[Do Not Merge] Preallocate more caches #470
Conversation
src/factorization.jl
Outdated
| const PREALLOCATED_NORMALCHOLESKY = ArrayInterface.cholesky_instance( | ||
| Symmetric(rand(1, 1)), NoPivot()) | ||
|
|
||
| function init_cacheval(alg::NormalCholeskyFactorization, A::Matrix, b, u, Pl, Pr, |
There was a problem hiding this comment.
| function init_cacheval(alg::NormalCholeskyFactorization, A::Matrix, b, u, Pl, Pr, | |
| function init_cacheval(alg::NormalCholeskyFactorization, A::Matrix{Float64}, b, u, Pl, Pr, |
src/default.jl
Outdated
| # DefaultAlgorithmChoice.GenericLUFactorization | ||
| DefaultAlgorithmChoice.RFLUFactorization |
There was a problem hiding this comment.
I'm not sure that's anything more than voodoo, since https://github.com/JuliaLinearAlgebra/RecursiveFactorization.jl/blob/master/src/lu.jl#L62-L69 . I don't think that uses more caches or anything?
There was a problem hiding this comment.
RFLU implementation here bypasses that to directly call https://github.com/JuliaLinearAlgebra/RecursiveFactorization.jl/blob/425e43622e8bb4a9e0a00f843791ab40df2f9466/src/lu.jl#L89
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #470 +/- ##
==========================================
+ Coverage 27.90% 28.11% +0.21%
==========================================
Files 27 27
Lines 2118 2134 +16
==========================================
+ Hits 591 600 +9
- Misses 1527 1534 +7 ☔ View full report in Codecov by Sentry. |
avik-pal
force-pushed
the
ap/overhead
branch
from
c9cb3cd to
6fd247c
Compare
|
These EnumX to Symbol mappings do have some unnecessary overhead, let me track them down |
julia> @btime solve(prob, $NewtonRaphson(; autodiff = AutoForwardDiff(; chunksize = 2)));
2.774 μs (46 allocations: 4.55 KiB)
julia> @btime solve(prob, $NewtonRaphson(; autodiff = AutoForwardDiff(; chunksize = 2), linsolve = \));
3.233 μs (49 allocations: 4.66 KiB)
julia> @btime solve(prob, $NewtonRaphson(; autodiff = AutoForwardDiff(; chunksize = 2), linsolve = GenericLUFactorization()));
2.620 μs (44 allocations: 4.23 KiB)
julia> @btime solve(prob, $NewtonRaphson(; autodiff = AutoForwardDiff(; chunksize = 2), linsolve = RFLUFactorization()));
2.607 μs (40 allocations: 3.91 KiB) |
avik-pal
force-pushed
the
ap/overhead
branch
from
4bb0210 to
085ba9d
Compare
|
I don't seem to have caused the Enzyme failures |
Testing SciML/SciMLBenchmarks.jl#858