Add shared Hessian tracer à la Walther #135
Merged
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
|
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## main #135 +/- ##
==========================================
+ Coverage 91.20% 91.84% +0.63%
==========================================
Files 37 37
Lines 1456 1484 +28
==========================================
+ Hits 1328 1363 +35
+ Misses 128 121 -7 ☔ View full report in Codecov by Sentry. |
Merged
|
Awesome! Did you already compare the performance to existing set types? |
No, nor did I include it in the benchmarks. Depending on what we decide for |
adrhill
reviewed
Jun 24, 2024
For mutable sets absolutely.
I think carrying around smaller sets instead of needlessly large ones could actually increase the performance of the tracing. Needless to say, the sparse AD pass will also be much faster due to increased sparsity. |
We don't have a static handle on which set types are mutable and which are not.
You might be misunderstanding my question. I'm not saying we should get rid of the "non-Walther" mode. I'm saying that in the "Walther" (shared) mode, we still allow the creation of empty Hessians, and I'm not sure it's necessary or even a good idea. |
|
Sorry, my review and my comment on This does make the question on empty tracer construction interesting. Maybe the solution is to add methods myempty(t::HessianTracer{G,H,true}) where {G, H} = t
myempty(::T) where {T<:AbstractTracer} = myempty(T)if we truly just want to have a single tracer instance. |
adrhill
reviewed
Jun 24, 2024
adrhill
reviewed
Jun 24, 2024
Benchmark resultJudge resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsA ratio greater than
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoTargetBaselineTarget resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoBaseline resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoRuntime information
|
|
Don't merge, the benchmarks show an epic slowdown in the shared version compared to the individual one. I'll investigate |
adrhill
reviewed
Jun 24, 2024
Benchmark resultJudge resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsA ratio greater than
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoTargetBaselineTarget resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoBaseline resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoRuntime information
|
|
I took a look at the flame graph for one of the instances where runtime and memory consumption exploded in shared mode. There are no unexpected copies, in fact the total number of allocations has gone down as expected. However, we spend the majority of the time in I suggest we keep this PR on the backburner for now, I'll open another one for a sparse matrix representation and we'll see whether that improves things. I'll also try to think about clever representations for a 2d set that only grows and never shrinks. |
Maybe Julia tries to avoid race conditions where all tracers try to write into the same memory simultaneously. |
|
I'd like to see this merged soon. |
gdalle
commented
Jun 28, 2024
Benchmark resultJudge resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsA ratio greater than
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoTargetBaselineTarget resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoBaseline resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoRuntime information
|
Benchmark resultJudge resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsA ratio greater than
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoTargetBaselineTarget resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoBaseline resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoRuntime information
|
|
The |
gdalle
commented
Jul 3, 2024
Benchmark resultJudge resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsA ratio greater than
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoTargetBaselineTarget resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoBaseline resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoRuntime information
|
I've added the missing test for |
Benchmark resultJudge resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsA ratio greater than
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoTargetBaselineTarget resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoBaseline resultBenchmark Report for /home/runner/work/SparseConnectivityTracer.jl/SparseConnectivityTracer.jlJob Properties
ResultsBelow is a table of this job's results, obtained by running the benchmarks.
Benchmark Group ListHere's a list of all the benchmark groups executed by this job:
Julia versioninfoRuntime information
|
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.
Source
sharedtype parameter toIndexSetHessianPattern, signifying that thehessianfield is shared between all scalar quantities1-to-1and2-to-1Hessian overloads to leverage sharedhessianfields inIndexSetHessianPatternand avoid creating new objectscreate_patternroutine which is needed for the creation of shared patternsifelseand the like)Tests
IndexSetHessianPatternto the test loopsIndexSetHessianPatternis sharedQuestions
Should
myemptybe defined forIndexSetHessianPattern{I,SG,SH,true}?create_tracers, and if we make another it means we screwed up somewhere and it may result in incorrect behavior / bad performance.SparseConnectivityTracer.jl/src/overloads/hessian_tracer.jl
Line 116 in 3a1df20