Create new sparse matrix for R factor in sparse least square solve (#…

…30289) instead of using a view to avoid slow fallback in back substitution. Fixes #30288
JuliaLang · Dec 7, 2018 · 87c5f36 · 87c5f36
1 parent c130002
commit 87c5f36
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/stdlib/SuiteSparse/src/spqr.jl b/stdlib/SuiteSparse/src/spqr.jl
@@ -394,7 +394,7 @@ function _ldiv_basic(F::QRSparse, B::StridedVecOrMat)
     X[rnk + 1:end, :] .= 0
 
     # Solve R*X = B
-    LinearAlgebra.ldiv!(UpperTriangular(view(F.R, Base.OneTo(rnk), Base.OneTo(rnk))),
+    LinearAlgebra.ldiv!(UpperTriangular(F.R[Base.OneTo(rnk), Base.OneTo(rnk)]),
                         view(X0, Base.OneTo(rnk), :))
 
     # Apply right permutation and extract solution from X