[LLVMbugs] [Bug 2297] New: Over-aggressive dead store elimination in instcombine

[bugzilla-daemon at cs.uiuc.edu]

http://llvm.org/bugs/show_bug.cgi?id=2297

           Summary: Over-aggressive dead store elimination in instcombine
           Product: new-bugs
           Version: unspecified
          Platform: PC
        OS/Version: Linux
            Status: NEW
          Severity: normal
          Priority: P2
         Component: new bugs
        AssignedTo: unassignedbugs at nondot.org
        ReportedBy: sharparrow1 at yahoo.com
                CC: llvmbugs at cs.uiuc.edu


int a() {
  int len;
  char* x = malloc(10);
  x[1] = 0;
  x[0] = 1;
  len =  strlen(x);
  x[0] = 0;
  b(x);
  return len;
}

A bit of a manufactured testcase, but it doesn't do anything especially
unusual.

Output from llvm-gcc -c -emit-llvm | opt -mem2reg -simplifycfg -instcombine |
llvm-dis:
; ModuleID = '<stdin>'
target datalayout =
"e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32"
target triple = "i386-pc-linux-gnu"

define i32 @a() nounwind  {
entry:
        %tmp1 = call i8* @malloc( i32 10 ) nounwind             ; <i8*>
[#uses=4]
        %tmp3 = getelementptr i8* %tmp1, i32 1          ; <i8*> [#uses=1]
        store i8 0, i8* %tmp3, align 1
        %tmp7 = call i32 @strlen( i8* %tmp1 ) nounwind readonly                
; <i32> [#uses=1]
        store i8 0, i8* %tmp1, align 1
        %tmp11 = call i32 (...)* @b( i8* %tmp1 ) nounwind               ; <i32>
[#uses=0]
        ret i32 %tmp7
}

declare i8* @malloc(i32) nounwind 

declare i32 @strlen(i8*) nounwind readonly 

declare i32 @b(...)

This is clearly not equivalent to the original program: the output depends on
whatever random junk happens to be in x[0] after malloc returns.

Another issue, which I haven't quite figured out: running the same program
through llvm-gcc -c -emit-llvm | opt -mem2reg -instcombine | llvm-dis produces
the following:

; ModuleID = '<stdin>'
target datalayout =
"e-p:32:32:32-i1:8:8-i8:8:8-i16:16:16-i32:32:32-i64:32:64-f32:32:32-f64:32:64-v64:64:64-v128:128:128-a0:0:64-f80:32:32"
target triple = "i386-pc-linux-gnu"

define i32 @a() nounwind  {
entry:
        %tmp1 = call i8* @malloc( i32 10 ) nounwind             ; <i8*>
[#uses=4]
        %tmp3 = getelementptr i8* %tmp1, i32 1          ; <i8*> [#uses=1]
        store i8 0, i8* %tmp3, align 1
        store i8 0, i8* %tmp1, align 1
        %tmp11 = call i32 (...)* @b( i8* %tmp1 ) nounwind               ; <i32>
[#uses=0]
        br label %return

return:         ; preds = %entry
        %tmp7 = call i32 @strlen( i8* %tmp1 ) nounwind readonly                
; <i32> [#uses=1]
        ret i32 %tmp7
}

declare i8* @malloc(i32) nounwind 

declare i32 @strlen(i8*) nounwind readonly 

declare i32 @b(...)

It's clearly wrong to reorder the strlen call in that way.


-- 
Configure bugmail: http://llvm.org/bugs/userprefs.cgi?tab=email
------- You are receiving this mail because: -------
You are on the CC list for the bug.