Backports for 1.11.4 #57183
Open
Backports for 1.11.4 #57183
+513
−261
Conversation
This file contains 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
The main motivation of this PR is to fix #55807. dc689fe tries to remove the slow `may_contain_union_decision` check by re-organizing the code path. Now the fast path has been removed and most of its optimization has been integrated into the preserved slow path. Since the slow path stores all inner ∃ decisions on the outer most R stack, there might be overflow risk. aee69a4 should fix that concern. The reported MWE now becomes ```julia 0.000002 seconds 0.000040 seconds (105 allocations: 4.828 KiB, 52.00% compilation time) 0.000023 seconds (105 allocations: 4.828 KiB, 49.36% compilation time) 0.000026 seconds (105 allocations: 4.828 KiB, 50.38% compilation time) 0.000027 seconds (105 allocations: 4.828 KiB, 54.95% compilation time) 0.000019 seconds (106 allocations: 4.922 KiB, 49.73% compilation time) 0.000024 seconds (105 allocations: 4.828 KiB, 52.24% compilation time) ``` Local bench also shows that 72855cd slightly accelerates `OmniPackage.jl`'s loading ```julia julia> @time using OmniPackage # v1.11rc4 20.525278 seconds (25.36 M allocations: 1.606 GiB, 8.48% gc time, 12.89% compilation time: 77% of which was recompilation) # v1.11rc4+aee69a4+72855cd 19.527871 seconds (24.92 M allocations: 1.593 GiB, 8.88% gc time, 15.13% compilation time: 82% of which was recompilation) ``` (cherry picked from commit f3a36d7)
(cherry picked from commit eed3365)
We were deleting code from external abstract interpreters since bdf8219. Fixes JuliaGPU/CUDA.jl#2637. Not sure how to write tests for this :( --------- Co-authored-by: Jeff Bezanson <jeff.bezanson@gmail.com> (cherry picked from commit 832f084)
Fixes #54560 (comment) (cherry picked from commit 79e98e3)
This was noted in the original review of that PR by the PR author, but was for some reason not fixed at that time: #46693 (comment) (cherry picked from commit db874ff)
Noticed in code reading, that 35e4a1f simplified this incorrectly resulting in all arguments being assigned to the function type, and then 7f8635f further broke the return type expected for the failure case. Co-authored-by: Shuhei Kadowaki <40514306+aviatesk@users.noreply.github.com> (cherry picked from commit 65de014)
We somehow just got (un)lucky that `DFS!` at Compiler/src/ssair/domtree.jl:184 just happened to store exactly the same value as this pointer in this particular memory location previously, so that this branch on `undef` hit exactly the right value to fail. What are the odds? Seen on a CI run (with rr) The odds of this happening seem somewhere around 2^60 against, to 1 for each time. So that seems impressive we hit this even this once. But we did, and the proof is here, caught in rr: https://buildkite.com/julialang/julia-master/builds/43366#019425d7-67fd-4f33-a025-6d7cd6181649 ``` From worker 6: julia: /cache/build/tester-amdci5-10/julialang/julia-master/src/gc-stock.h:492: gc_big_object_link: Assertion `node->header != gc_bigval_sentinel_tag' failed. 2025-01-02 07:47:22 UTC From worker 6: 2025-01-02 07:47:22 UTC From worker 6: [3877] signal 6 (-6): Aborted 2025-01-02 07:47:22 UTC From worker 6: in expression starting at none:1 2025-01-02 07:47:22 UTC From worker 6: gsignal at /lib/x86_64-linux-gnu/libc.so.6 (unknown line) 2025-01-02 07:47:22 UTC From worker 6: abort at /lib/x86_64-linux-gnu/libc.so.6 (unknown line) 2025-01-02 07:47:22 UTC From worker 6: unknown function (ip: 0x7fb9a4b5040e) at /lib/x86_64-linux-gnu/libc.so.6 2025-01-02 07:47:22 UTC From worker 6: __assert_fail at /lib/x86_64-linux-gnu/libc.so.6 (unknown line) 2025-01-02 07:47:22 UTC From worker 6: gc_big_object_link at /cache/build/tester-amdci5-10/julialang/julia-master/src/gc-stock.h:492 [inlined] 2025-01-02 07:47:22 UTC From worker 6: gc_setmark_big at /cache/build/tester-amdci5-10/julialang/julia-master/src/gc-stock.c:276 2025-01-02 07:47:22 UTC From worker 6: jl_gc_big_alloc_inner at /cache/build/tester-amdci5-10/julialang/julia-master/src/gc-stock.h:491 ``` (cherry picked from commit 36472a7)
The previous implementation was incorrect, leading to failing to mark variables correctly. The new implementation is more conservative. This simple analysis assumes that inference has normally run or that performance doesn't matter for a particular block of code. Fixes #56996 (cherry picked from commit b76fd9f)
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.
Backported PRs:
complete_methods!when max methods is hit #57138:enterblocks #57186tmerge(Const(s), Const(t))st.(s !== t) && (s == t)#56915@constprop, fix error message #56946Need manual backport:
Contains multiple commits, manual intervention needed:
String(::Memory)copy #54457using Foo: Foo, ...in package code (was: "using considered harmful") #42080structto hard #56755Non-merged PRs with backport label:
Pairspublic #56794@ccallablename before JIT registration #55813displaysizeto theIOContextused by the REPL #55499isfile_casesensitivefixes on Windows #55220@inboundsandBase.@propagate_inbounds#50157