v1 Opcodes
Ops have a 'cost' of 1 unless otherwise specified.
err¶
- Bytecode: 0x00
- Stack: ... → exits
- Fail immediately.
sha256¶
- Bytecode: 0x01
- Stack: ..., A: []byte → ..., [32]byte
- SHA256 hash of value A, yields [32]byte
- Cost: 7
keccak256¶
- Bytecode: 0x02
- Stack: ..., A: []byte → ..., [32]byte
- Keccak256 hash of value A, yields [32]byte
- Cost: 26
sha512_256¶
- Bytecode: 0x03
- Stack: ..., A: []byte → ..., [32]byte
- SHA512_256 hash of value A, yields [32]byte
- Cost: 9
ed25519verify¶
- Bytecode: 0x04
- Stack: ..., A: []byte, B: [64]byte, C: [32]byte → ..., bool
- for (data A, signature B, pubkey C) verify the signature of ("ProgData" || program_hash || data) against the pubkey => {0 or 1}
- Cost: 1900
- Mode: Signature
The 32 byte public key is the last element on the stack, preceded by the 64 byte signature at the second-to-last element on the stack, preceded by the data which was signed at the third-to-last element on the stack.
+¶
- Bytecode: 0x08
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A plus B. Fail on overflow.
Overflow is an error condition which halts execution and fails the transaction. Full precision is available from addw
.
-¶
- Bytecode: 0x09
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A minus B. Fail if B > A.
/¶
- Bytecode: 0x0a
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A divided by B (truncated division). Fail if B == 0.
divmodw
is available to divide the two-element values produced by mulw
and addw
.
*¶
- Bytecode: 0x0b
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A times B. Fail on overflow.
Overflow is an error condition which halts execution and fails the transaction. Full precision is available from mulw
.
<¶
- Bytecode: 0x0c
- Stack: ..., A: uint64, B: uint64 → ..., bool
- A less than B => {0 or 1}
>¶
- Bytecode: 0x0d
- Stack: ..., A: uint64, B: uint64 → ..., bool
- A greater than B => {0 or 1}
<=¶
- Bytecode: 0x0e
- Stack: ..., A: uint64, B: uint64 → ..., bool
- A less than or equal to B => {0 or 1}
>=¶
- Bytecode: 0x0f
- Stack: ..., A: uint64, B: uint64 → ..., bool
- A greater than or equal to B => {0 or 1}
&&¶
- Bytecode: 0x10
- Stack: ..., A: uint64, B: uint64 → ..., bool
- A is not zero and B is not zero => {0 or 1}
||¶
- Bytecode: 0x11
- Stack: ..., A: uint64, B: uint64 → ..., bool
- A is not zero or B is not zero => {0 or 1}
==¶
- Bytecode: 0x12
- Stack: ..., A, B → ..., bool
- A is equal to B => {0 or 1}
!=¶
- Bytecode: 0x13
- Stack: ..., A, B → ..., bool
- A is not equal to B => {0 or 1}
!¶
- Bytecode: 0x14
- Stack: ..., A: uint64 → ..., uint64
- A == 0 yields 1; else 0
len¶
- Bytecode: 0x15
- Stack: ..., A: []byte → ..., uint64
- yields length of byte value A
itob¶
- Bytecode: 0x16
- Stack: ..., A: uint64 → ..., [8]byte
- converts uint64 A to big-endian byte array, always of length 8
btoi¶
- Bytecode: 0x17
- Stack: ..., A: []byte → ..., uint64
- converts big-endian byte array A to uint64. Fails if len(A) > 8. Padded by leading 0s if len(A) < 8.
btoi
fails if the input is longer than 8 bytes.
%¶
- Bytecode: 0x18
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A modulo B. Fail if B == 0.
|¶
- Bytecode: 0x19
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A bitwise-or B
&¶
- Bytecode: 0x1a
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A bitwise-and B
^¶
- Bytecode: 0x1b
- Stack: ..., A: uint64, B: uint64 → ..., uint64
- A bitwise-xor B
~¶
- Bytecode: 0x1c
- Stack: ..., A: uint64 → ..., uint64
- bitwise invert value A
mulw¶
- Bytecode: 0x1d
- Stack: ..., A: uint64, B: uint64 → ..., X: uint64, Y: uint64
- A times B as a 128-bit result in two uint64s. X is the high 64 bits, Y is the low
intcblock¶
- Syntax:
intcblock UINT ...
where UINT ...: a block of int constant values - Bytecode: 0x20 {varuint count, [varuint ...]}
- Stack: ... → ...
- prepare block of uint64 constants for use by intc
intcblock
loads following program bytes into an array of integer constants in the evaluator. These integer constants can be referred to by intc
and intc_*
which will push the value onto the stack. Subsequent calls to intcblock
reset and replace the integer constants available to the script.
intc¶
- Syntax:
intc I
where I: an index in the intcblock - Bytecode: 0x21 {uint8}
- Stack: ... → ..., uint64
- Ith constant from intcblock
intc_0¶
- Bytecode: 0x22
- Stack: ... → ..., uint64
- constant 0 from intcblock
intc_1¶
- Bytecode: 0x23
- Stack: ... → ..., uint64
- constant 1 from intcblock
intc_2¶
- Bytecode: 0x24
- Stack: ... → ..., uint64
- constant 2 from intcblock
intc_3¶
- Bytecode: 0x25
- Stack: ... → ..., uint64
- constant 3 from intcblock
bytecblock¶
- Syntax:
bytecblock BYTES ...
where BYTES ...: a block of byte constant values - Bytecode: 0x26 {varuint count, [varuint length, bytes ...]}
- Stack: ... → ...
- prepare block of byte-array constants for use by bytec
bytecblock
loads the following program bytes into an array of byte-array constants in the evaluator. These constants can be referred to by bytec
and bytec_*
which will push the value onto the stack. Subsequent calls to bytecblock
reset and replace the bytes constants available to the script.
bytec¶
- Syntax:
bytec I
where I: an index in the bytecblock - Bytecode: 0x27 {uint8}
- Stack: ... → ..., []byte
- Ith constant from bytecblock
bytec_0¶
- Bytecode: 0x28
- Stack: ... → ..., []byte
- constant 0 from bytecblock
bytec_1¶
- Bytecode: 0x29
- Stack: ... → ..., []byte
- constant 1 from bytecblock
bytec_2¶
- Bytecode: 0x2a
- Stack: ... → ..., []byte
- constant 2 from bytecblock
bytec_3¶
- Bytecode: 0x2b
- Stack: ... → ..., []byte
- constant 3 from bytecblock
arg¶
- Syntax:
arg N
where N: an arg index - Bytecode: 0x2c {uint8}
- Stack: ... → ..., []byte
- Nth LogicSig argument
- Mode: Signature
arg_0¶
- Bytecode: 0x2d
- Stack: ... → ..., []byte
- LogicSig argument 0
- Mode: Signature
arg_1¶
- Bytecode: 0x2e
- Stack: ... → ..., []byte
- LogicSig argument 1
- Mode: Signature
arg_2¶
- Bytecode: 0x2f
- Stack: ... → ..., []byte
- LogicSig argument 2
- Mode: Signature
arg_3¶
- Bytecode: 0x30
- Stack: ... → ..., []byte
- LogicSig argument 3
- Mode: Signature
txn¶
- Syntax:
txn F
where F: txn - Bytecode: 0x31 {uint8}
- Stack: ... → ..., any
- field F of current transaction
txn¶
Fields (see transaction reference)
Index | Name | Type | Notes |
---|---|---|---|
0 | Sender | address | 32 byte address |
1 | Fee | uint64 | microalgos |
2 | FirstValid | uint64 | round number |
4 | LastValid | uint64 | round number |
5 | Note | []byte | Any data up to 1024 bytes |
6 | Lease | [32]byte | 32 byte lease value |
7 | Receiver | address | 32 byte address |
8 | Amount | uint64 | microalgos |
9 | CloseRemainderTo | address | 32 byte address |
10 | VotePK | [32]byte | 32 byte address |
11 | SelectionPK | [32]byte | 32 byte address |
12 | VoteFirst | uint64 | The first round that the participation key is valid. |
13 | VoteLast | uint64 | The last round that the participation key is valid. |
14 | VoteKeyDilution | uint64 | Dilution for the 2-level participation key |
15 | Type | []byte | Transaction type as bytes |
16 | TypeEnum | uint64 | Transaction type as integer |
17 | XferAsset | uint64 | Asset ID |
18 | AssetAmount | uint64 | value in Asset's units |
19 | AssetSender | address | 32 byte address. Source of assets if Sender is the Asset's Clawback address. |
20 | AssetReceiver | address | 32 byte address |
21 | AssetCloseTo | address | 32 byte address |
22 | GroupIndex | uint64 | Position of this transaction within an atomic transaction group. A stand-alone transaction is implicitly element 0 in a group of 1 |
23 | TxID | [32]byte | The computed ID for this transaction. 32 bytes. |
global¶
- Syntax:
global F
where F: global - Bytecode: 0x32 {uint8}
- Stack: ... → ..., any
- global field F
global¶
Fields
Index | Name | Type | Notes |
---|---|---|---|
0 | MinTxnFee | uint64 | microalgos |
1 | MinBalance | uint64 | microalgos |
2 | MaxTxnLife | uint64 | rounds |
3 | ZeroAddress | address | 32 byte address of all zero bytes |
4 | GroupSize | uint64 | Number of transactions in this atomic transaction group. At least 1 |
gtxn¶
- Syntax:
gtxn T F
where T: transaction group index, F: txn - Bytecode: 0x33 {uint8}, {uint8}
- Stack: ... → ..., any
- field F of the Tth transaction in the current group
for notes on transaction fields available, see txn
. If this transaction is i in the group, gtxn i field
is equivalent to txn field
.
load¶
- Syntax:
load I
where I: position in scratch space to load from - Bytecode: 0x34 {uint8}
- Stack: ... → ..., any
- Ith scratch space value. All scratch spaces are 0 at program start.
store¶
- Syntax:
store I
where I: position in scratch space to store to - Bytecode: 0x35 {uint8}
- Stack: ..., A → ...
- store A to the Ith scratch space
bnz¶
- Syntax:
bnz TARGET
where TARGET: branch offset - Bytecode: 0x40 {int16 (big-endian)}
- Stack: ..., A: uint64 → ...
- branch to TARGET if value A is not zero
The bnz
instruction opcode 0x40 is followed by two immediate data bytes which are a high byte first and low byte second which together form a 16 bit offset which the instruction may branch to. For a bnz instruction at pc
, if the last element of the stack is not zero then branch to instruction at pc + 3 + N
, else proceed to next instruction at pc + 3
. Branch targets must be aligned instructions. (e.g. Branching to the second byte of a 2 byte op will be rejected.) Starting at v4, the offset is treated as a signed 16 bit integer allowing for backward branches and looping. In prior version (v1 to v3), branch offsets are limited to forward branches only, 0-0x7fff.
At v2 it became allowed to branch to the end of the program exactly after the last instruction: bnz to byte N (with 0-indexing) was illegal for a TEAL program with N bytes before v2, and is legal after it. This change eliminates the need for a last instruction of no-op as a branch target at the end. (Branching beyond the end--in other words, to a byte larger than N--is still illegal and will cause the program to fail.)
pop¶
- Bytecode: 0x48
- Stack: ..., A → ...
- discard A
dup¶
- Bytecode: 0x49
- Stack: ..., A → ..., A, A
- duplicate A