[Bug c/84172] New: option "-O3" create slower code

[aotto1...@t-online.de]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84172

            Bug ID: 84172
           Summary: option "-O3" create slower code
           Product: gcc
           Version: 5.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P3
         Component: c
          Assignee: unassigned at gcc dot gnu.org
          Reporter: aotto1...@t-online.de
  Target Milestone: ---

here my test…

#:~/test> make test
gcc-5 -march=native -mtune=native -g -static -O0 -o test.0 main.c 
gcc-5 -march=native -mtune=native -g -static -O1 -o test.1 main.c 
gcc-5 -march=native -mtune=native -g -static -O2 -o test.2 main.c 
gcc-5 -march=native -mtune=native -g -static -O3 -o test.3 main.c 
for t in test.0 test.1 test.2 test.3; do ./$t; done
./test.0             → T1 = 673.300964 ms → HI = 0x1, LO = 0
./test.0             → T2 = 506.130981 ms → HI = 0x1, LO = 0
./test.1             → T1 = 136.671005 ms → HI = 0x1, LO = 0
./test.1             → T2 = 139.194000 ms → HI = 0x1, LO = 0
./test.2             → T1 = 139.225998 ms → HI = 0x1, LO = 0
./test.2             → T2 = 139.294998 ms → HI = 0x1, LO = 0
./test.3             → T1 = 217.908997 ms → HI = 0x1, LO = 0
./test.3             → T2 = 231.663010 ms → HI = 0x1, LO = 0

#:~/test> gcc-5 -v
Using built-in specs.
COLLECT_GCC=/usr/bin/gcc-5
COLLECT_LTO_WRAPPER=/usr/lib64/gcc/x86_64-suse-linux/5/lto-wrapper
Target: x86_64-suse-linux
Configured with: ../configure --prefix=/usr --infodir=/usr/share/info
--mandir=/usr/share/man --libdir=/usr/lib64 --libexecdir=/usr/lib64
--enable-languages=c,c++,fortran,ada,go --enable-checking=release
--with-gxx-include-dir=/usr/include/c++/5 --enable-ssp --disable-libssp
--disable-libvtv --enable-libmpx --disable-plugin
--with-bugurl=http://bugs.opensuse.org/ --with-pkgversion='SUSE Linux'
--disable-libgcj --with-slibdir=/lib64 --with-system-zlib --enable-__cxa_atexit
--enable-libstdcxx-allocator=new --disable-libstdcxx-pch
--with-default-libstdcxx-abi=gcc4-compatible
--enable-version-specific-runtime-libs --enable-linker-build-id
--enable-linux-futex --program-suffix=-5 --without-system-libunwind
--enable-multilib --with-arch-32=x86-64 --with-tune=generic
--build=x86_64-suse-linux --host=x86_64-suse-linux
Thread model: posix
gcc version 5.3.1 20160301 [gcc-5-branch revision 233849] (SUSE Linux) 
dev1usr@linux02:~/test> Selected  "main.c:40 [main]"

#my code

==========================================================================
#include <stdio.h>
#include <stdlib.h>
#include <stdint.h>
#include <time.h>

#define SIZE    100000000

#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
#   define  HI  1
#   define  LO  0
#else
#   define  HI  0
#   define  LO  1
#endif

int main(int argc, char *argv[])
{
    int                 i;
    clock_t             t1, t2;

    // T1
    if (1) {
        uint64_t            a = 0xffffffffffffffff;
        uint64_t            b = 0xfffffffffffffffe;
        unsigned __int128   r;
        t1 = clock();

        for (i=0; i<SIZE; i++) {
            r = ((unsigned __int128) a * (unsigned __int128) b);
            if (i%2==0) {
                a += (uint64_t) (r>>64);
                b -= (uint64_t) (r>>0);
            } else {
                a -= (uint64_t) (r>>64);
                b += (uint64_t) (r>>0);
            }
        }

        t2 = clock();
        float diff = ((float)(t2 - t1) / (float)CLOCKS_PER_SEC ) * 1000;
        printf("%-20s → T1 = %f ms → HI = %#x, LO = %#x\n", argv[0], diff, a,
b);
    }

    // T2
    if (1) {
        typedef union valU {
            unsigned __int128   ui128    ;
            uint64_t            ui64[2]  ;
        } valU_t;

        uint64_t            a = 0xffffffffffffffff;
        uint64_t            b = 0xfffffffffffffffe;
        valU_t              r;

        t1 = clock();

        for (i=0; i<SIZE; i++) {
            r.ui128 = ((unsigned __int128) a * (unsigned __int128) b);
            if (i%2==0) {
                a += r.ui64[HI];
                b -= r.ui64[LO];
            } else {
                a -= r.ui64[HI];
                b += r.ui64[LO];
            }
        }

        t2 = clock();
        float diff = ((float)(t2 - t1) / (float)CLOCKS_PER_SEC ) * 1000;
        printf("%-20s → T2 = %f ms → HI = %#x, LO = %#x\n", argv[0], diff, a,
b);
    }

    exit(0);
=======================================================================