Create your Gitee Account
Explore and code with more than 5 million developers,Free private repositories !:)
Sign up
Clone or download
Notice: Creating folder will generate an empty file .keep, because not support in Git

llvm-slicing: Symbolic Program Slicing with LLVM


Program slicing is a technique for simplifying programs by focusing on selected aspects of their behaviour. Current slicing techniques still have much room for improvement, such as handling programs written in multiple languages. Using the modern compilation framework LLVM (Low-Level Virtual Machine), we attempt in this project to meet this improvement by presenting a language-independent context-sensitive slicing approach, called Symbolic Program Slicing, including both backward and forward static slicing. In the symbolic slicing approach, slices are stored symbolically rather than procedure being re-analysed (cf. procedure summaries). For comparison, we systematically adapt SDG-based (system dependence graph) slicing methods with IFDS (Interprocedural Finite Distributive Subset analysis) to statically slice LLVM IR (intermediate representation).


Building from source

llvm-slicing is written in Haskell. It depends on LLVM 3.0-3.4 and llvm-config must be in your PATH. It is built and packaged using Cabal.

  • Install the package cabal-install from your system's package manager (with e.g. apt-get); Verify that cabal is installed and update its dependency list with cabal update.
  • git clone this repository, and cd to the llvm-slicing source directory (src) to build/install: cabal install. This will compile llvm-slicing and install it to your ~/.cabal/bin directory.
  • Add this directory to your PATH;
    Verify that your PATH is set up correctly with which llvm-slicing.

Using pre-built binary

Alternatively, download our pre-built binary for Ubuntu 12.04 LTS (64 bit) with LLVM 3.3:
Then unzip it and put the binary llvm-slicing in a directory that is on your PATH.

Creating docker image

For docker users, please visit the file Dockerfile or pull the image nuptzyz/llvm-slicing.


Currently, llvm-slicing includes four static slicers based on corresponding slicing methods, i.e. Symbolic slicing, Weiser slicing, SDG-based slicing (using intraprocedure slice result to generate summary edges), and IFDS-based slicing (using IFDS method to generate summary edges). To get detail help infomation:

   $ llvm-slicing -h

   llvm-slicing [-c|--criterion VARIABLES] [-d|--direction DIRECTION]
                [-m|--method SLICE_METHOD] [-p|--isParallel ISPARALLEL]
                [-t|--timeout TIMEOUT] [-g|--graph GRAPH_SHOW]
                [-o|--output FILE/DIR] FILE

   Available options:
     -h,--help                  Show this help text
     -c,--criterion VARIABLES   The criterion variables (with the form of Var@Fun,e.g. x@main) for slicing. 
                                If null, just output the slice table for all single variables.
     -d,--direction DIRECTION   The type of output to slice: Fwd, Bwd or Both. Default: Bwd
     -m,--method SLICE_METHOD   The slice algorithm: Symbolic,Weiser,SDG or IFDS. Default: Symbolic
     -p,--isParallel ISPARALLEL Whether or not travelling SCC in callgraph in parallel. Default: False
     -t,--timeout TIMEOUT       The timeout (sec.) for running slicer. Default: 1800
     -g,--graph GRAPH_SHOW      Print related graphs: Sdg,Cg,Cdg,Cfg,Icfg,Pdt or Dt.
     -o,--output FILE/DIR       The destination of a file output
     FILE                       The input file which can be bitcode,llvm assembly, or C/CPP sourcecode

For a multi-programs library, you can first compile these programs into corresponding IRs, then combine them into a single IR with the LLVM tool llvm-link, and finally use llvm-slicing to slice this library.


For a simple C program, sum3.c, its backward static slice table with symbolic slicing method:

  $ llvm-slicing sum3.c
  Backward Static SliceTable:
   Variable      SrcLineNumbers  
   a@add           {"sum3.c: [9,11,12,13,15,27,28,33,40,41]"}
   b@add           {"sum3.c: [9,12,13,15,27,28,33,40,41]"}
   i@main          {"sum3.c: [9,12,13,15,28,33,40,41]"}
   n@main          {"sum3.c: [9]"}
   sum@main        {"sum3.c: [9,11,12,13,15,27,28,33,40,41]"}
   tmp@inc         {"sum3.c: [9,12,13,15,28,33,40,41]"}
   x@A             {"sum3.c: [9,11,12,13,15,27,28,33,40,41]"}
   y@A             {"sum3.c: [9,12,13,15,28,33,40,41]"}
   z@inc           {"sum3.c: [9,12,13,15,28,33,40,41]"}

For simplicity, here backward slicing criteria can be automatically generated to slice on the last instruction (the first instruction for forward slicing criteria) of the main procedure for each global variable, and on the last instruction (the first definition instruction in forward slicing) of each procedure for each local variable allocated (declared) in the procedure.

To get the final IR slice result of the local variable, %z, in the procedure @inc (on its last instruction):

  $ llvm-slicing sum3.c -c z@inc

  Backward Static Slice for ["z@inc"]:
   <SourceLines> ["sum3.c: [9,12,13,15,28,33,40,41]"]: 
    %n = alloca i32 , align 4
    %i = alloca i32 , align 4
    %4 = call i32 @__isoc99_scanf ( i8* i8* getelementptr ( [3 x i8]* @.str1 ,  i32 0, i32 0 ), i32* %n )
    store i32 1 , i32* %i , align 4
    %7 = phi i32 [ [%.pre, %10], [1, %0] ]
    %8 = load i32* %n , align 4
    %9 = icmp sle i32 %7 , %8
    br i1 %9 , label %10 , label %11
    call void @A ( i32* %sum, i32* %i )
    %.pre = load i32* %i , align 4
    i32* %y
    call void @inc ( i32* %y )
    i32* %a
    i32* %b
    %3 = load i32* %a , align 4
    %4 = load i32* %b , align 4
    %5 = add nsw i32 %3 , %4
    store i32 %5 , i32* %a , align 4
    i32* %z
    %tmp = alloca i32 , align 4
    store i32* %z , i32** %1 , align 8
    store i32 1 , i32* %tmp , align 4
    call void @add ( i32* %z, i32* %tmp )

To get its forward static slice table with IFDS-based slicing method:

  $ llvm-slicing sum3.c -d Fwd -m IFDS
  Forward Static SliceTable:
   Variable      SrcLineNumbers  
   a@add           {"sum3.c: [13,15,16,18,19,20,23,27,28,29,31,33,35,37,40,41,42]"}
   b@add           {"sum3.c: [13,15,16,18,19,20,23,27,28,29,31,33,35,37,40,41,42]"}
   i@main          {"sum3.c: [12,13,15,16,18,19,20,23,27,28,29,31,33,35,37,40,41,42]"}
   n@main          {"sum3.c: [9,13,15,16,18,19,20,23,27,28,29,31,33,35,37,40,41,42]"}
   sum@main        {"sum3.c: [11,18,33]"}
   tmp@inc         {"sum3.c: [13,15,16,18,19,20,23,27,28,29,31,33,35,37,40,41,42]"}
   x@A             {"sum3.c: [18,33]"}
   y@A             {"sum3.c: [13,15,16,18,19,20,23,27,28,29,31,33,35,37,40,41,42]"}
   z@inc           {"sum3.c: [13,15,16,18,19,20,23,27,28,29,31,33,35,37,40,41,42]"}

To print its LLVM IR SDG:

  $ llvm-slicing sum3.c -g Sdg

SDG for sum3.c

For more slice results of sum3.c, please visit the folder test/C/sample.

Comments ( 0 )

Sign in for post a comment


C/C++程序切片器(arXiv:1903.05333)( spread retract


No release




load more
can not load any more