WIP improve frule inference#202
Open
aviatesk wants to merge 4 commits into
Open
Conversation
aviatesk
force-pushed
the
kf/genericinferencewip
branch
from
c39a1ce to
607ba18
Compare
Keno
reviewed
Jul 31, 2023
aviatesk
added a commit
to JuliaLang/julia
that referenced
this pull request
Jul 31, 2023
Currently, the code that updates `bestguess` using `ReturnNode` information includes hardcodes that relate to `Conditional` and `LimitedAccuracy`. These behaviors are actually lattice-dependent and therefore should be overloadable by `AbstractInterpreter`. Additionally, particularly in Diffractor, a clever strategy is required to update return types in a way that it takes into account information from both the original method and its rule method (xref: JuliaDiff/Diffractor.jl#202). This also requires such an overload to exist. In response to these needs, this commit introduces an implementation of a hook named `update_bestguess!`.
Codecov Report❌ Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #202 +/- ##
==========================================
- Coverage 55.01% 54.96% -0.06%
==========================================
Files 28 28
Lines 2790 2800 +10
==========================================
+ Hits 1535 1539 +4
- Misses 1255 1261 +6 ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
aviatesk
added a commit
to JuliaLang/julia
that referenced
this pull request
Aug 1, 2023
Currently, the code that updates `bestguess` using `ReturnNode` information includes hardcodes that relate to `Conditional` and `LimitedAccuracy`. These behaviors are actually lattice-dependent and therefore should be overloadable by `AbstractInterpreter`. Additionally, particularly in Diffractor, a clever strategy is required to update return types in a way that it takes into account information from both the original method and its rule method (xref: JuliaDiff/Diffractor.jl#202). This also requires such an overload to exist. In response to these needs, this commit introduces an implementation of a hook named `update_bestguess!`.
aviatesk
added a commit
to JuliaLang/julia
that referenced
this pull request
Aug 1, 2023
…50744) Currently, the code that updates `bestguess` using `ReturnNode` information includes hardcodes that relate to `Conditional` and `LimitedAccuracy`. These behaviors are actually lattice-dependent and therefore should be overloadable by `AbstractInterpreter`. Additionally, particularly in Diffractor, a clever strategy is required to update return types in a way that it takes into account information from both the original method and its rule method (xref: JuliaDiff/Diffractor.jl#202). This also requires such an overload to exist. In response to these needs, this commit introduces an implementation of a hook named `update_bestguess!`.
There was a problem hiding this comment.
Pull request overview
Updates ADInterpreter to support an alternate “generic inference” mode intended to work with Julia’s bestguess-overload branch by allowing a converged-inference pathway (current_level === missing) and hooking into Core.Compiler’s bestguess initialization/update points.
Changes:
- Extend
ADInterpreterwith agenericcache and widencurrent_leveltoUnion{Int,Missing}. - Add conditional overloads for
CC.InferenceStateandCC.update_bestguess!(guarded onCC.update_bestguess!existing) to override bestguess behavior whencurrent_level === missing. - Remove an older
abstract_call_gf_by_typecompatibility overload.
💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.
Comment on lines
61
to
63
| native_interpreter::NativeInterpreter | ||
| current_level::Int | ||
| current_level::Union{Int, Missing} | ||
| remarks::OffsetVector{RemarksCache} |
Comment on lines
73
to
75
| #=native_interpreter::NativeInterpreter=#NativeInterpreter(), | ||
| #=current_level::Int=#0, | ||
| #=remarks::OffsetVector{RemarksCache}=#OffsetVector([RemarksCache()], 0:0)) |
Comment on lines
+58
to
+60
| # Cache results for forward inference over a converged inference (current_level == missing) | ||
| generic::OptCache | ||
|
|
Comment on lines
+101
to
+120
| function CC.InferenceState(result::InferenceResult, cache::Symbol, interp::ADInterpreter) | ||
| sv = @invoke CC.InferenceState(result::InferenceResult, cache::Symbol, interp::AbstractInterpreter) | ||
| sv === nothing && return sv | ||
| if interp.current_level === missing | ||
| # override initial bestguess | ||
| arginfo = ArgInfo(nothing, result.argtypes) | ||
| si = StmtInfo(false) | ||
| sv.bestguess = CC.abstract_call(interp.native_interpreter, arginfo, si, sv).rt | ||
| end | ||
| return sv | ||
| end | ||
|
|
||
| function CC.update_bestguess!(interp::ADInterpreter, frame::InferenceState, | ||
| currstate::CC.VarTable, @nospecialize(rt)) | ||
| if interp.current_level === missing | ||
| rt = CC.getfield_tfunc(rt, Const(1)) | ||
| end | ||
| return @invoke CC.update_bestguess!(interp::AbstractInterpreter, frame::InferenceState, | ||
| currstate::CC.VarTable, rt::Any) | ||
| end |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Supposed to work with https://github.com/JuliaLang/julia/tree/avi/bestguess-overload.