Add function selector optimization

[fulldecent]

Test case

Compiler: 0.4.25+commit.59dbf8f1.Emscripten.clang
Optimization: on with default settings in Remix IDE

pragma solidity ^0.4.23;

contract Test
{
    function a() payable returns (address) {
        return address(this);
    }
}

608060405234801561001057600080fd5b5060c08061001f6000396000f300608060405260043610603f576000357c0100000000000000000000000000000000000000000000000000000000900463ffffffff1680630dbe671f146044575b600080fd5b604a608c565b604051808273ffffffffffffffffffffffffffffffffffffffff1673ffffffffffffffffffffffffffffffffffffffff16815260200191505060405180910390f35b6000309050905600a165627a7a72305820effc577e00e0dab11af3a8f067f6755dd66ef9191ab577e8ea00920b04f8429e0029

Opcodes

| Loc | Op | Data | | ---- | ------------ | ------------------------------------------------------------ | | 0 | PUSH1 | 80 | | 2 | PUSH1 | 40 | | 4 | MSTORE | | | 5 | PUSH1 | 04 | | 7 | CALLDATASIZE | | | 8 | LT | | | 9 | PUSH1 | 3f | | 11 | JUMPI | | | 12 | PUSH1 | 00 | | 14 | CALLDATALOAD | | | 15 | PUSH29 | 0100000000000000000000000000000000000000000000000000000000 | | 45 | SWAP1 | | | 46 | DIV | | | 47 | PUSH4 | ffffffff | | 52 | AND | | | 53 | DUP1 | | | 54 | PUSH4 | 0dbe671f | | 59 | EQ | | | 60 | PUSH1 | 44 | | 62 | JUMPI | | | 63 | JUMPDEST | | | 64 | PUSH1 | 00 | | 66 | DUP1 | | | 67 | REVERT | | | 68 | JUMPDEST | | | 69 | PUSH1 | 4a | | 71 | PUSH1 | 8c | | 73 | JUMP | | | 74 | JUMPDEST | | | 75 | PUSH1 | 40 | | 77 | MLOAD | | | 78 | DUP1 | | | 79 | DUP3 | | | 80 | PUSH20 | ffffffffffffffffffffffffffffffffffffffff | | 101 | AND | | | 102 | PUSH20 | ffffffffffffffffffffffffffffffffffffffff | | 123 | AND | | | 124 | DUP2 | | | 125 | MSTORE | | | 126 | PUSH1 | 20 | | 128 | ADD | | | 129 | SWAP2 | | | 130 | POP | | | 131 | POP | | | 132 | PUSH1 | 40 | | 134 | MLOAD | | | 135 | DUP1 | | | 136 | SWAP2 | | | 137 | SUB | | | 138 | SWAP1 | | | 139 | RETURN | | | 140 | JUMPDEST | | | 141 | PUSH1 | 00 | | 143 | ADDRESS | | | 144 | SWAP1 | | | 145 | POP | | | 146 | SWAP1 | | | 147 | JUMP | | | 148 | STOP | | | 149 | LOG1 | | | 150 | PUSH6 | 627a7a723058 | | 157 | SHA3 | | | 158 | INVALID | | | 159 | INVALID | | | 160 | JUMPI | | | 161 | PUSH31 | 00e0dab11af3a8f067f6755dd66ef9191ab577e8ea00920b04f8429e0029 |

Subjects

Check function signature

Currently this uses:

Loc	Op	Data
12	PUSH1	00
14	CALLDATALOAD
15	PUSH29	0100000000000000000000000000000000000000000000000000000000
45	SWAP1
46	DIV
47	PUSH4	ffffffff
52	AND

Optimizations

Function selector

Because words are 32 bytes, a divide by 0100000000000000000000000000000000000000000000000000000000 need not be followed by an and with ffffffff.

Maybe this case is an over optimization. But function selectors apply to every call of every contract. So they should be airtight.

New version:

Loc	Op	Data
12	PUSH1	00
14	CALLDATALOAD
15	PUSH29	0100000000000000000000000000000000000000000000000000000000
45	SWAP1
46	DIV

[chriseth]

Which exact logic would you propose? This kind of needs tracking of the range of all stack elements, which I think is a rather invasive step. Do you think this can be done a part of the peephole optimizer?

[chriseth]

Instead of adding an optimizer stage, we could also just fix this part of the code: If you right-shift after memory load, cleanup does not have to be performed.

[chriseth]

The relevant part is CompilerUtils::loadFromMemoryHelper - note that we only do not need cleanup for bytesNN and uintN types. It is required for signed integers, for example.

[fulldecent]

Specifically I recommend these two rules.

s/357c0100000000000000000000000000000000000000000000000000000000900463ffffffff16
/7c01000000000000000000000000000000000000000000000000000000003504/g

s/7c0100000000000000000000000000000000000000000000000000000000350463ffffffff16
/7c01000000000000000000000000000000000000000000000000000000003504/g

More general rules might do other things, but this is quickest.

Before I get more involved in optimization it would be really helpful if somebody had a list of all the contracts deployed on mainnet and how many billable times each instruction was executed. Maybe I should make that!