[Venom] Handling dynamic jumps with phi function

[2xic]

Hello!

If I then have some code like this ("psuedo" venom IR code)

; Select a target jump block based on the parent block
%block_selector = phi @parent_a, @block_a, @parent_b, @block_b
; Jump to the destination
djmp @block_a,@block_b,%block_selector

This won't compile because

AssertionError: Can't get lattice for non-variable (`block_b`)

Question: how would I implement my example in a way that compiles ?

I know that Vyper does use selector buckets for handling external calls, but in my case the phi function is to handle a shared internal code block with a dynamic destination.

[harkal]

Yeah, labels are not full first class citizens atm :(. Take a look at this PR until we merge it:

#4514

[2xic]

i fixed the issue in a9f184d. your example compiles now, to the following bytecode:

THANK YOU ❤️ Both for the quick replies and finding the bug.

@2xic i'm curious, would you mind sharing the source bytecode?

Sure, it's nothing fancy though 😅

    contract Hello {
        function test() public returns (uint256) {
            return bagel();
        }

        function bagel() public returns (uint256) {
            return 1;
        }
    }

Compiled with Solc 8.26 (no via ir or anything fancy). For context, I just needed a basic code block where the jump destination would be dynamic so I could add logic for dynamic jumps.

So the raw source bytecode is

6080604052348015600e575f80fd5b50600436106030575f3560e01c806366e41cb7146034578063f8a8fd6d14604e575b5f80fd5b603a6068565b604051604591906092565b60405180910390f35b60546070565b604051605f91906092565b60405180910390f35b5f6001905090565b5f60776068565b905090565b5f819050919050565b608c81607c565b82525050565b5f60208201905060a35f8301846085565b9291505056

Now that I have the phi functions working with djmps I'm onto the next challenge with a more complex contract 🫡

Thanks again for the help!

[charles-cooper]

thanks! are you sharing the tool you are building anywhere? would be interesting to see how well it can optimize existing bytecode.

[charles-cooper]

@2xic FYI i'm updating the text format for djmp to better match jnz, with the condition being the first operand: 574eae8

[2xic]

thanks! are you sharing the tool you are building anywhere? would be interesting to see how well it can optimize existing bytecode.

Currently it's not public, but I can make it public if you are curious, but want to preface it by it still being in still a very early stage, but I think the approach can scale.

I can give you a Github link to it over the weekend (if you prefer to be notified in a different way let me know).

[charles-cooper]

yes, definitely curious!