Merge pull request #989 from SciML/gd/sparsead

Add sparse AD comparison

benchmarks/AutomaticDifferentiationSparse/BrusselatorSparseAD.jmd

@@ -0,0 +1,169 @@
+---
+title: Sparse AD benchmarks
+author: Guillaume Dalle
+---
+
+```julia
+using ADTypes
+using LinearAlgebra, SparseArrays
+using BenchmarkTools, DataFrames
+import DifferentiationInterface as DI
+import SparseDiffTools as SDT
+using SparseConnectivityTracer: TracerSparsityDetector
+using SparseMatrixColorings: GreedyColoringAlgorithm
+using Symbolics: SymbolicsSparsityDetector
+using Test
+```
+
+## Definitions
+
+```julia
+const N = 32
+const xyd_brusselator = range(0; stop=1, length=N)
+const p = (3.4, 1.0, 10.0, step(xyd_brusselator))
+
+brusselator_f(x, y, t) = (((x - 0.3)^2 + (y - 0.6)^2) <= 0.1^2) * (t >= 1.1) * 5.0
+limit(a, N) =
+ if a == N + 1
+ 1
+ elseif a == 0
+ N
+ else
+ a
+ end;
+```
+
+```julia
+function brusselator_2d_loop(du, u, p, t)
+ A, B, alpha, dx = p
+ alpha = alpha / dx^2
+ @inbounds for I in CartesianIndices((N, N))
+ i, j = Tuple(I)
+ x, y = xyd_brusselator[I[1]], xyd_brusselator[I[2]]
+ ip1, im1, jp1, jm1 = limit(i + 1, N),
+ limit(i - 1, N), limit(j + 1, N),
+ limit(j - 1, N)
+ du[i, j, 1] =
+ alpha *
+ (u[im1, j, 1] + u[ip1, j, 1] + u[i, jp1, 1] + u[i, jm1, 1] - 4u[i, j, 1]) +
+ B +
+ u[i, j, 1]^2 * u[i, j, 2] - (A + 1) * u[i, j, 1] + brusselator_f(x, y, t)
+ du[i, j, 2] =
+ alpha *
+ (u[im1, j, 2] + u[ip1, j, 2] + u[i, jp1, 2] + u[i, jm1, 2] - 4u[i, j, 2]) +
+ A * u[i, j, 1] - u[i, j, 1]^2 * u[i, j, 2]
+ end
+end;
+```
+
+```julia
+function init_brusselator_2d(xyd)
+ N = length(xyd)
+ u = zeros(N, N, 2)
+ for I in CartesianIndices((N, N))
+ x = xyd[I[1]]
+ y = xyd[I[2]]
+ u[I, 1] = 22 * (y * (1 - y))^(3 / 2)
+ u[I, 2] = 27 * (x * (1 - x))^(3 / 2)
+ end
+ return u
+end;
+```
+
+```julia
+x0 = init_brusselator_2d(xyd_brusselator);
+y0 = similar(x0);
+
+f!(y, x) = brusselator_2d_loop(y, x, p, 0.0);
+```
+
+## Sparsity detection
+
+```julia
+S1 = ADTypes.jacobian_sparsity(f!, y0, x0, TracerSparsityDetector())
+S2 = ADTypes.jacobian_sparsity(f!, y0, x0, SymbolicsSparsityDetector())
+@test S1 == S2
+```
+
+```julia
+td1 = @belapsed ADTypes.jacobian_sparsity($f!, $y0, $x0, TracerSparsityDetector())
+println("Sparsity detection with SparseConnectivityTracer: $td1 s")
+```
+
+```julia
+td2 = @belapsed ADTypes.jacobian_sparsity($f!, $y0, $x0, SymbolicsSparsityDetector())
+println("Sparsity detection with Symbolics: $td2 s")
+```
+
+```julia
+println("Speedup from new sparsity detection method (>1 is better): $(td2 / td1)")
+```
+
+## Coloring
+
+```julia
+S = S1
+c1 = ADTypes.column_coloring(S, GreedyColoringAlgorithm())
+c2 = SDT.matrix_colors(S)
+@test c1 == c2
+```
+
+```julia
+tc1 = @belapsed ADTypes.column_coloring($S, GreedyColoringAlgorithm())
+println("Coloring with SparseMatrixColorings: $tc1 s")
+```
+
+```julia
+tc2 = @belapsed SDT.matrix_colors($S)
+println("Coloring with SDT: $tc2 s")
+```
+
+```julia
+println("Speedup from new coloring method (>1 is better): $(tc2 / tc1)")
+```
+
+## Compressed differentiation
+
+```julia
+backend = AutoSparse(
+ AutoForwardDiff();
+ sparsity_detector=TracerSparsityDetector(),
+ coloring_algorithm=GreedyColoringAlgorithm(),
+);
+```
+
+```julia
+extras = DI.prepare_jacobian(f!, similar(y0), backend, x0);
+J1 = DI.jacobian!(f!, similar(y0), similar(S, eltype(x0)), backend, x0, extras)
+
+cache = SDT.sparse_jacobian_cache(
+ backend, SDT.JacPrototypeSparsityDetection(; jac_prototype=S), f!, similar(y0), x0
+);
+J2 = SDT.sparse_jacobian!(similar(S, eltype(x0)), backend, cache, f!, similar(y0), x0)
+
+@test J1 == J2
+```
+
+```julia
+tj1 = @belapsed DI.jacobian!($f!, _y, _J, $backend, $x0, _extras) evals = 1 samples = 100 setup = (
+ _y = similar(y0);
+ _J = similar(S, eltype(x0));
+ _extras = DI.prepare_jacobian(f!, similar(y0), backend, x0)
+)
+println("Jacobian with DifferentiationInterface: $tj1 s")
+```
+
+```julia
+tj2 = @belapsed SDT.sparse_jacobian!(_J, $backend, _cache, $f!, _y, x0) evals = 1 samples = 100 setup = (
+ _y = similar(y0);
+ _J = similar(S, eltype(x0));
+ _cache = SDT.sparse_jacobian_cache(
+ backend, SDT.JacPrototypeSparsityDetection(; jac_prototype=S), f!, similar(y0), x0
+ )
+)
+println("Jacobian with SparseDiffTools: $tj2 s")
+```
+
+```julia
+println("Speedup from new differentiation method (>1 is better): $(tj2 / tj1)")
+```