EIPs/EIPS/eip-145.md

95 lines
3.3 KiB
Markdown
Raw Normal View History

2017-02-13 16:05:22 +00:00
## Preamble
2017-04-23 13:55:19 +00:00
EIP: 145
2017-02-13 16:05:22 +00:00
Title: Bitwise shifting instructions in EVM
Author: Alex Beregszaszi, Paweł Bylica
Type: Standard Track
2017-04-23 14:43:43 +00:00
Category: Core
2017-02-13 16:05:22 +00:00
Status: Draft
Created: 2017-02-13
## Simple Summary
To provide native bitwise shifting with cost on par with other arithmetic operations.
## Abstract
2017-09-22 13:04:45 +00:00
Native bitwise shifting instructions are introduced, which are more efficient processing wise on the host and are cheaper to use by a contract.
2017-02-13 16:05:22 +00:00
## Motivation
EVM is lacking bitwise shifting operators, but supports other logical and arithmetic operators. Shift operations can be implemented via arithmetic operators, but that has a higher cost and requires more processing time from the host. Implementing `SHL` and `SHR` using arithmetics cost each 35 gas, while the proposed instructions take 3 gas.
## Specification
The following instructions are introduced:
### `0x1b`: `SHL` (shift left)
2017-09-22 15:07:34 +00:00
The `SHL` instruction (shift left) pops 2 values from the stack, first `arg1` and then `arg2`, and pushes on the stack `arg2` shifted to the left by `arg1` number of bits. The result is equal to
2017-02-13 16:05:22 +00:00
```
(arg2 * 2^arg1) mod 2^256
2017-02-13 16:05:22 +00:00
```
Notes:
- The value (`arg2`) is interpreted as an unsigned number.
- The shift amount (`arg1`) is interpreted as an unsigned number.
- If the shift amount (`arg1`) is greater or equal 256 the result is 0.
- This is equivalent to `PUSH1 2 EXP MUL`.
2017-02-13 16:05:22 +00:00
### `0x1c`: `SHR` (logical shift right)
2017-09-22 15:07:34 +00:00
The `SHR` instruction (logical shift right) pops 2 values from the stack, first `arg1` and then `arg2`, and pushes on the stack `arg2` shifted to the right by `arg1` number of bits with zero fill. The result is equal to
2017-02-13 16:05:22 +00:00
```
floor(arg2 / 2^arg1)
2017-02-13 16:05:22 +00:00
```
Notes:
- The value (`arg2`) is interpreted as an unsigned number.
- The shift amount (`arg1`) is interpreted as an unsigned number.
- If the shift amount (`arg1`) is greater or equal 256 the result is 0.
- This is equivalent to `PUSH1 2 EXP DIV`.
2017-02-13 16:05:22 +00:00
### `0x1d`: `SAR` (arithmetic shift right)
2017-09-22 15:07:34 +00:00
The `SAR` instruction (arithmetic shift right) pops 2 values from the stack, first `arg1` and then `arg2`, and pushes on the stack `arg2` shifted to the right by `arg1` number of bits with sign extension. The result is equal to
2017-02-13 16:05:22 +00:00
```
floor(arg2 / 2^arg1)
2017-02-13 16:05:22 +00:00
```
Notes:
- The value (`arg2`) is interpreted as a signed number.
- The shift amount (`arg1`) is interpreted as an unsigned number.
- If the shift amount (`arg1`) is greater or equal 256 the result is 0 if `arg2` is non-negative or -1 if `arg2` is negative.
- This is **not** equivalent to `PUSH1 2 EXP SDIV`, since it rounds differently. See `SDIV(-1, 2) == 0`, while `SAR(-1, 1) == -1`.
2017-02-13 16:05:22 +00:00
2017-04-23 10:28:25 +00:00
The cost of the shift instructions is set at `verylow` tier (3 gas).
2017-02-13 16:05:22 +00:00
## Rationale
Instruction operands were chosen to fit the more natural use case of shifting a value already on the stack. This means the operand order is swapped compared to most arithmetic insturctions.
2017-02-13 16:05:22 +00:00
## Backwards Compatibility
The newly introduced instructions have no effect on bytecode created in the past.
## Test Cases
TBA
## Implementation
Client support:
- cpp-ethereum: https://github.com/ethereum/cpp-ethereum/pull/4054
2017-02-13 16:05:22 +00:00
Compiler support:
2017-09-22 13:04:45 +00:00
- Solidity/LLL: https://github.com/ethereum/solidity/pull/2541
2017-02-13 16:05:22 +00:00
## Copyright
Copyright and related rights waived via [CC0](https://creativecommons.org/publicdomain/zero/1.0/).