Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Feature/infer5 #153

Merged
merged 79 commits into from
Jan 31, 2023
Merged

Feature/infer5 #153

merged 79 commits into from
Jan 31, 2023

Conversation

leissa
Copy link
Member

@leissa leissa commented Nov 29, 2022

Type Inference 2: Electric Boogaloo

Okay, lots of changes here and there are still plenty of loose ends but at some point I wanted to merge into master before the changes are getting out of hands. Here is a list of the most important ones:

Curried Functions

These are now working as expected:

.lam f (i: .Nat)(j: .Nat) -> .Nat = /*...*/;
.con g (T: *)(x: T) = /*...*/;
.let k: .Cn [T: *][T, T] = /*...*/;

Note, that all curried parameter groups are in direct style even for .con/.cn/.Cn. Here, only the last parameter makes up the continuation.

Implicits and Type Inference

You can now mark a parameter group as implicit with a dot like this:

.ax %mem.load:  Π.Tas::[T: *, .Nat][%mem.M, %mem.Ptr Tas] -> [%mem.M, T], normalize_load;
.ax %core.wrap(add, sub, mul, shl): Π.[s: .Nat][.Nat][«2; .Idx s»] -> .Idx s, normalize_wrap;
.con .extern internal_diff_core_icmp_xYgLE !.(s: .Nat)!(ab::(a b: .Idx s), ret: .Cn[.Idx 2, .Cn[.Idx 2, .Cn«2; .Idx s»]]) =
    ret (%core.icmp.sle ab, .cn ![.Idx 2, pb_ret: .Cn«2; .Idx s»] = pb_ret ‹2; 0:(.Idx s)›);

The argument will be automatically inferred by Thorin:

.let acci = %core.wrap.add 0 (i, acc);

From the C++ API you can now use World::call/World::dcall like this:

w.dcall(dbg, core::wrap::add, mode, Defs{ldef, rdef});
w.call(core::wrap:: add, mode, Defs{ldef, rdef});

I ported core, mem, and math to make use of implicits and removed the now superfluous boilerplate from the dialect headers.

We had some discussions in Discord about the syntax for . and !. For now, I kept it like this because the other idea for using | just looked really odd to me but we can still change this.

In order to make this work, I'm using a class Ref that automatically looks through Infers. We have to use these Ref things even more aggressively, but for now I didn't want to have the diff going completely nuts.

The IR keeps track of implicits with the help of Def::meta. This can now be a nested pair that mirrors the curried type and indicates where to place an Infer.

Right now, the type inference will be triggered as soon as a function with implicit arguments is fully invoked. More complex things, won't work for now, but I will work on that in the future.

UMax/UInc

The type level doesn't have to be a constant anymore and can be adjusted with the umax/uinc operators. This is important for type inference and may be interesting for future universe polymorphism.

Minor Changes

  • Improved Def::dep & friends
  • Disabled Windows Debug workflow as MSVC Linker is going bonkers again :(
  • Removed this clumsy error handler in favor of just emitting an error.

Impala

Impala branch here.

Not needed anymore as the is_uniform checks have been rewritten to
always be fired in an "optimistic" spot.
Also refactored:
* core::conv::s2s -> core::conv::s
* core::conv::u2u -> core::conv::u

Rationale: s2u and u2s doesn't make sense.
@leissa leissa marked this pull request as ready for review January 27, 2023 17:04
@leissa leissa merged commit 961b7f6 into master Jan 31, 2023
@leissa leissa deleted the feature/infer5 branch January 31, 2023 16:41
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

None yet

2 participants