+ The source for the main model is in the model_parser directory.
+ New: MAPredict now does the total nodal prediction of LULESH. > 90% accuracy in Intel micro-architecutres.MAPredict, a static analysis driven framework that provides memory access prediction by gathering application and machines properties at compile time. MAPredict invokes an analytical model to predict LLC-DRAM traffic by combining the application model, the machine model, and user-provided hints for capturing dynamic information. MAPredict is tested on different micro-architectures of Intel and provided high accuracy for applications with different access patterns.
2.1 [applications]: Follow the installation procedure to get all the content of this folder.
2.1.1 [applications/memory_research_ornl/all_apps_experiments]: TAU-PAPI script to generate LLC-DRAM traffic
data for applications. Scripts are also available for different micro-architechtures of Intel and
also for OpenMP execution.
2.1.2 [applications/memory_research_ornl/aspen_model_generation]: This folder contains all model generation
source code (annotated source code) which is used by MAPredict to generate application model.
2.2 [aspen]: contains aspen source code. Aspen abstract models enable the creation of performance models.
2.3 [OpenARC-devel]: Contains the source code for OpenARC and COMPASS framework.
2.4 [model_parser]: Contains MAPredict's model parser that traverses the application and machine model
and invokes an appropriate analytical model.
2.5 [models]: This folder contains the application and machine models which are passes to MAPredict for prediction.
2.6 [scripts]: These scripts are used to compile MAPredict and invoke appropriate machine and application models.
3.1 Prerequisite
3.1.1 GCC 5.4 or higher.
3.1.2 Java
3.1.3 Bison
3.1.4 python 3.6 or higher.
3.1.5 cuda-10 or higher
3.2 Getting the code
3.2.1 git clone https://github.com/monil01/mapmc.git
3.2.2 cd mapmc
3.2.3 git checkout master
3.2.4 git submodule init
3.2.5 git submodule update --remote --merge
3.3 installing
3.3.1 cd [MAPredict_root]/aspen
3.3.2 ./configure
3.3.3 make all -j (it should build libaspen.a in the lib folder)
3.3.4 cd .. (come out of aspen directory)
3.3.5 cd OpenARC-devel
3.3.5 export openarc=`pwd`
3.3.6 make (it will create all openarc binaries and drivers)
3.3.7 cd ..
3.3.8 Modify the python path in the Makefile.
3.3.9 make (it should build the MAPredict binary)
3.4 testing
3.4.1 run: ./scripts/stream_100.sh , it compiles mapredict, generates the model and generate a prediction result.
3.4.2 convention of running MAPredict: ./[MAPredict_binary] [application_model] [machine_model]
4.1 Source Code annonation: Source code annotation examplese are in
[applications/memory_research_ornl/aspen_model_generation].
4.2 Memory prediction for all applications: [MAPredict_root]/scripts directory has scripts for all applications
that generate memory prediction.
4.3 run ./scripts/lulesh_full_nodal.sh
+ Source code lulesh with annotation :[master branch of the submodule: applications/memory_research_ornl/aspen_model_generation/lulesh_full_nodal]5.1 ISC21 submission data is available at [applications/memory_research_ornl/data_paper/Data_for_ICS21.xlsx].