Guess it's transposed?

Author: ChrisRackauckas

benchmarks/cudalu.jl

@@ -21,7 +21,7 @@ end
 algs = [MKLLUFactorization(), CUDAOffloadFactorization()]
 res = [Float32[] for i in 1:length(algs)]
 
-ns = 200:400:20000
+ns = 200:400:10000
 for i in 1:length(ns)
     n = ns[i]
     @info "$n × $n"

ext/LinearSolveMKLExt.jl

@@ -110,11 +110,11 @@ function SciMLBase.solve!(cache::LinearCache, alg::MKLLUFactorization;
     m, n = size(A, 1), size(A, 2)
     if m > n
         Bc = copy(cache.b)
-        getrs!('N', A.factors, A.ipiv, Bc; info)
+        getrs!('T', A.factors, A.ipiv, Bc; info)
         return copyto!(cache.u, 1, Bc, 1, n)
     else
         copyto!(cache.u, cache.b)
-        getrs!('N', A.factors, A.ipiv, cache.u; info)
+        getrs!('T', A.factors, A.ipiv, cache.u; info)
     end
 
     SciMLBase.build_linear_solution(alg, cache.u, nothing, cache)