Ethereum Cat Herders PEEPanEIP, 4 Apr 2022

Previous Meeting Next Meeting

⏯

youtube image

►

From YouTube: PEEPanEIP#66: EIP-3855: PUSH0 instruction with Alex Beregszaszi, Hugo De la cruz, Paweł Bylica

Description

Topics covered
-----------------------------
1:31 - Guests introduction
4:09 - Presentation
4:28 - Introduction
5:21 - Abstract
5:46 - Motivation
8:12 - Analysis & Findings
9:23 - General opcode usage chart
9:40 - PUSHn Usage
9:56 - PUSH 1 00
10:20 - Other commonly used constants
10:46 - Commonly used constants
11:39 - Inc/Dec
15:36 - Questions
16:04 - Why Push 0? Do you think there could be an advantage using a different mnemonic?
18:04 - Motivation - Can we get rid of EIP-2733 because it is Withdrawn
19:23 - Are you aware of any active implementation in the client?

Resources:
-----------------
Slides - https://notes.ethereum.org/@ipsilon/peepaneip-3855
EIP - https://eips.ethereum.org/EIPS/eip-3855
Discussion to - https://ethereum-magicians.org/t/eip-3855-push0-instruction/7014

Follow at Twitter
-----------------------------------------------
Alex Beregszaszi @alexberegszaszi | Hugo De la cruz @hugo_dc | Paweł Bylica @chfast | Pooja Ranjan @poojaranjan19

PEEPanEIP playlist - https://www.youtube.com/playlist?list=PL4cwHXAawZxqu0PKKyMzG_3BJV_xZTi1F

Shanghai playlist - https://www.youtube.com/playlist?list=PL4cwHXAawZxpok0smGmq-dFGVHQzW84a2

Check out upcoming EIPs in Peep an EIP series at https://github.com/ethereum-cat-herders/PM/projects/2

Contact Ethereum Cat Herders
---------------------------------------------------
Discord: https://discord.io/EthCatHerders
Twitter: https://twitter.com/EthCatHerders
Medium: https://medium.com/ethereum-cat-herders
GitHub: https://github.com/ethereum-cat-herders
Website: https://www.ethereumcatherders.com/

A

Hello and welcome to another episode of p penney pen, educational video series, where we discuss an ethereum improvement proposal, upcoming upgrades and the basics of ethereum blockchain, I'm puja ranjan from the ethereum cat herders and today we are going to discuss eip3855 push zero instruction with the co-authors, alex barrexazi, hugo de la cruz and paul balika.

A

A quick intro to the proposal. Eip3855 is a standard track. Core eip, which was initially proposed in february 2021 and currently, is in cfi. Consider for inclusion for ethereum shanghai upgrade for people who don't know cfi. It simply means there is a rough consensus by the client developers to implement the proposal and prepare it to be deployed with the following network. Upgrade we'll learn about what changes are brought by eip3855 in a while. First, let me welcome our guest today.

A

Thank you for joining us, alex hugo and paul.

A

May we begin with a quick introduction before we get to the proposal.

A

Anyone can go first.

B

um Okay, I I gotta go for some um yeah. My name is alex and I've been working um on ethereum for for several years now, um both on the solute language, as well as um with this team on various improvements to the protocol itself. um Initially, the team has been called ewasm, but last year we have rebranded to epsilon, um because the work we have been doing has been uh baby on what uh it was was about and for the past two years we have done actually not two.

B

At this point, it's like two and a half. uh We have done a lot of research on um the execution layer, um um more specifically on on evm improvements, um but we also, we not only create proposals on from our end, but we also do review other proposals made for the epm hugo. You want to go next.

C

Hi everyone, my name, is la cruz. I joined I'm part of the h9 team and I joined in on september 2017 when, when he wasn't started and currently mainly working on on.

C

Basically, for some kind of analysis regarding the proposal from the team on also from any other proposal,.

D

Hi my name is pavel. I started working on evm very early back in 2014 and since then the journey mostly follows what what alex explained. So I was working on evm in early days, uh then we had a period of uh leprosy. Web assembly excitement and we are back to evm right now um so yeah. My work is mostly maintaining c implementation of evm and proposing and reviewing other proposals for improvements.

A

Well, thank you so much for introduction. um I know people are looking forward to learning this proposal. So, let's just peep in okay.

C

Okay, hello, everyone. My name is.

C

I'm going to talk about either a 55 the push your instructions which is currently on review status and considered for inclusion in the shanghai network operate.

C

So the value 0 is a very frequently used value, especially by instructions expecting offset as input.

C

Here we have some examples of code dealing with officers and we have, for example, call that a copy call copy, et cetera. We have many of codes dealing with objects and in another common usage is when calling call of code.

C

The contract might prefer setting the return data parameters to zero and later make use of return data size so and we don't get a copy to receive the results of the previous call.

C

So this eip and a65 introduces a new op code for zero with a cost of two guards.

C

It is initially used as instruction number five f, which is continuous to the predicting push inspection.

C

One of the motivations of this eip is to reduce the gas usage and the contract size, for example. Currently, if we have, if we use the push one instruction to push the value of zero, we need two bytes uh costing 400 cars at deploy, time and g and c gas at runtime.

C

If we use this push serial instruction, it will already it would only require one byte costing 200 cash a deploy time and two guys when executing the instructions.

C

Another important motivation for this eip is to reduce the risk of contracts uh making use of other instructions in order to to achieve this, uh pushing the values you on the stack and maybe optimizing for that- and here's a list of some instructions that a contract might might use in order to achieve that.

C

But the main risk here is that these instructions might change in the future in the future in a way that is now guaranteed. That always is going to be expected. Zero value.

C

One example of this is eip2733.

C

Which, if implemented, would have changed the behavior of returned data size in a way that it is not. It is no longer ensured that the that this will return the zero value at the beginning of the call. So this, if this change of of any other change in the future in the future change any of the of the of the obstacles that we we see here.

C

It will potentially create existing contracts, making use of its functions. To forget the optimization when pushing the binding seal.

C

And for this erp we also did some analysis on of code usage and especially for the push instructions.

C

And we trace transactions on two different nodes uh that we're seeking at different at different block numbers. For example, one of the nodes was a vlog for thinking about millions when we were racing and the other node was sending a vlog first million and we we got um a histogram.

C

We collected our histogram with all the up-code usage, including the push the push instruction, and we also have a special column for when we found when we found um which one pushing the co value. Specific value of zero.

C

From this analysis, we found that push 1 and push 2 are the the most common, the most common use of codes from all of all.

C

And among, among the specific push up codes, we also have that push 1 and push 2 are the most common one.

C

And from which, of course, you search eleven percent of all the push-up code, 11.39 percent of all the push up codes are used to push the value zero to the stack.

C

Yeah, this is a specific for for the value zero, but we also need some analysis on other commonly used constants. For example, the.

C

0 1 2 and a three one and three two and we see that q is the most common one. We have one two and three two are also very commonly used in this case for the other, for the commonly used contacts, for example, for for the constant hero. This is already caused by this ic by adding the virtual instruction for the constant one it is commonly used in loops. This could be solved by adding the increment and decrement of code, which alex is going to talk about in the next slide and for the value 2.

C

It is commonly used by all contracts in order to emulate shifts, which is already solved by existing chip operators and but for values, 31 and 32. It is more commonly used by the solidity avm.

C

Now alex is going to talk about the incoming and determined structure.

B

Yeah, maybe a slight clarification um is that push 0, as explained by hugo. In the uh you know, the first half of the slides. um There is a certain number of motivation. Why put zero is is useful um because it definitely reduces the the cases where other instructions are kind of misused. In optimization cases- and um you know, given the design of the you, don't need to go back given the design of the evm.

B

These offsets are really common in every single contract, um but on the ethereum magicians. If you wanna go back to the final slide on the interior magicians um discussion thread uh when procedure was proposed, there was a question. You know how about that are constants, and that's why we have conducted this analysis uh looking at what others are common and we have found that potentially, the the constant one could have a improvement, because it's used in for loops for most of the time and what happens in for loops?

B

Is you need to increment the counter and then check whether you reach the desired end? But you could also go to go backwards and you know if there would be some appetite for it. One could consider introducing to new instructions on top of like to zero called income deck for increment and then decrement.

B

This could provide also like a slight improvement, but I would say this is more on the side of only improving gas-wise and it doesn't leave any of the other motivations. We have put zero. um So, as I stated here, the the benefit is just a gas reduction, but the the downside is that maybe this would move the evm too much in towards to a sisk instruction set, uh which is which is an instruction set, which has a lot of instructions for my new details and so far ethereum has an evm has avoided this.

B

Lastly, I wanted to mention a a fun fact is, um even though contract byte code once in the state can be deduplicated.

B

So if you have, for example, um proxy contracts which which have the same code- um and you have like a a hundred thousand of them with the same code likely, it is stored as a single, uh singular code on this on the disk of the node. So if you have like a 10 byte proxy contract, that's going to only take really just uh you know it's not it's not going to take hundred thousand times 10 bytes.

B

However, when you do actually deploy it, you have to pay for that cost every single time. um So we did like another calculation and checked all the contracts.

B

How many bytes are wasted on pushing uh the the number zero with the push one instruction, um and we found that uh at the time this was a year ago. At the time there have been 340 megabytes wasted on this, uh which equates to 68 billion gas wasted on deploying these extra bytes. um By wasted, I mean if there would have been put zero by um you know from the beginning. um This guys wouldn't have spent on on this. It's just a fun fact. um I think that's it from our side.

B

If there are any questions.

A

Thank you so much for sharing this presentation and the detailed analysis, charts and um charts for opcode usage. That was great and I think I saw it for the first time. So it was quite fascinating and knowing about how much gas it can probably save on contract is nice to know.

A

I do have a couple of questions so that we came up from the proposal discussion on fem and they could have issues. So my first question is related to the mnemonic. It's not a technical question, but uh definitely would like to understand the thought process bank. So a a fellowship of ethereum, magician user wrote love the idea, but maybe we should use a different mnemonic like I push hero for immediate and um that way we will have a room to grow.

A

So I'm wondering what is the thought behind using push zero and do you think there could be an advantage of using different mnemonics.

B

It's hard to say, uh and nobody has a really proposed anything uh super useful, um at least maybe in my opinion, but there are two reasons why we have gone with zero one. This has been partially inspired by.

B

You know many risk instruction sets, uh for example, mips, which do have like a um basically a constant zero register um and the the other uh reasoning is, um um I'm not sure, hugo. If you could go back to the the table with the push zero plus one and the the the instruction numbers um yeah.

B

So basically in in the the evm implementations, um there's a piece of code which, um uh for the the implementation of the push instruction and that's the reason we have chosen the the instruction 5f, because you can implement it with the same template. And if you do that, then just using the same mnemonic also makes sense.

A

Sounds good yeah all right. um My next question is related to motivation. So uh my question to you is like it's a general thought that we try to put as much as information which is relevant and needed for developers client developers to consider the proposal. This proposal 3855 has a mention of 2733, which is now withdrawn.

A

Do you think that uh un, under this special case, because now that has been withdrawn, uh we can probably get rid of this information? It may be redundant.

B

No, I think it's a good indication of uh I mean, even though, if like two two seven three three is uh is not currently pursued.

B

um Should there be any other proposal um which would influence the the semantics of existing up codes, then this this problem would come up again um and I think that's the the reason to to have it. There.

A

All right, yeah, I was just curious because, generally this is not the case and we usually find proposal either in stagnant status or somewhere in the draft review. But this we know that is not going to go any further. Nevertheless, uh I think I'm happy with the answer here. Okay, um I think that is good and um I don't know there was no mention of any active implementation in any of the project. Just wondering like.

A

uh Is there any active implementation that you are aware of now that it is in cfi, uh most likely every client will try to implement, but as of now, do you have any information on that.

B

It is implemented and implemented and merged in evm, 1 and ethereum gs, and it is implemented in goetherium, but it hasn't been merged. Yet.

A

All right that sounds good. Okay, that's all we have got a number of questions for this proposal and it's time to wrap up any final thing that you would like to add for this proposal.

B

Not for my side.

A

Well, thank you. Thanks alex hugo and pavel for providing uh the overview of eip3855, which may be expected with the shanghai upgrade. This proposal is up for review, so anyone listening to this recording, if you are having thoughts on this proposal, please join the discussion in the discussion tooling, provided in the description box.

A

Thanks to all our youtube viewers for watching and sharing feedback in comments. Should you have any questions on this or any other talk? Let us know in the comment and share it with us on eathcat herders discard subscribe to the ech youtube channel. If you haven't done it already see you next time with another ethereum improvement proposal till then have a good one. Everyone.