Pall haraldsson patch 1 #2
Closed
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
…liaLang#50374) * ensure GC_FINAL_STATS is consistent with new page metadata layout
* Set `VERSION` to `1.11.0-DEV` * move NEWS to HISTORY Co-authored-by: KristofferC <kristoffer.carlsson@juliacomputing.com>
* Extend ifelse lifting to regular SROA * Fix oracle violation This is a pre-existing bug, but was exposed by my improvements to SROA.
…Lang#50259) * fix(stdlib/Dates/periods.jl): conversion of empty CompoundPeriod to zero units * add(stdlib/Dates/test/periods.jl): add test for empty CompoundPeriod
Co-authored-by: Gabriel Baraldi <baraldigabriel@gmail.com>
The new commands are \guillemotleft and \guillemotright, respectively. These commands are in line with the corresponding commands defined in the LaTeΧ package csquotes. Co-authored-by: Steven G. Johnson <stevenj@mit.edu>
Our inlining cost model is extremely primitive, though surprisingly functional given its limitations. The basic idea for it was just that we'd give every intrinsic the approximate cost in cycles, such that for sufficiently large functions (>100 cycles), the cost of the extra call would be dwarfed by the cost of the function. However, there's a few problems with this. For one, the real issue is usually not the extra overhead of the call (which is small and well-predicated), but rather the inhibition of optimizations that inlining might have allowed. Additionally, the relevant cost comparison is not generally latency, but rather the size of the resulting binary. Lastly, the latency metric is misleading on modern superscalar architectures, because the core will perform other tasks while the operation is executing. In fact, somewhat counter-intuitively, this means that it is *more* important to inline high-latency instructions to allow the compiler to perform better latency hiding by spreading out the high-latency instructions. We probably need a full-on rethink of the inlining model at some point, but for the time being, this fixes a problem that I ran into in real code by reducing the inlining cost for floating point division to be the same as that of floating point multiplication. The particular case where I saw this was the batched forward AD rule for division, which had 6 calls to div_float. Inlining these provided substantially better performance.
Lower inlining cost of floating point div
…aLang#50421) Co-authored-by: Steven G. Johnson <stevenj@mit.edu>
Co-authored-by: Daniel Karrasch <daniel.karrasch@posteo.de>
* Add GC metric `last_incremental_sweep` * Update gc.c * Update gc.c
Note that this defines the lock order as `out` then `in` for streams which may try to take both locks. This is now a mandatory API convention for all future streams. Co-authored-by: Rafael Fourquet <fourquet.rafael@gmail.com>
The descriptions had `i <= 5` while the code block had `i <=3`.
…ompact!`-ion (JuliaLang#50767) In code like below ```julia Base.@assume_effects :nothrow function erase_before_inlining(x, y) z = sin(y) if x return "julia" end return z end let y::Float64 length(erase_before_inlining(true, y)) end ``` the constant prop' can figure out the constant return type of `erase_before_inlining(true, y)` while it is profitable not to inline expand it since otherwise we left some `!:nothrow` callees there (xref: JuliaLang#47305). In order to workaround this problem, this commit makes `compact!`move inlineable constants into argument positions so that the such "inlineable, but safe as a whole" calls to be erased during compaction. This should give us general compile-time performance improvement too as we no longer need to expand the IR for those calls. Requires: - JuliaLang#50764 - JuliaLang#50765 - JuliaLang#50768
…wing `InexactError`. (JuliaLang#50777)
If something odd happens during GC (the PC goes to sleep) or a very big transient the heuristics might make a bad decision. What this PR implements is if we try to make our target more than double the one we had before we fallback to a more conservative method. This fixes the new issue @vtjnash found in JuliaLang#40644 for me.
fixes JuliaLang#50780 caused by JuliaLang#47013.
``` julia> @CCall jl_dump_host_cpu()::Cvoid CPU: znver2 Features: sse3, pclmul, ssse3, fma, cx16, sse4.1, sse4.2, movbe, popcnt, aes, xsave, avx, f16c, rdrnd, fsgsbase, bmi, avx2, bmi2, rdseed, adx, clflushopt, clwb, sha, rdpid, sahf, lzcnt, sse4a, prfchw, mwaitx, xsaveopt, xsavec, xsaves, clzero, wbnoinvd julia> target = only(Base.current_image_targets()) znver2; flags=0; features_en=(sse3, pclmul, ssse3, fma, cx16, sse4.1, sse4.2, movbe, popcnt, aes, xsave, avx, f16c, fsgsbase, bmi, avx2, bmi2, adx, clflushopt, clwb, sha, rdpid, sahf, lzcnt, sse4a, prfchw, mwaitx, xsavec, xsaves, clzero, wbnoinvd) ``` Co-authored-by: Prem Chintalapudi <prem.chintalapudi@gmail.com> Co-authored-by: Jameson Nash <vtjnash@gmail.com>
…when partitioning (JuliaLang#50791)
Followup to JuliaLang#45964, JuliaLang#46506, and https://discourse.julialang.org/t/class-of-variables/83892. The error ``` julia> println(_) ERROR: syntax: all-underscore identifier used as rvalue ``` is hard to interpret if you are not familiar with the term `rvalue`, which is not used in any other context in Julia, and as discussed previously the use here is not clearly matching the wikipedia page referred to in the documentation either. This PR does away with the term `rvalue` by changing the error to ``` ERROR: syntax: all-underscore identifiers are write-only and their values cannot be used in expressions ``` and updates the documentation accordingly.
We don't really use anything meaningful from libm for this to matter much.
…uliaLang#50844) Detailed discussion and benchmarks by @oscardssmith in JuliaPackaging/Yggdrasil#7189
…#50851) Co-authored-by: Dilum Aluthge <dilum@aluthge.com>
This bumps the build numbers for stdlib and binary dependency JLLs, updates libssh2 to 1.11.0, libgit2 to 1.6.4, and objconv to 2.53. Julia's FreeBSD CI has been running on FreeBSD 13.2 for a while, but until more recently, Yggdrasil was still building FreeBSD binaries using the 12.2 sysroot. The sysroot was updated to 13.2 and I went through and rebuilt the dependencies that Julia uses. The updated build numbers correspond to these rebuilt but otherwise unchanged binaries. The actual version updates are because libssh2 in Yggdrasil was at 1.11.0 so I left it there (its [release notes](https://github.com/libssh2/libssh2/releases/tag/libssh2-1.11.0) suggest it's a safe update), libgit2 had a newer patch version available and needed to be fixed anyway since the Windows build was broken, and objconv needed its Yggdrasil build recipe fixed but Elliot's GitHub mirror of objconv was at 2.53 so I updated to use that.
All of Pkg's dependencies seemingly need to be in, at least :Markdown, even with Pkg excised. 30% smaller sysimage is possible, see at: JuliaLang#50833 (comment)
|
Something appears to be radically wrong with your git history. You appear to have merged a lot of random things in here. |
|
Yes, I'm not sure why, I didn't merge anything here intentionally, nor meant to make a PR at my own repo. I believe I synced (correctly?) my repo here though rather recently. I'm not sure why my PR didn't go to JuliaLang in the first place as intended, nor how to fix my repo here... |
ViralBShah
pushed a commit
that referenced
this pull request
Nov 2, 2023
…#51489) This exposes the GC "stop the world" API to the user, for causing a thread to quickly stop executing Julia code. This adds two APIs (that will need to be exported and documented later): ``` julia> @CCall jl_safepoint_suspend_thread(#=tid=#1::Cint, #=magicnumber=#2::Cint)::Cint # roughly tkill(1, SIGSTOP) julia> @CCall jl_safepoint_resume_thread(#=tid=#1::Cint)::Cint # roughly tkill(1, SIGCONT) ``` You can even suspend yourself, if there is another task to resume you 10 seconds later: ``` julia> ccall(:jl_enter_threaded_region, Cvoid, ()) julia> t = @task let; Libc.systemsleep(10); print("\nhello from $(Threads.threadid())\n"); @CCall jl_safepoint_resume_thread(0::Cint)::Cint; end; ccall(:jl_set_task_tid, Cint, (Any, Cint), t, 1); schedule(t); julia> @time @CCall jl_safepoint_suspend_thread(0::Cint, 2::Cint)::Cint hello from 2 10 seconds (6 allocations: 264 bytes) 1 ``` The meaning of the magic number is actually the kind of stop that you want: ``` // n.b. suspended threads may still run in the GC or GC safe regions // but shouldn't be observable, depending on which enum the user picks (only 1 and 2 are typically recommended here) // waitstate = 0 : do not wait for suspend to finish // waitstate = 1 : wait for gc_state != 0 (JL_GC_STATE_WAITING or JL_GC_STATE_SAFE) // waitstate = 2 : wait for gc_state != 0 (JL_GC_STATE_WAITING or JL_GC_STATE_SAFE) and that GC is not running on that thread // waitstate = 3 : wait for full suspend (gc_state == JL_GC_STATE_WAITING) -- this may never happen if thread is sleeping currently // if another thread comes along and calls jl_safepoint_resume, we also return early // return new suspend count on success, 0 on failure ``` Only magic number 2 is currently meaningful to the user though. The difference between waitstate 1 and 2 is only relevant in C code which is calling this from JL_GC_STATE_SAFE, since otherwise it is a priori known that GC isn't running, else we too would be running the GC. But the distinction of those states might be useful if we have a concurrent collector. Very important warning: if the stopped thread is holding any locks (e.g. for codegen or types) that you then attempt to acquire, your thread will deadlock. This is very likely, unless you are very careful. A future update to this API may try to change the waitstate to give the option to wait for the thread to release internal or known locks.
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.
32% smaller sysimage. It's 25% faster startup (with web browser running). Possibly more slimmed is coming, and more precise timing.