Memory Debugging
A Crash course
To get rid of stack errors, buffers, exceptions or segmentation faults , leaked memories, pointer bounds, etc while using pointers and files, and string operations is not straightforward. These bugs are inevitable while programming and can cause weird behaviour on different machines, etc. Most of the time these errors or bugs can be removed cleanly using simple debugging tools. I have been using "valgrind" which is simple, intuitive, and a powerful tool to debug memory / data / file operations generated...
Saturday, February 23, 2013
10:30 AM
No comments
"A User's Guide to MPI"
Peter S. Pacheco
http://www.academia.edu/2602564/_A_Users_Guide_to_MPI_by_Peter_S._Pacheco
...
Monday, February 18, 2013
11:40 AM
No comments
Basic Metrics of a CUDA application
After developing a CUDA application, the costly routines (in terms of runtime) need to be tuned or optimised for better performance. The primary metrics to identify whether the kernel is memory intensive or computationally intensive are L1 / L2 cache hit rates, the data throughput, and the computational throughput. This article discusses about the data and computational throughput.
Let's understand this using a simple example;
The kernel SUM adds two vectors a and b and stores the result in c.
_global__...
Tuesday, February 12, 2013
11:08 PM
No comments
Compilers, Debuggers & Profilers
Part A : Compilers
1. GNU compilers [ gcc / g++ / gfortran ] - free
2. Intel [ icc / icpc / ifort ]- free for linux
Part B : Debuggers
1. GNU debugger [ gdb ] - free
2. Intel debugger [ idb ] - free for linux
Part C : Profilers
1. ValGrind (Command line) , Valkyrie (GUI)
Purpose : memory debugging, memory leak detection, and profiling
Link : http://valgrind.org
2. nvprof (command line), computeprof / nvvp (GUI)
Purpose : profiling
Languages : CUDA
Some available lists;
A...
8:24 PM
No comments
CUDA Enabled Devices - Metrics Reference
(Using computeprof / nvvp)
This section contains detailed descriptions of the metrics that can be collected by the Visual Profiler. These metrics can be collected only from within the Visual Profiler. The command-line profiler and nvprof can collect low-level events but are not capable of collecting metrics.
Capability 2.x Metrics
Metric NameDescriptionFormula
sm_efficiencyThe ratio of the time at least one warp is active on a multiprocessor to the total time100 * (active_cycles...
11:32 AM
No comments
Fortran binding in C
Most of the libraries or useful routines are in FORTRAN and if you intend to use them in your C code, here are 2 ways of how to use them. Warning working with multi-dimensional arrays should include the row-major behaviour of C and column-major behaviour of FORTRAN. Also, take care of the data types compatibility across the two languages.
Part A - Using .f / .f90 directly
1. Compile the .f / .f90 files to .o object file using;
FC -c -O filename.f
This generates the filename.o object file...
Sunday, February 10, 2013
5:03 PM
No comments
Profiling a CUDA application
Tools required : NVIDIA's Visual profiler ( nvprof / computeprof )
Readme : http://docs.nvidia.com/cuda/profiler-users-guide/index.html
Firstly, identify an algorithms 'heavy' areas i.e. the most time consuming routines or kernels.
Then prepare the code for profiling,
1. Include these headers
cuda_profiler_api.h (or cudaProfiler.h for the driver API)
2. Add functions to start and stop profile data collection.
cudaProfilerStart() is used to start profiling
cudaProfilerStop()...
Friday, February 8, 2013
5:42 PM
No comments
Sparse (& dense) matrix libraries
A general survey
Useful for BLAS, LAPACK type implementations (but not limited to!) and possible iterative methods with the support of compressed sparse matrix formats. FORTRAN flavours can also be found analogously. Few libraries also include parallel features.
Intel MKL (highly recommended ! )
http://software.intel.com/en-us/intel-mkl
C++
http://seldon.sourceforge.net/
C++ / POSIX
http://plasimo.phys.tue.nl/TBCI/
C++ /
http://math.nist.gov/sparselib++/
Linear solvers (not sure if sparse functionality...
5:39 PM
No comments
A comprehensive list of Numerical libraries ......... for daily use
A library used is a hundred bugs eliminated
Multi-language
ALGLIB is an open source numerical analysis library which may be used from C++, C#, FreePascal, Delphi, VBA.
IMSL Numerical Libraries are libraries of numerical analysis functionality implemented in standard programming languages like C, Java, C# .NET, Fortran, and Python.
The NAG Library is a collection of mathematical and statistical routines for multiple programming languages (C, C++,...
Subscribe to:
Posts (Atom)
