[CI] Final integration of batch Hessian

exanauts · Jul 6, 2021 · f136813 · f136813
1 parent db1ca61
commit f136813
Show file tree

Hide file tree

Showing 5 changed files with 18 additions and 4 deletions.
diff --git a/benchmark/evaluators.jl b/benchmark/evaluators.jl
@@ -61,7 +61,6 @@ function run_reduced_evaluator(nlp, u; device=CPU())
     print("Hessian \t")
     H = similar(u, n, n)
     @btime ExaPF.hessian!($nlp, $H, $u)
-    @btime ExaPF.batch_hessian!($nlp, $H, $u)
     return
 end
 

diff --git a/deps/deps.jl b/deps/deps.jl
@@ -0,0 +1,5 @@
+# This file pulls the dependencies using git branches for development
+
+using Pkg
+cusolverrf = PackageSpec(url="https://github.com/exanauts/BlockPowerFlow.jl.git", rev="master")
+Pkg.add([cusolverrf])
diff --git a/src/Evaluators/bridge_evaluator.jl b/src/Evaluators/bridge_evaluator.jl
@@ -116,8 +116,8 @@ function ojtprod!(nlp::BridgeDeviceEvaluator, jv, u, σ, v)
 end
 
 function hessprod!(nlp::BridgeDeviceEvaluator, hv, u, v)
-    copyto!(nlp.bridge.v, v)
-    hessprod!(nlp.inner, nlp.bridge.jv, nlp.bridge.u, nlp.bridge.v)
+    copyto!(nlp.bridge.g, v)
+    hessprod!(nlp.inner, nlp.bridge.jv, nlp.bridge.u, nlp.bridge.g)
     copyto!(hv, nlp.bridge.jv)
     return
 end
@@ -128,6 +128,15 @@ function hessian!(nlp::BridgeDeviceEvaluator, H, u)
     return
 end
 
+function hessian_lagrangian_penalty_prod!(nlp::BridgeDeviceEvaluator, hv, u, y, σ, w, v)
+    copyto!(nlp.bridge.g, v)
+    copyto!(nlp.bridge.w, w)
+    copyto!(nlp.bridge.y, y)
+    hessian_lagrangian_penalty_prod!(nlp.inner, nlp.bridge.jv, nlp.bridge.u, nlp.bridge.y, σ, nlp.bridge.w, nlp.bridge.g)
+    copyto!(hv, nlp.bridge.jv)
+    return
+end
+
 function hessian_lagrangian_penalty!(nlp::BridgeDeviceEvaluator, H, u, y, σ, w)
     copyto!(nlp.bridge.w, w)
     copyto!(nlp.bridge.y, y)

diff --git a/test/cusolver.jl b/test/cusolver.jl
@@ -7,6 +7,7 @@ import ExaPF: LinearSolvers
 import BlockPowerFlow: CUSOLVERRF
 
 const LS = LinearSolvers
+const CUDA_ARCH = (CUDADevice(), CuArray, CuSparseMatrixCSR)
 
 # Overload factorization routine to use cusolverRF
 LS.exa_factorize(J::CuSparseMatrixCSR) = CUSOLVERRF.CusolverRfLU(J)

diff --git a/test/runtests.jl b/test/runtests.jl
@@ -18,8 +18,8 @@ const CASES = ["case9.m", "case30.m"]
 
 ARCHS = Any[(CPU(), Array, SparseMatrixCSC)]
 if has_cuda_gpu()
-    push!(ARCHS, (CUDADevice(), CuArray, CuSparseMatrixCSR))
     include("cusolver.jl")
+    push!(ARCHS, CUDA_ARCH)
 end
 
 # Load test modules