As many pointed out, it's a quite complicated question doesn't have a straightforward answer. I used to worked on RTL implementation of graphic algorithms for years and I can say there can be night and days between different implementations of the same algorithm.
Also unspecified is their performance requirement. What kind of input your algorithm is expecting? YUV? RGB? CMYK? What's the expecting throughput? How much memory and I/O bandwidth your algorithm is going to take? How many temporary registers are needed? Do they allow deep pipelining and longer latency? What's the fab process they are going to use? There are far too many variables need to be taken into consideration. Also some seemly minor tweaks can bring major differences on both area and speed. I once helped a friend to optimize a supposedly simple error diffusion pipeline. Took us 3 weeks to shrink original design into one third size of original implementation while improved its performance by 15% at the same time.
I would say simply tell them you don't know because you are software people unless they are only interested in synthesiz-able codes. Finding someone to write it in RTL for you can be a much heavier burden then you might expected because it's hard to manage something you don't understand. Chances are you may also need to change your algorithms a bit because some operations aren't feasible in hardware with reasonable cost, and some operations can't be removed or simplified.
Slashdot Mirror
As many pointed out, it's a quite complicated question doesn't have a straightforward answer. I used to worked on RTL implementation of graphic algorithms for years and I can say there can be night and days between different implementations of the same algorithm. Also unspecified is their performance requirement. What kind of input your algorithm is expecting? YUV? RGB? CMYK? What's the expecting throughput? How much memory and I/O bandwidth your algorithm is going to take? How many temporary registers are needed? Do they allow deep pipelining and longer latency? What's the fab process they are going to use? There are far too many variables need to be taken into consideration. Also some seemly minor tweaks can bring major differences on both area and speed. I once helped a friend to optimize a supposedly simple error diffusion pipeline. Took us 3 weeks to shrink original design into one third size of original implementation while improved its performance by 15% at the same time. I would say simply tell them you don't know because you are software people unless they are only interested in synthesiz-able codes. Finding someone to write it in RTL for you can be a much heavier burden then you might expected because it's hard to manage something you don't understand. Chances are you may also need to change your algorithms a bit because some operations aren't feasible in hardware with reasonable cost, and some operations can't be removed or simplified.