[Unity][Draft][WIP] Relax language specification #14148
Conversation
|
Thanks for contributing to TVM! Please refer to the contributing guidelines https://tvm.apache.org/docs/contribute/ for useful information and tips. Please request code reviews from Reviewers by @-ing them in a comment.
Generated by tvm-bot |
relax_spec.md
Outdated
|
|
||
| We also define the following special notation for datatypes, to be used in the rest of the specification: | ||
| 1. `Bool()`: This is shorthand for `UInt(bits=1, lanes=1)`, since TIR does not have a separate Boolean type. "True" refers to a value of 1 in this datatype and "false" refers to a value of 0. For convenience, we will refer to Boolean values as a separate datatype in the specification, due to their significance in `If` nodes. | ||
| 2. `Void()`: This is shorthand for `Handle(bits=64, lanes=0)`. TIR uses this datatype to refer to opaque objects; in Relax, it is used to denote an unknown datatype. |
There was a problem hiding this comment.
In TIR void has bits=0 and lanes=0. Why the deviation in Relax?
Also, void in TIR doesn't mean "opaque", it means "nothing". The use of "kHandle" for it doesn't carry any special meaning. If void in Relax is intended to mean something else, then we should pick a different name to avoid confusion.
There was a problem hiding this comment.
That was my understanding from TIR; from what I have gathered in discussions with others, void dtypes in TIR can be used to refer to objects that TIR should not operate on except through builtins/intrinsics. Whatever we use for Void in Relax should match TIR; I will check the resulting ASTs.
There was a problem hiding this comment.
As for whether we should use a different name, that would be worth discussing. My impression is that void was chosen for convenience (reusing the same dtype AST from TIR).
slyubomirsky
changed the title
slyubomirsky
changed the title
slyubomirsky
changed the title
slyubomirsky
force-pushed
the
relax-spec-draft
branch
from
e3c3734 to
567c910
Compare
slyubomirsky
force-pushed
the
relax-spec-draft
branch
from
5707a0d to
e1c24be
Compare
tqchen
force-pushed
the
unity
branch
2 times, most recently
from
a425bc7 to
5c8b7af
Compare
|
Updated to account for #14394, please review. |
slyubomirsky
force-pushed
the
relax-spec-draft
branch
from
b1e186f to
8742b5f
Compare
…identally), update to account for purity tracking
|
Specified |
…tion signature in any order
relax_spec.md
Outdated
|
|
||
| ### Checking Structural Information at the Start and End of a Function | ||
|
|
||
| «Shape variables are bound at the start and end of a function or in `MatchCast` bindings. This checking is done similarly to `MatchCast`, though with a slight difference: per rule #6 in under the [well-formedness criteria](#well-formedness-criteria), we allow shape variables to appear in arguments in any order so long as shape variables appear in a binding position at least once. This requires us to check the shapes of arguments dimension by dimension in a specific order. |
There was a problem hiding this comment.
This approach was discussed here: #15577 (comment)
I am not sure this is the best policy, since it makes the checking of function signatures more complex. However, I am led to understand that this is more convenient in certain situations. We should be sure the implementation follows this policy.
junrushao
force-pushed
the
unity
branch
2 times, most recently
from
c95d45f to
45eeb8c
Compare
|
Now that Unity is in mainline I will work to make this an RFC instead, so see the upcoming RFC for future updates. |
|
Closing the PR so discussion can move to the RFC instead. |
Rendered view.
(Transferred from tlc-pack/relax#273, see the old PR for older discussions and comments.)
This document is a draft language specification for Relax. The purpose of the language specification is to serve as a technical reference describing the language's behavior in sufficient detail as to clarify the intended behavior of the compiler, hence it is by design a very detailed document rather than an accessible tutorial for the language. Its focus is the "what" and "how" of Relax, but not always the "why," though we can add more sections giving design reasons if that is desired.
Note that «double caret marks» (guillemets) are used to denote parts of the specification discussing functionality that the present prototype doesn't yet support. This notation is somewhat cumbersome, but I wasn't sure how else to proceed because Github Markdown does not support changing the text color (which was how my initial document indicated these areas). The caret marks may look strange in the text, but they have the benefit of being easy to find by text search.
Out of scope, for now, in this document is the subject of parsing: We should eventually document how we intend to parse Python into Relax, but the parser itself is being greatly reworked. We can revisit the issue of documenting its behavior once that work has been completed. Additionally, this specification is intended (for now) to focus on the user-visible behavior of Relax rather than specifying lower-level interfaces or the precise mechanisms of Relax's implementation.
Aspects Requiring Review or Still to be Determined
Since this document is a draft, any part of it is up for review and open to revision, but certain parts of the document have proven particularly challenging to describe and could benefit the most from community discussions.
StructInfosystem has been a great challenge to specify and there are many questions as to how it should work. One question is whether "strong shaping" might lead to too many error messages for something that could be checked dynamically. Additionally, many potential shape mismatches could be eliminated using constant propagation or other transformations: If we check shapes without applying transformations, we would force users to add lots of redundant shape checks. On the other hand, if we require these transformations first, that might make the code harder for users to reason about.PackedFuncs can interact with Relax values. However, embedded targets do not support the TVM object system, so describing values in terms of the TVM object system directly may not work for all settings. Additionally, it should be determined how much detail about the representations should be included in the specification.call_tir) should be described in the specification. I am not certain the descriptions presently in the last section are entirely correct, so more review of them should be appreciated. Additionally, are there operators that should be there but are presently missing?There are also some more minor TODOs throughout the document.
The Future of This Document
Eventually, we will want this document to be part of the Relax documentation, in which case it will be placed into a different location in the repo and probably be formatted as an
rstfile. Before that, we will officially RFC the spec into TVM to allow for maximum public discussion as to the design decisions underlying the specification. Hence, I am going to leave this document as a "WIP PR" until it is officially RFC'd.