Essentially, PCA is a well implemented data-dependence analyzer targeting c program, in which we applied our optimization such as partial call-path analysis and integer encoding in reaching-definition computation to improve efficiency and effectiveness. PCA-Mem is a case study for evaluation, our tool also supports many other data-dependence based applications such as taint analysis, bug detection and so on.

Here we present two ways to fetch and use PCA:

1. Download PCA-VM.zip (the virtual disk image) through the link.
2. Launch the virtual machine in Virtual Box(username/password: pca/pca, root/pca).
3. Open a terminal, and entry the directory /home/pca/PCA.

1. Check prerequisites (UNIX: Ubuntu 16.04 LTS or Ubuntu 18.04 LTS).
2. Download source code through the link.
3. Enter directory PCA/llvm7 and run installLLVM.sh, which will install LLVM7 and configure environment variables automatically.
4. Enter directory PCA and build PCA with script build.sh.

We implemented a case study of PCA: PCA-Mem. This part shows how to run PCA-Mem against target program for memory leak detection.

To enable data-dependence analysis based on LLVM, target program needs to be compiled with clang and gold-plugin (details referred to here.

1. For the simple subject used in the demo, the command line forthis step is:

• clang -flto leak.c -c -o leak.bc

2. For the Slurm system subject used in the demo, the commandline for this step is:

• Specify environment variables for the compiler:
  • export CC="clang -flto"
  • export CXX="clang++ -flto"
  • export RANLIB=/bin/true
• Compile Slurm:
  • ./configure && make

In this step, we present how to run memory leak detection with PCA-Mem and generate the data-dependence graph (DDG) (for visual understanding purposes).

A small test case (leak.bc) is shown as Figure above, there are two partial free defects obviously, which are detected by PCA-Mem (with command: PCA-Mem -file leak.bc) as expected (shown as Figure below).

We present a parameter (--dump-DDG) to generate DDG.dot, which can be open by GVEdit as below (ICFG is painted with black color while DDG with red), in this case we could verify the correctness of DDG manually while debugging.

For large scale program like Slurm (Version 15.08.7), which usually contains multiple modules, PCA-Mem conducts two-step analysis:

1. Pre-process: Compute dependencies between modules (e.g., PCA-Mem -dir Slurm -pre=1).
2. Program analysis: Link all necessary IR of modules for the target executable and Perform data dependence analysis and memory leak detection sequentially (e.g., PCA-Mem -file Slurm/salloc.bc).