[llvm-bugs] [Bug 42384] New: miscompile with Loop Vectorizer on a simple double-nested loop

via llvm-bugs llvm-bugs at lists.llvm.org
Tue Jun 25 07:50:47 PDT 2019


https://bugs.llvm.org/show_bug.cgi?id=42384

            Bug ID: 42384
           Summary: miscompile with Loop Vectorizer on a simple
                    double-nested loop
           Product: libraries
           Version: trunk
          Hardware: PC
                OS: Linux
            Status: NEW
          Severity: normal
          Priority: P
         Component: Loop Optimizer
          Assignee: unassignedbugs at nondot.org
          Reporter: fedor.v.sergeev at gmail.com
                CC: llvm-bugs at lists.llvm.org

Created attachment 22137
  --> https://bugs.llvm.org/attachment.cgi?id=22137&action=edit
bad-vector.ll

Consider this C testcase:

] cat vector.c
extern int printf(const char*, ...);

long mainTest() {
  int byFld = 0;
  long instanceCount=0;
  int i24=0;
  for (int i25 = 0; ++i25 < 260;) {
    i24 += i25;
    for (int i30 = 1; i30 < 31; i30++) {
      instanceCount += (i30 ^ byFld);
      byFld -= (char)i24;
    }
  }
  return instanceCount;
}

int main() {
  printf("%ld\n", mainTest());
}
] gcc -m64 vector.c; ./a.out
-139279619
]

Attached is IR produced and reduced from an equivalent Java program,
snapshotted right before Loop Vectorizer. It looks basically as what you would
expect to see by translating the above C code into IR, yet I did not bother to
find exact sequence of clang optimizations that could show me the same result.

] grep -n define bad-vector.ll
6:define i64 @mainTest() {
49:define i32 @main() {
] bin/opt bad-vector.ll | bin/lli
-139279619

As you see, the result matches the expected output shown above.
Now adding loop-vectorizer:

] bin/opt -passes=loop-vectorize bad-vector.ll | bin/lli
-139446659

and here is a wrong result.

-- 
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/20190625/2131ffd9/attachment.html>


More information about the llvm-bugs mailing list