Gas for Execution Cost doesn't add up

[tomw1808]

Considering the following contract:

pragma solidity ^0.4.19;
contract A {
    
    function testme() public{
        
    }
}

It uses gas for input (transaction cost): 4*68 = 272 + 21000 = 21272 - this is correct in Remix. But I cannot verify the execution cost:

It uses the following instructions according to the debugger:

Gas	Instruction
3	000 PUSH1 60
3	002 PUSH1 40
3	004 MSTORE
3	005 PUSH1 04
2	007 CALLDATASIZE
3	008 LT
3	009 PUSH1 3f
10	011 JUMPI
3	012 PUSH1 00
3	014 CALLDATALOAD
3	015 PUSH29 0100000000000000000000000000000000000000000000000000000000
3	045 SWAP1
5	046 DIV
3	047 PUSH4 ffffffff
3	052 AND
3	053 DUP1
3	054 PUSH4 7fdcf8c4
3	059 EQ
3	060 PUSH1 44
10	062 JUMPI
1	068 JUMPDEST
2	069 CALLVALUE
3	070 ISZERO
3	071 PUSH1 4e
10	073 JUMPI
1	078 JUMPDEST
3	079 PUSH1 54
3	081 PUSH1 56
8	083 JUMP
1	086 JUMPDEST
8	087 JUMP
1	084 JUMPDEST
0	085 STOP

Which is in sum 119 gas, but the instruction details in the Remix logger shows 128.

status	0x1 Transaction mined and execution succeed
from	0xca35b7d915458ef540ade6068dfe2f44e8fa733c
to	A.testme() 0x692a70d2e424a56d2c6c27aa97d1a86395877b3a
gas	3000000 gas
transaction cost	21400 gas
execution cost	128 gas
hash	0xccca4e36cc65d4856f556ad56414848b82d6eccd9ea7a66dfa76e194042e555c
input	0x7fdcf8c4
decoded input	{}
decoded output	{}
logs	[]
value	0 wei

The instruction cost is the same as shown in Appendix G in the Yellowpaper, this is something I could verify. But I cannot verify where the 9 gas are coming from.

[tomw1808]

I found what causes the difference between the values shown and the values calculated: MSTORE is marked as "W_verylow" in the Yellowpaper (3 gas), shows 3 gas in the remix debugger, but takes up 12 gas on chain. I have released a similar contract on rinkeby, its's the third instruction and pretty obvious: http://rinkeby.etherscan.io/remix?txhash=0xb837a742727706e9b8f57f33477d3b5dfd3959f705f15f041c1ed6357178917b

I'd be very happy if someone can explain it to me before closing the issue. Thank you.

[yann300]

That's a bug in Remix:
The cost of MSTORE is 3 gas, but the general cost if the step can be higher if the memory needs to be extended.

[seandotau]

Thanks for this @yann300 . Could you elaborate under what situation would the memory need to be extended? Is it a linear increase as well? Just curious.

[yann300]

this is has been fixed for sometimes.
Sorry for not responding here.
you'll find the information here: https://github.com/wolflo/evm-opcodes/blob/main/gas.md#a0-1-memory-expansion
The following opcodes incur a memory expansion cost in addition to their otherwise static gas cost: RETURN, REVERT, MLOAD, MSTORE, MSTORE8.

[tomw1808]

@yann300 appreciate it. I already forgot about the issue, been 5 years. Again, really appreciate getting back to it and referencing more information here.