Writing Prefix Sum Library for Efficient Data Balancing
Learn how to write a prefix sum library to achieve efficient data balancing in parallel computing. Understand the initialization process, use cases, and migration considerations for seamless integration. Enhance your programming skills with this insightful tutorial.
Download Presentation
Please find below an Image/Link to download the presentation.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.
You are allowed to download the files provided on this website for personal or commercial use, subject to the condition that they are used lawfully. All files are the property of their respective owners.
The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author.
E N D
Presentation Transcript
MODULES AND LIBRARIES MODULES AND LIBRARIES Charm++ Tutorial
Example of library: data-balancing with prefix Remember in our MP1 (from the course cs598lvk) (data balancing using parallel prefix sum), We used parallel prefix to calculate which data goes to which processor. Ideally, the prefix should have been a separate library After all it has many other uses Instead, we just merged the application code and prefix algorithm in one module Now, let us do that properly
How to write prefix sum as a library What is the interface we want: It gets initialized as a bound array to the application s array Meant to be used with 1D chare array as clients (caller, application) Regular function call.. Say: prefixInit(clientArrayProxy) Each time we need to calculate prefix sum, the library should be called via a local call with the local value We assume it can be called multiple times Assuming a single-value prefix sum. We can generalize to vecs The call won t return, but later on, another entry method of the application will called (Why not a function call?) How to make prefix call application s entry method? Remember when prefix library is written, it doesn t know about any application that s using it So, can t use any name from the application Callback! That s flexible. Called can be threaded, or sdag, or
Initialization and usage Initialization: From the main chare of the application, after creating the application s 1D array (that will be the caller), with proxy in A Call function prefixInit(A) This will create a bound chare array in prefixLib, and return its proxy, say lib Call: (this can be made multiple times) libPtr = lib[thisIndex].ckLocal(); libPtr->computePrefix(myVal, callBack); I know ckLocal() won t return NULL: its bound to me
What to do when you migrate? With a bound array, you can be sure that the shadow chare in prefix library will migrate with you. But the pointer will be different If you use it via ckLocal() as in previous slide, there is no problem But, if you have stored the pointer, you have to re-obtain it That is where the migration constructor is useful App::App(CkMigrateMessage *m) { libPtr = lib[thisIndex].ckLocal() ; } or you can use just after AtSync(), but this may do it unnecessarily, because you might not have migrated
Exercise Make the above design work for your MP1 prefix-based data balancing For better understanding, do this in 2 steps First, write a program with 2 modules in the same folder Make sure prefixLib doesn t use any name from dataBalancer Next, do it with both modules in separate folders, AND precompile the library.. What information about prefixLib do you have to provide to the dataBalancer application ? In what form? .h? .ci? prefixLib.decl.h? What information is minimal necessary? What documentation should you provide?
