Add SWAP17-SWAP24, DUP17 - DUP24, SWAPN, DUPN, and EXCHANGE
instructions. The arbitrary depth operations must be preceded by PUSH1 instructions defining operands.
1 Like
The argument
Xmust be provided by aPUSH2operation immediately preceding theEXCHANGEinstruction
This argument must be provided by a
PUSH1operation immediately preceding theSWAPNandDUPNinstructions.
Why not just read from the stack and exceptional-halt if the param is too large?
Why not just read from the stack and exceptional-halt if the param is too large?
You do read them from the stack, but this simplifies static analysis i) by ensuring the operand is available directly before the instruction and ii) avoiding dynamic stack manipulation by using values from calldata or storage as the operand.
We really need this more than the entire EOF. Even though the Solidity compiler might have fixed its “stack-too-deep” errors by putting local variables into memory.
Why burn 17 opcodes for the fixed values and add 2 dynamic ones for same gas that do the same thing?
Also the 4 dynamic opcodes have 34 byte code look backs so 3 gas seems very low?
I think we have ample opcodes left and stack-too-deep is a common issue. The static N versions relieve some pressure efficiently while the dynamic N versions support much deeper stack manipulation at a slightly higher cost.
Yeah 3 has could be too low, will have to see how expensive the look back ends up being. Could also consider doing look after.
@benaadams btw I added a note in the EIP, but I remembered why the dynamic opcodes are actually not that expensive. You just need to look at pc-2 and see if it is a valid jump destination. If so, we know it executed since there couldn’t be any other operation in between. Another way to do this is just store an interpreter flag is_last_op_push1 and only set it when a PUSH1 is executed and unset after.