[LLVMbugs] [Bug 15707] New: StackColoring can wrongly merge stack slots due to incorrect liveness calculation

[bugzilla-daemon at llvm.org]

http://llvm.org/bugs/show_bug.cgi?id=15707

            Bug ID: 15707
           Summary: StackColoring can wrongly merge stack slots due to
                    incorrect liveness calculation
           Product: tools
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: llc
          Assignee: unassignedbugs at nondot.org
          Reporter: mseaborn at chromium.org
                CC: llvmbugs at cs.uiuc.edu
    Classification: Unclassified

StackColoring::calculateLiveIntervals() does not correctly handle the case
where the following sequence occurs in a basic block:

  llvm.lifetime.start(%buf)
  ...
  llvm.lifetime.end(%buf)
  ...
  llvm.lifetime.start(%buf)

calculateLiveIntervals() ignores the second lifetime.start() and regards %buf
as not being live after the lifetime.end().  This will cause StackColoring to
merge stack slots that shouldn't be merged.


Example:

declare void @llvm.lifetime.start(i64, i8*)
declare void @llvm.lifetime.end(i64, i8*)

declare void @ext(i8*)

define void @foo() {
  %buf1 = alloca i8, i32 10000, align 16
  %buf2 = alloca i8, i32 10000, align 16

  call void @llvm.lifetime.start(i64 -1, i8* %buf1)
  call void @llvm.lifetime.end(i64 -1, i8* %buf1)

  call void @llvm.lifetime.start(i64 -1, i8* %buf1)
  call void @llvm.lifetime.start(i64 -1, i8* %buf2)
  call void @ext(i8* %buf1)
  call void @ext(i8* %buf2)
  ret void
}


Expected output:

$ llc -mtriple=x86_64-none-gnu llvm_lifetime_bug.ll -O1 -o -
...
    subq    $20008, %rsp            # imm = 0x4E28
.Ltmp1:
    .cfi_def_cfa_offset 20016
    leaq    10000(%rsp), %rdi
    callq    ext
    leaq    (%rsp), %rdi
    callq    ext
    addq    $20008, %rsp            # imm = 0x4E28
    ret

Actual output:

$ llc -mtriple=x86_64-none-gnu llvm_lifetime_bug.ll -O1 -o -
...
    subq    $10008, %rsp            # imm = 0x2718
.Ltmp1:
    .cfi_def_cfa_offset 10016
    leaq    (%rsp), %rdi
    callq    ext
    leaq    (%rsp), %rdi
    callq    ext
    addq    $10008, %rsp            # imm = 0x2718
    ret


The following change fixes the bug:

@@ -434,10 +446,8 @@ void StackColoring::calculateLiveIntervals(unsigned
NumSlots) {
     if (Alive.any()) {
       for (int pos = Alive.find_first(); pos != -1;
            pos = Alive.find_next(pos)) {
-        if (!Starts[pos].isValid())
-          Starts[pos] = Indexes->getMBBStartIdx(MBB);
-        if (!Finishes[pos].isValid())
-          Finishes[pos] = Indexes->getMBBEndIdx(MBB);
+        Starts[pos] = Indexes->getMBBStartIdx(MBB);
+        Finishes[pos] = Indexes->getMBBEndIdx(MBB);
       }
     }

-- 
You are receiving this mail because:
You are on the CC list for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.llvm.org/pipermail/llvm-bugs/attachments/20130409/2cf30dd4/attachment.html>