was 5 times faster than assembly code and 30 times faster than a C implementation.