Deserializing a chip spec from JSON is slow in quilc (on the order of 1s for an Aspen-M-3 chip). In quilc, we got around that by adding a chip cache, so that only the first use of the chip was slow. We ought to include the same to alleviate that slow deserialization.
Alternatively, we could:
- make use of the caching already in quilc
- improve performance in quilc
The former is reasonable, but would be gated by review from quil-lang; if we instead implement it here, we can move quicker. The latter alternative is not a sure-thing and would take a good amount of effort. So I think managing the cache ourselves is the best option.
Deserializing a chip spec from JSON is slow in quilc (on the order of 1s for an Aspen-M-3 chip). In quilc, we got around that by adding a chip cache, so that only the first use of the chip was slow. We ought to include the same to alleviate that slow deserialization.
Alternatively, we could:
The former is reasonable, but would be gated by review from quil-lang; if we instead implement it here, we can move quicker. The latter alternative is not a sure-thing and would take a good amount of effort. So I think managing the cache ourselves is the best option.