• =?UTF-8?B?Tm9yZGzDtnc=?= (3/3) Dec 02 2015 Have anybody started working on a linear algebra library on top
• Ilya Yaroshenko (24/27) Dec 02 2015 I wrote tiny 'Guide for Slice/BLAS contributors'. It is located

=?UTF-8?B?Tm9yZGzDtnc=?= <per.nordlow gmail.com> writes:

Have anybody started working on a linear algebra library on top 
of ndslice?

If not, do you have any wishlist on implementation details?

Ilya Yaroshenko <ilyayaroshenko gmail.com> writes:

On Wednesday, 2 December 2015 at 17:09:01 UTC, Nordlöw wrote:
 Have anybody started working on a linear algebra library on top 
 of ndslice?

 If not, do you have any wishlist on implementation details?

I wrote tiny 'Guide for Slice/BLAS contributors'. It is located 
in top of ndslice.package. The guide covers base D implementation 
errors ( I hope so ).

Linear algebra can be splitter into BLAS and others.

About BLAS:
I plan to work on it after voting about ndslice. std.blas is very 
big project and multiple authors are good. Architecture should be 
described first.

Slice would be very cool instrument to build generic BLAS 
implementation like boost.nuBLAS but much more flexible and much 
more faster. Probably it would be world's most fastest BLAS 
without assembler code.

Ideas of implementation are described in Goto's paper: 
http://www.cs.utexas.edu/users/pingali/CS378/2008sp/papers/gotoPaper.pdf

Plus std.blas can use approach from OpenBLAS where are 
multiplication kernels optimised for special CPUs. Furthermore we 
can write generic SIMD optimisation logic based on cash level 
sizes (CASH LEVELS 1 2 3 and RAM).

About non-BLAS:
You may want to discus it in 
https://gitter.im/DlangScience/public . Lars wants or wanted to 
rewrite scid package.

-- Ilya