very interesting, dan!

I'm working on a julia library, so I dont think the solution of a patched 
Julia would work for me. 

However, do you think compiling SuiteSparse as a separate library and 
calling it from Julia (as if it were a seperate package) could work? Of 
course, it seems redundant, and hopefully this gets integrated into core 
Julia

Gabe

On Monday, July 11, 2016 at 12:34:55 PM UTC-7, Dan wrote:
>
> OK, details.
>
> Since SuiteSparse is part of the Julia dependencies, if you compile Julia 
> from source (recommended) then you already have KLU and SuiteSparse 
> usually. To use KLU we need to make it into a shared library (like cholmod 
> is). I've patched some makefiles to do this (just 3 makefiles).
>
> After installing the klu shared library it can be ccalled. I've put 
> together a test which basically replicates klu_simple.c from the tests in 
> KLU directory of SuiteSparse.
>
> Finally, added a little benchmark to see if there is actually a speedup 
> when pre-analyzing a matrix. There is - on my computer it was a 5x speedup, 
> but this speedup really depends on the use patterns.
>
> The attachments include a patch file detailing the makefile changes and 
> the klu_simple.jl file.
>
> Would love to hear if you put the KLU library to use and what kind of 
> speedup you get. Also, if you found a different solution it would be 
> interesting to hear about it.
>
> On Sunday, July 10, 2016 at 9:38:44 PM UTC-4, Gabriel Goh wrote:
>>
>> I'd love more details Dan!
>>
>> On Saturday, July 9, 2016 at 8:53:16 PM UTC-7, Dan wrote:
>>>
>>> Hi Gabe,
>>>
>>> SuiteSparse which comes together with Julia includes a library KLU 
>>> which does sparse LU factorization. It has the option to return the fill-in 
>>> reducing order it manages to find for a given sparsity pattern. With this 
>>> returned structure it can subsequently factorize many matrices (and solve 
>>> equations). After some tinkering, I have got it to work, so it is possible.
>>>
>>> If this is still an issue and using KLU fits the bill, I'll provide more 
>>> details of my tinkering.
>>>
>>> DG 
>>>
>>> On Friday, July 8, 2016 at 4:40:45 AM UTC-4, Gabriel Goh wrote:
>>>>
>>>> Hey,
>>>>
>>>> I have a sequence of sparse matrix factorizations I need to do, each 
>>>> one a different matrix but with the same sparsity structure. Is there a 
>>>> way 
>>>> I can save the AMD (or any other) ordering that sparsesuite returns, it 
>>>> does not need to be recomputed each time?
>>>>
>>>> Thanks a lot for the help!
>>>>
>>>> Gabe
>>>>
>>>

Reply via email to