[llvm-bugs] [Bug 49305] New: In O0 optimization, -nan cannot be calculated properly.

via llvm-bugs llvm-bugs at lists.llvm.org
Sun Feb 21 18:33:17 PST 2021


https://bugs.llvm.org/show_bug.cgi?id=49305

            Bug ID: 49305
           Summary: In O0 optimization, -nan cannot be calculated
                    properly.
           Product: libraries
           Version: 8.0
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: normal
          Priority: P
         Component: Backend: AArch64
          Assignee: unassignedbugs at nondot.org
          Reporter: 2077213809 at qq.com
                CC: arnaud.degrandmaison at arm.com,
                    llvm-bugs at lists.llvm.org, smithp352 at googlemail.com,
                    Ties.Stuij at arm.com

Simple test cases, target is aarch64-linux-gnu, The running environment is
aarch64 Linux.

#include <iostream>
#include <stdio.h>
#include <limits>

int main()
{
  double a = std::numeric_limits<double>::quiet_NaN();
  double b = -a;
  std::cout << b << std::endl;
  return 0;
}

clang++ -O0 nan_bug.cpp
./a.out
nan

clang++ -O2 nan_bug.cpp
./a.out
-nan

when we use -O0 optimize, d0 is nan, d1 is 0, and the result is still nan

15│    0x0000000000400910 <+52>:    stur    d0, [x29, #-16]
16│    0x0000000000400914 <+56>:    ldur    d0, [x29, #-16]
17│    0x0000000000400918 <+60>:    orr     x8, xzr, #0x8000000000000000
18│    0x000000000040091c <+64>:    fmov    d1, x8
19├>   0x0000000000400920 <+68>:    fsub    d0, d1, d0


use the (fsub 0, nan) may  can't get the correct result. the x86 is correct, 
it use the  xor to take the negative
11│    0x00000000004008cd <+45>:    movabs $0x8000000000000000,%rcx
12│    0x00000000004008d7 <+55>:    xor    %rcx,%rax

When O2 is used, -nan is directly calculated during optimization. So there's no
error.

-- 
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/20210222/057b4bad/attachment-0001.html>


More information about the llvm-bugs mailing list