Prefix sum/scan using global memory in cuda/OpenCL - opencl

I was looking for a global memory implementation of prefix sum/scan algorithm using CUDA or OpenCL.
All the implementation has been done using local memory.
Can anyone help me with the algorithm and how I should proceed?

Related

mpi4py or multiprocessing in Python ?

I am writing a machine learning toolkit to run algorithm with different settings in parallel (each process run the algorithm for one setting). I am thinking about either to use mpi4py or python's build-in multiprocessing ?
There are a few pros and cons I am considering about.
Easy-to-use:
mpi4py: It seems more concepts to learn and a bit more tricks to make it work well
multiprocessing: quite easy and clean API
Speed:
mpi4py: people say it is more low level, so I am expect it can be faster than python multiprocessing ?
multiprocessing: compared with mpi4py, much slower ?
Clean and short code:
mpi4py: seems more code to write
multiprocessing: preferred, easy to use API
The working context is I am aiming at running the code basically in one computer or a GPU server. Not really targeting at running in different machines in the network (which only MPI can do it).
And since the main goal is doing machine learning, so the parallelization is not really required to be very optimal, the key goal I want to have is to balance easy, clean and quick to maintain code base but at the same time like to exploit the benefits of parallelization.
With the background described above, is it recommended that using multiprocessing should just be enough ? Or is there a very strong reason to use mpi4py ?
By using mpi4py you can divide the task into multiple threads, but with a single computer with limited performance or number of cores the usability will be limited. However you might find it handy during training.
mpi4py is constructed on top of the MPI-1/2 specifications and provides an object oriented interface which closely follows MPI-2 C++ bindings.
MPI for Python provides MPI bindings for the Python language, allowing programmers to exploit multiple processor computing systems.
MPI for Python supports convenient, pickle-based communication of generic Python object as well as fast, near C-speed, direct array data communication of buffer-provider objects

PETSc with shared memory

I have an MPI parallel code using PETSc to solve a linear equation system with a matrix-free GMRES method. It works fine, but each process uses about the same amount of memory, independent of the number of processes I use. So when using many processes, memory usage gets excessive. I am wondering if there is a way around this and I think using a shared memory approach might be the way to go.
As I understand the PETSc website shared memory is supported by PETSc (MPI shared memory is used for simplicity), but I can't find any information on how to enable or use it. Is using PETSc with shared memory a possible solution to my problem and if yes, is there any documentation on how to do this? Or is MPI shared memory used in PETSc by default if possible without any need for additional coding?

OpenCL scan code

I'm looking for a fast implementation of scan(prefixsum) in OpenCL. The best thing that I found is in the Nvidia SDK but it's old(2010).
Does anyone know any other implementation of Scan in OpenCL?
There are several open-source implementations of scan operation in OpenCL:
CLOGS, a library for higher-level operations on top of the OpenCL C++ API.
Boost.Compute, a C++ GPU Computing Library for OpenCL.
VexCL, a C++ vector expression template library for OpenCL/CUDA.
Bolt, a C++ template library optimized for GPUs.
The author of CLOGS wrote a paper comparing performance of scan (and sort) operations in these implementations.
if your device supports 2.0 then, use builtin operations for that.
https://stackoverflow.com/a/32394920/4877550
http://developer.amd.com/community/blog/2014/11/17/opencl-2-0-device-enqueue/

Rust on grid computing

I'm looking to create Rust implementations of some small bioinformatics programs for my research. One of my main considerations is performance, and while I know that I could schedule the Rust program to run on a grid with qsub - the cluster I have access to uses Oracle's GridEngine - I'm worried that the fact that I'm not calling MPI directly will cause performance issues with the Rust program.
Will scheduling the program without using an MPI library hinder performance greatly? Should I use an MPI library in Rust, and if so, are there any known MPI libraries for Rust? I've looked for one but I haven't found anything.
I have used several supercomputing facilities (I'm an astrophysicist) and have often faced the same problem: I know C/C++ very well but prefer to work with other languages.
In general, any approach other than MPI will do, but consider that often such supercomputers have heavily optimised MPI libraries, often tailored for the specific hardware integrated in the cluster. It is difficult to tell how much the performance of your Rust programs will be affected if you do not use MPI, but the safest bet is to stay with the MPI implementation provided on the cluster.
There is no performance penalty in using a Rust wrapper around a C library like a MPI library, as the bottleneck is the time needed to transfer data (e.g. via a MPI_Send) between nodes, not the negligible cost of an additional function call. (Moreover, this is not the case for Rust: there is no additional function call, as already stated above.)
However, despite the very good FFI provided by Rust, it is not going to be easy to create MPI bindings. The problem lies in the fact that MPI is not a library, but a specification. Popular MPI libraries are OpenMPI (http://www.open-mpi.org) and MPICH (http://www.mpich.org). Each of them differs slightly in the way they implement the standard, and they usually cover such differences using C preprocessor macros. Very few FFIs are able to deal with complex macros; I don't know how Rust scores here.
As an instance, I am implementing an MPI Program in Free Pascal but I am not able to use the existing MPICH bindings (http://wiki.lazarus.freepascal.org/MPICH), as the cluster I am using provides its own MPI library and I prefer to use this one for the reason stated above. I was unable to reuse MPICH bindings, as they assumed that constants like MPI_BYTE were hardcoded integer constants. But in my case they are pointers to opaque structures that seem to be created when MPI_Init is called.
Julia bindings to MPI (https://github.com/lcw/MPI.jl) solve this problem by running C and Fortran programs during the installation that generate Julia code with the correct values for such constants. See e.g. https://github.com/lcw/MPI.jl/blob/master/deps/make_f_const.f
In my case I preferred to implement a middleware, I.e., a small C library which wraps MPI calls with a more "predictable" interface. (This is more or less what the Python and Ocaml bindings do too, see https://forge.ocamlcore.org/projects/ocamlmpi/ and http://mpi4py.scipy.org.) Things are running smoothly, so far I haven't got any problem.
Will scheduling the program without using an MPI library hinder performance greatly?
There are lots of ways to carry out parallel computing. MPI is one, and as comments to your question indicate you can call MPI from Rust with a bit of gymnastics.
But there are other approaches, like the PGAS family (Chapel, OpenSHMEM, Co-array Fortran), or alternative messaging like what Charm++ uses.
MPI is "simply" providing a (very useful, highly portable, aggressively optimized) messaging abstraction, but as long as you have some way to manage the parallelism, you can run anything on a cluster.

Getting OpenCL program code from GPU

I have program which use OpenCL do do math, how i can get source code of opencl, that execute on my gpu when this program do calculations?
The most straightforward approach is to look for the kernel string in the application. Sometimes you'll be able to just find its source lying in some .cl file, otherwise you can try to scan the application's binaries with something like strings. If the application is not purposefully obfuscating the kernel source, you're likely to find it using one of those methods.
A more bulletproof approach would be to catch the strings provided to the OpenCL API. You can even provide your own OpenCL implementation that just prints out the kernel strings in the relevant cl function. It's actually pretty easy: start with pocl and change the implementation of clCreateProgramWithSource to print out the input strings - this is a trivial code change.
You can then install that modified version as an OpenCL implementation and make sure the application uses it. This might be tricky if the application requires certain OpenCL capabilities, but your implementation can of course lie about those.
Notice that in the future, SPIR can make this sort of thing impossible - you'll be able to get an IR of the kernel, but not its source.
clGetProgramInfo(..., CL_PROGRAM_BINARIES, ...) gets you the compiled binary, but interpreting that is dependent upon the architecture. Various SDK's have different tools that might get you GPU assembly though.

Resources