contracts: Upgrade to wasmi 0.28 by athei · Pull Request #13312 · paritytech/substrate

athei · 2023-02-04T20:15:24Z

Also upgrading to newest wasmparser to keep up with wasmi.

Performance uplift for instructions execution looks pretty good from wasmi 0.20. As we expected.

Please note that this is without wasm-opt as applying this to the runtime didn't improve performance as a whole (#13408). It also did remove all the benefits of this new wasmi-version. Hence we just merge without wasm-opt.

This is how this PR performed with wasmt-opt vs. no wasm-opt. this is with -O2but -O4 performed the same:

Cumulus companion: paritytech/cumulus#2350

frame/contracts/src/wasm/runtime.rs

frame/contracts/src/weights.rs

koute · 2023-02-06T06:30:27Z

Since this is enabling new optimizations in wasm-opt ideally I'd like to run a full burn-in before we consider merging this. Basically hack the WASM executor so that it dynamically runs wasm-opt on every runtime upgrade (so essentially apply it retroactively to every existing on-chain runtime) and see if everything successfully syncs.

I'll set it up on my GCP node.

koute · 2023-02-06T08:36:21Z

Burn-in is running; will probably take a few days to finish:

2023-02-06 08:32:26 ⚙️  Syncing 362.0 bps, target=#16514561 (40 peers), best: #57137 (0x4f91…6757), finalized #57127 (0x560b…089b), ⬇ 268.6kiB/s ⬆ 142.6kiB/s    
2023-02-06 08:32:31 ⚙️  Syncing 353.6 bps, target=#16514561 (40 peers), best: #58905 (0x8eea…203a), finalized #58882 (0x4aeb…beb0), ⬇ 508.6kiB/s ⬆ 156.3kiB/s    
2023-02-06 08:32:36 ⚙️  Syncing 150.6 bps, target=#16514561 (40 peers), best: #59658 (0xeb84…bc63), finalized #59470 (0x7e2c…dfa6), ⬇ 182.1kiB/s ⬆ 133.4kiB/s    
2023-02-06 08:32:41 ⚙️  Syncing  0.0 bps, target=#16514561 (40 peers), best: #59658 (0xeb84…bc63), finalized #59470 (0x7e2c…dfa6), ⬇ 149.0kiB/s ⬆ 150.1kiB/s    
2023-02-06 08:32:46 ⚙️  Syncing  0.0 bps, target=#16514561 (40 peers), best: #59658 (0xeb84…bc63), finalized #59470 (0x7e2c…dfa6), ⬇ 167.2kiB/s ⬆ 168.6kiB/s    
2023-02-06 08:32:51 ⚙️  Syncing  0.0 bps, target=#16514561 (40 peers), best: #59658 (0xeb84…bc63), finalized #59470 (0x7e2c…dfa6), ⬇ 216.7kiB/s ⬆ 218.9kiB/s    
2023-02-06 08:32:56 ⚙️  Syncing  0.0 bps, target=#16514561 (40 peers), best: #59658 (0xeb84…bc63), finalized #59470 (0x7e2c…dfa6), ⬇ 179.8kiB/s ⬆ 181.1kiB/s    
2023-02-06 08:32:59 Optimized WASM blob with opt level 4 in 26.30s: 1177008 -> 1075157    
2023-02-06 08:33:01 ⚙️  Syncing 72.0 bps, target=#16514561 (40 peers), best: #60018 (0xf001…d268), finalized #59904 (0x782a…05ff), ⬇ 189.0kiB/s ⬆ 166.3kiB/s    
2023-02-06 08:33:06 ⚙️  Syncing 371.2 bps, target=#16514561 (40 peers), best: #61874 (0x5851…951a), finalized #61772 (0xd671…ebc5), ⬇ 336.8kiB/s ⬆ 218.3kiB/s    
2023-02-06 08:33:11 ⚙️  Syncing 384.8 bps, target=#16514561 (40 peers), best: #63798 (0xb7e7…abd8), finalized #63489 (0xb1bd…2d3c), ⬇ 326.4kiB/s ⬆ 197.6kiB/s    
2023-02-06 08:33:16 ⚙️  Syncing 348.6 bps, target=#16514561 (40 peers), best: #65541 (0x8caa…515b), finalized #65536 (0x3db2…1579), ⬇ 513.4kiB/s ⬆ 161.9kiB/s    
2023-02-06 08:33:21 ⚙️  Syncing 391.2 bps, target=#16514561 (40 peers), best: #67497 (0x94b8…f05f), finalized #67072 (0x9d70…4481), ⬇ 369.5kiB/s ⬆ 165.9kiB/s

Robbepop · 2023-02-06T08:59:58Z

Since this is enabling new optimizations in wasm-opt ideally I'd like to run a full burn-in before we consider merging this. Basically hack the WASM executor so that it dynamically runs wasm-opt on every runtime upgrade (so essentially apply it retroactively to every existing on-chain runtime) and see if everything successfully syncs.

I'll set it up on my GCP node.

@koute Can you elaborate on why an update to wasmi enables new optimizations in wasm-opt?

koute · 2023-02-06T09:03:55Z

@koute Can you elaborate on why an update to wasmi enables new optimizations in wasm-opt?

I'm not entirely sure I understand the question. (:

Here's the patch from this PR which changes wasm-opt to actually optimize the runtime blobs we produce:

https://github.com/paritytech/substrate/pull/13312/files#diff-90c7859f5054a6b99320e4e6a8391a6e7d7135e993a40ce451a68c44c471e171R694

Robbepop · 2023-02-06T09:06:06Z

I didn't notice that this PR was also updating wasm-opt besides wasmi. Maybe it would have been better to do both in separate steps to have a better understanding of what exactly caused performance differences in the contracts-pallet if there are any.

koute · 2023-02-06T09:07:50Z

to have a better understanding of what exactly caused performance differences in the contracts-pallet if there are any.

Well, the last time we tried there was visible impact on performance when using wasm-opt, so I'm expecting that it's probably still there.

athei · 2023-02-06T11:29:19Z

I just added it to this PR because I suspected the degradation was from not using wasm-opt (wasmi is sensitive to that). But it was about the benchmarking machine I guess.

Thanks for doing the burn in. If everything goes well I propose that we will remove the change from this PR and put it into a standalone PR.

athei · 2023-02-07T13:21:03Z

I removed the wasm-opt changes from this PR. If your burn-in doesn't show any problems I will make a new PR with only the wasm-opt change and generate a new weight baseline in that PR. Then we can see the performance impact and can decide the next steps.

koute · 2023-02-07T13:55:13Z

I removed the wasm-opt changes from this PR. If your burn-in doesn't show any problems I will make a new PR with only the wasm-opt change and generate a new weight baseline in that PR. Then we can see the performance impact and can decide the next steps.

Sounds good to me. Current progress:

2023-02-07 13:53:29 ⚙️  Syncing 21.6 bps, target=#16514571 (40 peers), best: #7999611 (0x9b7f…8efd), finalized #7999488 (0x8b28…9874), ⬇ 387.1kiB/s ⬆ 186.0kiB/s

koute · 2023-02-13T07:09:40Z

@athei The burn-in finished successfully.

Robbepop · 2023-02-13T08:18:16Z

In the meantime I released version 0.26.0 featuring fuel metering. So maybe you could insta upgrade in this PR.

athei · 2023-02-13T16:37:49Z

Yeah will do that. There is a queue of PRs that need to go in before as soon as metering is fixed, though.

athei · 2023-03-18T11:48:31Z

Performance uplift without wasm-opt looks nice. Let's try with wasm-opt enabled.

athei · 2023-03-19T10:52:13Z

frame/contracts/src/wasm/prepare.rs

-		if matches!(determinism, Determinism::Deterministic) {
-			contract_module.ensure_no_floating_types()?;
-		}


We no longer need this because the newest wasmparser rejects floating types for us if floats: false. Before it only rejected float instructions.

wasmparser rejects floating types for us if floats: false

just curious, where do you define this setting?

Just a bit above. Argument to Validator::new_with_features.

athei · 2023-03-19T11:00:01Z

wasm-opt turned out to be regressing the performance of the runtime as a whole and of wasmi specifically (see PR description). Hence we just go ahead with only upgrading wasmi. I rebased the PR to remove all wasm-opt experiments. It is ready for review now.

pgherveou

lgtm

pgherveou · 2023-03-20T08:39:17Z

frame/contracts/src/wasm/prepare.rs

-		if matches!(determinism, Determinism::Deterministic) {
-			contract_module.ensure_no_floating_types()?;
-		}


wasmparser rejects floating types for us if floats: false

just curious, where do you define this setting?

pgherveou · 2023-03-20T08:41:06Z

frame/contracts/src/wasm/prepare.rs

 				(func (export "deploy"))
 			)
 			"#,
-			Err("use of floating point type in globals is forbidden")


don't we want to keep the more specific error message?

We can't. The function returns a &'static str. We opted for logging the error message to the console instead of returning it. Not optimal. The error handling there is a mess. But the whole thing will become much simpler once we have #13639. Then we can have a hard look on the error handling there.

frame/contracts/src/wasm/prepare.rs

agryaznov

LGTM

frame/contracts/src/wasm/prepare.rs

athei · 2023-03-20T17:40:52Z

I am doing the update to wasmi 0.29 in a follow up PR. I don't want to re-run benchmarks.

athei · 2023-03-20T18:38:02Z

Okay the rename needed a cumulus companion. Please also approve the trivial companion: paritytech/cumulus#2350

athei · 2023-03-20T23:09:13Z

bot merge

* Upgrade to wasmi 0.28 * ".git/.scripts/commands/bench/bench.sh" pallet dev pallet_contracts * Update stale comment * Renamed variants of `Determinism` * Compile fix --------- Co-authored-by: command-bot <>

Robbepop reviewed Feb 4, 2023

View reviewed changes

frame/contracts/src/wasm/runtime.rs Outdated Show resolved Hide resolved

athei requested a review from koute as a code owner February 5, 2023 03:21

athei commented Feb 5, 2023

View reviewed changes

frame/contracts/src/weights.rs Outdated Show resolved Hide resolved

athei mentioned this pull request Feb 17, 2023

Enable wasm-opt optimizations in wasm-builder #13408

Closed

athei changed the title ~~contracts: Upgrade to wasmi 0.25~~ contracts: Upgrade to wasmi 0.28 Mar 14, 2023

athei and others added 2 commits March 19, 2023 11:45

Upgrade to wasmi 0.28

ae8e48b

".git/.scripts/commands/bench/bench.sh" pallet dev pallet_contracts

c1ab31d

athei force-pushed the at/wasmi branch from 573b54b to c1ab31d Compare March 19, 2023 10:49

paritytech deleted a comment from mordamax Mar 19, 2023

athei commented Mar 19, 2023

View reviewed changes

athei requested review from Robbepop, agryaznov and pgherveou March 19, 2023 10:58

This was referenced Mar 19, 2023

Add linguist-generated to gitattributes #13628

Merged

contracts: Follow up for benchmark simplification #13635

Merged

contracts: Switch to wasmi's builtin metering #13639

Closed

pgherveou approved these changes Mar 20, 2023

View reviewed changes

Robbepop reviewed Mar 20, 2023

View reviewed changes

frame/contracts/src/wasm/prepare.rs Outdated Show resolved Hide resolved

agryaznov approved these changes Mar 20, 2023

View reviewed changes

frame/contracts/src/wasm/prepare.rs Outdated Show resolved Hide resolved

athei added 4 commits March 20, 2023 16:08

Merge branch 'master' into at/wasmi

b2a8c14

Update stale comment

e000381

Renamed variants of Determinism

3a31926

Compile fix

159e87c

Robbepop approved these changes Mar 20, 2023

View reviewed changes

athei mentioned this pull request Mar 20, 2023

Companion for substrate #13312: Rename Deterministic to Enforce paritytech/cumulus#2350

Merged

paritytech-processbot bot merged commit 8f0b0f6 into master Mar 20, 2023

paritytech-processbot bot deleted the at/wasmi branch March 20, 2023 23:09

kacperzuk-neti mentioned this pull request Jun 23, 2023

Polkadot v0.9.43 liberland/liberland_substrate#295

Merged

15 tasks

Conversation

athei commented Feb 4, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

koute commented Feb 6, 2023

Uh oh!

koute commented Feb 6, 2023

Uh oh!

Robbepop commented Feb 6, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

koute commented Feb 6, 2023

Uh oh!

Robbepop commented Feb 6, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

koute commented Feb 6, 2023

Uh oh!

athei commented Feb 6, 2023

Uh oh!

athei commented Feb 7, 2023

Uh oh!

koute commented Feb 7, 2023

Uh oh!

koute commented Feb 13, 2023

Uh oh!

Robbepop commented Feb 13, 2023

Uh oh!

athei commented Feb 13, 2023

Uh oh!

athei commented Mar 18, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

athei Mar 19, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pgherveou Mar 20, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

athei Mar 20, 2023

Choose a reason for hiding this comment

Uh oh!

athei commented Mar 19, 2023

Uh oh!

pgherveou left a comment

Choose a reason for hiding this comment

Uh oh!

pgherveou Mar 20, 2023 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

pgherveou Mar 20, 2023

Choose a reason for hiding this comment

Uh oh!

athei Mar 20, 2023

Choose a reason for hiding this comment

Uh oh!

Uh oh!

agryaznov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

athei commented Mar 20, 2023

Uh oh!

athei commented Mar 20, 2023

Uh oh!

athei commented Mar 20, 2023

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

athei commented Feb 4, 2023 •

edited

Loading

Robbepop commented Feb 6, 2023 •

edited

Loading

Robbepop commented Feb 6, 2023 •

edited

Loading

athei commented Mar 18, 2023 •

edited

Loading

athei Mar 19, 2023 •

edited

Loading

pgherveou Mar 20, 2023 •

edited

Loading

pgherveou Mar 20, 2023 •

edited

Loading