I’m analyzing the performance of julia linear solver (A,B) vs the Matlab solver. Julia is fast only on the second call of the solver and only for small (A) matrix. Over size of A 500x500, julia solver is always slower than Matlab. Anyone did a similar performance comparison ? I looked for a chart…

See the very lengthy discussion here: [image] Benchmark MATLAB & Julia for Matrix Operations Performance Hi, My first try with Julia was to compare its speed on Matrix Operations (Linear Algebra oriented) to MATLAB. Benchmark MATLAB & Julia for Matrix Operations As any …

Seems a nice discussion, very useful. Anyway, I’d like to reconstruct the algorithm used by julia’s solver… no one seems to know it and can be interesting have a flow chart of it. (Like these Solve systems of linear equations Ax = B for x - MATLAB mldivide \ - MathWorks Italia ) Anyone know which f…

As an informative example of how to find this: julia> A = rand(4,4) 4×4 Array{Float64,2}: 0.791559 0.706394 0.914927 0.00942085 0.20433 0.414622 0.209776 0.747403 0.79516 0.356356 0.691833 0.433256 0.557565 0.131694 0.352743 0.843174 julia> b = rand(4) 4-element Array{Float64,1}:…

Reading the code (thanks ChrisRackauckas) i’ve seen the most of the time is spent during LU factorization made by LAPACK. To check if Intel MKL is faster i’ve wrote this script… but it doesn’t work… The idea is to call the routine solver from MKL ( Documentation Library ) # Test MKL solve linear …

Don’t quote me on this, but to test it out you may need to build with MKL. Instructions are here: https://github.com/JuliaLang/julia#intel-compilers-and-math-kernel-library-mkl That by default will build master (pre-0.6). So to A/B test you’ll either need to make sure you build the v0.5.0 tag or t…

He just wants to call a function from an external library so not sure why he would need to rebuild anything?

Yes, i want call MKL as external library. The execution has no error… but seems that the result (if the calculation is made) isn’t stored.

I think you need some “&” in the ccall.

Julia’s arrays are column major so you probably want to use 102. s is for single precision i.e. Float32 and your input is Float64 so you should use the d version of the LAPACK routine If you want to measure the factorization you should call dgetrf where f is for factorization instead of the dgetrs …

[image] andreasnoack: If you want to measure the factorization you should call dgetrf where f is for factorization instead of the dgetrs which is a solver routine for which takes the output of dgetrf as input (as well as a right-hand-side.) Yes i want measure the LU factorization time, but fo…

Linear solver \(A, B) performance vs Matlab A\b

General Usage

Tamas_Papp February 13, 2017, 2:36pm 2

See the very lengthy discussion here:

Basically, most points there apply here too.

Topic		Replies	Views
LU factorization performance issue New to Julia linearalgebra	30	889	June 6, 2022
Performance gotcha in linear algebra lu() General Usage performance , linearalgebra	33	3829	February 11, 2020
Performance of Julia binary wrapper behind Matlab mex-file interface performance Performance matlab	38	1592	May 17, 2026
Is the linear system solver \ also multi threaded in Julia as in Matlab? And how to “multithread” it in Julia? Performance linearalgebra , sparse	34	2839	May 1, 2026
Benchmark MATLAB & Julia for Matrix Operations Performance	148	20825	October 15, 2019

Linear solver \(A, B) performance vs Matlab A\b

Related topics