12-19-2013, 02:34 AM
Counting "logical operations" is based on the frame of reference of the observer. If, instead, the number of ticks per net function (e.g. adding, multiplying, etc.) was considered, it would then be easy to compare one device to another. For example, if I made an ALU which could perform functions x,y,z in n ticks, then I would compare it to other ALUs which could do the same tasks and have a measure of relative performance. This, however, is exactly what we already do, and is why I do not think the herp is a useful unit. The herp, however interesting, is not useful for comparing devices.
More useful, in my opinion, is the idea of comparing tick growth as a function of throughput growth to parallelism growth. For example, If I had a device which could perform its task in 3 ticks, but gained 2 ticks every 8 bits which were added, I could write that as 3,2. 3 ticks plus 2 ticks per additional byte. This, in my opinion, is a more useful assessment of the logical efficiency of a device.
More useful, in my opinion, is the idea of comparing tick growth as a function of throughput growth to parallelism growth. For example, If I had a device which could perform its task in 3 ticks, but gained 2 ticks every 8 bits which were added, I could write that as 3,2. 3 ticks plus 2 ticks per additional byte. This, in my opinion, is a more useful assessment of the logical efficiency of a device.