Uniswap
gas optimizations for the FullRange hook
Related Issue
#84
Description of changes
This PR attempts to improve the FullRange hook code.
The rationale for increasing the tick spacing is described in #84.
The existing tests show some gas improvements, for most of the tests. However, the main benefits would for large swaps, something that is not captured by the existing tests. A new test testFullRange_swap_LargeSwap is added, with a gas snapshot of its own. With 60 tick spacing the gas cost is 175,523. With the new 0x7000 tick spacing it is just 150,173.
Unfortunately the large tick spacing adds some rounding errors to the liquidity distribution, so most of the existing tests were failing. I modified the tests, mostly by changing from assertEq to assertApproxEqAbs with the accuracy of DUST (30 wei).
Let me know how you feel about increasing the tick spacing, if this is contentious I can make a simple PR that just adds the immutable fields and the other minor changes.
Also I did not attempt to do the more tricky optimizations like adding unchecked to the code, so I believe there's room for further improvements.
I have signed and updated the PR @saucepoint
Some of the less common operations, like the second swap in a tx, actually got a little bit more expensive, but there are big gas savings on the first swap, and the code is simpler as well.
About why use 0x7000 as tick spacing -- good question.
I touched this a bit in issue #84. To expand further:
- the max value supported by the v4-core is 0x7fff (max of a signed 16-bit integer)
- however, we can save some gas in calldata by sending zero bytes, so 0x7f00 or 0x7000 are going to be slightly more efficient
- there's not that big difference from the liquidity perspective between 0x7fff, 0x7f00, or 0x7000 ticks.
- however, there's a big difference between 0x7000 and a more "round" number such as 0x1000 or (decimal) 10000.
The reasoning above caused me to prefer 0x7000 (28672 in decimal).