Interprocedural slicing failed

[Justme0]

I use llvm-slicer to slice the following example. The criterion is assert function, and statements about sum should be sliced but failed.

// a.c
#include <assert.h>
#include <stdio.h>

int add(int x, int y) {
    return x + y;
}

int main() {
    int sum = 0;
    int i = 1;
    while (i < 11) {
        sum = add(sum, i);
        i = add(i, 1);
    }
    assert(11 == i);

    return 0;
}

Firstly, compile it to LLVM IR using clang -g -c -emit-llvm a.c;
Secondly, run slicer using ./llvm-slicer -c=__assert_fail a.bc;
Thirdly, get readable IR using llvm-dis a.sliced.
The slice about two funtions is

; Function Attrs: nounwind uwtable
define i32 @add(i32 %x, i32 %y) #0 !dbg !6 {
entry:
  %x.addr = alloca i32, align 4
  %y.addr = alloca i32, align 4
  store i32 %x, i32* %x.addr, align 4
  store i32 %y, i32* %y.addr, align 4
  %0 = load i32, i32* %x.addr, align 4, !dbg !10
  %1 = load i32, i32* %y.addr, align 4, !dbg !11
  %add = add nsw i32 %0, %1, !dbg !12
  ret i32 %add, !dbg !13
}

; Function Attrs: nounwind uwtable
define i32 @main() #0 !dbg !14 {
entry:
  %sum = alloca i32, align 4
  %i = alloca i32, align 4
  store i32 0, i32* %sum, align 4, !dbg !17
  store i32 1, i32* %i, align 4, !dbg !18
  br label %while.cond

while.cond:                                       ; preds = %while.body, %entry
  %0 = load i32, i32* %i, align 4, !dbg !19
  %cmp = icmp slt i32 %0, 11, !dbg !21
  br i1 %cmp, label %while.body, label %while.end, !dbg !22

while.body:                                       ; preds = %while.cond
  %1 = load i32, i32* %sum, align 4, !dbg !23
  %2 = load i32, i32* %i, align 4, !dbg !25
  %call = call i32 @add(i32 %1, i32 %2), !dbg !26
  store i32 %call, i32* %sum, align 4, !dbg !27
  %3 = load i32, i32* %i, align 4, !dbg !28
  %call1 = call i32 @add(i32 %3, i32 1), !dbg !29
  store i32 %call1, i32* %i, align 4, !dbg !30
  br label %while.cond

while.end:                                        ; preds = %while.cond
  %4 = load i32, i32* %i, align 4, !dbg !31
  %cmp2 = icmp eq i32 11, %4, !dbg !31
  br i1 %cmp2, label %safe_return, label %cond.false, !dbg !31

cond.false:                                       ; preds = %while.end
  call void @__assert_fail(i8* getelementptr inbounds ([8 x i8], [8 x i8]* @.str, i32 0, i32 0), i8* getelementptr inbounds ([4 x i8], [4 x i8]* @.str.1, i32 0, i32 0), i32 15, i8* getelementptr inbounds ([11 x i8], [11 x i8]* @__PRETTY_FUNCTION__.main, i32 0, i32 0)) #2, !dbg !32
  unreachable, !dbg !32

safe_return:                                      ; preds = %while.end
  ret i32 0
}

The statements about variable sum should be deleted. It seems that dg use SDG(System Dependence Graph) to do interprocedural slicing.

[mchalupa]

dg does not use exactly SDG -- it does not use the summary edges. That is also the reason why the sum is not deleted, because without summary edges we can not perform the two-pass algorithm. We simply find dependent statements in one backward traverse through the graph. In other words, dg is atm completely context-insensitive (if you manually try to duplicate the add function, then it will work as you expect). However, it is one of planned future extensions.

[mchalupa]

You can find some more info here: #107

[Justme0]

OK. Thanks a lot!

[mchalupa]

As this question is superseded by #107, I'm closing it in the favor of the later one.