[Bug c/103670] New: Incorrect optimization of loop termination: Early exit with any optimization

[Bug c/103670] New: Incorrect optimization of loop termination: Early exit with any optimization

[robert.muench at saphirion dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103670

            Bug ID: 103670
           Summary: Incorrect optimization of loop termination: Early exit
                    with any optimization
           Product: gcc
           Version: 10.3.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: robert.muench at saphirion dot com
  Target Milestone: ---

Created attachment 51981
  --> https://gcc.gnu.org/bugzilla/attachment.cgi?id=51981&action=edit
Assembly output with & without -Os optimization

Whenever any -Ox option is used, the loop exits after 3 rounds, even the
counter should run for 6 rounds. Without any -O option, the code works as
expected.

[Bug c/103670] Incorrect optimization of loop termination: Early exit with any optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103670

--- Comment #1 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Can you provide the preprocessed source that is used to generate the assembly
file?

[Bug middle-end/103670] Incorrect optimization of loop termination: Early exit with any optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103670

Andrew Pinski <pinskia at gcc dot gnu.org> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
   Last reconfirmed|                            |2021-12-12
             Status|UNCONFIRMED                 |WAITING
           Keywords|                            |wrong-code
     Ever confirmed|0                           |1
          Component|c                           |middle-end

[Bug middle-end/103670] Incorrect optimization of loop termination: Early exit with any optimization

[robert.muench at saphirion dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103670

--- Comment #2 from Robert M. Münch <robert.muench at saphirion dot com> ---
That's the code snippet. The for(...) loop is where things start to become
strange. 

The thing is, if we use the code from a pure C program, it works. We are using
this code from a Windows DLL which we call via an FFI from an interpreter.
However, omitting any optimization makes it work.


if (pAdapter != NULL) {
  int i = 0;
  int offset = 0;

  // loop for as many bytes as given by the address length
  // sprintf returns num char written, take: out buffer, max size of buffer,
format, ...
  // format: %[flags][width][.precision][size]type
  // .2 = two characters
  // type X = Unsigned hexadecimal integer; uses "ABCDEF"
  // %.2X 

  LOG(" AddressLenth: %i", pAdapter->AddressLength); // MAC-48 = 48 bit, 6
bytes, 6 two char blocks
  LOG(" mac: %.12X", pAdapter->Address);
  for (i = 0; i < pAdapter->AddressLength; i++) {
    LOG(" i: %i offset: %i, AddressLenth: %i", i, offset,
pAdapter->AddressLength);

    // omit outputting the - char for the last byte
    if (i == (pAdapter->AddressLength - 1)) {
      LOG("%s", " finalizing MAC address buidling");
      offset += sprintf_s(MacAddress + offset, MAX_ADAPTER_ADDRESS_LENGTH,
"%02X",pAdapter->Address[i]);
    } else {
      offset += sprintf_s(MacAddress + offset, MAX_ADAPTER_ADDRESS_LENGTH,
"%02X-",pAdapter->Address[i]);
    }
  }

  LOG("<- OK: %s AddressLength: %i", MacAddress, i);
  return MacAddress;
}

[Bug middle-end/103670] Incorrect optimization of loop termination: Early exit with any optimization

[robert.muench at saphirion dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103670

--- Comment #3 from Robert M. Münch <robert.muench at saphirion dot com> ---
I updated to GCC 11.2, which has the same problem but after 4 rounds, not 3.

GCC 10.3 Output:

9512] ==> GetAdapterMACAddress
[9512]  parameter check dwRetVal: 0 number: 0
[9512]  AddressLenth: 6
[9512]  i: 0 offset: 0, AddressLenth: 6
[9512]  i: 1 offset: 3, AddressLenth: 6
[9512]  i: 2 offset: 6, AddressLenth: 6
[9512] <- OK: 00-1C-42- AddressLength: 3

GCC 11.2 Output:

[7128] ==> GetAdapterMACAddress
[7128]  parameter check dwRetVal: 0 number: 0
[7128]  AddressLenth: 6
[7128]  i: 0 offset: 0, AddressLenth: 6
[7128]  i: 1 offset: 3, AddressLenth: 6
[7128]  i: 2 offset: 6, AddressLenth: 6
[7128]  i: 3 offset: 9, AddressLenth: 6
[7128] <- OK: 00-1C-42-10- AddressLength: 4

[Bug middle-end/103670] Incorrect optimization of loop termination: Early exit with any optimization

[pinskia at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103670

--- Comment #4 from Andrew Pinski <pinskia at gcc dot gnu.org> ---
Please read https://gcc.gnu.org/bugs/. A code snippet is not what we want. We
need a full compilable (not always runable) testcase for wrong code.

[Bug middle-end/103670] Incorrect optimization of loop termination: Early exit with any optimization

[robert.muench at saphirion dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103670

Robert M. Münch <robert.muench at saphirion dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
         Resolution|---                         |INVALID
             Status|WAITING                     |RESOLVED

--- Comment #5 from Robert M. Münch <robert.muench at saphirion dot com> ---
This was a wrong positive, caused by some weird side effect originating in some
far away place.