Reuse Existing Compiled Code from Equivalent Program if Exists #67

rauljordan · 2023-05-06T01:00:33Z

One important optimization that can lead to massive gas savings pointed out by @rachel-bousfield is reusing compiled code that already exists in the state DB when a user is trying to compile a program they deploy. If they have deployed the same keccak program, for example, at address 0xA than one that already exists at another address, we can short circuit the compilation and reuse the existing compiled code from the other program address.

Here's the difference in speed for keccak

Time to compile: 220.795153ms // Newly compiled
Time to compile: 23.957493ms // Reusing other program address' equal compiled code

This PR also depends on stylus-geth's PR here: OffchainLabs/stylus-geth#17

This PR changes functions associated with WASM programs and the state DB to deal with codehashes internally instead of purely program addresses. This allows us to reuse compiled programs. Changes are also added to the recording DB to ensure we can reuse code in there as well.

We include a comprehensive test that checks several invariants we expect from these changes. Here's how our test works:

We deploy the same keccak WASM code to two different program addresses
We attempt to call them and expect a failure as they have not been compiled
We then:

Make a multicall that: (a) tries to compile keccak A and succeeds, (b) revert right after
We then try to call keccak A, but expect it to fail due to it not being compiled
We then compile keccak A and try to call B

Here are some of the details of our changelog:

We change the machine versions mapping to operate based on codehash instead of program address. When we attempt to compile, this will succeed early if the same code has already been compiled in Stylus regardless of address
StateDB changes in stylus-geth are changed to use codehash based compiled wasm caching

Thanks to @rachel-bousfield for the fantastic guidance along the way.

…r compilation costs

rachel-bousfield

Nice! Just a couple of tiny things

arbos/programs/programs.go

rauljordan · 2023-06-27T20:57:07Z

system_tests/program_test.go

 	defer cleanup()

+	programAddress := deployWasm(t, ctx, auth, l2client, rustFile("fallible"))


Separating the deploy function from the setup so that I can test reusing existing compilation at two different program addresses

tsahee · 2023-08-18T23:55:03Z

system_tests/common_test.go

@@ -929,6 +929,7 @@ func deployContract(
 ) common.Address {
 	deploy := deployContractInitCode(code, false)
 	basefee := GetBaseFee(t, client, ctx)
+	basefee.Mul(basefee, big.NewInt(2))


note: I don't understand why it's necessary specifically for this change.

TestProgramMemory fails about 50% of the time while trying to deploy the grow-120 contract with an error that PricePerGas is lower than baseFee. This is specifically for this test and contract (the contract cannot compile - but I don't understand how that's relevant)

Increasing baseFee like that does the trick, but there might be better solutions.

instead of deploy params

rachel-bousfield · 2023-08-25T00:53:58Z

precompiles/ArbWasm.go

-func (con ArbWasm) ProgramVersion(c ctx, _ mech, program addr) (uint16, error) {
-	return c.State.Programs().ProgramVersion(program)
+func (con ArbWasm) ProgramVersion(c ctx, evm mech, program addr) (uint16, error) {
+	return c.State.Programs().ProgramVersion(evm.StateDB.GetCodeHash(program))


We should charge for this

rachel-bousfield · 2023-08-25T01:13:28Z

arbos/programs/programs.go

+		contractAddress: scope.Contract.Address(),
+		msgSender:       scope.Contract.Caller(),
+		msgValue:        common.BigToHash(scope.Contract.Value()),


contract is scope.Contract

rachel-bousfield · 2023-08-25T01:15:40Z

arbos/programs/programs.go

 func (p Programs) getProgram(contract *vm.Contract) (Program, error) {
-	address := contract.Address()
-	if contract.CodeAddr != nil {
-		address = *contract.CodeAddr
-	}
-	return p.deserializeProgram(address)
+
+	return p.deserializeProgram(contract.CodeHash)
 }


Can be collapsed

rachel-bousfield · 2023-08-25T01:19:08Z

system_tests/program_test.go

+func TestProgramKeccakArb(t *testing.T) {
+	keccakTest(t, false)
+}


This is duplicative of TestProgramArbitratorKeccak

rachel-bousfield · 2023-08-25T01:22:50Z

system_tests/program_test.go

+func TestProgramCompilationReuse(t *testing.T) {
+	t.Parallel()
+	testCompilationReuse(t, true)
+}


Analogue is missing in stylus_test.go

rachel-bousfield · 2023-08-25T01:25:14Z

system_tests/program_test.go

+func setupProgramTest(t *testing.T, jit bool) (
+	context.Context, *arbnode.Node, *BlockchainTestInfo, *ethclient.Client, bind.TransactOpts, func(),


Instead of 1 or 0, we should probs take a list of filenames for generality

rachel-bousfield · 2023-08-25T01:26:53Z

system_tests/program_test.go

+	arbOwner, err := precompilesgen.NewArbOwner(types.ArbOwnerAddress, l2client)
+	Require(t, err)
+
+	ensure(arbOwner.SetInkPrice(&auth, 1e2))


Was 1e2 intentional?

rachel-bousfield

Nice work! I've added a few commits that are worth taking a look at that address most of the feedback. I decided not to change programs.go to minimize merge conflicts with

Onchain compiled #95

rauljordan added 4 commits May 5, 2023 20:36

reuse compiled attempt

ce2141a

reuse compiled program code from another program successfully, cheape…

790557c

…r compilation costs

fix ref

d231f04

fix confs

ec01891

rauljordan marked this pull request as draft May 6, 2023 01:04

rauljordan marked this pull request as ready for review May 6, 2023 01:04

geth ref

3257e2f

rauljordan marked this pull request as draft May 6, 2023 01:06

rauljordan marked this pull request as ready for review May 6, 2023 01:07

rauljordan requested review from joshuacolvin0 and rachel-bousfield May 6, 2023 01:08

rauljordan added 3 commits May 5, 2023 22:03

added in programs

d51cfa5

edit behavior

679e178

update ref

5b7ac7c

rachel-bousfield suggested changes May 6, 2023

View reviewed changes

arbos/programs/programs.go Outdated Show resolved Hide resolved

arbos/programs/programs.go Show resolved Hide resolved

rauljordan added 7 commits May 5, 2023 22:35

fix up

3e73ee3

code review comments

25ebea1

rem

48310da

update geth ref

16f2730

fix up program test

7a2415f

update geth dep

872634d

keccak reuse test

49fae6a

rauljordan commented Jun 27, 2023

View reviewed changes

rauljordan added 7 commits June 27, 2023 18:00

fix build

0d90c4e

update yml

4006def

modify jit prover to understand codehash boundaries

ee6f17e

comments

1d23653

no fmt

c417dda

advanced tests

b20e0aa

failure

c4b2865

tsahee added 4 commits August 18, 2023 12:50

remove address from Program struct

ce4a4d9

program version is uint16

186f5a3

update geth: use u16 for wasm version

8190e52

system_tests: increase baseFee in deployContract

9a64a96

tsahee reviewed Aug 18, 2023

View reviewed changes

tsahee added 3 commits August 18, 2023 17:55

Merge remote-tracking branch 'origin/stylus' into reuse-compiled-code

dedb470

system_test: reduce ink price in testMemory

87f3673

instead of deploy params

fix retryableList

ea84382

rachel-bousfield suggested changes Aug 25, 2023

View reviewed changes

rachel-bousfield reviewed Aug 25, 2023

View reviewed changes

rachel-bousfield suggested changes Aug 25, 2023

View reviewed changes

rachel-bousfield reviewed Aug 25, 2023

View reviewed changes

rachel-bousfield removed the ready for review label Aug 26, 2023

rachel-bousfield added 4 commits August 27, 2023 14:48

Merge branch 'stylus' into reuse-compiled-code

ef75885

improve tests

1f2e671

pay for codehash

634d7c3

revert ink-price change

b296607

rachel-bousfield approved these changes Aug 27, 2023

View reviewed changes

repin geth

5c1eb0a

rachel-bousfield enabled auto-merge August 27, 2023 21:37

rachel-bousfield disabled auto-merge August 27, 2023 22:02

rachel-bousfield added 3 commits August 27, 2023 21:04

fix native_keccak pricing divergence

6ccb423

feature-gate benchmarks

8231c3d

reintroduce prover test for activation failures

7779316

rachel-bousfield enabled auto-merge August 28, 2023 03:13

rachel-bousfield merged commit 81e3ea0 into stylus Aug 28, 2023
7 checks passed

rachel-bousfield deleted the reuse-compiled-code branch August 28, 2023 03:38

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Reuse Existing Compiled Code from Equivalent Program if Exists #67

Reuse Existing Compiled Code from Equivalent Program if Exists #67

rauljordan commented May 6, 2023 •

edited

Loading

rachel-bousfield left a comment

rauljordan Jun 27, 2023

tsahee Aug 18, 2023

rachel-bousfield Aug 25, 2023

rachel-bousfield Aug 25, 2023

rachel-bousfield Aug 25, 2023

rachel-bousfield Aug 25, 2023

rachel-bousfield Aug 25, 2023

rachel-bousfield Aug 25, 2023

rachel-bousfield Aug 25, 2023

rachel-bousfield left a comment •

edited

Loading

		defer cleanup()

		programAddress := deployWasm(t, ctx, auth, l2client, rustFile("fallible"))

		func setupProgramTest(t *testing.T, jit bool) (
		context.Context, arbnode.Node, BlockchainTestInfo, *ethclient.Client, bind.TransactOpts, func(),

Reuse Existing Compiled Code from Equivalent Program if Exists #67

Reuse Existing Compiled Code from Equivalent Program if Exists #67

Conversation

rauljordan commented May 6, 2023 • edited Loading

rachel-bousfield left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

rachel-bousfield left a comment • edited Loading

Choose a reason for hiding this comment

rauljordan commented May 6, 2023 •

edited

Loading

rachel-bousfield left a comment •

edited

Loading