[llvm-bugs] [Bug 27372] New: -fno-associative-math is not honored

[via llvm-bugs]

https://llvm.org/bugs/show_bug.cgi?id=27372

            Bug ID: 27372
           Summary: -fno-associative-math is not honored
           Product: new-bugs
           Version: unspecified
          Hardware: PC
                OS: Windows NT
            Status: NEW
          Severity: normal
          Priority: P
         Component: new bugs
          Assignee: unassignedbugs at nondot.org
          Reporter: warren_ristow at playstation.sony.com
                CC: llvm-bugs at lists.llvm.org
    Classification: Unclassified

Re-association enabled via '-ffast-math' is not suppressed with
'-fno-associative-math'.  Consider the following test-case:

        // -O1 -ffast-math -fno-associative-math
        float foo(float a, float x)
        {
          return ((a + x) - x);
        }

Compiling at -O1, results in an Add and a Subtract, as expected.  Enabling
fast-math, allows re-association to happen, and the arithmetic ops are folded
away, also as expected.  But enabling fast-math while explicitly disabling
associative-math doesn't suppress the re-association (the arithmetic ops are
still folded away).  Specifically:

        $ clang++ --version
        clang version 3.9.0 (trunk 266451)
        Target: x86_64-unknown-linux-gnu
        Thread model: posix
        InstalledDir: /home/warren/llvm/bin
        $ clang++ -c -O1 test.cpp
        $ objdump -d test.o

        test.o:     file format elf64-x86-64
          ...
        0000000000000000 <_Z3fooff>:
           0:   f3 0f 58 c1             addss  %xmm1,%xmm0
           4:   f3 0f 5c c1             subss  %xmm1,%xmm0
           8:   c3                      retq
        $ clang++ -c -O1 -ffast-math test.cpp
        $ objdump -d test.o
          ...
        0000000000000000 <_Z3fooff>:
           0:   c3                      retq
        $ clang++ -c -O1 -ffast-math -fno-associative-math test.cpp
        $ objdump -d test.o
          ...
        0000000000000000 <_Z3fooff>:
           0:   c3                      retq
        $

GCC does honor the '-fno-associative-math' switch (tested with GCC 4.9.2):

        $ g++ -c -O1 test.cpp
        $ objdump -d test.o

        test.o:     file format elf64-x86-64
          ...
        0000000000000000 <_Z3fooff>:
           0:   f3 0f 58 c1             addss  %xmm1,%xmm0
           4:   f3 0f 5c c1             subss  %xmm1,%xmm0
           8:   c3                      retq
        $ g++ -c -O1 -ffast-math test.cpp
        $ objdump -d test.o
          ...
        0000000000000000 <_Z3fooff>:
           0:   f3 c3                   repz retq
        $ g++ -c -O1 -ffast-math -fno-associative-math test.cpp
        $ objdump -d test.o
          ...
        0000000000000000 <_Z3fooff>:
           0:   f3 0f 58 c1             addss  %xmm1,%xmm0
           4:   f3 0f 5c c1             subss  %xmm1,%xmm0
           8:   c3                      retq
        $

This appears to be target-independent behavior (noticed on X86, but tried with
a few other architectures, and observed the same issue).

Also, tested with a handful of Clang versions, and it's a longstanding
behavior.

-- 
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/20160415/56e6d624/attachment.html>