On x86 architectures, when the target is -march=pentium or higher, the compiler
seems to replace (x || y) by ((x | y) != 0) too soon for optimizations to apply,
when x and y are integers.

Testcase (compiled with "gcc -O3 -march=something"):

typedef int T;
void g(T);
void f(T x, T y) {
  if (x || y) return;
  g(!x);
}

When the target is -march=i386, GCC is doing a great job and it is able to
replace !x by 1 when optimizing:
  if (x || y) return;
  g(1);

However, when the target is -march=pentium, the generated code is equivalent to:
  if ((x | y) != 0) return; /* this is a bit better than previously */
  g(x != 0);                /* this is a lot worse */

If int is replaced by _Bool in the typedef, then the code is correctly
optimized: !x is replaced by 1 whatever the target. So the misoptimization is
triggered only when boolean values are passed through integers (quite a common
situation in C) and when some specific x86 architectures are targeted.

$ gcc-snapshot -v
Using built-in specs.
Target: i686-linux-gnu
Configured with: ../src/configure -v
--enable-languages=c,c++,java,f95,objc,obj-c++,treelang
--prefix=/usr/lib/gcc-snapshot --enable-shared --with-system-zlib --disable-nls
--enable-__cxa_atexit --enable-libstdcxx-allocator=mt --enable-clocale=gnu
--enable-libstdcxx-debug --enable-java-gc=boehm --enable-java-awt=gtk
--enable-gtk-cairo --with-java-home=/usr/lib/gcc-snapshot/jre --enable-mpfr
--disable-werror i686-linux-gnu
Thread model: posix
gcc version 4.1.0 20050904 (experimental)

-- 
           Summary: Missed optimization with integer booleans when compiling
                    for pentium and higher
           Product: gcc
           Version: 4.1.0
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: guillaume dot melquiond at ens-lyon dot fr
                CC: gcc-bugs at gcc dot gnu dot org
GCC target triplet: i686-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=23869

Reply via email to