Optimize `type(uint256).max` into `PUSH0 NOT`

[radeksvarz]

Abstract and Motivation

type(uint256).max is often used as the indicator for something unlimited, such as approvals.

Solidity compiles type(uint256).max into PUSH32 0xffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffffff

This requires 33 bytes in the code (3 gas) for each situation type(uint256).max is used (such as comparison).

I propose much shorter approach for the case the optimiser is set to optimise bytecode length.

Compile to this instead (in the case the optimiser is set to optimise bytecode length):

PUSH0
NOT

This results in to 2 bytes code and 5 gas.

Backwards Compatibility

The optimised bytecode is shorter (-31 bytes), but has +2 gas. Functionally the same.

[cameel]

Any particular case where you see this optimization not being applied? This is generally what the compiler does already:

test.sol

contract C {
    function f() public returns (uint) {
        return type(uint).max;
    }
}

solc test.sol --ir-optimized --optimize --asm

legacy pipeline:

        /* "test.sol":17:90  function f() public returns (uint) {... */
    tag_3:
      not(0x00)

IR pipeline:

mstore(_1, /** @src 0:69:83  "type(uint).max" */ not(0))

Unless you don't enable optimizer or tell it that you don't care about bytecode size that much (by passing a large number into --optimize-runs). Then you'll get the original constant there.

[radeksvarz]

@cameel Thank you for your time & effort to explain. You are very much correct. To confirm (using default legacy pipeline):

unoptimised code:
After forge init.
The forge docs state: By default, the optimizer is enabled and runs for 200 cycles.

So it looks like this might be the issue of the Foundry, not providing the default optimiser params.

optimised code:
(optimiser = true, 200 runs -> short bytecode preference ):

note: here the PUSH1 0x46 JUMP JUMPDEST is redundant, but that is a different issue.

optimised code:
(optimiser = true, 2000000 runs -> low run gas preference ):