[Bug c++/61169] New: [4.6,4.7,4.8,4.9] unnecessarily honors bracket in mathematical statements

[Bug c++/61169] New: [4.6,4.7,4.8,4.9] unnecessarily honors bracket in mathematical statements

[hannesroest at gmx dot ch]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61169

            Bug ID: 61169
           Summary: [4.6,4.7,4.8,4.9] unnecessarily honors bracket in
                    mathematical statements
           Product: gcc
           Version: 4.6.3
            Status: UNCONFIRMED
          Severity: minor
          Priority: P3
         Component: c++
          Assignee: unassigned at gcc dot gnu.org
          Reporter: hannesroest at gmx dot ch

I hope this is the right place to report this. During some code review of a C++
project, we have noticed that GCC seems to honor brackets added in mathematical
statements in the code unnecessarily and produce slightly suboptimal code under
some circumstances. I am not sure whether this is a feature or not but it can
lead to suboptimal code as we discovered in our codebase and I hoped that I
could get some answers here.

Specifically, the issue is how GCC treats an expression like (x*x) with
brackets vs an expression like x*x without the brackets and prevents
optimization in the first case.  Specifically, one can see this in the
following example where gcc >= 4.6 uses 4 instructions for the second function
"test4ops" but only 3 instructions for the first function "test3ops".
Interestingly, gcc < 4.6 used 3 instructions for both functions:

$ cat test.C
int test3ops(int x, int c, int d) {
  return c * x + d * x * x;
}

int test4ops(int x, int c, int d) {
  return c * x + d * (x * x);
}
$ gcc -S -O3 test.C 
$ cat test.s 
    .file    "test.C"
    .text
    .p2align 4,,15
    .globl    _Z8test3opsiii
    .type    _Z8test3opsiii, @function
_Z8test3opsiii:
.LFB0:
    .cfi_startproc
    imull    %edi, %edx
    leal    (%rdx,%rsi), %eax
    imull    %edi, %eax
    ret
    .cfi_endproc
.LFE0:
    .size    _Z8test3opsiii, .-_Z8test3opsiii
    .p2align 4,,15
    .globl    _Z8test4opsiii
    .type    _Z8test4opsiii, @function
_Z8test4opsiii:
.LFB1:
    .cfi_startproc
    imull    %edi, %esi
    imull    %edi, %edi
    imull    %edx, %edi
    leal    (%rsi,%rdi), %eax
    ret
    .cfi_endproc
.LFE1:
    .size    _Z8test4opsiii, .-_Z8test4opsiii
    .ident    "GCC: (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3"
    .section    .note.GNU-stack,"",@progbits    
$ gcc --version
gcc (Ubuntu/Linaro 4.6.3-1ubuntu5) 4.6.3

[Bug tree-optimization/61169] [4.6,4.7,4.8,4.9] unnecessarily honors bracket in mathematical statements

[paolo.carlini at oracle dot com]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61169

Paolo Carlini <paolo.carlini at oracle dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
          Component|c++                         |tree-optimization

--- Comment #1 from Paolo Carlini <paolo.carlini at oracle dot com> ---
Doesn't seem a C++ front-end issue to me.

[Bug tree-optimization/61169] [4.6,4.7,4.8,4.9] unnecessarily honors bracket in mathematical statements

[rguenth at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=61169

Richard Biener <rguenth at gcc dot gnu.org> changed:

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

--- Comment #2 from Richard Biener <rguenth at gcc dot gnu.org> ---
You want -ffast-math, or more specificall -fassociative-math.  Re-association
can change FP values which is not allowed and the first expression is
(d * x) * x.