Re: [ublas] Status of development /Benchmarks

Subject: Re: [ublas] Status of development /Benchmarks
From: Petros (pmamales_at_[hidden])
Date: 2013-12-08 16:04:29

Hi,
I would like to suggest :
- the ability of the usage of tbb as a threading model. OpenMP is basically a FORTRAN tool. OpenMP does not integrate well with 
other threading models. OpenMP on a library level is going to become a  maintenance nightmare.
- the creation of solvers and more general lapack wrappers as classes, wrapping Lapack calls
(i.e. something like RogueWave libraries provided).
- Also, especially for the LAPACK fortran style matrices, the incorporation of BLAS calls as member functions.
So, instead of axpy type calls, the usage of operator= and operator += overloading with the specific template expressions.
This cannot be generic, i.e. for all kind of matrices, but nevertheless, is still very useful for the ones for which BLAS should be called.
- traits classes that would provide for whether the matrix is dense, banded etc, row/column major, what is the LAPACK leading dimension
etc.
Thak you for your consideration,
Petros P. Mamales

From: oswin krause 
Sent: Sunday, December 08, 2013 4:18 AM
To: ublas_at_[hidden] 
Subject: Re: [ublas] Status of development /Benchmarks

Hi, 

I agree with you and I think a rewrite wouldn't change the "high-level" syntax. I like the way expressions are formed right now. There is no real way to get rid of noalias as there is no real cheap solution for alias-tests. And honestly i like prod() over op* because prod() scrams in your face "this is going to take a while".

  for this outer parallelization to be useful, one needs guarantees that openmp is not used in the inner loop... otherwise problems may arise with nested parallelism.
I don't see the issue here. OpenMP has in most configurations a fixed size thread pool and would never spawn more threads than allowed. Further it is allowed for an OpenMP thread to spawn more threads.

The real issue is, OpenMP does not and will never support C++. There is OpenMP support in gcc but in fact there is no guarantee that this works as the C++11 memory model is not compatible with the OpenMP model. AFAIK there are no plans for a C++11 compatible OpenMP standard. Therefore we can't expect that there ever is OpenMP support for clang or other compilers and I really think that uBLAS would need to use proper threading models, for example using asio thread_pool

On 07.12.2013 23:35, Riccardo Rossi wrote:

  Dear Oswin,

  my first point is really trivial. i do not doubt that the current expression template mechanism is not optimal.

  the only thing i propose is to mantain a small backward compatibility layer so that code that works with the old ublas is not unnecessarily broken.

  for example if you assume that the novel syntax is

  C += A*B

  i believe that it could be useful to define some wraper between the old and new syntax. the easiest way (dirty) would be to define two macros

  "noalias"                   and                  " prod"

  so that 

  noalias(C) += prod(A,B)
    

  expands to the new format... indeed using macros is not a good idea but it gives the point

  my second point is mostly about  matrices of size say 30*30, whose size is NOT known at compile time.

  for such matrix size openmp is not useful, however one may perform in parallel a lot of operations on different  small matrices, parallelizing some sort of outer iteration.

  for this outer parallelization to be useful, one needs guarantees that openmp is not used in the inner loop... otherwise problems may arise with nested parallelism.

  ...i only implied this...

  ciao

  Riccardo

  On Sat, Dec 7, 2013 at 10:10 PM, oswin krause <oswin.krause_at_[hidden]> wrote:

    Hi,

    I like your ideas. But i am not sure what you mean with 1. and 2.
    The problem I see is, that uBLAS right now doesn't offer the desired quality and speed for single core, let alone multi core. Therefore I will focus on the internal infrastructure instead of motivating new algorithms.

    short description:
    1. Rock solid implementation of the BLAS algorithms( "compute kernels")
    2. Rewriting of the expression mechanic.
    2.1 automatic use of the compute kernels for the operations
    2.2 automatic optimisation of expressions

    long description
    1. Rock solid implementation of the BLAS algorithms: matrix x vector and matrix x matrix in all useful combinations (sparse/sparse dense/dense dense/sparse ... triangular/dense...etc there are a lot), but also: matrix assignment. uBLAS will not be compared in #Features but in runtime compared to Armadillo/Eigen/Atlas/GotoBlas. Having a fast implementation of this is also crucial for multi-core and distributed computations. Also all higher level decompositions (Cholesky, LU, SVD, RankDecomp,...) and triangular solvers rely on fast BLAS. 
     
    2. Rewriting of the expression mechanic. The way uBLAS is implemented right now, writing 

    noalias(A)+=prod(B,C)+D
    or
    vec = prod(prod<matrix>(B,C),vec2);

    is in most cases a bad decision. The second expression wouldn't even compile without the template parameter.
    What I would like to have is the following:

    2.1 automatic use of the fast algorithms. That is the above expression should be translated to
    axpy_prod(B,C,A,false);
    noalias(A) +=D;

    right now the matrix_assign algorithms are used which do a terrible job on prod(B,C).

    2.2 automatic optimization of expressions. The second expression above is really inefficient. First the matrix-matrix prod is evaluated and the result is afterwards used for a single matrix-vector prod.

    now compare to this:
    prod(B,prod<vector>(C,d));

    only 2 matrix-vector products are performed which might save 99% of the FLOPS of the first expression. So it would be better if this would be transformed automatically. Normally the user could do that himself but sometimes it is not possible. Consider a Conjugate Gradient Solver:

    x=CG(A,b);

    a usual pattern of A is A=XX^T+c*Identity. CG uses internally only matrix-vector products - and possibly not many of them. Therefore you don't want to compute A because it is a) slow b) might require more memory than is available. Without expression optimizations it would be impossible to write a good CG that could do that.

    Greetings,
    Oswin 

    On 07.12.2013 11:38, David Bellot wrote:

      Hi,

      it has been a long discussion we all had for many months now. Should we rewrite ublas from scratch or simply improve it.

      Joaquim and Oswin wants a brand new ublas

      Nasos was more in favor of improving it.

      I personally find very exciting the idea of writing something new, but ublas is very well known now. On the other hand, Eigen and Armadillo took the crown of the main C++ blas library in users' hearts.

      On my todo list for ublas, there are things that will require ublas to be deeply changed. At this stage, we can almost talk about a new library.

      Christmas is very close now, so maybe it's a good time to start talking about the features we wish for ublas and see if they can be implemented with the current version or if a new uBLAS 2.0 is necessary.
      After all, Boost::signal did the same a few years ago. We can definitively do the transition.

      I begin:

      - unified representation of vectors and matrices to represent the fact that a vector IS a matrix. Matlab does the same

      - automated use of different algorithm to let the compiler "chooses" the best implementation (if possible) and switch on SSE, distributed or whateve we need

      - implementation of solvers and decompositions algorithms

      and this is what Nasos and I think should be integrated too:

      1. Matrix multiplication

      2. Algorithms infrastructure (so that we can have real useful features)

      3. Matrix/vector views for interoperability <- I think this is ultra critical because now ublas is monolithic in the sense that you have to use it everywhere you manipulate data. This would really help into letting people for example have a list of vectors (they are plotting) and ublas working on top of that to do for example transformations

      4. NEW DOCUMENTATION - examples and the rest

      5. Incorporate some critical bindings (i.e. mumps bindings which is currently probably the most efficient smp and distributed open source linalg solver)

      6. matlab binding?

      7. distributed ublas

      Please add and ESPECIALLY, please tell me your view on the current infrastructure of uBLAS. It seems many people are not happy with the current "expression template" grammar.

      I'm open to everything

      Best,
      David

      On Thu, Dec 5, 2013 at 11:14 AM, Joaquim Duran <jduran.gm_at_[hidden]> wrote:

        I think that al stuff pending of merge listed by David, should be merged and migrate to uBlas 2.0 and while uBlas 2.0 is in development/maintenance then design from scratch uBlas 3.0.

       

_______________________________________________
ublas mailing list
ublas_at_[hidden]
http://lists.boost.org/mailman/listinfo.cgi/ublas
Sent to: Oswin.Krause_at_[hidden]

    _______________________________________________
    ublas mailing list
    ublas_at_[hidden]
    http://lists.boost.org/mailman/listinfo.cgi/ublas
    Sent to: rrossi_at_[hidden]

  -- 

  Dr. Riccardo Rossi, Civil Engineer

  Member of Kratos Team 

  International Center for Numerical Methods in Engineering - CIMNE
  Campus Norte, Edificio C1 

  c/ Gran Capitán s/n

  08034 Barcelona, España

  Tel:        [phone]

  Fax:       [phone]

  web:       www.cimne.com 

   

_______________________________________________
ublas mailing list
ublas_at_[hidden]
http://lists.boost.org/mailman/listinfo.cgi/ublas
Sent to: Oswin.Krause_at_[hidden]

--------------------------------------------------------------------------------
_______________________________________________
ublas mailing list
ublas_at_[hidden]
http://lists.boost.org/mailman/listinfo.cgi/ublas
Sent to: pmamales_at_[hidden]