fluidex / circuits Goto Github PK

plonkit analyse -c ./testdata/Block_2_2_2_2/circuit.r1cs 

before reuse https://github.com/Fluidex/circuits/tree/af124e57ca6f4c259999b83a18884e24124addb3

"num_constraints": 78813,
"num_gates": 265703,

after reuse https://github.com/Fluidex/circuits/tree/3c0cdae0bbccf4593a89f07c6f127172aa1a48f7

"num_constraints": 52555,
 "num_gates": 161903,

use camelCase

#41 (comment)

Maybe we wrote something badly? Rescue is not faster inside circuit.

On the other hand, I remember poseidon is faster when doing offline cpu computing...

Rescue needs pow5/invpow5 on field, which is slower than simple multiply on field

ref:

https://github.com/arnaucube/poseidon-rs

https://github.com/Fluidex/rescue-hash-js/tree/master/tools/rescue_wasm

https://github.com/Fluidex/state_keeper/blob/master/src/state/types.rs

By sha256 all the tx data

// do it later

in fact this repo contains following parts:

1. circuits itself
2. some scripts that generate input/output test data for circuits. like class TestPlaceOrder implements SimpleTest
3. the tester, return true/false with input / output / circuit
4. the global state manager, which is core of the future rollup module

this function will be auto run when the wasm module is loaded?

like https://github.com/Fluidex/rescue-hash-js/blob/b06d3ca4bd0aea13916c806d4240d5622bd81a4c/tools/rescue_wasm/src/lib.rs#L16

When testing block.circom in #31

[CompileError: WebAssembly.compile(): data segments count of 104626 exceeds internal limit of 100000 @+650474]

It seems that there are too data segments being used

helpers will be used in both this repo's tests and Fluidex's frontend. We should use a "commonjs" repo in the future.

gates profile result of Block(4, 2, 2):

main.processTransfer.newSenderHash.hash.mimcs.invpow.pow5 1716.28
main.processTransfer.oldReceiverHash.hash.mimcs.invpow.pow5 1716.28
main.processTransfer.newReceiverHash.hash.mimcs.invpow.pow5 1716.28
main.processWithdraw.oldAccountHash.hash.mimcs.invpow.pow5 1716.28
main.processWithdraw.newAccountHash.hash.mimcs.invpow.pow5 1716.28
main.processDepositToNew.newAccountHash.hash.mimcs.invpow.pow5 1716.57
main.processTransfer.sigVerifier.mulFix.segments.adders 2354.35
main.processWithdraw.sigVerifier.mulFix.segments.adders 2354.35
main.processTransfer.sigVerifier.hash.mimcs.pow 2415.89
main.processWithdraw.sigVerifier.hash.mimcs.pow 2415.89
main.processTransfer.sigVerifier.hash.mimcs.invpow.pow5 2589.78
main.processWithdraw.sigVerifier.hash.mimcs.invpow.pow5 2589.78
main.processTransfer.sigVerifier.mulAny.segments.bits.selector 3870.84
main.processWithdraw.sigVerifier.mulAny.segments.bits.selector 3870.84
main.processTransfer.sigVerifier.mulFix.segments.windows.mux 4332.03
main.processWithdraw.sigVerifier.mulFix.segments.windows.mux 4332.03
main.processTransfer.sigVerifier.mulAny.segments.bits 5329.30
main.processWithdraw.sigVerifier.mulAny.segments.bits 5329.30
main.processTransfer.sigVerifier.mulAny.segments.bits.doubler 5440.06
main.processWithdraw.sigVerifier.mulAny.segments.bits.doubler 5440.06
main.processTransfer.sigVerifier.mulAny.segments.bits.adder 5616.06
main.processWithdraw.sigVerifier.mulAny.segments.bits.adder 5616.06
one 34781.95

total_gates 274159.00

TODO: check whether the gate cost of sigVerifier is reasonable.

wasm tester will be helper for debugging

Add test/circuits & test/global_state to requitred status checks.

We may need 1 more weeks until github recognize them...

The meaning of 'data availability' is to enable proving account has balance of token in the tree. So order details are not needed.

SpotTrade:

AccountID1 4bytes
AccountID2 4bytes
tokenID_1to2 2bytes
tokenID_2to1 2bytes
Amount1to2 3bytes
Amount2to1 3bytes
// NewAccount1OrderRoot 
// NewAccount2OrderRoot
// although hash root is only 1 field element, it consumes many bytes in fact. So we prefer use original data
Order1Pos 2bytes
Order1ID 4bytes
Order1BuyAmount 3bytes
Order1SellAmount 3bytes
// Order1SellFilled 
// Order1BuyFilled
.. and for order2 ..

So total 42 bytes.

DepositToOld:

AccountID1 4bytes
tokenID1 2bytes
Amount1 3bytes

DepositToNew

AccountID1 4bytes
tokenID1 2bytes
Amount1 3bytes
ethAddr: 20bytes
sign 1bit
ay field: 32 bytes

total: 61bytes

Transfer

AccountID1 4bytes
AccountID2 4bytes
tokenID2bytes
Amount 3bytes

Withdraw

AccountID1 4bytes
tokenID1 2bytes
Amount1 3bytes

eth_address is handled in contract level

ERROR:  6 Signal assigned twice 0 0 0 0
Error: Signal assigned twice 0 0 0 0
    at error (/home/chris/gitfiles/Fluidex/circuits/node_modules/circom_runtime/js/witness_calculator.js:51:23)
    at setSignal (wasm://wasm/000f95ca:0:27522)
    at null.<anonymous> (wasm://wasm/000f95ca:0:28963)
    at null.<anonymous> (wasm://wasm/000f95ca:0:26916)
    at init (wasm://wasm/000f95ca:0:27116)
    at WitnessCalculator._doCalculateWitness (/home/chris/gitfiles/Fluidex/circuits/node_modules/circom_runtime/js/witness_calculator.js:130:31)
    at WitnessCalculator.calculateWitness (/home/chris/gitfiles/Fluidex/circuits/node_modules/circom_runtime/js/witness_calculator.js:158:20)
    at WasmTester.calculateWitness (/home/chris/gitfiles/Fluidex/circuits/node_modules/circom/ports/wasm/tester.js:57:45)
    at Object.testWithInputOutput (/home/chris/gitfiles/Fluidex/circuits/test/tester/wasm.ts:31:33)
    at main (/home/chris/gitfiles/Fluidex/circuits/test/test_all.ts:23:5)

c_tester works good

Choose a circuit ( either spotTrade or Block including spotTrade ), get a table with following schema:

Both one core and multi core ( 4 core or 8 core ) ( or real time / user time ) results are useful :

plonkit should be upgraded to bellman beta branch ( If fully upgrading including solidity stuff costs much time, we can create a branch rather than push unfinished code to plonkit master )

The bench script code may be executed later so it is needed to be commited into the repo. ( We don't need a full auto script. Some manual operations are ok )

https://github.com/Fluidex/circuits/blob/master/tools/replace_hash.sh may be useful.

rapidsnark need a zkey, we may have to generate it.

we should encode amount to Scientific notation (Mantisa & Exponent) for compression

https://github.com/Fluidex/circuits/runs/1892280300

fails but it doesn't report error correctly.

Seems we can fix this following https://github.com/Fluidex/dingir-exchange/blob/94b673330395ef655f933728e025ca4e55ded0df/examples/js/print_orders.js#L35-L39

iden3/circomlib@cf853c1

1. Default nodejs core dump for large circuirt(2**24). We have to pass some cmd flags or even patch and build nodejs ourselves.
2. Its very slow.

TODO: create zkey from zkutil output

https://github.com/Fluidex/circuits/blob/c28a169e20891320552e98a33d6d1f17ff833d10/src/lib/hash_state.circom#L96

add a template CalculateRootFromDefaultLeaf(n_levels, leaf)

decouple order id and merkle tree idx

Order id is like nonce, always incremental. User signed this id.

Meanwhile trade circuit uses idx of order merkle tree to fetch the order info then check trade price/amount against order info.

They are not same thing. After decoupling these two items, we can implement this #104

rewrite the order state transition login

These are the only valid transition methods:

1. from empty slot to a new order
2. from an old slot to a new order. Order with bigger id is newer.

The above two can be combined if we assume order_id:0 means empty order, real order starts with order id 1.

how to reproduce

when testing TestBlock or TestEmptyBlock

seems we may have infinite recursive loop

also depends on #35

1. compiled circuit
2. nlohmann/json (#90)
3. node_modules

Our recent goal is to run a 300tps rollup test on testnet ( for 1 hour? ), including matching engine, witness generation, proving, contract verification.

here I list some must-to-do for this goal:

1. design bitwise tx encoding mechanism to save space and gas
2. sha the inputs and make all other circuit input private. Only the hash in public input.
3. rewrite global state manager in rust.
4. implement parallel merkle tree update
5. we may even need to use poseidon ( seems faster than rescue for offline cpu calculation?)

Better to do:

1. Reuse order tree slot
2. Encode big int, signficand + exponent

Depedency of other repo:

1. prover cluster
2. matchengine can generate rollup event

For all things above, we can support 'SpotTrade' first. Implement 'Withdraw/DepositToNew' etc later.

if we want to fit current spot_trade circuit into block circuit, block circuit will be too complicated and tedious

We have different types of op DepositToNew DepositToOld Transfer etc.

As we all know the if else in circuits is not same to if else in common programming languages where either if branch or else branch will run. if else in circuits run both branch. It will take more resources ( constraints/gates here ).

So there may be a method to split the op by type then combine them using recursion... More research is needed here.

atm we use exsiting path_elements for "whatever" path_elements
if we can generate empty path_elements, it will benefit nop tx

https://github.com/Fluidex/circuits/runs/2431804240

Seems processTransfer.sigVerifier has some problems...