Yul, eWasm, Solidity: Progress and Future Plans by Christian Reitwiessner (Devcon5)

[Applause] so this is more or less a continuation of the talk I gave at DEFCON 4 let's guess community it's not problem if you haven't seen that talk just context not talk I explained how we want to want to to leverage an intermediate language so that and in several involve accessibility it was which back-end you have to comply to IBM to be Muslim in one run five or and on the bottom you can see some example new code is in allocation or it's a standard sylheti alligator memory allocator and i think is obvious what it does okay what did we do so this was basically we have up the top of last year and what happened after that we started the rewrite of the salinity generator this would be the first third of the talk then I will tell you how he wasn't called from solidity and the last part of the perpetrator we are actually far with that we had to stop at some point because we want to really insulate you zero sixes we are the reason the main reason why we didn't do that in the o2 villages because it yeah these checks will have promised Ofsted you know he has those variables and these are rather direct translation from variables to a state machine but in general you have to worry about mystically also there is one of the most complex parts in the celebrity culture it is false because the product foot has so many parameters and it also depends whether one whether you want ivory or a regular function or whether you specify the gas over they specify the value and so on so there is different ways to use the follow up code and all the time you have to keep track of where everything is on the stack in the old code generator and if you consider it just reference stuff by their names it's just yes and another thing is that so when the Opel generator we forgot to the world did not yet implement a certain feature but by accident the stacked layout is correct then this will need to code that will actually generate it and will of course do something about it in the new code generator some expressions are always referenced by their identifier and a full generation of that sub expressions are implemented then the identifier logistics is in the universe an additional level of checks on the intermediate language and this will be detected and how can you actually use it I'm going to show you how to use it on the command line the standard Chasen also works so - my art will generate you vote from facility code and for safety reason we do not support a direct translation from civility through you'll TBM because it's only an experimental stage so you have to yeah copy the ultimate on the first stage and then rerun the sudanese compiler in assembly mode and then transform the video recorder - yeah that's a gonna get how that looks in practice so this is an example years and 20 content already a hard early years in 20 contract we have an event we have the mapping of the balances to supply a mint function that generates tokens which is conscious which is called in the instructor so center gets 20 points and the mid function this required also Everest mint to cannot be the similarities and then towards my analysis is modified and the transfer event now we compile that to unity and so this is not the quote directly generated particle generator but this is the code after optimization I think it's really nice that this is still nice detail and you will see but you see that it still have expressions and yeah it so you would even see the mid function call here and he was recognized the name mint the the issue here is just that the mint function in the init context is only called once and because of that is first in line and so the function eternal but we so it starts with initializing the memory allocator then we have the core value checks people and then the second thing here is required that the and as we meant to it's not zero you actually see the error string here here's 20 in minute to the zero edges so this this block here is the require home and then we update the tour so this this is this is a common I am menu yeah you take the stuff that is take the day that is currently at storage vacation you have 29 useful now it it calls for the user-defined function called check add this is the the overflow check I was talking about earlier and that is one of the two functions that's not visible on the screen here and the rest of the rest of the coast all the all there is to the instructors instructor called missing to school functions and now we update the balances this is the kind of problem there is a function access team mapping and so on it's a little bit longer so the reason is that so this is the the in x-axis helper function for storage mappings and the name is generated from the type and I think we still have to come up with a way to shorten it so it's very cool yeah and another issue here is you see that not even a success the function was just talking about it's called here and it's also called here again with the same parameters usually the optimizer is able to combine them and call it only once the problem here is that it calls the Kaptchuk function which works on memory because of that it has to access the memory and the optimizer cannot yet recognize that writing this stuff so that the stuff that is written in the first call and the stuff that was written said Paul did no conflict because it exactly the same so they write to the same memory location because of that it can't really swap them but it should detect that it's actually the same step this really yeah talking about exercise and the truth you can this is 256-bit EVM code and I think it's also a lot more obviously 20 then it was a big challenge but we can prepare the compiler that goes why you will and does not kill why are you so there the particle politics 1747 bytes that's been unfair because it doesn't alter como flow checks if we add the objects it's fine and if you do that why it's 2062 so it shows that are still some work to do actually I did not you find the time to see where it has these additional bytes so it might be just amenities ok webassembly similar to assault see my expense I already knew SSD - - he wasn't it will so and this will actually generate he wasn't text representation in one run thing is it's not tested at all it's not tested you know because we couldn't get find execution environment we can Callaghan took my contracts and the reason for that is that we transform web-centric text representation tumor center by the representation and you need that for deploying the fraction deploying the contract smooth because it means the return of ivory I'm representation how does the compiler - II wasn't work it takes the regular IBM you know so it starts exactly at the point where the civility to usually be a provider went and so you have to you have to transform 256 bit code to 64-bit code and this is that by splitting all the you'll variables into four modules and at that point we have some kind of a mix of web assembly because we still use the EBM opcodes but we already use 64-bit variables and then we add kind of a library code that implements all EVM opcodes has used different functions using the EVM LED wasn't within functions and at that point we have it wasn't favorable and we run the regular optimizer will also have an EVM you with slight modifications because we have different approaches and yeah it turns out that we are able to reuse almost all of the proponents of the PM's you will optimizer yeah in terrorist academy yeah pretty straightforward transformation everything yeah that's secretive that is specific some example we just use we just store the current block element at position zero in memory this yeah I have to warn you because of those 256 bits 64-bit thing this will be complicated okay what we do is guesstimate returns a 256 bit value so we have to split that into four variables that's what is done here in the beginning so we call guesstimate and selected for variables and then we for M store that may be kind of sweet new hybrid EVM it was new function that takes eight parameters with one for the address and proper value and then we have the implementations of the idiom of codes in some built-ins of gasoline for example just cause you don't get of Testament and an M store our wheat for minion swapping because what assembly is a machine is not very complicated but if you apply the optimizer then it will of course in line the guess that we call here and also remove all the other varieties we don't get anymore also the M store it has all the constants here so it's probably make sense and then and like this only the engines watches okay now the three tarts third part of the talk the original optimizer all all components of the optimizer take your coat transform it and I'll put you in yes sir it's so all the motives are simple that can be combined in an arbitrary rating if all the modules are correct and the Veloster it turned out that we need a little bit more notice that we initially thought we currently have roughly 25 but most of the motives are really tiny and there are only two to three components that we store a data structure so so for example the condition of 35 is that it takes a look at branches so for some you see we start the body of this is almost earnings so after the body you need you know that C is false or so so the only thing that this asses injects assignments of this form after terminating branches and other motifs in the optimizer can pick that up and use their information but the information is visible below this is not stored in some Hey talk about the future memory times we already saw that we have some problems with memory this will hopefully be solved by adding memory chunks as a built-in type into you that way different memory stores two different data structure are fully independent because it uses kind of a different hand into excess memory also we're thinking about hanging a super optimizer so in optimizer that takes a really long time and tries to really crunch the numbers and find a good way to optimize the code and that can have that can maybe have shortcuts if you want to recompile stuff that it's it's faster and when thinking about adding a genetic I really need to find a good order for applying steps it has already said each step individually should be correct the order doesn't matter for correctness but it matters for efficiency and yes fortunately come up with a good order to apply them the truth was that [Applause]